Power prediction method based on abnormal data detection and frost ice optimization
The BiLSTM model was optimized by KD-DBSCAN and frost optimization algorithm, which solved the time-consuming and parameter combination problems of DBSCAN clustering and achieved high efficiency and accuracy in wind power prediction.
Patent Information
- Application Number
- CN202510570530.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-30
- Publication Date
- 2025-09-19
AI Technical Summary
The DBSCAN clustering method in the existing technology takes too long and cannot adaptively find the optimal clustering parameter combination. Simple neural network prediction cannot obtain the optimal parameters for wind power prediction, which affects the prediction results.
The KD-DBSCAN model combined with the Silhouette score is used to evaluate clustering quality, the KD tree is used to accelerate DBSCAN clustering, and the parameters of the BiLSTM model are optimized using the Frost Ice Optimization algorithm to automatically find the optimal parameter combination.
The clustering time is effectively reduced, and the accuracy and efficiency of wind power prediction are improved. The prediction effect of the optimized BiLSTM model is better than that of the unoptimized and other optimization algorithms.
Smart Images
Figure CN120671027A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of power prediction, and in particular to a power prediction method based on abnormal data detection and frost and ice optimization. Background Art
[0002] At present, in the field of wind power prediction, the DBSCAN clustering method is used to clean the data to detect abnormal data, and then the neural network model is simply combined to predict wind power.
[0003] The DBSCAN clustering method mentioned above takes a significant amount of time to perform clustering and cannot adaptively find the optimal clustering parameter combination. Furthermore, simply using a neural network to predict power fails to yield the optimal parameter combination for the prediction model, which can negatively impact the final wind power prediction results. Summary of the Invention
[0004] In view of this, the purpose of the present invention is to propose a power prediction method based on abnormal data detection and frost optimization to solve the problems in data cleaning that the DBSCAN clustering method takes too long and cannot automatically obtain the optimal prediction parameters of the prediction model.
[0005] Based on the above objectives, the present invention provides a power prediction method based on abnormal data detection and frost and ice optimization, comprising:
[0006] S1. Create a KD tree using wind power data, set the parameter values of Minpts and Eps in the KD-DBSCAN model, calculate the distance ε between the wind power data point and the KD tree node, and for each KD tree node, calculate its boundary range according to the Eps parameter input by the KD-DBSCAN model. Then compare the calculated distance ε with the corresponding boundary range, cluster the wind power data with a distance less than the boundary range into the corresponding KD tree node to form a set, delete the wind power data points with a distance greater than the boundary range from the set, and determine whether the number of wind power data points in the set is greater than Minpts. If so, retain the set for clustering. If not, delete the set, and traverse all KD tree nodes to obtain the final clustering result.
[0007] S2. Set the Minpts and Eps ranges in the KD-DBSCAN model to [5, 15] and [0.1, 3] respectively, with a Minpts step size of 1 and an Eps step size of 0.01. Obtain all final cohesion results obtained under different Minpts and Eps parameter values. Use the Silhouette score to evaluate the quality of all KD-DBSCAN final cohesion results. Select the optimal Minpts and Eps parameter combination corresponding to the cohesion result with the highest score to complete the training of the KD-DBSCAN model.
[0008] S3. Detect and delete abnormal data in real-time data using the trained KD-DBSCAN model, and obtain clustered data. Then, use linear interpolation to fill in the data to obtain processed wind power data.
[0009] S4. Input the processed wind power data into the frost optimization algorithm to optimize the BiLSTM combination model to obtain the predicted wind power results and perform error calculation.
[0010] Optionally, in step S1, the formula for calculating the distance ε between the wind power data point and the KD tree node is:
[0011]
[0012] Among them, ε represents the distance between two points, x i ,y i and z i represents the i-th wind power data observation value of the three variables, x j ,y j and z j Represents the j-th wind power data observation value of the three variables.
[0013] Optionally, the mathematical formula of the Silhouette score is:
[0014]
[0015] Among them, s(i) represents the Silhouette score, a(i) represents the average distance from other data points in the same cluster to the data point, which is called cohesion, and b(i) represents the average distance from the data points in the closest other clusters, which is called separation.
[0016] Optionally, the mathematical formula of the linear interpolation method is:
[0017]
[0018] Where y is the value to be interpolated, x is the known value, y1 is the value to the left of the missing value, y2 is the value to the right of the missing value, and x1 and x2 are the positions corresponding to y1 and y2.
[0019] Optionally, optimizing the BiLSTM combination model using the frost optimization algorithm includes:
[0020] S401, initialization group and parameter stage, in this stage, the initial position of the frost group composed of a certain number of frost particles is randomly generated, the parameters of the RIME algorithm are initialized, and the initial learning rate of the search space is defined: [10 -5 , 10 -1 ], Regularization coefficient L2: [10 -5 , 10-1 ], Number of neurons: [4, 128], Define the fitness function;
[0021] S402, soft frost search strategy stage: In this stage, the parameters of the initial learning rate, regularization coefficient L2, and number of neurons are updated, the condensation process of each frost particle is simulated, the position of the frost particle after condensation is calculated using a formula, and the fitness value of the current position is calculated;
[0022] S403, hard frost ice puncture mechanism stage, in this stage, the parameters of the initial learning rate, regularization coefficient L2 and number of neurons are updated to cause displacement between particles, and the updated fitness value is calculated;
[0023] S404, in a positive greedy selection mechanism stage, the fitness value of the frost ice crystal after the update is compared with the fitness value before the update. If the fitness value after the update is better than the fitness value before the update, the fitness value before the update is replaced by the updated value, and the initial learning rate, regularization coefficient L2 and number of neurons parameters before the update are replaced by the updated initial learning rate, regularization coefficient L2 and number of neurons parameters;
[0024] S405 , looping steps S403 - S404 to obtain the optimal initial learning rate, regularization coefficient L2 and number of neurons parameters.
[0025] Optionally, in step S402, the mathematical formula for calculating the position of frost particles is:
[0026]
[0027] Where i and j represent the jth particle in the i-th frost crystal, is the new position of the updated particle, R best,j is the jth particle of the best frost crystal in the frost swarm R, parameter r1 is a random number in the range (-1, 1), cosθ varies with the number of iterations, β is an environmental factor, h represents the adhesion, which is a random number in the range (0, 1), U and L are the upper and lower bounds of the iteration space, E is the coefficient of adhesion, which increases with the number of iterations, r2 is a random number in the range (0, 1), which together with E controls whether the particle position is updated;
[0028] The mathematical formulas for θ, β, and E in the mathematical formula for the position of frost particles are:
[0029]
[0030] Where t is the current iteration number and T2 is the maximum iteration number of the algorithm;
[0031]
[0032] The mathematical model of β is a step function, the value of w defaults to 5, and [] indicates rounding operation;
[0033]
[0034] Where t is the current iteration number and T2 is the maximum iteration number of the algorithm.
[0035] Optionally, in step S403, in the stage of implementing the hard frost ice puncture mechanism, the mathematical formula for the replacement between particles is:
[0036]
[0037] Among them, r3 is a random number in the range of (-1, 1), S i represents the i-th frost crystal, F normr (S i ) represents the normalized value of the current crystal fitness value.
[0038] Optionally, in step S4, BiLSTM is used as a prediction model, and the forward and backward propagation hidden states and output mathematical formulas of its neural network are:
[0039]
[0040] in, and Represent the forward and backward propagation hidden states at time t, W hf 、W if and b t They represent the forward propagation hidden state h t-1 Weight, input X t The weight and bias term, W hb 、W ib and b′ t They represent the forward propagation hidden state h′ t+1 Weight, input X t The weight and bias term, σ is the activation function;
[0041]
[0042] Among them, y t represents the output, U and U′ represent the forward propagation state at time t and back-propagation state The weight of b y is the bias term, and σ is the activation function.
[0043] Optionally, in step S4, after obtaining the predicted wind power result, the method further includes:
[0044] The mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R2) were used to calculate the error.
[0045] (1) The technical solution of the present invention can avoid a large number of distance calculations and domain search operations for each data point by clustering with the nodes of the KD tree as the core points, and judge whether the number of wind power data points in the set is greater than Minpts. If so, the set is retained for clustering; if not, the set is deleted. The KD tree accelerates the DBSCAN clustering time and effectively reduces the clustering time. The Silhouette score is used to adaptively find the optimal clustering parameters Eps and Minpts of DBSCAN, and the effect is better than the clustering effect of manually setting the parameters.
[0046] (2) The technical solution of the present invention optimizes the BiLSTM prediction model through the frost optimization algorithm, and can adaptively find the optimal parameters of the prediction model (initial learning rate, regularization coefficient L2 and number of neurons). The effect is better than the prediction effect of BiLSTM optimized without the optimization algorithm, and is also better than the prediction effect of BiLSTM optimized with other optimization algorithms. BRIEF DESCRIPTION OF THE DRAWINGS
[0047] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only for the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.
[0048] Figure 1 is a flow chart of the overall method of an embodiment of the present invention;
[0049] Figure 2 A flowchart of KD tree construction according to an embodiment of the present invention;
[0050] Figure 3 Flowchart of the RIME algorithm according to an embodiment of the present invention;
[0051] Figure 4 This is a basic structure diagram of the BiLSTM neural network according to an embodiment of the present invention;
[0052] Figure 5 A comparison diagram of fitness curves of different optimization algorithms according to an embodiment of the present invention;
[0053] Figure 6 This is a comparison chart of the prediction curves before and after cleaning according to an embodiment of the present invention;
[0054] Figure 7 2 is a comparison chart of prediction curves of different models according to an embodiment of the present invention. DETAILED DESCRIPTION
[0055] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to specific embodiments.
[0056] It should be noted that, unless otherwise defined, the technical or scientific terms used in the present invention should have the usual meanings understood by people with ordinary skills in the field to which the present invention belongs. The "first", "second" and similar words used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the object being described changes, the relative positional relationship may also change accordingly.
[0057] like Figure 1-Figure 7 As shown, a power prediction method based on abnormal data detection and frost and ice optimization includes:
[0058] S1. Create a KD tree using wind power data, set the parameter values of Minpts and Eps in the KD-DBSCAN model, calculate the distance ε between the wind power data point and the KD tree node, and for each KD tree node, calculate its boundary range according to the Eps parameter input by the KD-DBSCAN model. Then compare the calculated distance ε with the corresponding boundary range, cluster the wind power data with a distance less than the boundary range into the corresponding KD tree node to form a set, delete the wind power data points with a distance greater than the boundary range from the set, and determine whether the number of wind power data points in the set is greater than Minpts. If so, retain the set for clustering. If not, delete the set, traverse all KD tree nodes, and obtain the final clustering result.
[0059] S2. Set the Minpts and Eps ranges in the KD-DBSCAN model to [5, 15] and [0.1, 3] respectively, with a Minpts step size of 1 and an Eps step size of 0.01. Obtain all final cohesion results obtained under different Minpts and Eps parameter values. Use the Silhouette score to evaluate the quality of all KD-DBSCAN final cohesion results. Select the optimal Minpts and Eps parameter combination corresponding to the cohesion result with the highest score to complete the training of the KD-DBSCAN model.
[0060] S3. Detect and delete abnormal data in real-time data using the trained KD-DBSCAN model, and obtain clustered data. Then, use linear interpolation to fill in the data to obtain processed wind power data.
[0061] S4. Input the processed wind power data into the frost optimization algorithm to optimize the BiLSTM combination model to obtain the predicted wind power results and perform error calculation.
[0062] (1) The technical solution of the present invention can avoid a large number of distance calculations and domain search operations for each data point by clustering with the nodes of the KD tree as the core points, and judge whether the number of wind power data points in the set is greater than Minpts. If so, the set is retained for clustering; if not, the set is deleted. The KD tree accelerates the DBSCAN clustering time and effectively reduces the clustering time. The Silhouette score is used to adaptively find the optimal clustering parameters Eps and Minpts of DBSCAN, and the effect is better than the clustering effect of manually setting the parameters.
[0063] (2) The technical solution of the present invention optimizes the BiLSTM prediction model through the frost optimization algorithm, and can adaptively find the optimal parameters of the prediction model (initial learning rate, regularization coefficient L2 and number of neurons). The effect is better than the prediction effect of BiLSTM optimized without the optimization algorithm, and is also better than the prediction effect of BiLSTM optimized with other optimization algorithms.
[0064] In some embodiments, in step S1, the formula for calculating the distance ε between the wind power data point and the KD tree node is:
[0065]
[0066] Among them, ε represents the distance between two points, x i ,y i and z i represents the i-th wind power data observation value of the three variables, x j ,y j and z j Represents the j-th wind power data observation value of the three variables.
[0067] In some embodiments, the mathematical formula for the Silhouette score is:
[0068]
[0069] Among them, s(i) represents the Silhouette score, a(i) represents the average distance from other data points in the same cluster to the data point, which is called cohesion, and b(i) represents the average distance from the data points in the closest other clusters, which is called separation.
[0070] In some embodiments, in order to avoid the influence of abnormal data on the prediction results, after the abnormal data value is detected, the abnormal data is deleted and then the data is filled using linear interpolation to ensure the accuracy and completeness of the data. The mathematical formula of the linear interpolation method is:
[0071]
[0072] Where y is the value to be interpolated, x is the known value, y1 is the value to the left of the missing value, y2 is the value to the right of the missing value, and x1 and x2 are the positions corresponding to y1 and y2.
[0073] In some embodiments, optimizing the BiLSTM combination model using the frost optimization algorithm includes:
[0074] S401, initialization group and parameter stage, in this stage, the initial position of the frost group composed of a certain number of frost particles is randomly generated, the parameters of the RIME algorithm are initialized, and the initial learning rate of the search space is defined: [10 -5 , 10 -1 ], Regularization coefficient L2: [10 -5 , 10 -1 ], Number of neurons: [4, 128], Define the fitness function;
[0075] S402, soft frost search strategy stage: In this stage, the parameters of the initial learning rate, regularization coefficient L2, and number of neurons are updated, the condensation process of each frost particle is simulated, the position of the frost particle after condensation is calculated using a formula, and the fitness value of the current position is calculated;
[0076] S403, hard frost ice puncture mechanism stage, in this stage, the parameters of the initial learning rate, regularization coefficient L2 and number of neurons are updated to cause displacement between particles, and the updated fitness value is calculated;
[0077] S404, in a positive greedy selection mechanism stage, the fitness value of the frost ice crystal after the update is compared with the fitness value before the update. If the fitness value after the update is better than the fitness value before the update, the fitness value before the update is replaced by the updated value, and the initial learning rate, regularization coefficient L2 and number of neurons parameters before the update are replaced by the updated initial learning rate, regularization coefficient L2 and number of neurons parameters;
[0078] S405 , looping steps S403 - S404 to obtain the optimal initial learning rate, regularization coefficient L2 and number of neurons parameters.
[0079] In order to avoid the problem of manual selection of single prediction simulation parameters, the Frost Ice algorithm is used to optimize the BiLSTM prediction model parameters, which can automatically obtain the optimal initial learning rate, regularization coefficient L2 and number of neurons.
[0080] In some embodiments, in step S402 , the mathematical formula for calculating the position of frost particles is:
[0081]
[0082] Where i and j represent the jth particle in the i-th frost crystal, is the new position of the updated particle, R best,j is the jth particle of the best frost crystal in the frost swarm R. Parameter r1 is a random number in the range (-1, 1). cosθ varies with the number of iterations. β is an environmental factor. h represents the adhesion, a random number in the range (0, 1). U and L are the upper and lower bounds of the iteration space, respectively. E is the coefficient of adhesion, which increases with the number of iterations. r2 is a random number in the range (0, 1). Together with E, it controls whether the particle position is updated.
[0083] The mathematical formulas for θ, β, and E in the mathematical formula for the position of frost particles are:
[0084]
[0085] Where t is the current iteration number and T2 is the maximum iteration number of the algorithm;
[0086]
[0087] The mathematical model of β is a step function, the value of w defaults to 5, and [] indicates rounding operation;
[0088]
[0089] Where t is the current iteration number and T2 is the maximum iteration number of the algorithm.
[0090] In some embodiments, in step S403, during the hard frost ice penetration mechanism phase, the mathematical formula for the replacement between particles is:
[0091]
[0092] Among them, r3 is a random number in the range of (-1, 1), S i represents the i-th frost crystal, F normr (S i ) represents the normalized value of the current crystal fitness value.
[0093] In some embodiments, in step S4, BiLSTM is used as a prediction model, and the mathematical formulas of the forward and backward propagation hidden states and outputs of its neural network are:
[0094]
[0095] in, and Represent the forward and backward propagation hidden states at time t, W hf 、W if and b t They represent the forward propagation hidden state h t-1 Weight, input X t The weight and bias term, W hb 、W ib and b′ t They represent the forward propagation hidden state h′ t+1 Weight, input X t The weight and bias term, σ is the activation function;
[0096]
[0097] Among them, y t represents the output, U and U′ represent the forward propagation state at time t and back-propagation state The weight of b y is the bias term, and σ is the activation function.
[0098] In some embodiments, in step S4, after obtaining the predicted wind power result, the method further includes:
[0099] The mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R) are used. 2 Perform error calculation.
[0100]
[0101] where y i Represents the predicted wind power value at the i-th moment.
[0102] Table 1 below shows the comparison results of prediction indicators before and after data cleaning.
[0103] Table 1
[0104]
[0105] As can be seen from Table 1, the cleaning method provided in the examples of this specification is better than the effect of not cleaning.
[0106] Table 2 below shows the BiLSTM model parameters manually selected and optimized using different optimization algorithms.
[0107] Table 2
[0108]
[0109] Table 3 below shows the prediction indicators of different models.
[0110] Table 3
[0111]
[0112] From the comparison of the above prediction structures, it can be seen that the prediction method RIME-BiLSTM provided in the embodiments of this specification is better than other existing methods.
[0113] Those skilled in the art will understand that the discussion of any of the above embodiments is merely illustrative and is not intended to limit the scope of the present invention to these examples. Within the spirit and principles of the present invention, the technical features of the above embodiments or different embodiments may be combined, the steps may be implemented in any order, and many other variations exist for the various aspects of the present invention described above, which are not provided in detail for the sake of clarity. Any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.
[0114] Those skilled in the art should understand that the discussion of any of the above embodiments is merely illustrative and is not intended to imply that the scope of the present invention (including the claims) is limited to these examples. Within the scope of the present invention, the technical features in the above embodiments or different embodiments may be combined, the steps may be implemented in any order, and there are many other variations of the above aspects of the present invention, which are not provided in detail for the sake of simplicity.
[0115] The present invention is intended to cover all such substitutions, modifications and variations that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A power prediction method based on abnormal data detection and frost optimization, characterized in that: include: S1. Create a KD tree using wind power data, set the parameter values of Minpts and Eps in the KD-DBSCAN model, calculate the distance ε between the wind power data point and the KD tree node, and for each KD tree node, calculate its boundary range according to the Eps parameter input by the KD-DBSCAN model. Then compare the calculated distance ε with the corresponding boundary range, cluster the wind power data with a distance less than the boundary range into the corresponding KD tree node to form a set, delete the wind power data points with a distance greater than the boundary range from the set, and determine whether the number of wind power data points in the set is greater than Minpts. If so, retain the set for clustering. If not, delete the set, and traverse all KD tree nodes to obtain the final clustering result. S2. Set the Minpts and Eps ranges in the KD-DBSCAN model to [5, 15] and [0.1, 3] respectively, with a Minpts step size of 1 and an Eps step size of 0.
01. Obtain all final cohesion results obtained under different Minpts and Eps parameter values. Use the Silhouette score to evaluate the quality of all KD-DBSCAN final cohesion results. Select the optimal Minpts and Eps parameter combination corresponding to the cohesion result with the highest score to complete the training of the KD-DBSCAN model. S3. Detect and delete abnormal data in real-time data using the trained KD-DBSCAN model, and obtain clustered data. Then, use linear interpolation to fill in the data to obtain processed wind power data. S4. Input the processed wind power data into the frost optimization algorithm to optimize the BiLSTM combination model to obtain the predicted wind power results and perform error calculation.
2. The power prediction method based on abnormal data detection and frost and ice optimization according to claim 1, characterized in that: In step S1, the formula for calculating the distance ε between the wind power data point and the KD tree node is: Among them, ε represents the distance between two points, x i ,y i and z i represents the i-th wind power data observation value of the three variables, x j ,y j and z j Represents the j-th wind power data observation value of the three variables.
3. The power prediction method based on abnormal data detection and frost and ice optimization according to claim 1, characterized in that: The mathematical formula for the Silhouette score is: Among them, s(i) represents the Silhouette score, a(i) represents the average distance from other data points in the same cluster to the data point, which is called cohesion, and b(i) represents the average distance from the data points in the closest other clusters, which is called separation.
4. The power prediction method based on abnormal data detection and frost and ice optimization according to claim 1, characterized in that: The mathematical formula of the linear interpolation method is: Where y is the value to be interpolated, x is the known value, y1 is the value to the left of the missing value, y2 is the value to the right of the missing value, and x1 and x2 are the positions corresponding to y1 and y2.
5. The power prediction method based on abnormal data detection and frost and ice optimization according to claim 1, characterized in that: Optimizing the BiLSTM combination model using the Frost Ice Optimization Algorithm includes: S401, initialization group and parameter stage, in this stage, the initial position of the frost group composed of a certain number of frost particles is randomly generated, the parameters of the RIME algorithm are initialized, and the initial learning rate of the search space is defined: [10 -5 , 10 -1 ], Regularization coefficient L2: [10 -5 , 10 -1 ], Number of neurons: [4, 128], Define the fitness function; S402, soft frost search strategy stage: In this stage, the parameters of the initial learning rate, regularization coefficient L2, and number of neurons are updated, the condensation process of each frost particle is simulated, the position of the frost particle after condensation is calculated using a formula, and the fitness value of the current position is calculated; S403, hard frost ice puncture mechanism stage, in this stage, the parameters of the initial learning rate, regularization coefficient L2 and number of neurons are updated to cause displacement between particles, and the updated fitness value is calculated; S404, in a positive greedy selection mechanism stage, the fitness value of the frost ice crystal after the update is compared with the fitness value before the update. If the fitness value after the update is better than the fitness value before the update, the fitness value before the update is replaced by the updated value, and the initial learning rate, regularization coefficient L2 and number of neurons parameters before the update are replaced by the updated initial learning rate, regularization coefficient L2 and number of neurons parameters; S405 , looping steps S403 - S404 to obtain the optimal initial learning rate, regularization coefficient L2 and number of neurons parameters.
6. The power prediction method based on abnormal data detection and frost and ice optimization according to claim 5, characterized in that: In step S402, the mathematical formula for calculating the position of frost particles is: Where i and j represent the jth particle in the i-th frost crystal, is the new position of the updated particle, R best,j is the jth particle of the best frost crystal in the frost swarm R, parameter r1 is a random number in the range (-1, 1), cosθ varies with the number of iterations, β is an environmental factor, h represents the adhesion, which is a random number in the range (0, 1), U and L are the upper and lower bounds of the iteration space, E is the coefficient of adhesion, which increases with the number of iterations, r2 is a random number in the range (0, 1), which together with E controls whether the particle position is updated; The mathematical formulas for θ, β, and E in the mathematical formula for the position of frost particles are: Where t is the current iteration number and T2 is the maximum iteration number of the algorithm; The mathematical model of β is a step function, the value of w defaults to 5, and [] indicates rounding operation; Where t is the current iteration number and T2 is the maximum iteration number of the algorithm.
7. The power prediction method based on abnormal data detection and frost and ice optimization according to claim 5, characterized in that: In step S403, the hard frost ice puncture mechanism is implemented, and the mathematical formula for the replacement between particles is: Among them, r3 is a random number in the range of (-1, 1), S i represents the i-th frost crystal, F normr (S i ) represents the normalized value of the current crystal fitness value.
8. The power prediction method based on abnormal data detection and frost and ice optimization according to claim 1, characterized in that: In step S4, BiLSTM is used as the prediction model, and the mathematical formulas of the forward and backward propagation hidden states and outputs of its neural network are as follows: in, and Represent the forward and backward propagation hidden states at time t, W hf 、W if and b t They represent the forward propagation hidden state h t-1 Weight, input X t The weight and bias term, W hb 、W ib and b t ′ represents the forward propagation hidden state h′ t+1 Weight, input X t The weight and bias term, σ is the activation function; Among them, y t represents the output, U and U′ represent the forward propagation state at time t and back-propagation state The weight of b y is the bias term, and σ is the activation function.
9. The power prediction method based on abnormal data detection and frost and ice optimization according to claim 1, characterized in that: In step S4, after obtaining the predicted wind power result, the method further includes: The mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R2) were used to calculate the error.