CN118297239B - A chiller load forecasting method driven by data and model fusion - Google Patents

A chiller load forecasting method driven by data and model fusion Download PDF

Info

Publication number
CN118297239B
CN118297239B CN202410569004.5A CN202410569004A CN118297239B CN 118297239 B CN118297239 B CN 118297239B CN 202410569004 A CN202410569004 A CN 202410569004A CN 118297239 B CN118297239 B CN 118297239B
Authority
CN
China
Prior art keywords
data
model
past
prediction
follows
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202410569004.5A
Other languages
Chinese (zh)
Other versions
CN118297239A (en
Inventor
商亮亮
杨柳
张宇超
赵凡一
严浩
陆天奇
陈万
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Nantong University
Original Assignee
Nantong University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Nantong University filed Critical Nantong University
Priority to CN202410569004.5A priority Critical patent/CN118297239B/en
Publication of CN118297239A publication Critical patent/CN118297239A/en
Application granted granted Critical
Publication of CN118297239B publication Critical patent/CN118297239B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0499—Feedforward networks
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Energy or water supply
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
    • Y04S10/00—Systems supporting electrical power generation, transmission or distribution
    • Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Business, Economics & Management (AREA)
  • Mathematical Physics (AREA)
  • Data Mining & Analysis (AREA)
  • Health & Medical Sciences (AREA)
  • Economics (AREA)
  • Computing Systems (AREA)
  • General Health & Medical Sciences (AREA)
  • Human Resources & Organizations (AREA)
  • Software Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Strategic Management (AREA)
  • Computational Linguistics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • General Business, Economics & Management (AREA)
  • Evolutionary Computation (AREA)
  • Marketing (AREA)
  • Molecular Biology (AREA)
  • Pure & Applied Mathematics (AREA)
  • Mathematical Optimization (AREA)
  • Tourism & Hospitality (AREA)
  • Computational Mathematics (AREA)
  • Mathematical Analysis (AREA)
  • Algebra (AREA)
  • Primary Health Care (AREA)
  • Water Supply & Treatment (AREA)
  • Public Health (AREA)
  • Databases & Information Systems (AREA)
  • Development Economics (AREA)
  • Game Theory and Decision Science (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

本发明提供了一种数据与模型融合驱动的冷水机组负荷预测方法，属于冷水机组负荷预测技术领域。解决了冷水机组传统负荷预测方法精度不高的技术问题。其技术方案为：该方法分为三个部分：分别完成CVs‑ELM模型部分和CVA状态空间模型部分后，根据各自预测结果的平均绝对百分比误差策略，输出融合驱动负荷预测结果，与真实数据对比，得出在可接受范围内的预测准确度。本发明的有益效果为：将规范变量分析和极限学习机融合，完成负荷预测，实现了多变量多模型的融合驱动预测，提高负荷预测的准确度，对负荷预测方法的一次有效的挖掘和补充。

The present invention provides a chiller load forecasting method driven by data and model fusion, which belongs to the technical field of chiller load forecasting. The technical problem of low accuracy of traditional chiller load forecasting methods is solved. The technical scheme is as follows: the method is divided into three parts: after completing the CVs‑ELM model part and the CVA state space model part respectively, the fusion-driven load forecasting results are output according to the mean absolute percentage error strategy of each prediction result, and compared with the real data, the prediction accuracy within an acceptable range is obtained. The beneficial effects of the present invention are: the canonical variable analysis and the extreme learning machine are integrated to complete the load forecasting, and the fusion-driven forecasting of multiple variables and multiple models is realized, the accuracy of load forecasting is improved, and an effective mining and supplement to the load forecasting method is achieved.

Description

Data and model fusion driven chiller unit load prediction method
Technical Field
The invention relates to the technical field of chiller load prediction, in particular to a data and model fusion driving chiller load prediction method.
Background
With the continuous development of technology, the energy consumption of large public buildings is increased year by year, and a central air conditioner is necessary equipment of the large public buildings and is also main energy consumption equipment, wherein a water chilling unit is a main component part of the central air conditioner and has large energy consumption, so that the load prediction of the water chilling unit is particularly important for energy resource management, especially facing the complex running condition and relating to the load prediction of multiple variables. The accurate prediction has great effect on planning demand scheduling, and can save a lot of resources. Therefore, research on a load prediction method driven by fusion of data and a model is helpful to improve prediction accuracy.
Canonical variable analysis (Canonical VARIATE ANALYSIS, CVA) is a data-driven-based multivariate analysis method aimed at processing the autocorrelation in multivariate data and the correlation between variables, and by analyzing the correlation between variables, the data dimension can be effectively reduced, which is suitable for the dimension-reducing processing of data before the load prediction of an extreme learning machine. The subspace identification method (Subspace Identification Methods, SIMs) can directly determine the model order and state sequence of the dynamic system by using the input and output observation values, thereby obtaining a state space matrix, building a model, and mainly identifying a high-dimensional process model with a large number of inputs, outputs and states. And the CVA in the subspace identification method has superior performance and accuracy. Recent studies have shown that Juricek and Larimore et al demonstrate that CVAs perform better than the numerical algorithm (N4 SID) model of subspace state space system recognition, over recursive models based on least squares and constraint classification regression, and that they also compare results based on CVAs, N4 SIDs and ARX (Auto-REGRESSIVE WITH Exogenous) models in terms of chemical process dynamic model recognition, indicating that CVAs can provide more accurate recognition results and can result in some improvement in prediction accuracy. Therefore, modeling can be performed based on CVA in the subspace identification method, and a state space model is established, so that load prediction can be performed.
As a single hidden layer feedforward neural network model, an Extreme learning machine (Extreme LEARNING MACHINE, ELM) is used, and compared with a traditional feedforward neural network, the input weight matrix and the bias matrix of a hidden layer of the feedforward neural network are randomly generated, and the output weight is adjusted by a least square method. The problem is solved with less execution time, the learning speed is high, the generalization performance is good, the structure is simple, the realization is easy, and the global optimum can be achieved. 2022, wang Qiujiang et al have invented a method, system and product patent for predicting the cooling load of building air conditioner, and the patent number (CN 202211574119.0) proposes a method for determining the structural parameters of an ELM network model by using the main influencing factors of the cooling load of building air conditioner, and then outputting a prediction result after optimizing by using a longhorn beetle swarm algorithm, wherein the main influencing factors are determined by principal component analysis. The method has the advantages that the integrity of input information is reserved, the operation rate is improved, the adaptability to environmental changes is strong, the prediction effect is stable, the generalization capability is high, the convergence speed is low, the iteration times are high, the calculation load is high, the prediction results are inconsistent each time, and the method is not suitable for online prediction.
Disclosure of Invention
The invention provides a method for predicting the load of a water chilling unit by using open-source water chilling unit data, which aims at the technical problem of low precision of the load prediction of the water chilling unit and is driven by fusing data and a model, and comprises the following steps: after the CVs-ELM model part and the CVA state space model part are respectively completed, a fusion driving load prediction result is output according to an average absolute percentage error strategy of each prediction result, and then the fusion driving load prediction result is compared with real data, so that the prediction accuracy in an acceptable range is obtained.
The invention provides a method for predicting the load of a water chilling unit driven by fusion of data and a model, which comprises the steps of dividing the data into a CVs-ELM model part, dividing the data into past data and future data, constructing a past and future combined matrix, calculating a cross covariance matrix, carrying out singular value decomposition on a Hankel matrix to obtain a conversion matrix, extracting multivariable data into a small number of standard variables CVs, reserving the integrity of the data, training an extreme learning machine ELM, constructing a CVs-ELM model, carrying out singular value decomposition on the Hankel matrix, further utilizing standard variable analysis CVA based on a subspace identification method to divide the standard variable space into a main space and a residual space, defining a linear combination of past observation vectors according to the residual space, obtaining a model coefficient matrix through least square recursion, obtaining a CVA state space model, and carrying out fusion driving prediction part, feeding input data into the CVs-ELM model and the CVA state space model, and carrying out a selection strategy of an average percentage error on respective prediction result to complete the fusion driving prediction. According to the invention, the standard variable analysis and the extreme learning machine are fused to complete load prediction, so that the fusion driving prediction of multiple variables and multiple models is realized, and the accuracy of load prediction is improved, which is an effective excavation and supplement of a load prediction method. In addition, the invention has wide application prospect in the aspect of energy resource management, and is expected to provide powerful support for sustainable development of the energy industry.
In order to achieve the aim of the invention, the technical scheme adopted by the invention is that the method for predicting the data and model fusion driving load comprises the following steps:
Step S1, dividing the acquired data in a period of time into past data y p(r) and future data y f(r), determining a past observation length p and a future observation length f, and setting initial values of an order n of a predictive system model.
Step S2, constructing a past and future combination matrix Y p、Yf, and calculating a covariance matrix Sigma pp、∑ff of past and future observations and a cross covariance matrix Sigma fp between the past and future observations.
And S3, constructing a Hankel matrix, carrying out singular value decomposition on the Hankel matrix, and obtaining a canonical variable Z r and a conversion matrix J based on past measured values.
And S4, projecting the conversion matrix J to the original data u to obtain the input data u' with reduced dimensions.
And S5, dividing the data x' into training data p 1、t1 and test data p 2、t2, carrying out normalization processing, creating and training an extreme learning machine model, setting the number of input layers and hidden layers N, L, and randomly initializing the connection weight IW and hidden layer bias B 1 between the input layers and the hidden layers.
And S6, calculating an output value tempH of the hidden layer, obtaining a mapping value h of the hidden layer to the sample characteristic by using a mapping function, and then calculating a connection weight LW of the hidden layer and the output layer.
And S7, repeating the step S6 for the trained model, and calculating tempH 1 and h 1 to obtain a predicted result Y ELM of the CVs-ELM model.
Step S8, the canonical variable space is further divided into a main space and a residual space by the step S3, the canonical variable Z r is obtained and expanded into an expansion vector, and then the linear combination of the past observation vectors is defined according to the residual space, so that a state vector x r、xr-1 is obtained, and the system state is determined.
And S9, a model coefficient matrix A, B, C, D is estimated through linear least square recursion, a model is built, and a prediction result Y CVA of the CVA state space model is obtained.
Step S10, for the two model prediction results, adopting a mean absolute percentage error (Mean Absolute Percentage Error, MAPE for short) selection strategy, and selecting that the error between each sample is small, so as to obtain a fusion driving load prediction result Y Parallel.
And S11, comparing the three types of prediction results with real data, and comparing the predicted performance indexes, namely average absolute Error (Mean Absolute Error, abbreviated as MAE), average relative Error (MEAN RELATIVE Error, abbreviated as MRE) and root mean square Error (Root Mean Square Error, abbreviated as RMSE), so as to obtain a conclusion.
Further, in step S1, the collected data is divided into past data y p(r) and future data y f(r), where r is expressed as a class of ordinal numbers, and p and f refer to the lengths of past and future observations, respectively.
Further, in step S2, r=p+1, p+2,..p+n is set, and a past and future combination matrix Y p、Yf is constructed, as follows.
Yp＝[yp(p+1) yp(p+2) … yp(p+N)] (3)
Yf＝[yf(p+1) yf(p+2) … yf(p-N)] (4)
Wherein Y p∈R、Yf e R, setting l observation variables, the last element of Y p(p+1) is Y (1),yf(p+N) and the last element is Y (l), i.e. the largest column number of Hankel matrix is n=l-p-f+1. Covariance matrices of past and future observations and cross-covariance matrices therebetween are calculated as follows.
Further, in step S3, the CVA attempts to find the best linear combination between y p(r) and y f(r) to maximize the correlation between past and future observations, which can be achieved by singular value decomposition of the Hankel matrix, as follows.
Based on past measurements, the canonical variable Z r and the transformation matrix J are derived as follows.
Further, in step S4, the transformation matrix J is projected onto the original data u to obtain the reduced-dimension input data u'.
u'=uJ (11)
Further, in step S5, the data u' is divided into training data p 1、t1 and test data p 2、t2, and normalized. The formula is as follows, where the input data p= [ P 1 p2 ], the output data t= [ T 1 t2 ].
[p1,minp,maxp,t1,mint,maxt]=premnmx(P,T) (12)
An extreme learning machine is created, the number of input layers and hidden layers is set S, L, and the connection weight IW and the hidden layer bias B 1 between the input layers and the hidden layers are initialized randomly, and the formula is as follows.
IW=2*rand(L,S)-1 (13)
B1=rand(L,1) (14)
Further, in step S6, an output value tempH of the hidden layer is calculated, a mapping value h of the hidden layer to the sample feature is obtained by using the mapping function, and then a connection weight L W of the hidden layer and the output layer is calculated, where the formula is as follows.
tempH=IW*p1+B1 (15)
LW=pinv(h′)*t1 (17)
Further, in step S7, the trained model is repeated in step S6, and tempH 1 and h 1 are calculated to obtain the predicted result Y ELM of the CVs-ELM model.
tempH1＝IW*P+B1 (18)
YELM＝h1′*LW (20)
Further, in step S8, the past measurement values are converted into a canonical variable space according to the conversion matrix J in step S3, which is divided into a main space and a residual space, the first Q main singular values determine the main space x r, and the remaining singular values determine the residual space r r, so that the canonical variable Z r is further expanded into an expansion vector, as follows, where V x contains the first Q columns of V,
A linear combination of past observation vectors is defined according to the residual space, so that a state vector x r、xr-1 is obtained, and the system state is determined. The formula is as follows, where 0 e R, the matrix is all zero elements,Is the past observation vector at r sampling time in the recursive modeling process.
Further, in step S9, the model coefficient matrix A, B, C, D is recursively estimated by linear least squares. The formula is as follows, wherein Xr＝x(r：r+n)、Xr-1＝x(r-1:r+n-1)、Yr＝y(r:r+n)、Yr-1＝y(r-1:r+n-1)、Ur＝u(r:r+n),ur is the raw data.
xr+1Axr+Bur (28)
And constructing a state space model to obtain a predicted result Y CVA of the CVA state space model.
YCVA＝Cxr+Dur (29)
Further, in step S10, an average absolute percentage error MAPE selection strategy is adopted for the two model predictions, whereinIs the predicted result, and y i is the true value. The formula is as follows.
And calculating MAPE 1 of the CVs-ELM model and MAPE 2 of the CVA state space model, and selecting a small MAPE value between each sample between the two models by using a condition conditional statement to obtain a fusion driving load prediction result Y Parallel.
Further, in step S11, the three types of prediction results are compared with the real data, and the predicted performance index MAE, MRE, RMSE is compared to obtain a conclusion. The formula is as follows.
In the above method, steps S1 to S7 are steps of building a CVs-ELM model, steps S8 to S9 are steps of building a CVA state space model, and steps S10 to S11 are steps of fusion driving prediction and concluding.
Compared with the prior art, the invention has the beneficial effects that:
1. The load prediction method provided by the invention originally introduces the idea of fusion driving prediction, and provides the water chilling unit load prediction method driven by fusion of data and a model, which obviously improves the accuracy of load prediction, realizes effective management of energy and injects new energy into the energy utilization field.
2. The invention skillfully fuses the standard variable analysis and the extreme learning machine, replaces the traditional method, creatively builds the CVs-ELM model, and can realize the effective and rapid extraction of the multi-variable complex information, thereby greatly improving the efficiency and the accuracy of information processing and providing a new way for the research and the application of the related fields.
3. According to the invention, based on the reduction of data dimension by standard variable analysis, a CVA state space model is further built, and the average absolute percentage error MAPE is used as a key selection strategy to complete fusion driving load prediction. Meanwhile, the method effectively ensures the accuracy of load prediction, provides powerful support for the research direction of load prediction, and shows great potential in practical application.
Drawings
The accompanying drawings are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate the invention and together with the embodiments of the invention, serve to explain the invention.
FIG. 1 is a general flow chart of the load prediction of the method of the present invention.
Figure 2 is a flow chart of the part of the method of the invention for building a model.
FIG. 3 is a graph showing the load prediction results of a CVs-ELM model building portion according to an embodiment of the present invention.
Fig. 4 is a load prediction result diagram of a CVA state space model building portion according to an embodiment of the present invention.
Fig. 5 is a graph of load prediction results using the fusion driving of the present invention according to an embodiment of the present invention.
FIG. 6 is a comparison of three load prediction result graphs according to an embodiment of the present invention.
FIG. 7 is a graph comparing two models with a fusion-driven prediction in accordance with an embodiment of the present invention.
Wherein, (a) is a comparison graph of absolute errors of two models and fusion driving prediction in the embodiment of the invention, and (b) is a comparison graph of relative errors of two models and fusion driving prediction in the embodiment of the invention.
Detailed Description
The present invention will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present invention more apparent. Of course, the specific embodiments described herein are for purposes of illustration only and are not intended to limit the invention.
Example 1
The american society of heating, cooling and air conditioning engineers (ASHRAE) was established in new york in 1894 in united states, and in 2019 ASHRAE held a field of energy prediction competition that provided open source real data about water chiller units, as shown in table 1. Based on this, the method for predicting the load of the water chilling unit driven by fusion of data and a model provided by the embodiment is as follows:
TABLE 1
The number of input variables of the real data of the water chilling unit is 8, the number of output data is only 1, and the number of samples is 13615. The embodiment visually expresses the effectiveness of the load prediction method of the fusion driving water chiller, and takes part of sample data as model training test data.
The specific steps of this embodiment are as follows:
Step 1, firstly dividing the sample data u into past data y p(r) and future data y f(r), setting an observation value length p and future observation value lengths f, wherein the values of p=1, f=1, setting an initial value of an order n of a predictive system model, wherein the n=1, r represents a class of ordinal numbers, the first data y (r-1) of the embodiment is the data u (1), and the like.
Setting r=p+1, p+2, & gt, p+N, constructing a past and future combination matrix Y p、Yf,Yp∈R、Yf E R, setting l observation variables, namely that the last element of Y p(p+1) is Y (1),yf(p+N), the last element is Y (l), and the maximum column number N=l-p-f+1 of the Hankel matrix, wherein N=8;
Yp＝[yp(p+1) yp(p+2) … yp(p+N)],Yf＝[yf(p+1) yf(p+2) … yf(p+N)];
and 3, calculating covariance matrixes of past and future observation values and a cross covariance matrix between the covariance matrixes, and then carrying out singular value decomposition on the Hankel matrix.
Step 4, obtaining a standard variable Z r and a conversion matrix J according to the step 3, and further projecting the conversion matrix J to sample data u to obtain dimension-reduced data u';
u′=uJ;
And 5, dividing the data u' into training data P 1、t1 and test data P 2、t2, and carrying out normalization processing on input data p= [ P 1 p2 ] and output data T= [ T 1 t2 ].
[p1,minp,maxp,t1,mint,maxt=premnmx(P,T);
And 6, creating an extreme learning machine ELM, setting the number of input layers S and hidden layers L, wherein S is the number 5,L of main space dimensions and is set to be 100 in the embodiment, and randomly initializing the connection weight IW and the hidden layer bias B 1 between the input layers and the hidden layers.
IW=2*rand(L,S)-1,B1=rand(L,1);
And 7, training by using training data p 1、t1, calculating an output value tempH of the hidden layer, obtaining a mapping value h of the hidden layer to the sample characteristics by using a mapping function, and calculating a connection weight LW of the hidden layer and the output layer.
tempH=IW*p1+B1,LW=pinv(h′)*t1;
And 8, after testing by using the test data, building a CVs-ELM model, and feeding the input data P into the model to obtain a predicted result Y ELM of the CVs-ELM model.
Step 9, according to the transformation matrix J in step 3, transforming the past measured values into a canonical variable space, dividing the canonical variable space into a main space and a residual space, determining the main space x r by the first Q main singular values, and determining the residual space r r by the residual singular values, so that the canonical variable Z r is further expanded into an expansion vector, wherein V x comprises the front Q column of V.
Step 10, defining the linear combination of past observation vectors according to residual space to obtain a state vector x r、xr-1, determining the system state, 0E R, the matrix is all zero elements,Is the past observation vector at r sampling time in the recursive modeling process.
And 11, constructing a state space model after a model coefficient matrix A, B, C, D is estimated through linear least square recursion, and feeding sample data u into the model to obtain a prediction result Y CVA of the CVA state space model.
Xr＝x(r：r+n)、Xr-1＝x(r-1：r+n-1)、Yr＝y(r：r+n)、Yr-1＝y(r-1：r+n-1)、Ur＝u(r：r+n);
xr+1＝Axr+Bur,YCVA＝Cxr+Dur;
And 12, for the two model prediction results, adopting an average absolute percentage error MAPE selection strategy, and selecting that the MAPE value between each sample between the two models is small to obtain a fusion driving load prediction result Y Parallel.
In this embodiment, simulation is performed according to the above steps, so as to obtain the load prediction accuracy of the chiller with multiple models within the receiving range, as shown in table 2.
TABLE 2 prediction accuracy with errors within 5 kWh
To more intuitively compare the three load prediction effects, the three prediction effects were compared using three indexes of Mean Absolute Error (MAE), mean Relative Error (MRE), root Mean Square Error (RMSE), as shown in table 3.
Table 3 comparison of predicted results
Evaluation index MAE/kWh MRE/% RMSE/kWh
CVs-ELM model 6.661 5.13% 3.584
CVA state space model 2.777 2.08% 0.479
Fusion driving of the present invention 1.963 1.48% 0.348
As can be seen from the above tables 2 and 3, the fusion driving proposed in this embodiment is significantly superior to the prediction results of the CVs-ELM model part and the CVA state space model part. The load prediction accuracy of the water chilling unit is greatly improved, and breakthroughs are brought to the field of energy management. The innovative data and model fusion driving method adopted by the embodiment not only organically combines the data and the model, creates a new way for reducing the dimension, effectively digs the potential value of multiple variables, but also further builds a state space model part to carry out load prediction, and remarkably improves the accuracy of the prediction. The outstanding performance of this approach allows fusion driven predictions to achieve good results. In summary, the method for predicting the load of the water chiller driven by fusion of data and the model in the embodiment shows high adaptability and value in the aspect of energy resource prediction.
The foregoing description of the preferred embodiments of the invention is not intended to limit the invention to the precise form disclosed, and any such modifications, equivalents, and alternatives falling within the spirit and scope of the invention are intended to be included within the scope of the invention.

Claims (9)

1.一种数据与模型融合驱动的冷水机组负荷预测方法，其特征在于，包括以下步骤：1. A chiller load forecasting method driven by data and model fusion, characterized by comprising the following steps: 步骤S1：对采集到的一段时间内的数据进行分割，将其划分为过去数据yp(r)和未来数据yf(r)，确定过去观测值长度p和未来观测值长度f，并设定预知系统模型的阶数n的初始值；Step S1: Segment the data collected within a period of time into past data yp(r) and future data yf(r) , determine the length of past observations p and the length of future observations f, and set the initial value of the order n of the prediction system model; 步骤S2：构造过去和未来的组合矩阵Yp、Yf，计算过去和未来观测值的协方差矩阵Σpp、Σff以及它们之间的互协方差矩阵Σfp；Step S2: construct the past and future combination matrices Y p , Y f , calculate the covariance matrices Σ pp , Σ ff of the past and future observations and the cross-covariance matrix Σ fp between them; 步骤S3：构造Hankel矩阵，对其进行奇异值分解，基于过去测量值得到规范变量Zr和转换矩阵J；Step S3: construct the Hankel matrix, perform singular value decomposition on it, and obtain the canonical variable Z r and the transformation matrix J based on the past measurement values; 步骤S4：将转换矩阵J投影到原始数据u，以获取降低维度后的输入数据u'；Step S4: Project the transformation matrix J onto the original data u to obtain the input data u' after the dimension is reduced; 在步骤S4中，将转换矩阵J投影到原始数据u，以获取降低维度后的输入数据u'，In step S4, the transformation matrix J is projected onto the original data u to obtain the input data u' after dimensionality reduction. u'＝uJ (1)；u'＝uJ (1); 步骤S5：将数据x'分为训练数据p1、t1和测试数据p2、t2，并进行归一化处理，创建并训练极限学习机模型，设定输入层和隐含层数量N、L，随机初始化输入层与隐含层间的连接权值IW和隐含层偏置B1；Step S5: Divide the data x' into training data p1 , t1 and test data p2 , t2 , and perform normalization processing, create and train the extreme learning machine model, set the number of input layers and hidden layers N, L, and randomly initialize the connection weight IW between the input layer and the hidden layer and the hidden layer bias B1 ; 步骤S6：计算隐含层的输出值tempH，利用映射函数得到隐含层对样本特征的映射值h，然后计算隐含层与输出层的连接权值LW；Step S6: Calculate the output value tempH of the hidden layer, use the mapping function to obtain the mapping value h of the hidden layer to the sample feature, and then calculate the connection weight LW between the hidden layer and the output layer; 步骤S7：对训练好的模型，重复步骤S6，计算tempH1和h1，得到CVs-ELM模型的预测结果YELM；Step S7: Repeat step S6 for the trained model, calculate tempH 1 and h 1 , and obtain the prediction result Y ELM of the CVs-ELM model; 步骤S8：由步骤S3进一步将规范变量空间分成主空间和残差空间，得到规范变量Zr扩展成扩展矢量，再根据残差空间定义过去观测矢量的线性组合，从而得到状态矢量xr、xr-1，确定系统状态；Step S8: Further divide the canonical variable space into the main space and the residual space according to step S3, expand the canonical variable Zr into an extended vector, and then define the linear combination of the past observation vectors according to the residual space, thereby obtaining the state vectors xr , xr-1 , and determining the system state; 步骤S9：通过线性最小二乘递归估计出模型系数矩阵A、B、C、D，搭建模型，得到CVA状态空间模型的预测结果YCVA；Step S9: Recursively estimate the model coefficient matrices A, B, C, and D through linear least squares, build the model, and obtain the prediction result Y CVA of the CVA state space model; 在步骤S9中，通过线性最小二乘递归估计出模型系数矩阵A、B、C、D，公式如下，其中Xr＝x(r:r+n)、Xr-1＝x(r-1:r+n-1)、Yr＝y(r:r+n)、Yr-1＝y(r-1:r+n-1)、Ur＝u(r:r+n)，ur是原始数据，In step S9, the model coefficient matrices A, B, C, and D are estimated by linear least squares recursion, and the formula is as follows, where Xr = x (r:r+n) , Xr -1 = x (r-1:r+n-1) , Yr = y (r:r+n) , Yr -1 = y (r-1:r+n-1) , Ur = u (r:r+n) , ur is the original data, xr+1＝Axr+Bur (29)x r+1 =Ax r +Bu r (29) 搭建状态空间模型，得到CVA状态空间模型的预测结果YCVA；Build a state space model and obtain the prediction result Y CVA of the CVA state space model; YCVA＝Cxr+Dur(30)；Y CVA =Cx r +Du r (30); 步骤S10：对于两个模型预测结果，采取平均绝对百分比误差选择策略，选取每一个样本间误差小的，得到融合驱动负荷预测的结果YParallel；Step S10: For the prediction results of the two models, the mean absolute percentage error selection strategy is adopted to select the one with the smallest error between each sample to obtain the result Y Parallel of the fusion drive load prediction; 在步骤S10中，对于两个模型预测结果，采取平均绝对百分比误差MAPE选择策略，其中是预测结果，yi是真实值，公式如下；In step S10, for the prediction results of the two models, the mean absolute percentage error (MAPE) selection strategy is adopted, where is the prediction result, yi is the true value, and the formula is as follows; 计算CVs-ELM模型的MAPE1、CVA状态空间模型的MAPE2，用condition条件语句选取两个模型间每一个样本间MAPE值小的，得到融合驱动负荷预测的结果YParallel；Calculate the MAPE 1 of the CVs-ELM model and the MAPE 2 of the CVA state space model, and use the conditional statement to select the one with the smallest MAPE value between each sample of the two models to obtain the result Y Parallel of the fusion drive load prediction; 步骤S11：对比三类预测结果与真实数据，对比预测的性能指标：平均绝对误差、平均相对误差、均方根误差，得出结论。Step S11: Compare the three types of prediction results with the actual data, compare the prediction performance indicators: mean absolute error, mean relative error, root mean square error, and draw conclusions. 2.根据权利要求1所述的数据与模型融合驱动的冷水机组负荷预测方法，其特征在于，在步骤S1中，对采集到的一段时间内的数据进行分割，将其划分为过去数据yp(r)和未来数据yf(r)，公式如下，式中r表示为一类序数，p和f分别指过去和未来观测值的长度；2. The chiller load forecasting method driven by data and model fusion according to claim 1 is characterized in that, in step S1, the data collected within a period of time is segmented and divided into past data yp(r) and future data yf(r) , and the formula is as follows, where r represents a type of ordinal number, and p and f refer to the lengths of past and future observations, respectively; 3.根据权利要求1所述的数据与模型融合驱动的冷水机组负荷预测方法，其特征在于，在步骤S2中，设置r＝p+1,p+2,…,p+N，构造过去和未来的组合矩阵Yp、Yf，公式如下：3. The chiller load forecasting method driven by data and model fusion according to claim 1 is characterized in that, in step S2, r=p+1, p+2, ..., p+N is set to construct the past and future combination matrices Y p and Y f , and the formula is as follows: Yp＝[yp(p+1)yp(p+2)…yp(p+N)] (5)Y p =[y p(p+1) y p(p+2) ...y p(p+N) ] (5) Yf＝[yf(p+1)yf(p+2)…yf(p+N)] (6)Y f =[y f(p+1) y f(p+2) ...y f(p+N) ] (6) 其中Yp∈R、Yf∈R，设置l个观测变量，则yp(p+1)最后一个元素是y(1)，yf(p+N)最后一个元素是y(l)，即Hankel矩阵最大列数是N＝l-p-f+1，计算过去和未来观测值的协方差矩阵和它们之间的互协方差矩阵，公式如下：Where Yp∈R , Yf∈R , set l observation variables, then the last element of yp(p+1) is y (1 ), and the last element of yf(p+N) is y (l) , that is, the maximum number of columns of the Hankel matrix is N=lp-f+1. Calculate the covariance matrix of past and future observations and the cross-covariance matrix between them. The formula is as follows: 4.根据权利要求1所述的数据与模型融合驱动的冷水机组负荷预测方法，其特征在于，在步骤S3中，CVA试图发现yp(r)和yf(r)之间的最佳线性组合，以最大化过去和未来观测值之间的相关性，通过对Hankel矩阵进行奇异值分解来实现，公式如下，4. The chiller load forecasting method driven by data and model fusion according to claim 1 is characterized in that, in step S3, CVA attempts to find the best linear combination between yp(r) and yf(r) to maximize the correlation between past and future observations, which is achieved by performing singular value decomposition on the Hankel matrix, and the formula is as follows, 基于过去测量值得到规范变量Zr和转换矩阵J，公式如下，Based on past measurements, the canonical variables Z r and the transformation matrix J are obtained as follows: 5.根据权利要求1所述的数据与模型融合驱动的冷水机组负荷预测方法，其特征在于，在步骤S5中，将数据u'分为训练数据p1、t1和测试数据p2、t2，并进行归一化处理，公式如下，其中输入数据P＝[p1p2]，输出数据T＝[t1t2]，5. The chiller load forecasting method driven by data and model fusion according to claim 1 is characterized in that, in step S5, the data u' is divided into training data p 1 , t 1 and test data p 2 , t 2 , and normalized, and the formula is as follows, wherein input data P = [p 1 p 2 ], output data T = [t 1 t 2 ], [p1,minp,maxp,t1,mint,maxt]＝premnmx(P,T) (13)[p 1 ,minp,maxp,t 1 ,mint,maxt]=premnmx(P,T) (13) 创建极限学习机，设定输入层和隐含层数量S、L,随机初始化输入层与隐含层间的连接权值IW和隐含层偏置B1，公式如下，Create an extreme learning machine, set the number of input layers and hidden layers S and L, randomly initialize the connection weights IW between the input layer and the hidden layer and the hidden layer bias B 1 , the formula is as follows, IW＝2*rand(L,S)-1 (14)IW＝2*rand(L,S)-1 (14) B1＝rand(L,1) (15)。B 1 =rand(L,1) (15). 6.根据权利要求1所述的数据与模型融合驱动的冷水机组负荷预测方法，其特征在于，在步骤S6中，计算隐含层的输出值tempH,利用映射函数得到隐含层对样本特征的映射值h，然后计算隐含层与输出层的连接权值LW,公式如下，6. The chiller load prediction method driven by data and model fusion according to claim 1 is characterized in that, in step S6, the output value tempH of the hidden layer is calculated, the mapping value h of the hidden layer to the sample feature is obtained by using a mapping function, and then the connection weight LW between the hidden layer and the output layer is calculated, and the formula is as follows: tempH＝IW*p1+B1 (16)tempH＝IW*p 1 +B 1 (16) LW＝pinv(h')*t1 (18)。LW = pinv(h')* t1 (18). 7.根据权利要求1所述的数据与模型融合驱动的冷水机组负荷预测方法，其特征在于，在步骤S7中，对训练好的模型，重复步骤S6，计算tempH1和h1，得到CVs-ELM模型的预测结果YELM，7. The chiller load forecasting method driven by data and model fusion according to claim 1 is characterized in that, in step S7, for the trained model, step S6 is repeated to calculate tempH 1 and h 1 to obtain the forecast result Y ELM of the CVs-ELM model, tempH1＝IW*P+B1 (19)tempH 1 =IW*P+B 1 (19) YELM＝h1'*LW (21)。Y ELM =h 1 '*LW (21). 8.根据权利要求1所述的数据与模型融合驱动的冷水机组负荷预测方法，其特征在于，在步骤S8中，根据步骤S3中转换矩阵J，转换过去测量值到规范变量空间划分成：主空间和残差空间，前Q个主要奇异值确定主空间xr，剩余的奇异值确定残差空间rr，规范变量Zr进一步扩展成扩展矢量，公式如下，其中Vx包含V的前Q列， 8. The chiller load forecasting method driven by data and model fusion according to claim 1 is characterized in that, in step S8, according to the conversion matrix J in step S3, the past measured values are converted into the canonical variable space and divided into: a main space and a residual space, the first Q main singular values determine the main space x r , the remaining singular values determine the residual space r r , and the canonical variable Z r is further expanded into an extended vector, the formula is as follows, where V x contains the first Q columns of V, 根据残差空间定义过去观测矢量的线性组合，从而得到状态矢量xr、xr-1，确定系统状态，公式如下，其中0∈R，矩阵全是零元素，是递推建立模型过程中r采样时刻的过去观测矢量，Define the linear combination of past observation vectors according to the residual space to obtain the state vectors x r , x r-1 , and determine the system state. The formula is as follows, where 0∈R, the matrix is full of zero elements, is the past observation vector at the r sampling time in the process of recursive model building, 9.根据权利要求1所述的数据与模型融合驱动的冷水机组负荷预测方法，其特征在于，在步骤S11中，对比三类预测结果与真实数据，对比预测的性能指标MAE、MRE、RMSE，得出结论，公式如下；9. The chiller load forecasting method driven by data and model fusion according to claim 1 is characterized in that, in step S11, the three types of forecast results are compared with the real data, and the forecast performance indicators MAE, MRE, and RMSE are compared to draw a conclusion, and the formula is as follows; 步骤S1至步骤S7为搭建CVs-ELM模型部分，步骤S8至步骤S9为搭建CVA状态空间模型阶段，步骤S10至步骤S11为融合驱动预测部分并得出结论。Steps S1 to S7 are for building the CVs-ELM model, steps S8 to S9 are for building the CVA state space model, and steps S10 to S11 are for the fusion drive prediction part and drawing conclusions.
CN202410569004.5A 2024-05-09 2024-05-09 A chiller load forecasting method driven by data and model fusion Active CN118297239B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202410569004.5A CN118297239B (en) 2024-05-09 2024-05-09 A chiller load forecasting method driven by data and model fusion

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202410569004.5A CN118297239B (en) 2024-05-09 2024-05-09 A chiller load forecasting method driven by data and model fusion

Publications (2)

Publication Number Publication Date
CN118297239A CN118297239A (en) 2024-07-05
CN118297239B true CN118297239B (en) 2024-12-03

Family

ID=91688281

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202410569004.5A Active CN118297239B (en) 2024-05-09 2024-05-09 A chiller load forecasting method driven by data and model fusion

Country Status (1)

Country Link
CN (1) CN118297239B (en)

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101542407A (en) * 2006-09-29 2009-09-23 费舍-柔斯芒特系统股份有限公司 Multivariate detection of abnormal conditions in a process plant
AU2020104000A4 (en) * 2020-12-10 2021-02-18 Guangxi University Short-term Load Forecasting Method Based on TCN and IPSO-LSSVM Combined Model

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104299043A (en) * 2014-06-13 2015-01-21 国家电网公司 Ultra-short-term load prediction method of extreme learning machine
CN108876054B (en) * 2018-07-06 2022-03-01 国网河南省电力公司郑州供电公司 Short-term power load prediction method based on improved genetic algorithm optimization extreme learning machine

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101542407A (en) * 2006-09-29 2009-09-23 费舍-柔斯芒特系统股份有限公司 Multivariate detection of abnormal conditions in a process plant
AU2020104000A4 (en) * 2020-12-10 2021-02-18 Guangxi University Short-term Load Forecasting Method Based on TCN and IPSO-LSSVM Combined Model

Also Published As

Publication number Publication date
CN118297239A (en) 2024-07-05

Similar Documents

Publication Publication Date Title
CN116448419B (en) Zero-Shot Bearing Fault Diagnosis Method Based on High-Dimensional Parameter Multi-Objective Efficient Optimization Using Deep Models
CN113449919B (en) Power consumption prediction method and system based on feature and trend perception
CN110069467A (en) System peak load based on Pearson's coefficient and MapReduce parallel computation clusters extraction method
CN115472233B (en) Semi-supervised integrated industrial process soft sensor modeling method and system based on heat diffusion label propagation
CN106650774A (en) A method of obtaining the regression relationship between dependent variable and independent variable in data analysis
Nichiforov et al. Deep learning techniques for load forecasting in large commercial buildings
CN114021483A (en) Ultra-short-term wind power prediction method based on time domain characteristics and XGboost
CN113255900A (en) Impulse load prediction method considering improved spectral clustering and Bi-LSTM neural network
CN111861002A (en) Building cooling and heating load prediction method based on data-driven Gaussian learning technology
CN119128423B (en) A method for calculating electrical appliance energy consumption based on digital twin model
CN120671471B (en) A Smart Optimization Design Method for Curtain Wall Frame Structure
CN114357870A (en) Metering equipment operation performance prediction analysis method based on local weighted partial least squares
CN118569892A (en) A method for predicting second-hand housing prices based on deep learning
CN119806063A (en) An intelligent planning method for assembly sequence of high-packing-density aviation equipment products
CN105912773A (en) Novel intelligent stamping process design method based on data mining technology
CN118297239B (en) A chiller load forecasting method driven by data and model fusion
CN107451684A (en) Stock market's probability forecasting method based on core stochastic approximation
CN118821874A (en) An intelligent system evaluation method based on feedback feature extraction and causal inversion identification
Kaligambe et al. Indoor Room Temperature and Relative Humidity Estimation in a Commercial Building Using the XGBoost Machine Learning Algorithm
CN105913153A (en) Target predication method for measuring characteristic similarity of multidirectional data
CN118410292A (en) Multi-element time sequence long-term prediction method based on depth FCM
Karijadi et al. Electricity Load Prediction using Fuzzy c-means Clustering EMD based Support Vector Regression for University Building
CN111008783B (en) Factory processing flow recommendation method based on singular value decomposition
Arias-Requejo et al. Energy demand prediction in smart buildings using advanced machine learning techniques
CN113191082B (en) Model parameter obtaining method and system based on machine learning and readable medium

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant