JP3214876B2 - Neural network configuration method and neural network construction support system - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a neural network system, and more particularly to a high-speed learning / recall suitable for real-time processing of time-series information, that is, processing for predicting a future value based on a transition of past data. The present invention relates to a method of determining a structure of a neural network to be performed, and an inference method and apparatus including a time-series information generating unit.

[0002]

2. Description of the Related Art A neural network has a multilayer structure including an input layer, an intermediate layer, and an output layer. Signals input to the input layer are determined by the degree of coupling between the layers of neurons constituting each layer. A unique output signal is output from the output layer. The degree of connection of the neurons can be appropriately determined by learning. Therefore, in order to use a neural network, it is necessary to perform learning for determining the degree of connection between each neuron and each neuron in another layer so that a desired output signal can be obtained for a certain input signal. . Conventionally, the most typical learning algorithm as a learning method for a neural network is a method using an error back propagation learning method. This is a method in which an input (learning pattern) is given to a neural network, and the degree of coupling is adjusted so that the output at that time from an ideal value (teacher pattern) is minimized.

Incidentally, in order to actually use a neural network in control or the like, it is necessary to be able to perform processing at a high speed similarly to other functional elements, that is, to have high processing performance and high output accuracy. Is required. The performance of the neural network changes depending on the configuration mode, learning method, and the like. Therefore, how to improve the processability and the accuracy is the greatest practical problem of the neural network.

In order to improve the processing performance, that is, the processing speed, (1) improving the operation speed itself by increasing the speed of the elements, and (2) reducing the number of logically time-consuming products and additions. And (3) minimizing the configuration of the network, that is, optimizing the configuration of the neural network. Among them, (1) and (2) are the same as the problems in the conventional information processing, but (3) is a problem unique to the neural network.

As one application of the neural network, a system for predicting a future value from a past time series is considered. In such a system, the input neurons of the neural network are given time-series data that is retroactive for a certain period of time from the current value, and the output neuron value is used as the predicted value. For this reason, it is required to obtain high-speed and high-precision results, and an optimal network configuration is required.

However, conventionally, the input layer of the neural network determines the number of neurons according to constraints such as a measuring instrument, a sample interval and a sample period, that is, corresponding to an input variable, and the output layer predicts. The number of neurons is determined according to the output variable, and the hidden layer (intermediate layer) is empirically determined by the number of neurons in the input layer and the output layer. That is, no effective research has been conducted on a method of configuring a neural network that can be used as an inference device applicable to a process control device. Particularly in the field of real-time prediction using time-series data, no effective method has been proposed for speeding up and increasing accuracy so that neural networks can be applied to process control devices. It is the current situation.

[0007] For the neural network, for example, "Pattern recognition and learning algorithm"
(Sentence-Published by Sogo Publishing), JP-A-1-201764, and the like.

[0008]

Problems to be solved when the above-described neural network is put to practical use, particularly when prediction is performed based on time-series data, will be further described.

FIG. 3 shows a neural network for outputting a predicted value from time series data. The input neuron is input with past values (time-series data) sampled n times at equal intervals at a sampling interval d within a sampling period T from continuous data. this is,
It is represented by the following equation.

X (t), x (t−d), x (t−2d),...
.., X (t−nd) where T = (n * d) Further, the predicted value is represented by x (t + P).

In FIG. 3, the horizontal axis t represents time, and the vertical axis x is an actual measurement value at that time. The sampling period T represents a past time that affects the recall result (predicted value).

Here, a set of time series data P ₁ , P ₂ ,..., P for a certain sampling period measured at certain intervals.
_By learning _m and a measurement value at a time P (a fixed time) after t (current time) for each pair as a learning pattern / teacher pattern, a neural network neuron that outputs a predicted value is learned. Determine the weight of the connection.

When the sampling interval T is increased while the sampling period T is kept constant, the number of input neurons decreases. As a result, the structure of the neural network is reduced, and the processability of the learning / recall processing is improved. On the other hand, the recall result faithfully represents a measurement point with a large sampling interval, and therefore differs from the waveform of the actual measurement value, and the prediction accuracy is deteriorated. When the sampling interval T is reduced while the sampling period T is fixed, the number of input neurons increases. As a result, the structure of the neural network becomes large, and the processability of the learning / recall processing decreases. On the other hand, the recall result faithfully represents a measurement point with a small sampling interval, so that it is closer to the waveform of the actual measurement value, and the prediction accuracy is improved.

As described above, the determination of the sampling interval d in the prediction of time-series data has a direct effect on processability and accuracy.

On the other hand, when the sampling interval d is constant,
When the sampling period T is reduced, the number of input neurons decreases. As a result, the structure of the neural network becomes smaller, and the processing of learning / recalling is improved.

However, to what extent the size is reduced,
At present, no effective solution has been proposed for the problem of affecting accuracy degradation. However, application to actual problem solving is difficult, and no effective decision method for general tasks has been proposed.

A first object of the present invention is to provide a method of constructing a neural network having an optimum sampling interval and sampling period in predicting time-series data as described above.

A second object of the present invention is to provide a neural network construction support system for configuring a neural network having an optimal sampling interval and sampling period in predicting time-series data.

Further, a third object of the present invention is to provide a neural network system which can converge early in learning, perform high-speed and high-precision operation in recall, and can process prediction of time-series data in real time. Is to provide.

[0020]

According to a first aspect of the present invention, there is provided a computer system having a plurality of input neurons and at least one output neuron,
When constructing a neural network in which past time-series data is sampled and input to an input neuron and a predicted value for predicting a future value of the time-series data is obtained from an output neuron, a sampling interval and sampling corresponding to a predetermined number of input neurons are used. A teacher pattern for a period is generated, a learning of the neural network is performed on the generated teacher pattern, and a prediction accuracy is evaluated for a prediction value obtained by the neural network. The generation of the teacher pattern sequentially changed for at least one of the sampling interval and the sampling period and the learning and the evaluation of the prediction accuracy are repeated, and when the prediction accuracy set in advance is obtained, the sampling interval of the teacher pattern at that time and Sun From the ring period, to determine the number of input neurons of the neural network, configuring the neural network, characterized in that to define the configuration of the neural network is provided.

According to a second aspect of the present invention, there is provided a neural network comprising: means for defining a configuration of a neural network; a plurality of defined input neurons and at least one output; Neural network holding means for storing a neural network including neurons; learning means for learning the stored neural network using a teacher pattern; teacher pattern generating means for generating the teacher pattern; Prediction accuracy evaluation means for evaluating the prediction accuracy of the prediction value obtained by the neural network, and control means for controlling the prediction accuracy. The control means obtains a prediction value that reaches a preset accuracy. When not, the above sampling interval and For at least one of the sampling periods, a changed teacher pattern is generated, the learning means is made to perform learning using the changed teacher pattern, and the prediction accuracy evaluation means is evaluated for accuracy. Means for controlling until the value reaches a preset accuracy, means for defining the configuration of the neural network, when the predicted value reaches the preset accuracy, the sampling interval and sampling period at that time A neural network construction support system having a function of determining a corresponding number of input neurons as the number of input neurons of the neural network is provided.

[0022]

The evaluation of the prediction accuracy can be performed, for example, by obtaining a correlation coefficient between the predicted value and the actually measured value, and comparing the correlation coefficient with a predetermined satisfaction value. here,
As the actually measured value, for example, data at the next time after the data input by the teacher pattern may be used.

It is preferable that the change of at least one of the sampling interval and the sampling period be sequentially changed in a direction in which the number of input neurons of the neural network is small to large. This is because the smaller the number of input neurons, the smaller the amount of calculation, and the faster the learning.

[0025]

[0026]

In the logic for determining the values of the sampling interval d and the sampling period T, the sampling interval and the sampling period (time) necessary for obtaining the desired prediction accuracy in the shortest time can be optimally determined. As a result, the structure of the neural network is defined and time-series data is generated. Therefore, redundant operations in learning are reduced, and as a result, learning is completed in a short time. Also, high processing performance can be obtained in recall.

Furthermore, since the prediction evaluation is performed using the correlation coefficient between the final predicted value and the measured value in the logic for determining the values of the sampling interval d and the sampling period T, a desired prediction accuracy is guaranteed. Will be. Therefore,
The prediction value of the time-series data can be determined with high accuracy and at high speed.

[0028]

DESCRIPTION OF THE PREFERRED EMBODIMENTS One embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is a functional diagram for realizing a time series neural network configuration method according to an embodiment of the present invention.
Shown in With reference to FIG. 1, an outline of the neural network configuration method of the present embodiment will be described.

In this embodiment, in a neural network for predicting a future value from past time series data, values x (t) sampled at regular intervals from time series data corresponding to a plurality of input neurons 1; x (t−d), x
When inputting (t−2d),..., X (t−nd), the sampling interval d and the period from the present to the past, that is, the sampling period (T = n * d) are optimal.
That is, the configuration of the neural network having the minimum number of input neurons within the range in which the desired prediction accuracy can be obtained is determined.

First, the number of input neurons corresponding to the sampling interval d and the sampling period T is temporarily determined, and the configuration of the neural network is defined. Then, the time series data generated by the teacher pattern generating means 1 is stored in the teacher pattern (input) memory 6 and the teacher pattern (output) memory 7, and learning is performed by the learning means 4 based thereon. After the learning has converged, the prediction accuracy evaluation means 3 calculates a correlation coefficient between the predicted value by the neural network and a corresponding value of the teacher pattern (a value corresponding to an actually measured value). If the correlation coefficient satisfies the desired prediction accuracy, the time-series neural network defining means 2 determines a time-series neural network 5 having input neurons corresponding to the sampling interval d and the sampling period T at that time. When the desired prediction accuracy is not satisfied, a teacher pattern (input / output) is generated by sequentially changing at least one of the sampling interval d and the sampling period T until the desired prediction accuracy is satisfied. , And repeat the evaluation of the prediction accuracy. The reason why the teacher pattern is used as the learning pattern will be described later.

The number of input neurons of the neural network is changed so that the number of input neurons is sequentially changed from a smaller number to a larger number. Therefore, when initializing the neural network, the number of input neurons is set in consideration of this.

Next, FIG. 2 shows a configuration diagram of a learning / recall system according to an embodiment of the present invention.

This learning / recall system comprises a computer system 10, an input / output device 11 for inputting desired prediction accuracy and displaying a definition state, an input device 12 for inputting data, programs, etc., and a neural network. And an output device 13 for outputting the configuration data, the recall result, and the like. The computer system 10 includes, for example, a central processing unit (not shown), an arithmetic processing unit 10 a having a main memory for storing operation programs and data, and an interface 9. The computer system 10 executes a program to define a neural network 5 corresponding to an optimal sampling interval and a sampling period.
A teacher pattern generating means 1 for generating a teacher pattern corresponding to the current sampling interval and sampling period, a learning means 4 for executing learning based on the teacher patterns,
Prediction accuracy evaluation means 3 for evaluating the prediction accuracy of the learning result
And control means 20 for controlling the execution of each of the above means. Each of the above units exchanges information with the input / output device 11, the input device 12, and the output device 13 via the interface 9.

The neural network used in this embodiment is constructed such that the number of output neurons is 1 and the number of input neurons is (n + 1). As shown in FIG. 3, when the sampling period is T and the sampling interval is d, the number of input neurons is determined to be (n + 1) from the relationship of T = nd. The intermediate layer is appropriately set according to the number of input neurons. For example, the number may be equal to or less than the number of input neurons. The number of neurons in the intermediate layer may be increased or decreased in accordance with the number of input neurons, or may be fixed regardless of the number of input neurons.

The learning means 4 comprises, for example, a learning calculation means 41 and a weight calculation correction means 42 as shown in FIG. A learning pattern is input to the learning operation unit 41. In response to this, the learning operation unit 41 sequentially performs a non-linear function operation of sum of products of the neuron output value and the synapse weight for the neural network stored in the neural network holding unit 5 for each layer, and outputs the output value h (x ) Is output. The weight calculation correcting means 42 includes
An error between the teacher pattern d (x) and the output value h (x) is obtained, and the weight coefficient is corrected using the back propagation method so as to minimize the error.

Here, in the case of the present embodiment, the same data can be used for the learning pattern and the teacher's pattern because they are time-series data. That is, the learning is performed by using data from the specific point in time of the teacher pattern up to one sampling interval before as the learning pattern, and learning so that the predicted value approaches the teacher pattern data at the specific point in time. It is.

Next, an example in which the present invention is applied to a process of predicting a VI (transmittance) value in a tunnel ventilation process by a neural network will be described.
Here, the prediction accuracy of the predicted value is evaluated by obtaining a correlation with the measured value.

The present invention evaluates and compares the prediction accuracy of a time-series neural network using a correlation graph and a correlation coefficient between a VI actual measurement value (%) and a VI prediction value (%) as shown in FIG. The optimum sampling interval and sampling period T are obtained, and the configuration of the time-series neural network is determined based on them.

The correlation function used in this embodiment is represented by the following equation.

[0041]

(Equation 1)

FIG. 4 shows the number of employed data (279 data) with the vertical axis representing the VI predicted value (%) and the horizontal axis representing the VI actual measured value (%).
, And this graph is called a correlation diagram.

In the correlation diagram, when the measured VI value (%) increases, the predicted VI value (%) also tends to increase (positive correlation).
When the measured value and the predicted value are aligned on a straight line and coincide with each other, the strongest correlation is shown, and r = 1.

Accordingly, when the correlation coefficients = 0.34614 and 0.667299 shown in FIGS. 4A and 4B,
0.667299 is a stronger correlation, and can be evaluated as having higher prediction accuracy.

FIG. 5 shows a satisfactory value of the correlation coefficient (r _S ).
, The sampling interval d of the neural network
4 shows a flowchart of a logic for determining a sampling period T and a sampling period T.

Sampling interval d and sampling period T
Is determined according to the flowchart in the computer system 10.
Is processed in the following order.

S101: The following initial values are set.

(1) Set the initial value of the sampling period.

(Sufficiently large period in which the effect is expected on the prediction after a certain time from the present time) (2) Set the initial value of the sampling interval of the time series data.

(Sufficiently large interval for measurable sampling interval) S102: For the following S103 to S107, S10
At 5, the process is repeated until the correlation coefficient for the correction of the sampling interval d satisfies the desired correlation coefficient.

S103: Time-series data is sampled based on the sampling period T and the sampling interval d, and the time-series data is input and learned to obtain a weight coefficient of the neural network.

S104: A correlation coefficient r is calculated from the measured value sampled in S103 and the predicted value determined in the above step.

S105: The correlation coefficient calculated in the above step is compared with a predetermined degree of correlation r _S given in advance.

When r ≧ r _S, the condition is satisfied (satisfied) S106; When the condition is satisfied (satisfied) in S105, the correction of the sampling interval d ends.

S107: Condition not satisfied at S105 (other than)
, The sampling interval d is narrowed by Δd, and S1
Return to 03.

S108: Regarding the following S109 to S114, in S111, the limit value at which the correlation coefficient for the correction of the sampling period T satisfies the desired correlation coefficient (unsatisfactory 1)
Repeat).

S109: Same as S103.

S110: Same as S104.

S111: Same as S105.

S112: If the condition is satisfied (satisfied) in S111, the sampling period T is narrowed by ΔT, and S10
Return to 9.

S113 to S114: When the condition is not satisfied (other than S111) in S111, the sampling period T is widened by ΔT, and the correction of the sampling period T ends.

To summarize the above processing, S101 to S1
In the processing up to 07, the desired prediction accuracy, that is, the sampling interval required to obtain the degree of correlation between the actually measured value and the predicted value is determined, and the processing in S108 to S114 is required to obtain the above-mentioned accuracy. From the present to the past,
That is, it is the logic for determining the shortest sampling period that affects the future value prediction.

The relationship between the sampling interval and the correlation coefficient will be described with reference to FIGS.

First, a fixed time P from the current point to the prediction point
Will be described.

A predicted value x for a certain time P after the current point
The sequence of (t + P) (marked with ○) is the measured values x (t), x (t−d),
.., And a graph connecting points at equal intervals of x (t-nd) (marked by ●) is represented as faithfully as possible. Therefore,
For example, when P = d / 2, by calculating the correlation coefficient between all the corresponding predicted values and the measured values corresponding to the predicted values,
The degree of correlation between the graph (dotted line) connecting the predicted values and the measured value graph (solid line) corresponding to the predicted value can be determined.

Here, the satisfaction value (r _S ) of the correlation coefficient is a constant set in advance depending on the application field and the amount to be predicted. For example, in VI prediction, 0.9 or more is sufficient.

FIGS. 7 and 8 show the difference in the correlation coefficient depending on the sampling interval. Here, the sampling period is fixed.

FIG. 7 shows that in the embodiment, when the sampling interval d is large, the correlation coefficient is a small value. On the other hand, FIG. 8 shows that the correlation coefficient was improved by reducing the sampling interval to half that of FIG. 7, and the prediction graph became similar to the measurement graph.

7 and 8, the number of input neurons is larger in the method of the example shown in FIG. 8 than in the example shown in FIG. The correlation coefficient at the sampling interval shown in FIG. 7 is as shown in FIG. 4A, and the correlation coefficient at the sampling interval shown in FIG. 8 is as shown in FIG.

From this, by making the sampling interval smaller and increasing the sampling data,
It can be seen that the prediction accuracy is further improved. When the sampling interval is increased, the number of input neurons is correspondingly increased.
There is a problem that processing time is required. However, in the case of the present invention, the number of input neurons can be determined when the required accuracy is reached by evaluating the accuracy of the predicted value in advance. For this reason, in the present invention, the configuration of the neural network can be set to the minimum scale while ensuring necessary accuracy.

The present invention can be applied to a process control system for controlling a process by obtaining a predicted value for time series data of a measured value. This allows, for example,
A neural network that includes a predefined number of input neurons and output neurons, and outputs a predicted value for the time series data of the measured value; and, for the measured value, a sampling period and It is possible to realize a process control system including a time-series data generating unit that generates time-series data by sampling measurement values at sampling intervals and supplies the time-series data to input neurons.

FIG. 9 shows an example of this process control system, showing an embodiment of a process control device in a tunnel ventilation process provided with the neural network of FIG.

In FIG. 9, a control device 16 such as a dust collector is provided on the process side, and a measuring device 15 for measuring the state of each part of the process is provided.

Further, the process control device 17 includes the measuring device 1
5, an input device 12 which takes in the measured values from 5 and converts it into digital data, a time-series data generator 8 which samples input data to generate time-series data, a computer system 10, and outputs from the computer system 10. An operation amount calculating unit 14 for obtaining an operation amount based on a signal to be operated, an obtained operation amount calculating unit, and an output device 13 for outputting the calculated operation amount.

The time-series process amount input by the measuring device 15 is input by the input device 12, and the corresponding time-series process amount is converted into a time-series data for generating time-series data corresponding to a predetermined optimum sampling interval and sampling period. The computer system 1 is interfaced with the interface through the data generator 8.
By inputting 0, the prediction amount is accurately determined by the time-series neural network, which is a high-speed inference device built in the computer system, a predetermined operation amount for the prediction amount is calculated by the operation calculation unit 14, and the process output is output. The output of the control device 16 is performed by the device 13.

As shown in FIGS. 7 and 8, by gradually reducing the sampling interval and then decreasing the sampling interval, in the tunnel ventilation process, for a desired correlation coefficient of 0.9 or more, the sampling interval is 5 minutes. Since a sampling period of 30 minutes is satisfied, a time-series data generator corresponding to seven input neurons is provided according to the corresponding time-series sampling condition.

That is, the VI value (t) and the VI value (t−
5) The VI value (t-10), the VI value (t-15), the VI value (t-20), the VI value (t-25), and the VI value (t-30)
A VI value (t + 5) is input as an input value to an input neuron, and a VI value (t + 5) is determined as an output neuron value of the corresponding neural network after learning, and a predetermined operation amount is output to a control device according to the value to give a driving instruction. It is like that.

As described above, according to the present invention, in a neural network for predicting a future value from past time series data, values sampled at regular intervals from time series data corresponding to a plurality of input neurons. When inputting, the neural network is built in because the sampling interval and sampling period are optimal, that is, time series data corresponding to a plurality of input neurons that are minimum within a range that can obtain the desired prediction accuracy can be automatically generated. With such an inference apparatus, there is an effect that accurate learning can be performed in a short time and high-speed prediction can be performed.

[0079]

According to the present invention, prediction and evaluation are performed based on the correlation coefficient between the actually measured pattern and the predicted pattern, and the structure of the neural network is defined based on the result. There is an effect that it becomes possible.

[Brief description of the drawings]

FIG. 1 is a functional diagram of a time-series neural network configuration device.

FIG. 2 is a block diagram showing a hardware configuration of a time-series neural network.

FIG. 3 is an explanatory diagram of a neural network in predicting time-series data.

FIG. 4 is a graph showing a VI actual measured value / predicted value correlation graph and a correlation coefficient.

FIG. 5 is a flowchart showing a logic for determining a sampling interval d and a sampling period T;

FIG. 6 is an explanatory diagram showing a relationship between a sampling interval d and a fixed time P from a current point to a predicted point.

FIG. 7 is an explanatory diagram showing time-series prediction of the neural network when the sampling interval d is large.

FIG. 8 is an explanatory diagram showing time-series prediction of a neural network when the sampling interval d is small (when correction is performed).

FIG. 9 is a block diagram illustrating a configuration of an example of a process control device.

FIG. 10 is a block diagram showing a configuration of an example of a learning unit.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... Teacher pattern generation means, 2 ... Time series neural network definition means, 3 ... Prediction accuracy evaluation means, 4 ... Learning means, 5 ... Neural network, 6 ... Teacher pattern (input) memory, 7 ... Teacher pattern (output) memory , 8
... time-series data generator, 9 ... interface, 10 ...
Computer system, 10a arithmetic processing unit, 11 input / output device, 12 input device, 13 output device, 14 operation amount arithmetic unit, 15 measuring instrument, 16 control device, 17 process control device.

──────────────────────────────────────────────────続 き Continuing on the front page (72) Inventor Masakazu Yahiro 5-2-1 Omikacho, Hitachi City, Ibaraki Prefecture Inside the Hitachi, Ltd. Omika Plant (72) Inventor Noboru Abe 5-2-2 Omikamachi, Hitachi City, Ibaraki Prefecture No. 1 Hitachi, Ltd. Omika Plant (56) References JP-A-4-318656 (JP, A) JP-A-2-28701 (JP, A) Tsuboka, Takada, Wakita, "has a time-series processing function Hierarchical Neural Network "IEICE Technical Report, Vol. 91, No. 95 (SP91-6-17), pp. 63-70 (Patent Office CSDB literature number: CCNT199900735006) (58) Fields investigated (Int. Cl. ⁷ , DB name) G06N 1/00-7/00 G05B 13/00-13/04 G06G 7/60 JICST File (JOIS) CSDB (Japan Patent Office)

Claims (5)

(57) [Claims] An input neuron includes at least one output neuron, samples past time-series data and inputs the input neuron, and predicts a future value of the time-series data from the output neuron. When constructing a neural network to obtain, a training pattern of a sampling interval and a sampling period corresponding to a predetermined number of input neurons is generated, learning of the neural network is performed on the generated training pattern, and a prediction value obtained by the neural network is obtained. For the evaluation of the prediction accuracy, until a predetermined accuracy is obtained, repeatedly generating teacher patterns sequentially changed for at least one of the sampling interval and the sampling period, and repeatedly learning and evaluation of the prediction accuracy, Pre-set schedule A method for constructing a neural network, comprising determining the number of input neurons of a neural network from a sampling interval and a sampling period of a teacher pattern at the time when measurement accuracy is obtained, and defining a configuration of the neural network. 2. A means for defining a configuration of a neural network, a neural network holding means for storing a neural network including a plurality of defined input neurons and at least one output neuron, and the stored neural network. About the network
Learning means for learning using a teacher pattern; teacher pattern generation means for generating the teacher pattern; prediction accuracy evaluation means for evaluating the prediction accuracy of a predicted value obtained by the learned neural network; Control means for performing, when the obtained predicted value does not reach a preset accuracy, the teacher pattern generation means, at least one of the sampling interval and the sampling period, the changed teacher pattern Generating, causing the learning means to perform learning using the changed teacher pattern, and causing the prediction accuracy evaluation means to evaluate the accuracy until the predicted value reaches a preset accuracy. Means for controlling, and means for defining the configuration of the neural network,
A neural network construction support system having a function of determining the number of input neurons corresponding to a sampling interval and a sampling period at the time when a predicted value reaches a preset accuracy as the number of input neurons of the neural network. 3. The method according to claim 1, wherein the evaluation of the prediction accuracy is performed in advance.
Find the correlation coefficient between the measured value and the actual measured value, and
This is done by comparing with the set satisfaction value.
How to configure a neural network. 4. The method according to claim 2, wherein the prediction accuracy evaluation means comprises:
The evaluation of prediction accuracy involves calculating the correlation coefficient between the predicted value and the measured value.
To compare this correlation coefficient with a preset satisfaction value.
Neural network construction support
Support system. 5. The method according to claim 1, wherein the sampling interval and
Change for at least one of
Furthermore, the number of input neurons in the neural network is small.
A neural that changes sequentially from the direction that does not have more
How to configure a network.