CN106920008A - A kind of wind power forecasting method based on Modified particle swarm optimization BP neural network - Google Patents

Abstract

The invention discloses a wind power prediction method based on improved particle swarm optimization BP neural network, (1): the weights and thresholds in the BP neural network are compiled into particles and initialized; (2): the BP neural network is used to The difference between the training result and the expected value is used as the fitness function to calculate the fitness value of each particle; (3): compare the fitness value of each particle with the individual optimal particle to obtain the global optimal particle; (4): Update the particle speed and position; (5): judge whether the obtained global optimal particle meets the termination condition, if it meets the set termination condition, stop the operation and output the optimal weight threshold, otherwise return to step (2) iterative operation; (6) : Use the optimal weight threshold obtained in (5) to connect the input layer, hidden layer and output layer of the BP neural network, and obtain the wind power prediction result according to the BP neural network result prediction. The convergence speed is faster, the precision is high, and it is not easy to fall into local extremum.

Description

A Wind Power Prediction Method Based on Improved Particle Swarm Optimization BP Neural Network

technical field

The invention relates to new energy predictive control, in particular to a wind power prediction method based on improved particle swarm optimization BP neural network.

Background technique

As a renewable energy source, wind energy has the advantages of no pollution, no need for fuel, and no occupation of cultivated land. It has attracted more and more people's attention, and its reserves are also very large. According to relevant statistics, the annual radiation energy from outer space is 1.5 *10 ¹⁸ kWh, of which 2.5%, or 3.8*10 ¹⁶ kWh of energy is absorbed by the atmosphere, generating approximately 4.3*l0 ^l2 kWh of wind energy. my country's actual exploitable wind energy resource reserves are 253 million kW, accounting for 7.8% of the total wind energy resources at 10m altitude.

According to the semi-annual report issued by the World Wind Energy Association, the global wind power industry newly installed 63,013MW in 2015, with an annual market growth rate of 22%. The US market reached 8.6GW for the year, and Germany exceeded expectations with 6GW of new additions, including 2.3GW of offshore projects. By the end of 2015, the cumulative installed capacity of wind power in the world reached 432,419MW, with a cumulative annual growth rate of 17%. At the same time, as of 2015, as the largest country in wind power installed capacity, my country's cumulative installed wind power capacity reached 145.1GW, and the annual wind power generation capacity reached 186.3 billion kWh, accounting for 3.3% of the total power generation.

With the maturity of wind power technology and the continuous reduction of its operating costs, wind power has shown a strong growth momentum, and its proportion in the electricity market demand is also increasing, and the negative impact of wind power on the grid will also follow. highlight. Due to the strong intermittence, randomness and uncontrollability of wind energy, the wind power also has large fluctuations, which will seriously affect the safe and stable operation of the power system. Prediction of wind power generation can turn the unknown into the known, effectively reduce the impact of such power generation systems connected to the grid, improve the utilization rate of renewable energy, and obtain greater economic benefits. Therefore, it is a work of far-reaching significance to carry out research on wind power forecasting and gradually improve its forecasting accuracy.

BP neural network (Back Propagation Neural Network, BPNN) is widely used in power generation prediction of wind power generation systems due to its good nonlinear mapping ability and self-adaptive learning ability. The two aspects of running speed and precision are not ideal enough.

Contents of the invention

The purpose of the present invention is to solve the above problems, to provide a wind power prediction method based on improved particle swarm optimization BP neural network, the prediction result has better nonlinear fitting ability, faster convergence speed and higher prediction accuracy , and it is not easy to fall into a local optimum.

In order to achieve the above object, the present invention adopts the following technical solutions:

A wind power prediction method based on improved particle swarm optimization BP neural network, comprising the following steps:

Step (1): Code the weights and thresholds in the BP neural network into particles and initialize them;

Step (2): Use the difference between the wind power prediction output and the wind power expected value obtained from the training set of BP neural network training as the fitness function to calculate the fitness value of each particle;

Step (3): Compare the fitness value of each particle with the individual optimal particle to obtain the global optimal particle;

Step (4): update particle velocity and position according to preset rules;

Step (5): Judging whether the obtained global optimal particle satisfies the set termination condition, if the set termination condition is met, then stop the operation and output the optimal weight threshold, otherwise return to step (2) iterative operation;

Step (6): Use the optimal weight threshold obtained in step (5) to connect the input layer, hidden layer and output layer of the BP neural network, and obtain the wind power prediction result according to the BP neural network result prediction.

The initialization content in the step (1) includes particle population size, initial position and initial velocity.

In the described step (2), the fitness function is:

In the formula: O _k is the expected wind power output of the kth node, and y _k is the predicted wind power output of the kth node.

In the step (3), if the fitness value of each particle is better than the individual optimum, then the particle is set as the individual optimum, and then the individual optimum is compared with the global optimum, and if it is better than the global optimum, then the The particle is set as the global optimum.

The formula according to which the particle speed is updated in the step (4) is as follows:

v _i (t+1)＝ωv _i (t)+c ₁ r ₁ (p _b (t)-x _i (t))+c ₂ r ₂ (p _g (t)-x _i (t))

In the formula: v _i (t) is the velocity of the i-th particle in the t-th iteration; x _i (y) is the position of the i-th particle in the t-th iteration; ω is the inertia weight; c ₁ , c ₂ are cognitive learning factors and social learning factors respectively; r ₁ and r ₂ are random numbers between [0, 1], p _g (t) is the individual optimal value in the t-th iteration; p _g (t ) is the global optimal value in the tth iteration.

The values of c ₁ and c ₂ are c ₁ =c ₂ =2.

The formula according to which the particle position is updated in the step (4) is as follows:

x _i (t+1)＝ _xi (t)+v _i (t+1)

In the formula: x _i (t+1) is the position of the i-th particle in the t+1 iteration.

where ω is defined as follows:

Where n is the total number of particles, α is a random number in the interval [0,1], f _t (X _i ) is the fitness value corresponding to the i-th particle at the t-th iteration, f _t (X _best ) is the t-th time The fitness value corresponding to the optimal particle during iteration.

In the step (6), the input layer nodes are determined by the input data, and the input layer can select the wind power values of the previous moment, the first two moments and the first three moments of the forecast moment.

The selection of the number of hidden layer nodes in the step (6) is determined by the following formula:

Among them, m is the number of nodes in the input layer, q is the number of nodes in the output layer, a is a trial and error number, and the value range is 1-10.

The invention combines the improved particle swarm algorithm and BP neural network to predict the output power of the wind farm, and the improved particle swarm algorithm judges the speed and step size of the next particle search through the fitness value obtained by each iteration, thereby improving the efficiency and Accuracy, to avoid it falling into local optimum, optimize the weight and threshold of the neural network with the optimization ability of the improved particle swarm optimization algorithm, and achieve better prediction results than the BP neural network.

Beneficial effects of the present invention:

The present invention uses the improved particle swarm algorithm that can self-adaptively adjust the weight to replace the error function according to the gradient descent in the BP neural network to adjust the weight threshold, which can quickly search for the optimal weight threshold, and the improved particle swarm algorithm can avoid its Trapped in a local extremum during the optimization process, the prediction result obtained by using the prediction algorithm of the present invention has better nonlinear fitting ability, faster convergence speed and higher prediction accuracy.

Description of drawings

Fig. 1 is the schematic diagram of BP neural network structure;

Fig. 2 is the flow chart of improved particle swarm optimization BP algorithm;

Figure 3 is the prediction of wind power by BP neural network;

Figure 4 is the particle swarm optimization algorithm to optimize the BP neural network to predict wind power;

Figure 5 is the improved particle swarm optimization algorithm to optimize the BP neural network to predict wind power;

Figure 6 is the absolute error comparison of the three models;

Fig. 7 is the convergence when the standard particle swarm optimization algorithm is applied to the BP network;

Figure 8 shows the convergence of the improved particle swarm optimization algorithm applied to the BP network.

detailed description

The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

Aiming at the problems existing in the prior art, the present invention proposes a wind power prediction method based on the improved particle swarm optimization BP neural network. Compared with the standard particle swarm optimization BP neural network prediction algorithm, the prediction result of the proposed optimized prediction model has Better nonlinear fitting ability, faster convergence speed and higher prediction accuracy, and not easy to fall into local optimum. MATLAB simulation verifies the effectiveness and practicability of the proposed algorithm and model of the present invention.

Artificial neural network is a simulation of biological nervous system based on bionics. Similar to the biological nervous system, there are a large number of neurons in the artificial neural network to form a complete neural network through complex connections. The input of each neuron corresponds to a single output, so as to perform distributed parallel information processing. At the same time, the neural network also has the ability to learn and adaptive capabilities. With the various needs of engineering, scholars have proposed a variety of neural network models, among which the backpropagation network, or BP network, is the most typical and widely used of various artificial network models that have been proposed. It has good Non-linear mapping ability and self-adaptive learning ability are very suitable for power generation prediction of wind power and other power generation systems.

As shown in Figure 1, the BP neural network generally consists of three layers, including: input layer, hidden layer, and output layer, where the input layer has m nodes, the hidden layer has n nodes, and the output layer has q nodes. ω _mn and ω _nq are the connection weights between the input layer and the hidden layer, and between the hidden layer and the output layer, respectively, b _n and b _q are the thresholds between the hidden layer and the output layer nodes.

Suppose a set of input vectors is X=(x ₁ ,x ₂ ,…x _m ), the output vector of the output layer is Y=(y ₁ ,y ₂ ,…y _q ), and the expected output vector is O=(o ₁ ,o ₂ ,... o _q ). The workflow of BP neural network is as follows:

First, train the neural network. The training process is divided into two processes: information forward transmission and error back propagation. For information forward transmission, there are:

The nth node input of the hidden layer is:

Its output expression is:

Where f(h _n ) is the activation function of the hidden layer (the activation function is set as a Sigmoid function here) to describe the relationship between layers, and the input of the qth node of the output layer:

Its output expression is:

When the actual output of the neural network is inconsistent with the expected output or the error is greater than the set value, the BP neural network distributes the error backpropagation to all units in each layer, and corrects the weights and thresholds of each unit through the error signal of the units in each layer. The BP neural network uses the error function of gradient descent to adjust the weight threshold. For K training samples, the error function is as follows:

The correction amount of each weight threshold (D is collectively referred to as w _mn , w _nq , b _n , and b _q )

The training stops when the error output by the network decreases to the required range or reaches the preset learning times.

It is worth noting that for the BP neural network with a certain model, the number of nodes in the input layer and output layer has been determined, while the initial number of nodes in the hidden layer neurons needs to be determined according to the following formula based on experience, and then gradually increase by using the "trial and error method". Enlarge and reduce the number of hidden layer nodes to minimize the error of the neural network, finally determined.

The value of a ranges from 1 to 10.

Particle Swarm Optimization (PSO) was proposed by Dr. Eberhart and Dr. Kennedy in 1995. Its basic idea is to start from a random solution and find the optimal solution through collaboration and information sharing between individuals and groups. . When using PSO to solve optimization problems, the algorithm initializes a group of random particles, each particle has its own speed and position, and the particles update their own speed and position through each iteration, and each particle is updated separately according to two simple rules Its velocity and position:

1) Move according to the best plan found by the particle itself, and thus obtain the individual optimal value;

2) Follow the particle with the best performance among all the particles, so as to obtain the global optimal value. Calculate the fitness value of the particle according to the particle position, so as to measure the quality of the particle. Through this method, all the particles reach the global optimal solution.

The speed and position update formula of the i-th particle is as follows:

v _i (t+1)＝ωv _i (t)+c ₁ r ₁ (p _b (t)-x _i (t))+c ₂ r ₂ (p _g (t)-x _i (t))

x _i (t+1)＝ _xi (t)+v _i (t+1)

In the formula: v _i (t) is the speed of the i-th particle in the t-th iteration; ω is the inertia weight; c ₁ and c ₂ are cognitive learning factors and social learning factors respectively, and generally take c ₁ =c ₂ = ₂ ; r ₁ and r ₂ are random numbers between [0,1]; p _g (t) is the individual optimal value in the t-th iteration; p _g (t) is the global optimal value in the t-th iteration The optimal value.

Among them, the size of ω determines whether the particle is more inclined to global search or local search, c ₁ determines the degree of particle dependence on itself, and c ₂ determines the degree of particle dependence on the global.

The modified particle swarm optimization algorithm (Modified Particle Swarm Optimization, MPSO) proposed by the present invention: the main characteristics of the particle swarm algorithm are simple principle, few parameters, fast convergence speed, and easy implementation, but it is difficult to solve it if it is easy to fall into a local optimal solution. Jumping out and other defects, the current improvement of the particle swarm algorithm mainly focuses on making the inertia weight ω linearly decrease with the number of iterations. This method makes the following improvements to the particle swarm algorithm according to the defects of the particle swarm:

Among them, ω _max and ω _min are the maximum and minimum values of the inertial weight respectively, t is the current iteration number, and T is the maximum iteration number.

This improvement makes the particle search step size gradually decrease as the number of iterations increases, which can prevent the particles from falling into the local extremum to a certain extent, but once trapped, it is still difficult to jump out, and the change of ω in this algorithm is only linear with the number of iterations Correlation, cannot better adapt to optimization problems with complex nonlinear changing characteristics. For this reason, the present invention proposes the following particle swarm optimization algorithm adjusted according to the fitness value:

Where n is the total number of particles, f _t (X _i ) is the fitness value corresponding to the i-th particle at the t-th iteration, and f _t (X _best ) is the fitness value corresponding to the optimal particle at the t-th iteration.

In the formula, K _t is used to judge the degree of dispersion of the fitness value of the i-th particle in the t-th iteration. The larger the K _t is, the more divergent the particle is in this iteration, and the smaller the particle is, the closer it is to the particle in this iteration. the best particles. ω _t is based on e, and the ratio of K _t to K _t-1 is used as a reference. According to the nature of the exponential function, ω can be limited to the [0,1] interval. In order to make the particle swarm optimization algorithm more efficient, in the global search The step size is larger, adding in the formula Where α is a random number in the [0,1] interval, and the range of ω is widened to [0,1.5] interval.

For ω _t , if K _t >K _t-1 , that is, K _t /K _t-1 >1, it means that the t-th iteration diverges relative to the t-1-th iteration, and the larger the ratio of K _t /K _t-1 at this time The smaller ω _t is, that is, the smaller the search step size, the more the particle swarm tends to search locally; if K _t <K _t-1 , that is, K _t /K _t-1 <1, it means that the tth iteration is relatively 1 convergence, the smaller the ratio of K _t /K _t-1 at this time, the larger ω _t is, that is, the larger the search step, the more the particle swarm tends to global search. This makes ω not blindly decrease as the number of iterations increases, but is determined according to the specific function fitness value. The particle swarm optimization algorithm improved in this way accelerates the convergence speed and is not easy to fall into local extremum.

The BP neural network adopts the error function of gradient descent to adjust the connection weight ω and the threshold b, so there are defects such as slow convergence speed in the learning process, network training is easy to fall into local extremum, network structure is difficult to determine, and network generalization ability is difficult to guarantee. . In response to the above problems, this paper combines the BP neural network with the improved particle swarm algorithm, and uses the particle swarm optimization algorithm to replace the gradient descent method in the BP algorithm to train the parameters of the neural network to improve the performance of the BP algorithm and make it difficult to fall into the local minimum. Enhance the generalization ability to achieve the purpose of optimizing the network.

As shown in Figure 2, a wind power prediction method based on improved particle swarm optimization BP neural network, the specific steps are as follows:

Step (1): Code the weights and thresholds in the BP neural network into particles and initialize them. The initialization content includes particle population size, initial position and initial velocity;

Step (2): Using the difference between the training result of BP neural network and the expected value as the fitness function, calculate the fitness value of each particle;

Step (3): Compare the fitness value of each particle with the individual optimal particle, if it is better than the individual optimal, set the particle as the individual optimal, and then compare the individual optimal with the global optimal, if If it is better than the global optimum, then the particle is set as the global optimum;

Step (4): Update the particle velocity and position according to the rules;

Step (5): Judging whether the obtained global optimal particle satisfies the termination condition, if the set condition is met, stop the operation and output the optimization result, otherwise return to step (2) iterative operation;

Step (6): Connect the input layer, hidden layer and output layer of the BP neural network using the optimal weight threshold obtained above, and obtain the result according to the prediction of the BP neural network result.

The following is proved by simulation:

In order to prove the effectiveness of the proposed algorithm, this paper takes Guodian Jinan Changqing Wind Farm as an example, uses MATLAB language to write the algorithm program, and constructs three prediction models: BP neural network prediction model, particle swarm optimization optimization BP neural network prediction model ( PSO-BP), an improved particle swarm optimization algorithm to optimize the BP neural network prediction model (MPSO-BP), to predict and simulate the output power of wind power within a period of time. The power sample is the data collected every 15 minutes for the next 72 hours from 00:00 on January 1, 2015, by Guodian Jinan Changqing Wind Farm. The first 260 data are used to train the network, and the last 28 data are used for prediction, that is, 260 sets of training samples and 28 sets of test samples K. After experimental debugging, the hidden layer is finally determined to be 8 nodes. Among them, the maximum training times of BP network is set to 2000, the training accuracy is 0.002, and the learning rate is 0.01. The performance evaluation of the prediction results uses the mean absolute error MAPE, the root mean square error MSE and the correlation R ² .

Where ESS is the regression sum of squares and TSS is the overall sum of squares.

Three models of BP, PSO-BP and MPSO-BP are simulated respectively, and the prediction results of the three models are shown in Fig. 3, Fig. 4 and Fig. 5 respectively. Figure 6 is a comparison of the absolute errors of the three models. Figure 7 and Figure 8 show the convergence of particle swarm optimization and improved particle swarm optimization when applied to BP network respectively. The results of each of the three models running 20 times are shown in Table 1.

Table 1 Analysis of the results of the three models running 20 times

Fig. 3 is the prediction result of wind power by BP neural network prediction method, the average absolute error MAPE is 0.0507, the root mean square error MSE is 0.0692, and the fitting degree R2 between the prediction curve and the expected output curve is 0.8980; Fig. 4 is the result of prediction using the particle swarm algorithm to optimize the BP neural network method. The average absolute error MAPE is 0.0327, the root mean square error MSE is 0.0469, and the fitting degree R2 between the prediction curve and the expected output curve is 0.9655; 5 is the result of using the improved particle swarm algorithm to optimize the BP neural network method for prediction. The average absolute error MAPE is calculated to be 0.0248, the root mean square error MSE is 0.0346, and the fitting degree R2 of the predicted curve and the expected output curve is 0.9809 . Apparently, the result of optimized prediction using the improved particle swarm algorithm to optimize BP neural network is better than the results of particle swarm optimization optimized BP neural network and single BP neural network prediction. It can be seen from Figure 6 that the result of using the improved particle swarm optimization algorithm to optimize the BP neural network for optimization prediction is smaller and more stable than the other two errors. The comparison of Figure 7 and Figure 8 shows that the improved particle swarm optimization algorithm proposed in this paper has a faster convergence speed than the basic particle swarm optimization algorithm. It can also be seen from Table 1 that the results of using the improved particle swarm optimization algorithm to optimize the BP neural network for optimization prediction are more stable than the other two.

Although the specific implementation of the present invention has been described above in conjunction with the accompanying drawings, it does not limit the protection scope of the present invention. Those skilled in the art should understand that on the basis of the technical solution of the present invention, those skilled in the art do not need to pay creative work Various modifications or variations that can be made are still within the protection scope of the present invention.

Claims (10)

1. a wind power prediction method based on improved particle swarm optimization BP neural network, is characterized in that, comprises the steps: Step (1): Code the weights and thresholds in the BP neural network into particles and initialize them; Step (2): Use the difference between the wind power prediction output and the wind power expected value obtained from the training set of BP neural network training as the fitness function to calculate the fitness value of each particle; Step (3): Compare the fitness value of each particle with the individual optimal particle to obtain the global optimal particle; Step (4): update particle velocity and position according to preset rules; Step (5): Judging whether the obtained global optimal particle satisfies the set termination condition, if the set termination condition is met, then stop the operation and output the optimal weight threshold, otherwise return to step (2) iterative operation; Step (6): Use the optimal weight threshold obtained in step (5) to connect the input layer, hidden layer and output layer of the BP neural network, and obtain the wind power prediction result according to the BP neural network result prediction. 2. a kind of wind power prediction method based on improved particle swarm optimization BP neural network as claimed in claim 1, is characterized in that, in the described step (1), initialization content comprises particle population size, initial position and initial velocity. 3. a kind of wind power prediction method based on improved particle swarm optimization BP neural network as claimed in claim 1, is characterized in that, in described step (2), fitness function is: ${\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; K</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}\mathrm{<span\; class="notranslate">\; a</span>}\mathrm{<span\; class="notranslate">\; b</span>}\mathrm{<span\; class="notranslate">\; the\; s</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; o</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; )</span>$ In the formula: O _k is the expected output of the kth node, and _yk is the predicted output of the kth node. 4. a kind of wind power prediction method based on improved particle swarm optimization BP neural network as claimed in claim 1, is characterized in that, in described step (3), if the fitness value of each particle is better than individual optimal then will The particle is set as the individual optimum, and then the individual optimum is compared with the global optimum, and if it is better than the global optimum, the particle is set as the global optimum. 5. a kind of wind power prediction method based on improved particle swarm optimization BP neural network as claimed in claim 1, is characterized in that, the formula of particle speed update basis in the described step (4) is as follows: v _i (t+1)＝ωv _i (t)+c ₁ r ₁ (p _b (t)-x _i (t))+c ₂ r ₂ (p _g (t)-x _i (t)) In the formula: v _i (t) is the velocity of the i-th particle in the t-th iteration; x _i (t) is the position of the i-th particle in the t-th iteration; ω is the inertia weight; c ₁ , c ₂ are cognitive learning factors and social learning factors respectively; r ₁ and r ₂ are random numbers between [0, 1], p _g (t) is the individual optimal value in the t-th iteration; p _g (t ) is the global optimal value in the tth iteration. 6 . A wind power prediction method based on improved particle swarm optimization BP neural network according to claim 5 , wherein the values of c ₁ and c ₂ are c ₁ =c ₂ =2. 7. a kind of wind power prediction method based on improved particle swarm optimization BP neural network as claimed in claim 5, is characterized in that, the formula according to which particle position is updated in the described step (4) is as follows: x _i (t+1)＝ _xi (t)+v _i (t+1) In the formula: x _i (t+1) is the position of the i-th particle in the t+1 iteration. 8. a kind of wind power prediction method based on improved particle swarm optimization BP neural network as claimed in claim 5, is characterized in that, wherein ω is defined as follows: ${\mathrm{<span\; class="notranslate">\; \&omega;</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}}\mathrm{<span\; class="notranslate">\; \&alpha;</span>}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; e</span>}}^{<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; K</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; K</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}}$ ${\mathrm{<span\; class="notranslate">\; K</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; no</span>}}\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; no</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; b</span>}\mathrm{<span\; class="notranslate">\; e</span>}\mathrm{<span\; class="notranslate">\; the\; s</span>}\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}$ Where n is the total number of particles, α is a random number in the interval [0,1], f _t (X _i ) is the fitness value corresponding to the i-th particle at the t-th iteration, f _t (X _best ) is the t-th time The fitness value corresponding to the optimal particle during iteration. 9. a kind of wind power prediction method based on improved particle swarm optimization BP neural network as claimed in claim 1, is characterized in that, in the described step (6), the input layer node is determined by the input data. 10. a kind of wind power prediction method based on improved particle swarm optimization BP neural network as claimed in claim 1, is characterized in that, the selection of hidden layer node number is determined by following formula in described step (6): $\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; =</span>\sqrt{\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; q</span>}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; a</span>}$ Among them, m is the number of nodes in the input layer, q is the number of nodes in the output layer, a is a trial and error number, and the value range is 1-10.