Power grid new energy power prediction method combining static parameters and dynamic parameters
By combining static and dynamic parameters of the power grid new energy power prediction method, the problem of limitations in new energy dynamic power capture in the existing technology is solved, and more accurate power prediction and grid management are achieved.
Patent Information
- Application Number
- CN202411678624.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-22
- Publication Date
- 2025-03-11
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The prior art has limitations in capturing the dynamic changes in new energy power, resulting in inaccurate predictions.
A power prediction method for power grid new energy is adopted that combines static parameters and dynamic parameters. By collecting and preprocessing static and dynamic parameters, differential statistics and joint evaluation are performed using cross-verification method and Akagi information criterion to determine the optimal power prediction model.
It realizes more accurately capturing the fluctuations in new energy power, enhances the flexibility of grid scheduling, reduces energy waste, reduces operating costs, and improves the accuracy and reliability of new energy power prediction.
Smart Images
Figure CN119598196A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of optimization management and efficient utilization of new energy sources in power grids, and in particular relates to a power grid new energy power prediction method combining static parameters and dynamic parameters. Background Art
[0002] With the widespread application of renewable energy in power systems, accurate prediction of renewable energy power has become the key to optimizing grid operation. Traditional prediction models mainly rely on static parameters, such as geographic location, equipment model, etc., which reflect the inherent characteristics of renewable energy power generation equipment. However, renewable energy power is significantly affected by dynamic factors, such as wind speed, solar radiation, etc., which change with time and environmental conditions. Therefore, existing models have limitations in capturing the dynamic changes of renewable energy power.
[0003] The limitations of existing technologies mainly include:
[0004] Existing technologies rely too much on the static parameter of the location of renewable energy power plants, ignoring the fact that the actual impact of location on power output may change with environmental conditions (such as climate change). For example, a wind power plant located in an area with historically high wind speeds may experience lower-than-expected actual wind speeds due to climate change, and existing models cannot accurately predict such changes.
[0005] Existing models only consider the static parameter of the device model, but do not take into account that the device will age or degrade over time. For example, the conversion efficiency of a solar panel may decrease due to long-term use and environmental factors (such as temperature and humidity), and existing models cannot capture the impact of such performance changes on power output.
[0006] Existing technologies cannot effectively capture dynamic factors such as wind speed and solar radiation that change over time. For example, in wind power generation, a sudden increase or decrease in wind speed will directly affect the power generation. If the model cannot respond to such changes in real time, it will not be able to accurately predict the power output.
[0007] Existing models do not fully consider the impact of environmental conditions, such as temperature and humidity, on the efficiency of renewable energy generation. For example, in solar power generation, high temperatures may reduce the efficiency of solar panels, but existing models may not accurately predict the specific impact of this efficiency reduction on power output.
[0008] Existing technologies treat time information (such as hours, weeks, and months) as static parameters, without considering the dynamic characteristics of renewable energy power in different time periods. For example, solar power generation has significantly different power outputs at different times of the day (morning, noon, and evening), and existing models may not accurately reflect the dynamic nature of this time change.
[0009] To this end, we propose a new energy power prediction method for power grids that combines static and dynamic parameters. Summary of the invention
[0010] In view of the above deficiencies in the prior art, the technical problem to be solved by the present invention is: to provide a method for predicting the power of new energy in a power grid by combining static parameters and dynamic parameters, so as to solve the problem that the capture of dynamic power of new energy in the prior art is limited, resulting in inaccurate prediction.
[0011] In order to achieve the above objectives, the present invention adopts the following technical solutions:
[0012] A method for predicting the power of new energy sources in power grids by combining static and dynamic parameters:
[0013] S1. Collect static parameters and dynamic parameters within a certain period of time, and pre-process the collected static parameters and dynamic parameters;
[0014] S2. Use cross-validation method to perform statistical analysis and evaluation on the difference between static power parameters and dynamic power parameters after preprocessing;
[0015] S3. Use the Akaike information criterion to jointly evaluate the difference between static power parameters and dynamic power parameters to determine the optimal power prediction model;
[0016] S4. Utilize the determined optimal model to capture the fluctuation of renewable energy power and predict the renewable energy power of the power grid.
[0017] Furthermore, the method of using the cross-validation method to perform statistical analysis on the difference between the trained static power parameters and the dynamic power parameters comprises the following steps:
[0018] S21. Cross-validation settings: divide the parameters obtained after preprocessing into two groups, namely, static parameter set and dynamic parameter set; set the cross-validation fold k, and divide the static parameter set and dynamic parameter set into k subsets: I1, I2, I3, ... I k , for each subset I i (i=1,2,...,n): Take the i-th subset as the test set, denoted as V i , except I i All subsets except i , train static parameter sets and dynamic parameters;
[0019] S22. Cross-performance validation calculation, calculate the average MSE on all k-fold test sets CV ;
[0020] S23. Difference statistics, calculate the MSE difference between static parameters and dynamic parameters.
[0021] Furthermore, in step S21, the method for training the static parameter set and the dynamic parameter set includes:
[0022] Train the model using the training set T i The model is trained using the data points on X. Assume that the model is a function M that fits the training data through parameters θ: M(X; θ)
[0023] Where X is the input feature set (including static parameters and dynamic parameters), θ model parameters;
[0024] Parameter estimation, estimate the parameter θ through optimization problem, and use the loss function L to measure the difference between the model prediction and the actual output of the training set. Specifically, find a set of θ values that minimizes the sum of the squares of L. The formula is:
[0025]
[0026] Where L is the loss function and y is the true output in the test set;
[0027] Model evaluation, using the trained model M and parameters θ i To predict V i The output of , using mean squared error to calculate the performance metric:
[0028]
[0029] Where y is the true output in the test set, L is the loss function, and |V i | is the size of the test set;
[0030] Furthermore, in step S22, the average MSE on all k-fold test sets is calculated. CV The calculation formula is:
[0031]
[0032] Furthermore, in step S2, the method of evaluating static power parameters and dynamic power parameters using a cross-validation method includes:
[0033] Using the training set T i The model is trained based on static parameters only, and the model parameters θ are obtained. static ; In the test set V i Using θ static Test and get the predicted value Using the training set T i The model is trained based on static parameters only, and the model parameters θ are obtained. dynamic ; In the test set V i Using θdynamic Test and get the predicted value
[0034] Use mean squared error to evaluate the model:
[0035]
[0036] Among them, Y ij is the true power value of the jth sample in the i-th fold test set, and are the predicted power values of the static and dynamic models, respectively;
[0037]
[0038] β0 is the intercept term, β1, β2, ···β m are model parameters, corresponding to different static characteristics, X static,ij1 , X static,ij2 ,···X static,ijn is the static eigenvalue of the jth sample in the i-th fold test set;
[0039]
[0040] X static,ijn represents a specific static feature, which can be geographical location, equipment model, installed capacity, etc. γ0 is the intercept term, γ1, γ2, ···γ n are model parameters, corresponding to different dynamic characteristics, X dynamic,ij1 , X dynamic,ij2 ,···X dynamic,ijn is the dynamic eigenvalue of the jth sample in the i-th fold test set, X dynamic,ijn Represents specific dynamic characteristics, which can be wind speed, solar radiation, etc.;
[0041] Performance evaluation, calculate the average MSE of all k tests to get the average performance index of the model:
[0042]
[0043] Comparing MSE CV,dynamic and MSE CV,static To evaluate which type of parameters contributes more to the model prediction performance.
[0044] Furthermore, in the step S23, the MSE difference between the static parameter model and the dynamic parameter model is calculated;
[0045] D i =|MSE statiic -MSE dynamic |;
[0046] MSE static is the average MSE calculated in cross validation for models trained using only static parameters, MSE dynamic is the average MSE calculated in cross validation for models trained using only dynamic parameters. If D i The value of is positive and large, indicating that the prediction accuracy of the static parameter model is lower than that of the dynamic parameter model; if D i The value of is negative and large, indicating that the prediction accuracy of the static parameter model is higher than that of the dynamic parameter model; if D i is close to zero, indicating that the prediction accuracy of the two models is similar.
[0047] Furthermore, in step S3, the step of jointly evaluating the difference between the prediction result of each model and the static parameters and the dynamic parameters using the Akaike Information Criterion is as follows:
[0048] S31. Akaike Information Criterion AIC calculation; AIC calculation is performed for each model;
[0049] S32. Model evaluation and selection: For each model, AIC and difference statistics are used for joint evaluation;
[0050] S33. Determine the optimal model through the combined evaluation results.
[0051] Furthermore, in step S31:
[0052]
[0053] where τ is the number of model parameters, is the maximum likelihood estimate of the model;
[0054] Furthermore, in step S32, the joint evaluation index is defined as:
[0055] S c =AIC+λ·D i ;
[0056] Among them, AIC is the Akaike information criterion, which is used to measure the goodness of fit and complexity of the model;
[0057] λ is a weight parameter used to balance the importance of the difference between AIC and MSE;
[0058] D i is the difference in MSE (mean square error) between the static parameter model and the dynamic parameter model. Further, in step S33,
[0059] M * =argmin M (S c );
[0060] Among them, M * represents the optimal model, S c It is a joint evaluation indicator.
[0061] In summary, due to the adoption of the above technical solution, the beneficial technical effects of the invention are:
[0062] A method for predicting power of renewable energy sources in a power grid by combining static and dynamic parameters can more accurately capture the fluctuation of renewable energy power and enhance the flexibility of power grid dispatching. Accurate power prediction makes power grid dispatching more flexible and helps optimize the allocation of power resources.
[0063] A method for predicting renewable energy power in power grids that combines static and dynamic parameters reduces energy waste caused by inaccurate predictions and reduces grid operating costs. At the same time, it effectively improves the accuracy and reliability of renewable energy power predictions, improves grid stability, and better balances supply and demand, providing strong technical support for optimized grid management and efficient use of renewable energy. BRIEF DESCRIPTION OF THE DRAWINGS
[0064] Figure 1 The figure is a flow chart of a method for predicting power of new energy sources in a power grid by combining static parameters and dynamic parameters.
[0065] Figure 2 The figure is a flowchart of a method for performing statistical analysis on the trained static power parameters and dynamic power parameters using the cross-validation method.
[0066] Figure 3 Flowchart of the method for training static parameter sets and dynamic parameter sets.
[0067] Figure 4 Flow chart of the method for jointly evaluating the difference between static power parameters and dynamic power parameters using the Akaike Information Criterion. DETAILED DESCRIPTION
[0068] In order to make the purpose, technical solution and advantages of the invention more clear, the invention is further described in detail below in combination with the embodiments. It should be understood that the specific embodiments described here are only used to explain the invention and are not used to limit the invention.
[0069] Example
[0070] In this embodiment, a method for predicting the power of new energy sources in a power grid combining static parameters and dynamic parameters is provided:
[0071] like Figure 1 As shown, a method for predicting the power of new energy sources in a power grid combining static parameters and dynamic parameters is as follows:
[0072] S1. Collect static parameters and dynamic parameters within a certain period of time, and pre-process the collected parameters by data cleaning, feature engineering and normalization;
[0073] Among them, static parameters include: geographical location, including the latitude and longitude of the renewable energy power plant; equipment type, the model of wind turbines or solar panels used in the renewable energy power plant; installed capacity, the total power capacity of renewable energy power generation equipment, etc.;
[0074] Dynamic parameters include: wind speed: real-time wind speed data; solar radiation: real-time solar radiation intensity; temperature: ambient temperature; humidity: ambient humidity; time information: hour, week, month, etc.;
[0075] Parameter preprocessing, using data cleaning, feature engineering and normalization to preprocess parameters; data cleaning: processing missing values and outliers to ensure the quality of the data set; feature engineering: creating new features based on the original data to enhance the predictive ability of the model; normalization / standardization: since different features may have different dimensions and magnitudes, normalization or standardization is required to facilitate model processing;
[0076] S2. Use cross-validation method to perform statistical analysis and evaluation on the difference between static power parameters and dynamic power parameters after preprocessing;
[0077] Determine the specific impact of static and dynamic parameters on the performance of the prediction model, and quantify the difference in the contribution of the two types of parameters to the model accuracy; through difference statistics, deeply understand the impact of different parameters on model performance, and provide a basis for model selection and optimization;
[0078] S3. Use the Akaike information criterion to jointly evaluate the difference between static power parameters and dynamic power parameters to determine the optimal power prediction model;
[0079] The prediction results of each model and the differences between static power parameters and dynamic power parameters were jointly evaluated by the Akaike information criterion, which improved the goodness of fit and selected the model that best fits the training data; the joint evaluation model of the differences by the Akaike information criterion is concise and avoids overfitting; the model with lower MSE on the test data is selected, indicating that it has better prediction performance; the optimal balance between complexity and prediction performance is found through the joint evaluation of AIC and MSE differences.
[0080] By using AIC for joint evaluation, we can ensure that the selected model not only performs well on historical data but also has good generalization ability and interpretability, which is crucial for renewable energy power forecasting and grid management.
[0081] S4. Use the determined optimal model to capture the fluctuation of renewable energy power and predict the renewable energy power of the power grid.
[0082] like Figure 2 As shown, the method of using the cross-validation method to perform statistical analysis on the difference between the trained static power parameters and the dynamic power parameters includes the following steps:
[0083] S21. Cross-validation settings: divide the parameters obtained after preprocessing into two groups, namely, static parameter set and dynamic parameter set; set the cross-validation fold k, and divide the static parameter set and dynamic parameter set into k subsets: I1, I2, I3, ... I k , for each subset I i (i=1,2,...,n): Take the i-th subset as the test set, denoted as V i , except I i All subsets except i , train static parameter sets and dynamic parameters;
[0084] S22. Cross-performance validation calculation, calculate the average MSE on all k-fold test sets CV ;
[0085] S23. Difference statistics, calculate the MSE difference between static parameters and dynamic parameters.
[0086] like Figure 3 As shown, in step S21, the method for training the static parameter set and the dynamic parameter set includes:
[0087] Train the model using the training set T i The model is trained based on the data points on the dataset. Assume that the model is a function M that fits the training data through parameters θ: M(X; θ), where X is the input feature set (including static and dynamic parameters), and θ is the model parameter.
[0088] Parameter estimation, estimate the parameter θ through optimization problem, and use the loss function L to measure the difference between the model prediction and the actual output of the training set. Specifically, find a set of θ values that minimizes the sum of the squares of L. The formula is:
[0089]
[0090] Where L is the loss function and y is the true output in the test set;
[0091] Model evaluation, using the trained model M and parameters θ i To predict V i The output of , using mean squared error to calculate the performance metric:
[0092]
[0093] Where y is the true output in the test set, L is the loss function, and |V i | is the size of the test set;
[0094] In step S22, the average MSE on all k-fold test sets is calculated. CV The calculation formula is:
[0095]
[0096] In step S2, the method of evaluating static power parameters and dynamic power parameters using a cross-validation method includes:
[0097] Using the training set T i The model is trained based on static parameters only, and the model parameters θ are obtained. static ; In the test set V i Using θ static Test and get the predicted value Using the training set T i The model is trained based on static parameters only, and the model parameters θ are obtained. dynamic ; In the test set V i Using θ dynamic Test and get the predicted value
[0098] Use mean squared error to evaluate the model:
[0099]
[0100] Among them, Y ij is the true power value of the jth sample in the i-th fold test set, and are the predicted power values of the static and dynamic models, respectively;
[0101]
[0102] β0 is the intercept term, β1, β2, ···β m are model parameters, corresponding to different static characteristics, X static,ij1 , X static,ij2 ,···X static,ijn is the static eigenvalue of the jth sample in the i-th fold test set;
[0103]
[0104] X static,ijn represents a specific static feature, which can be geographical location, equipment model, installed capacity, etc. γ0 is the intercept term, γ1, γ2, ···γ n are model parameters, corresponding to different dynamic characteristics, Xdynamic,ij1 , X dynamic,ij2 ,···X dynamic,ijn is the dynamic eigenvalue of the jth sample in the i-th fold test set, X dynamic,ijn Represents specific dynamic characteristics, which can be wind speed, solar radiation, etc.;
[0105] Performance evaluation, calculate the average MSE of all k tests to get the average performance index of the model:
[0106]
[0107] Comparing MSE CV,dynamic and MSE CV,static To evaluate which type of parameters contributes more to the model prediction performance.
[0108] In the step S23, the MSE difference between the static parameter model and the dynamic parameter model is calculated;
[0109] D i =|MSE statiic -MSE dynamic |;
[0110] MSE static is the average MSE calculated in cross validation for models trained using only static parameters, MSE dynamic is the average MSE calculated in cross validation for models trained using only dynamic parameters. If D i The value of is positive and large, indicating that the prediction accuracy of the static parameter model is lower than that of the dynamic parameter model; if D i The value of is negative and large, indicating that the prediction accuracy of the static parameter model is higher than that of the dynamic parameter model; if D i is close to zero, indicating that the prediction accuracy of the two models is similar.
[0111] like Figure 4 As shown, in step S3, the step of jointly evaluating the difference between the static power parameter and the dynamic power parameter using the Akaike information criterion is:
[0112] S31. Akaike Information Criterion AIC calculation; AIC calculation is performed for each model;
[0113] S32. Model evaluation and selection: For each model, AIC and difference statistics are used for joint evaluation;
[0114] S33. Determine the optimal model through the combined evaluation results.
[0115] In step S31:
[0116]
[0117] where τ is the number of model parameters, is the maximum likelihood estimate of the model;
[0118] In step S32, the joint evaluation index is defined as:
[0119] S c =AIC+λ·D i ;
[0120] Among them, AIC is the Akaike information criterion, which is used to measure the goodness of fit and complexity of the model, λ is a weight parameter used to balance the importance of the difference between AIC and MSE, and D i is the difference in MSE (mean square error) between the static parameter model and the dynamic parameter model.
[0121] In the step S33,
[0122] M * =argmin M (S c );
[0123] Among them, M * represents the optimal model, S c It is a joint evaluation indicator.
[0124] This paper innovatively combines the differences between AIC and MSE to propose a new model evaluation index that comprehensively considers the model's goodness of fit, complexity, and the importance of parameter types;
[0125] In-depth analysis of dynamic parameters: Special emphasis is placed on the role of dynamic parameters in new energy power prediction, and their differences from static parameters are quantified through statistical methods.
[0126] The above description is a preferred embodiment of the invention and is not intended to limit the invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the invention should be included in the protection scope of the invention.
Claims
1. A method for predicting new energy power of a power grid combining static parameters and dynamic parameters, characterized in that: The following steps are involved: S1. Collect static parameters and dynamic parameters within a certain period of time, and pre-process the collected static parameters and dynamic parameters; S2. Use cross-validation method to perform statistical analysis and evaluation on the difference between static power parameters and dynamic power parameters after preprocessing; S3. Use the Akaike information criterion to jointly evaluate the difference between static power parameters and dynamic power parameters to determine the optimal power prediction model; S4. Utilize the determined optimal model to capture the fluctuation of renewable energy power and predict the renewable energy power of the power grid.
2. A method for predicting new energy power of a power grid combining static parameters and dynamic parameters according to claim 1, characterized in that: In step S2, the method of using the cross-validation method to perform statistical analysis on the difference between the trained static power parameters and the dynamic power parameters includes: S21. Cross-validation settings: divide the parameters obtained after preprocessing into two groups, namely, static parameter set and dynamic parameter set; set the cross-validation fold k, and divide the static parameter set and dynamic parameter set into k subsets: I1, I2, I3, ... I k , for each subset I i (i=1,2,...,n): Take the i-th subset as the test set, denoted as V i , except I i All subsets except i , train static parameter sets and dynamic parameters; S22. Cross-performance validation calculation, calculate the average MSE on all k-fold test sets CV ; S23. Difference statistics, calculate the MSE difference between static parameters and dynamic parameters.
3. The method for predicting new energy power of a power grid combining static parameters and dynamic parameters according to claim 2 is characterized in that: In step S21, the method for training the static parameter set and the dynamic parameter set includes: Train the model using the training set T i The model is trained based on the data points on the dataset. Assume that the model is a function M that fits the training data through parameters θ: M(X; θ), where X is the input feature set (including static and dynamic parameters), and θ is the model parameter. Parameter estimation, estimate the parameter θ through optimization problem, and use the loss function L to measure the difference between the model prediction and the actual output of the training set. Specifically, find a set of θ values that minimizes the sum of the squares of L. The formula is: L is the loss function, and y is the true output in the test set; Model evaluation, using the trained model M and parameters θ i To predict V i The output of , using mean squared error to calculate the performance metric: y is the true output in the test set, L is the loss function, |V i | is the size of the test set.
4. The method for predicting new energy power of a power grid combining static parameters and dynamic parameters according to claim 2 is characterized in that: In step S22, the average MSE on all k-fold test sets is calculated. CV The calculation formula is:
5. The method for predicting new energy power of a power grid combining static parameters and dynamic parameters according to claim 2 is characterized in that: In step S2, the method of evaluating static power parameters and dynamic power parameters using a cross-validation method includes: Using the training set T i The model is trained based on static parameters only, and the model parameters θ are obtained. static ; In the test set V i Using θ static Test and get the predicted value Using the training set T i The model is trained based on static parameters only, and the model parameters θ are obtained. dynamic ; In the test set V i Using θ dynamic Test and get the predicted value Use mean squared error to evaluate the model: Among them, Y ij is the true power value of the jth sample in the i-th fold test set, and are the predicted power values of the static and dynamic models, respectively; β0 is the intercept term, β1, β2, ···β m are model parameters, corresponding to different static characteristics, X static,ij1 , X static,ij2 ,···X static,ijn is the static eigenvalue of the jth sample in the i-th fold test set; X static,ijn represents a specific static feature, γ0 is the intercept term, γ1, γ2, ···γ n are model parameters, corresponding to different dynamic characteristics, X dynamic,ij1 , X dynamic,ij2 ,···X dynamic,ijn is the dynamic eigenvalue of the jth sample in the i-th fold test set, X dynamic,ijn Represents specific dynamic characteristics; Performance evaluation, calculate the average MSE of all k tests to get the average performance index of the model: Comparing MSE CV,dynamic and MSE CV,static To evaluate which type of parameters contributes more to the model prediction performance.
6. A method for predicting new energy power of a power grid combining static parameters and dynamic parameters according to claim 2, characterized in that: In step S23, the formula for calculating the MSE difference between the static parameter model and the dynamic parameter model is: D i =|MSE statiic -MSE dynamic |; MSE static is the average MSE calculated in cross validation for models trained using only static parameters, MSE dynamic is the average MSE calculated in cross validation for models trained using only dynamic parameters. If D i The value of is positive and large, indicating that the prediction accuracy of the static parameter model is lower than that of the dynamic parameter model; if D i The value of is negative and large, indicating that the prediction accuracy of the static parameter model is higher than that of the dynamic parameter model; if D i is close to zero, indicating that the prediction accuracy of the two models is similar.
7. The method for predicting power of new energy sources in a power grid by combining static parameters and dynamic parameters according to claim 1, characterized in that: In step S3, the step of jointly evaluating the difference between the prediction result of each model and the static parameters and the dynamic parameters using the Akaike Information Criterion is as follows: S31. Akaike Information Criterion AIC calculation; AIC calculation is performed for each model; S32. Model evaluation and selection: For each model, AIC and difference statistics are used for joint evaluation; S33. Determine the optimal model based on the results of the joint evaluation.
8. A method for predicting new energy power of a power grid combining static parameters and dynamic parameters according to claim 7, characterized in that: In step S31, the Akaike Information Criterion AIC calculation formula is: where τ is the number of model parameters, is the maximum likelihood estimate of the model.
9. The method for predicting power of new energy sources in a power grid by combining static parameters and dynamic parameters according to claim 1, characterized in that: In step S32, AIC and difference statistics are used for joint evaluation, and the joint evaluation index is defined as: S c =AIC+λ·D i ; Where AIC is the Akaike information criterion, which is used to measure the goodness of fit and complexity of the model; λ is a weight parameter used to balance the importance of the difference between AIC and MSE; D i is the difference in MSE between the static parameter model and the dynamic parameter model.
10. The method for predicting new energy power of a power grid combining static parameters and dynamic parameters according to claim 1, characterized in that: In step S33, the optimal model is determined as: M * =argmin M (S c ); Where M * represents the optimal model, S c It is a joint evaluation indicator.