Wind power prediction method and system and computer equipment

Abstract

The invention belongs to the technical field of wind power generation, and particularly relates to a wind power prediction method and system and computer equipment. According to the method, the index data required for predicting the output power of the wind power plant at the prediction moment is acquired, and the index data is input into the trained machine learning model, so that the output power of the wind power plant at the prediction moment is obtained; wherein the initial values of the parameters of the machine learning model are obtained by optimizing an improved particle swarm algorithm, and the improvement of the improved particle swarm algorithm comprises the following steps: along with the increase of the number of iterations, an individual learning factor used for updating the individual speed is gradually increased, and a social learning factor is gradually reduced; wherein the absolute value of the difference between the initial value of the individual learning factor and the initial value of the social learning factor is greater than a set difference threshold, and the initial value of the individual learning factor is smaller than the initial value of the social learning factor. The problem that an existing wind power prediction method is low in prediction result precision is solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of wind power generation, and in particular relates to a wind power prediction method, system and computer equipment. Background Art

[0002] With the development of economic globalization, the development of new energy has become an inevitable choice for sustainable development in all countries. Compared with new energy sources such as solar energy and biomass energy, which have obvious limitations, wind power generation has the advantages of less pollution and renewable nature. It has the potential to replace primary energy in the short term and has great commercial development prospects. However, due to the randomness, intermittency and strong nonlinearity of wind power, it seriously affects the safety of the power system. In order to ensure the smooth operation of the power system, cope with the uncertain impact of wind power fluctuations on the power grid system, and avoid the phenomenon of "wind abandonment" caused by the difficulty in absorbing wind power resources, it is necessary to carry out accurate prediction research on wind power to provide an accurate scientific basis for power trading and provide strong support for power grid dispatching.

[0003] The Chinese invention patent application with the existing application publication number CN114386718A discloses a short-term prediction algorithm for wind farm output power combined with a particle swarm neural network. The method predicts wind power by analyzing the uncertainty, conditionality and regionality of wind power and the factors affecting wind power generation. Although the method optimizes the BP neural network through the particle swarm algorithm and overcomes the problem of slow convergence of the neural network algorithm, the learning factor of the particle swarm algorithm used in the dynamic prediction model is a fixed value, which makes it easy to fall into the local optimum when searching for the optimal solution, and too many particles stay in the search area of ​​their own area, so that the prediction result is low in accuracy. Summary of the invention

[0004] The object of the present invention is to provide a wind power prediction method, system and computer equipment to solve the problem that the learning factor of the particle swarm algorithm used in the existing wind power prediction method is a fixed value, resulting in low prediction result accuracy.

[0005] In order to solve the above technical problems, the present invention provides a wind power prediction method, which comprises:

[0006] Obtain the index data required for predicting the output power of the wind farm at the predicted time, input the index data into the trained machine learning model, and obtain the output power of the wind farm at the predicted time; the initial values ​​of the parameters of the machine learning model are obtained by optimizing using an improved particle swarm algorithm, which is an improvement on the traditional particle swarm algorithm, and the improvement includes:

[0007] As the number of iterations increases, the individual learning factor used in updating the individual speed gradually increases, and the social learning factor used in updating the individual speed gradually decreases; the absolute value of the difference between the initial value of the individual learning factor and the initial value of the social learning factor is greater than the set difference threshold, and the initial value of the individual learning factor is less than the initial value of the social learning factor.

[0008] Furthermore, the individual learning factor and / or social learning factor changes nonlinearly.

[0009] Furthermore, the iterative update formula of the individual learning factor is:

[0010]

[0011] in, is the individual learning factor, a is a constant between 0 and 1, sinh(x) is the hyperbolic sine function, sin(x) is the sine function, and k is the number of iterations.

[0012] Furthermore, the iterative update formula of the social learning factor is:

[0013]

[0014] in, is the social learning factor, b is a constant between 0 and 1, tanh(x) is the hyperbolic tangent function, sin(x) is the sine function, and k is the number of iterations.

[0015] Furthermore, the improvement of the improved particle swarm algorithm also includes: the inertia weight used in updating the individual speed changes continuously with the increase of the number of iterations, and the iterative update formula of the inertia weight is:

[0016]

[0017] in, is the inertia weight, ω min is the minimum value of the inertia weight, ω max is the maximum value of the inertia weight, ω range =ω max -ω min , k is the number of iterations, ξ and λ are both positive constants.

[0018] Furthermore, the improvement of the improved particle swarm algorithm also includes: obtaining the initial population of the improved particle swarm algorithm by using chaos mapping, specifically:

[0019]

[0020] Among them, α, β, δ, ζ, and p are control parameters, and xi is the ith chaos number.

[0021] Furthermore, the index data includes a data sequence consisting of at least one of environmental indexes, wind farm operation indexes, daily type values ​​and power consumption per unit time in a period of time before the prediction moment.

[0022] Furthermore, the environmental indicators include at least one of: air pressure, topography, light radiation intensity, maximum temperature, minimum temperature, average temperature, temperature in the turbine cabin, wind level, humidity and meteorological conditions.

[0023] Furthermore, the wind farm operation indexes include: blade pitch angles, blade depression angles and yaw angle of the nacelle.

[0024] Furthermore, the power data in the training data used to train the machine learning model is preprocessed data, and the preprocessing specifically includes:

[0025] In the case where it is detected that a certain data in the power data is abnormal or missing, if the interval of the moments corresponding to the power data is less than the first set interval threshold, a linear interpolation method is used for processing to fill or replace the data accordingly; if the interval of the moments corresponding to the power data is greater than or equal to the first set interval threshold and less than the second set interval threshold, a horizontal processing method is used for processing to fill or replace the data accordingly; if the interval of the moments corresponding to the power data is greater than or equal to the second set interval threshold, a vertical processing method is used for processing to fill or replace the data accordingly; wherein the first set interval threshold is less than the second set interval threshold;

[0026] The horizontal processing method is: taking the average value of the power at adjacent moments corresponding to the data as the power at that moment; the vertical processing method is: obtaining the power at that moment based on the power data of the same period as the data.

[0027] Furthermore, the machine learning model is a BP neural network.

[0028] In order to solve the above technical problem, the present invention further provides a computer device, including a processor, wherein the processor executes a computer program to implement the steps of the above wind power prediction method.

[0029] In order to solve the above technical problems, the present invention further provides a wind power prediction system, wherein the prediction server comprises a processor, and the processor executes a computer program to implement the steps of the above wind power prediction method.

[0030] Furthermore, the wind power prediction system also includes a big data server, which is used to install a wind farm database system to store real-time data and historical data, and forward the data to the prediction server.

[0031] Its beneficial effects are: the present invention is an improved invention, which uses a trained machine learning model to predict the wind power of a wind farm. Considering that the initial values ​​of the parameters of the machine learning model affect the accuracy and training speed of the machine learning model, the initial values ​​of the parameters of the machine learning model are optimized. Specifically, when optimizing the machine learning model, the traditional particle swarm algorithm is improved. The specific improvements include making the learning factor no longer a fixed value, but a value that changes with the number of iterations. Specifically, the individual learning factor is set to a relatively small value, and the social learning factor is set to a relatively large value. Moreover, as the number of iterations increases, the individual learning factor used in updating the individual speed gradually increases, and the social learning factor used in updating the individual speed gradually decreases, so that it has a strong local search ability in the early stage. As the number of iterations increases, its global search ability is enhanced in the later stage, avoiding the problem of easily falling into the local optimum when searching for the optimal solution and too many particles staying in the search range of their own area, so that the initial values ​​of the parameters of the machine learning model obtained by optimizing with the improved particle swarm algorithm are more accurate, and the accuracy of the machine learning model in predicting the output power of the wind farm at the prediction time is improved, and the problem of low accuracy of the prediction result caused by the fixed learning factor of the particle swarm algorithm used in the existing wind power prediction method is solved. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 is an algorithm flow chart of the IPSO-BP prediction model of an embodiment of the present invention;

[0033] Figure 2 This is a chaotic value effect diagram of the optimized chaotic mapping according to an embodiment of the present invention;

[0034] Figure 3 is a frequency distribution diagram of an optimized chaotic map according to an embodiment of the present invention;

[0035] Figure 4 is a flow chart of the IPSO algorithm of an embodiment of the present invention;

[0036] Figure 5 This is a weight optimization comparison diagram of an embodiment of the present invention;

[0037] Figure 6 is a comparison chart of individual and social factors changing over time in an embodiment of the present invention;

[0038] Figure 7 is a flow chart of a BP neural network algorithm according to an embodiment of the present invention;

[0039] Figure 8 It is a partial wind power data sample of an embodiment of the present invention;

[0040] Fig. 9 It is a comparison chart of the predicted value and the target value before and after optimization of the embodiment of the present invention. DETAILED DESCRIPTION

[0041] This embodiment obtains the index data required for predicting the output power of the wind farm at the prediction time, inputs the index data into the trained machine learning model, and obtains the output power of the wind farm at the prediction time; wherein, the initial value of the parameter of the machine learning model is obtained by optimizing using the improved particle swarm algorithm, and the improved particle swarm algorithm is obtained by improving the traditional particle swarm algorithm, and the improvement includes: as the number of iterations increases, the individual learning factor used in updating the individual speed gradually increases, and the social learning factor used in updating the individual speed gradually decreases; wherein the absolute value of the difference between the initial value of the individual learning factor and the initial value of the social learning factor is greater than the set difference threshold, and the initial value of the individual learning factor is less than the initial value of the social learning factor. This solves the problem that the learning factor of the particle swarm algorithm used in the existing wind power prediction method is a fixed value, resulting in low prediction result accuracy.

[0042] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention is further described in detail below with reference to the accompanying drawings and embodiments.

[0043] Method Example:

[0044] A wind power prediction method of the present embodiment optimizes the inertia weight, individual learning factor and group learning factor of the improved particle swarm algorithm by optimizing the update iteration mode, so that the individual learning factor and group learning factor of the particle swarm algorithm are no longer fixed values, but change with the increase of the number of iterations, so as to optimize the optimization speed and accuracy of the particle swarm algorithm; wherein the improvement of the traditional particle swarm algorithm mainly lies in optimizing the initial population, learning factor and inertia weight of the particle swarm algorithm, so as to obtain the improved particle swarm algorithm; then the improved particle swarm algorithm is used to optimize and obtain the initial values ​​of the parameters (including weights and thresholds) of the machine learning model, and the machine learning model is trained and tested; then the indicator data required for predicting the output power of the wind farm at the prediction time is input into the trained machine learning model to complete the wind power prediction of the wind farm at the prediction time.

[0045] Among them, the machine learning model selected in this embodiment is the BP neural network. As other implementation methods, they can be selected according to actual needs, such as the CNN neural network.

[0046] like Figure 1As shown in Figure 2, the specific process of wind power forecasting includes:

[0047] Step 1: Use the improved particle swarm algorithm to optimize and obtain the initial values ​​of the weights and thresholds of the BP neural network.

[0048] Particle Swarm Optimization (PSO) simulates bird hunting behavior by designing a massless particle. The particle (also called an individual) has two properties: speed and position. In each iteration, the particle is updated by tracking two optimal values. After searching for these two optimal values, the particle updates its speed and position by using formulas (1) and (2). This process is repeated until the maximum number of iterations is reached or the termination condition is met, and the calculation stops.

[0049]

[0050] Among them, ω is the inertia weight, C1 is the individual learning factor, and C2 is the social learning factor, which respectively adjust the step size of the particle to its own best position and the global best position; rand1 and rand2 are two random numbers in the range of [0,1]. represents the d-th dimension component of the flight velocity vector of particle i at the k-th iteration, Represents the d-th dimension component of the position vector of particle i at the k-th iteration. represents the individual extreme value of particle i in the kth iteration, Represents the population extreme value p of the entire particle population in the first k iterations best The d-th dimension position component.

[0051] Since the inertia weight ω, individual learning factor C1 and social learning factor C2 of the standard PSO algorithm are all fixed, a C1 value that is too large will cause too many particles to stay in their own area to search, and a C2 value that is too large will cause the particles to converge too early and fall into the local optimum, resulting in low prediction accuracy. Therefore, the present invention optimizes the initial values ​​of the parameters of the BP neural network by improving the particle swarm algorithm, thereby avoiding the problem of low prediction accuracy caused by the fixed inertia weight coefficient and learning factor of the standard PSO algorithm.

[0052] The improvement of the particle swarm algorithm involves three specific aspects. The first is to use chaos mapping to obtain the initial population of the particle swarm algorithm. The second is to improve the inertia weight of the algorithm. The third is to design the individual learning factor of the algorithm to gradually increase and the social learning factor to gradually decrease. The following are introduced respectively:

[0053] 1) Use chaos mapping to obtain the initial population of the particle swarm algorithm.

[0054] This embodiment uses IPWLCM (Improved Piecewise Linear Chaotic Map) to obtain the initialization population; the chaotic map allows different function forms to be used in different intervals, that is, the mapping behavior of different intervals can be adjusted according to needs. IPWLCM has significant advantages in flexibility, parameter diversity, easy implementation and control of chaotic behavior, such as Figure 2 The optimized chaotic mapping chaos value effect diagram shown, the x-axis represents the range, and the y-axis represents the chaos value; Figure 3 The frequency distribution diagram of the optimized chaotic mapping shown in the figure has the x-axis representing the chaotic value and the y-axis representing the frequency. The specific expression is as follows:

[0055]

[0056] Among them, α, β, δ, ζ, and p are control parameters, and x i is the ith chaos number.

[0057] 2) Improve the inertia weight of particle swarm algorithm.

[0058] The inertia weight strategy of exponential decay mode is adopted, so that the inertia weight value gradually decreases as the iteration procedure progresses, and this decrease presents nonlinear characteristics, showing a rapid decrease in the early stage and a slow decrease in the later stage. The early stage focuses on extensive exploration, while the later stage focuses on detailed mining, which helps to enhance the performance of the algorithm in terms of convergence speed and accuracy. Figure 5 The weight optimization comparison chart shown in the figure shows that Original represents the change of the inertia weight of the traditional particle swarm algorithm, and Transformed represents the change of the inertia weight of the improved particle swarm algorithm. The specific adjustment strategies are as follows:

[0059]

[0060] Among them, ω(t) represents the inertia weight, k represents the number of iterations, and k max Represents the preset maximum number of iterations, λ is a positive constant that determines the speed at which ω(t) decreases.

[0061] In order to prevent the algorithm from exponential decay in the early stage of training, which causes the learning rate or inertia weight to drop too quickly, and makes the algorithm converge to an undesirable solution too early in the exploration phase, ω(t) is mapped to the input of the Sigmoid function, and a min to max Smooth transition function. It is necessary to change the range of ω(t) from [ω min ,ω max ] is converted to the range of [0,1], and then the converted value is used as the input of the Sigmoid function. Note the basic form of the Sigmoid function:

[0062]

[0063] Among them, x is the input.

[0064] To ensure that the output curve is steep enough, the converted value is multiplied by a larger number, and the optimized formula is:

[0065]

[0066] in,

[0067]

[0068] ω range =ω max -ω min (8)

[0069] Then from formulas (4) to (8), we can get:

[0070]

[0071] in, is the inertia weight, ω min is the minimum value of the inertia weight, ω max is the maximum value of the inertia weight, and ξ is a positive constant.

[0072] 3) Improve the individual learning factor and social learning factor of particle swarm algorithm.

[0073] By design The nonlinear growth and The nonlinear reduction strategy of makes the algorithm have stronger local search ability in the early stage, and enhances its global search ability in the later stage. Non-linear growth, It decreases nonlinearly. As other implementation methods, the change mode can be selected according to actual needs, such as It increases linearly, Decreases linearly; Figure 6 The comparison chart of the value of individual and social factors over time is as follows:

[0074]

[0075] in, is the individual learning factor, is the social learning factor, a and b are constants between 0 and 1, which are used to control the degree of change of the function; sinh(x) represents the hyperbolic sine function, tanh(x) represents the hyperbolic tangent function, and k represents the number of iterations.

[0076] At the beginning of the algorithm, set to a lower value (i.e. C1), and Set to a higher value (i.e. C2) to enhance the local exploration ability of the algorithm. As the number of iterations increases, It increases nonlinearly. It decreases nonlinearly in order to enhance the global development ability of the algorithm.

[0077] According to the above three improvements, the formula for updating particle speed and position of the improved particle swarm algorithm is:

[0078]

[0079] in, is the d-th dimension component of the flight velocity vector of particle i at the k-th iteration, Represents the d-th dimension component of the position vector of particle i at the k-th iteration.

[0080] In summary, if Figure 4 As shown, the specific process of using the improved particle swarm algorithm to optimize the initial values ​​of the weights and thresholds of the BP neural network includes the following steps:

[0081] Step S1: Use IPWLCM to initialize the population, set the maximum number of iterations, the number of independent variables, the population size, the learning factor, the inertia weight coefficient, set the maximum speed range, and randomly initialize the speed and position of the population.

[0082] Step S2: Obtain the fitness value of each particle according to the current particle position.

[0083] Step S3: Update the optimal individual particle p according to the fitness value in step S2 id and the optimal position p gd , where p id The dth dimension representing the individual extreme value of the ith variable, p gd represents the dth dimension of the global optimal solution.

[0084] Step S4: Update the speed and position of the population particles according to formulas (12) and (13) until a new population is generated.

[0085] Step S5: Determine whether the current value is better than the historical value. If so, update the current value to the historical value; otherwise, do not make any changes.

[0086] Step S6: Determine whether the maximum number of iterations has been reached or the difference between generations has reached the set training threshold. If so, stop iterating and output the optimal solution. Otherwise, return to step S2.

[0087] The training threshold in this embodiment is set to 0.0001, and the number of iterations is set to 200. Other implementations can be set according to actual needs.

[0088] Step 2: Construct the IPWLCM-IPSO-BP (Improved Piecewise Linear Chaotic Map-Improved Particle Swarm Optimization-Back Propagation) prediction model. Figure 7 As shown, specifically including:

[0089] ① BP neural network initialization, determine the model network structure, including network structure parameters such as input layer, hidden layer, and output layer.

[0090] ② Initialize the velocity, position, weight and threshold of each particle in the IPWLCM-IPSO-BP model.

[0091] ③ Calculate the fitness value of each particle, and adjust and record the optimal solution of the particle and the optimal solution of the population. The calculation formula is:

[0092]

[0093] Among them, n represents the number of samples; q represents the number of neurons; represents the expected output value of the jth output node of the i-th sample, y ij Represents the actual output value of the jth output node of the i-th sample.

[0094] ④ Track individual extreme value p id Group extreme value p gd , if so, the current fitness value f i Better than p id , then update particle p id Otherwise, update the speed and position of the particle. If the current fitness value of all particles in the population is better than p gd , then update p g d.

[0095] ⑤ Update the speed and position of each particle according to formulas (12) and (13).

[0096] ⑥ Check the individual extreme value p of the search id and the group extreme value p gd Whether the termination condition is met, if so, the algorithm iteration is terminated and the optimal weight and threshold of the BP neural network are obtained, otherwise, continue to return to step ③.

[0097] ⑦ Determine whether the algorithm meets the termination condition. If the error result reaches the set training threshold or meets the maximum number of iterations, the algorithm operation stops. Otherwise, return to step ③ and continue the iterative operation. In this embodiment, the set training threshold is 0.0001 and the set number of iterations is 200. As other implementation methods, they can be set according to actual needs.

[0098] ⑧Build the IPWLCM-IPSO-BP prediction model, determine the BP neural network topology, train the model using the weights and thresholds obtained through the IPSO algorithm (i.e., improved particle swarm algorithm), and output the final prediction results.

[0099] The parameters of the BP neural network include: training function trainlm, output layer activation function purelin, hidden layer activation function tansig, and loss function MSE. In this embodiment, the learning rate is set in the range of [0.01, 0.8]. As other implementation methods, it can be set according to actual needs.

[0100] Step 3: Read data from the wind power database and preprocess the sample data (i.e., index data) for training the IPWLCM-IPSO-BP prediction model.

[0101] Obtain sample data for training and testing the IPWLCM-IPSO-BP prediction model, such as Figure 8 As shown, each input sample data includes a data sequence consisting of at least one of environmental indicators, wind farm operation indicators, daily type values ​​and power consumption per unit time in a period of time before the prediction moment; wherein the environmental indicators include: air pressure, topography, light radiation intensity, maximum temperature, minimum temperature, average temperature, temperature in the turbine cabin, wind force level, humidity and meteorological conditions; the wind farm operation indicators include: blade pitch, pitch angle and yaw angle of the cabin; the output is the predicted output power of the wind farm at the prediction moment.

[0102] Not all acquired data is suitable for sample prediction and needs to be preprocessed. The amount of data should be moderate, and data missing and data anomalies caused by power equipment failure or other human factors should be considered. Problem data cannot be used directly for wind power forecasting, otherwise it will lead to large deviations in forecasting, so it is necessary to fill or correct the problem data. The processing of abnormal or missing data includes the following:

[0103] When the interval of the moments corresponding to the acquired power data is less than the first set interval threshold, a linear interpolation method is used to process the missing or abnormal data;

[0104] When the interval of the time corresponding to the acquired power data is greater than or equal to the first set interval threshold and less than the second set interval threshold, the missing or abnormal data is processed by the horizontal processing method, that is, the average value of the power of the adjacent time at the moment is taken as the power at the moment;

[0105] When the interval corresponding to the moment of the acquired power data is greater than or equal to the second set interval threshold, the vertical processing method is used to process the missing or abnormal data, that is, the power at that moment is obtained based on the power data of the same period as the data; for example, the power at the same moment one day or two days before the moment is taken as the power at the moment.

[0106] The first set interval threshold is smaller than the second set interval threshold.

[0107] In addition, the horizontal processing method of this embodiment focuses on using data from adjacent time points on the same date to process abnormal or missing values; the vertical processing method focuses on using data from different dates but the same time points to process abnormal or missing values.

[0108] Normalize the sample data after filling or correction. Specifically:

[0109]

[0110] in, represents the normalized data, y is the input data, y max ,y min Respectively represent the maximum and minimum values ​​of the source data.

[0111] Step 4: Divide the data processed in step 3 into a training set and a test set, and train and test the constructed IPWLCM-IPSO-BP prediction model.

[0112] In order to verify the effectiveness of the IPWLCM-IPSO-BP prediction model, the root mean square error (RMSE), mean absolute percentage error (MAPE) and mean absolute value error (MAE) are introduced to evaluate the prediction results of the IPWLCM-IPSO-BP prediction model. Specifically:

[0113]

[0114] Where n is the size of the sample data; y i and They represent the actual value and predicted value of the prediction model respectively; RMSE is the sample standard deviation between the predicted value and the actual value; MAPE is the average of the ratios between the errors of all prediction points and the actual values; MAE is the average of the absolute values ​​of the deviations of all sample observations from the arithmetic mean. The smaller the value, the higher the prediction accuracy of the model.

[0115] Step 5: Obtain the index data required for predicting the output power of the wind farm at the predicted time and input the values ​​into the trained IPWLCM-IPSO-BP prediction model to predict the wind power. Fig. 9 The figure shows the comparison between the predicted value and the target value before and after optimization of this embodiment. Fig. 9 It can be clearly seen that the output power prediction result obtained by using the IPWLCM-IPSO-BP prediction model of the present invention (i.e., the curve corresponding to IPWLCM-IPSO-BP) is closer to the actual output power of the wind farm at the corresponding moment (i.e., the curve corresponding to Target), and the prediction error is smaller.

[0116] The present invention improves the traditional particle swarm algorithm so that the inertia weight and learning factor of the improved particle swarm algorithm are no longer fixed values, but change with the increase of the number of iterations, and adopts the optimized chaotic mapping (IPWLCM) to obtain the initial population of the improved particle swarm algorithm; then the improved particle swarm algorithm is used to optimize the initial weight and threshold of the BP neural network, and the speed and position of each particle between the populations are continuously updated, and finally the optimal fitness value is searched. Compared with the traditional training model, the combined model has higher prediction accuracy. It solves the problems of low prediction accuracy of the existing prediction model under the influence of multiple factors, and the algorithm is easy to fall into the local optimum, optimizes the management and operation level of wind farms, improves the economic benefits of wind power enterprises, and improves the wind power resource consumption capacity of the power grid system.

[0117] Computer equipment example:

[0118] A computer device of the present invention includes a processor, which executes a computer program to implement the steps of a wind power prediction method as described in the above method embodiment. The principle, implementation process, and effect that can be achieved of the method have been fully described in the method embodiment and will not be repeated here.

[0119] System Example:

[0120] A wind power prediction system of the present invention includes a big data server and a prediction server. The big data server is used to install a wind farm database system to store real-time data every day, receive NBW data, parameters and historical data, and finally forward the data to the wind power prediction server through a reverse isolation device; the prediction server is used to receive real-time data to perform wind power prediction to implement the steps of a wind power prediction method as described in the above method embodiment. The principle, implementation process and achievable effects of the method have been fully introduced in the method embodiment and will not be repeated here.

[0121] In addition, in order to ensure the security of the power grid system, when the prediction server receives NBW data from the outside, corresponding isolation devices and firewalls are installed to ensure the safety of the power system.

Claims

1. A wind power prediction method, characterized in that: The method includes: Obtain the index data required for predicting the output power of the wind farm at the predicted time, input the index data into the trained machine learning model, and obtain the output power of the wind farm at the predicted time; the initial values ​​of the parameters of the machine learning model are obtained by optimizing using an improved particle swarm algorithm, which is an improvement on the traditional particle swarm algorithm, and the improvement includes: As the number of iterations increases, the individual learning factor used in updating the individual speed gradually increases, and the social learning factor used in updating the individual speed gradually decreases; the absolute value of the difference between the initial value of the individual learning factor and the initial value of the social learning factor is greater than the set difference threshold, and the initial value of the individual learning factor is less than the initial value of the social learning factor.

2. The wind power prediction method according to claim 1, characterized in that: The individual learning factor and / or social learning factor changes nonlinearly.

3. The wind power prediction method according to claim 2, characterized in that: The iterative update formula of the individual learning factor is: in, is the individual learning factor, a is a constant between 0 and 1, sinh(x) is the hyperbolic sine function, sin(x) is the sine function, and k is the number of iterations.

4. The wind power prediction method according to claim 2, characterized in that: The iterative update formula of the social learning factor is: in, is the social learning factor, b is a constant between 0 and 1, tanh(x) is the hyperbolic tangent function, sin(x) is the sine function, and k is the number of iterations.

5. The wind power prediction method according to claim 1, characterized in that: The improvement of the improved particle swarm algorithm also includes: the inertia weight used in updating the individual speed changes continuously with the increase of the number of iterations, and the iterative update formula of the inertia weight is: in, is the inertia weight, ω min is the minimum value of the inertia weight, ω max is the maximum value of the inertia weight, ω range =ω max -ω min , k is the number of iterations, ξ and λ are both positive constants.

6. The wind power prediction method according to claim 1, characterized in that: The improvement of the improved particle swarm algorithm also includes: using chaos mapping to obtain the initial population of the improved particle swarm algorithm, specifically: Among them, α, β, δ, ζ, and p are control parameters, and x i is the ith chaos number.

7. The wind power prediction method according to claim 1, characterized in that: The index data includes a data sequence consisting of at least one of environmental indexes, wind farm operation indexes, daily type values ​​and power consumption per unit time in a period of time before the prediction moment.

8. The wind power prediction method according to claim 7, characterized in that: The environmental indicators include at least one of air pressure, topography, light radiation intensity, maximum temperature, minimum temperature, average temperature, temperature in the turbine cabin, wind force level, humidity and meteorological conditions.

9. The wind power prediction method according to claim 7, characterized in that: The wind farm operation indicators include: blade pitch angle, blade pitch angle and nacelle yaw angle.

10. The wind power prediction method according to claim 1, characterized in that: The power data in the training data used to train the machine learning model is preprocessed data, and the preprocessing specifically includes: In the case where it is detected that a certain data in the power data is abnormal or missing, if the interval of the moments corresponding to the power data is less than the first set interval threshold, a linear interpolation method is used for processing to fill or replace the data accordingly; if the interval of the moments corresponding to the power data is greater than or equal to the first set interval threshold and less than the second set interval threshold, a horizontal processing method is used for processing to fill or replace the data accordingly; if the interval of the moments corresponding to the power data is greater than or equal to the second set interval threshold, a vertical processing method is used for processing to fill or replace the data accordingly; wherein the first set interval threshold is less than the second set interval threshold; The horizontal processing method is: taking the average value of the power at adjacent moments corresponding to the data as the power at that moment; the vertical processing method is: obtaining the power at that moment based on the power data of the same period as the data.

11. The wind power prediction method according to claim 1, characterized in that: The machine learning model is a BP neural network.

12. A computer device comprising a processor, characterized in that: The processor executes a computer program to implement the steps of the wind power prediction method according to any one of claims 1 to 11.

13. A wind power prediction system, characterized in that: The method comprises a prediction server, wherein the prediction server comprises a processor, and the processor executes a computer program to implement the steps of the wind power prediction method according to any one of claims 1 to 11.

14. The wind power prediction system according to claim 13, characterized in that: The wind power prediction system also includes a big data server, which is used to install a wind farm database system to store real-time data and historical data, and forward the data to the prediction server.

Images