Regional generating capacity prediction method and prediction system

Abstract

The invention discloses a regional power generation capacity prediction method and prediction system. The method comprises the steps of obtaining preprocessed historical power generation time sequence data of a to-be-predicted region and regional meteorological data of a prediction time period; inputting the historical power generation time sequence data into a trained time sequence feature extraction model based on a self-attention mechanism, and extracting to obtain advanced time sequence features; according to the advanced time sequence features, fusing the advanced time sequence features with regional meteorological data of a prediction time period to obtain fused features; and inputting the fusion features into a trained generating capacity prediction model to obtain a regional generating capacity prediction value in a prediction time period. According to the method, accurate and reliable prediction of the generating capacity of a future region is realized, particularly, the prediction robustness under the condition that the meteorological condition fluctuates greatly is improved, a key technical guarantee is provided for safe and stable operation of a power system, and the method has a good industrial application prospect.

Description

Technical Field

[0001] This invention relates to a regional power generation prediction method and system, belonging to the field of power generation prediction technology. Background Technology

[0002] Power generation forecasting, a core component of power system operation and planning, aims to accurately estimate the next day's power generation by analyzing historical data and factors such as current weather conditions. Accurate power generation forecasting contributes to the safety and stability of the power grid. For complex energy systems, especially those with renewable energy sources, power generation exhibits strong intermittency, volatility, and uncertainty, rendering traditional forecasting methods inaccurate. Therefore, the development of high-precision power generation forecasting technology is crucial for optimizing unit allocation and ensuring a balance between power supply and demand.

[0003] Existing technologies for predicting power generation suffer from drawbacks such as slow training speed, difficulty in parallelization, and the risk of overfitting due to excessive model complexity. These shortcomings typically make it difficult for models to fully learn all the key patterns affecting power generation, thus hindering further improvements in prediction accuracy. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art and provide a regional power generation prediction method and prediction system. It integrates high-level time series features extracted by the time series feature extraction model with regional meteorological data for the prediction period, and then makes predictions through the power generation prediction model. This achieves accurate and reliable prediction of future regional power generation, especially improving the prediction robustness under conditions of large fluctuations in meteorological conditions. It provides key technical support for the safe and stable operation of the power system and has good prospects for industrial application.

[0005] To achieve the above objectives, the present invention is implemented using the following technical solution: On one hand, this invention discloses a method for predicting regional power generation, comprising the following steps: Obtain preprocessed historical power generation time series data for the area to be predicted, as well as regional meteorological data for the prediction period; The historical power generation time series data is input into a trained time series feature extraction model based on a self-attention mechanism to extract high-level time series features; Based on the advanced time-series features, a fusion operation is performed with the regional meteorological data for the predicted time period to obtain fused features; The fused features are input into the trained power generation prediction model to obtain the regional power generation prediction value for the prediction period.

[0006] Furthermore, the steps for obtaining the preprocessed historical power generation time-series data of the region to be predicted are as follows: Obtain the original historical power generation time series data for the region to be predicted; Identify missing values ​​in the original historical power generation time series data and perform imputation operations to obtain historical power generation time series data after missing value processing; Identify outliers in the historical power generation time series data after missing value processing and perform correction operations to obtain historical power generation time series data after outlier processing. Based on the historical power generation time series data after outlier processing, a standardization operation is performed to obtain preprocessed historical power generation time series data.

[0007] Furthermore, the original historical power generation time series data includes hydropower generation data, new energy power generation data, imported power data, total regional power generation data, and corresponding time series labels for preset historical time periods.

[0008] Furthermore, the filling operation includes linear interpolation based on time series, filling with the previous valid value, or filling with the mean. The correction operations include mean replacement or boundary value truncation. The standardization process includes normalization or Z-Score standardization.

[0009] Furthermore, the time-series feature extraction model based on the self-attention mechanism includes: The data input unit is used to receive preprocessed historical power generation time-series data; The location encoding unit is used to embed and encode the preprocessed historical power generation time series data according to time series labels to form a feature vector sequence containing time series information. The feature encoding unit is used to obtain high-level temporal features that integrate contextual information of all time steps by parallel computing the attention weight matrix of the global sequence based on the feature vector sequence containing temporal information. The data output unit is used to output the advanced timing features.

[0010] Furthermore, the power generation prediction model adopts a gradient boosting decision tree ensemble model; The gradient boosting decision tree ensemble model includes multiple decision tree base predictors and an additive ensemble module. The additive ensemble module is communicatively connected to the multiple decision tree base predictors and is used to accumulate the preliminary prediction scores output by the multiple decision tree base predictors to obtain the final regional power generation prediction value.

[0011] Furthermore, the regional meteorological data for the predicted time period includes ambient temperature, horizontal wind speed, wind direction angle, and total horizontal solar irradiance on the land surface; The fusion operation includes vector concatenation or feature concatenation to form multidimensional fused features.

[0012] Furthermore, the training steps for the time-series feature extraction model and the power generation prediction model are as follows: Obtain a training set, which includes preprocessed training historical power generation time series data of the area to be predicted, meteorological data of the training area during the prediction period, and corresponding actual power generation labels. Based on the training set, the pre-constructed time-series feature extraction model and power generation prediction model are jointly trained end-to-end to obtain the trained time-series feature extraction model and the trained power generation prediction model. The end-to-end joint training steps are as follows: The training historical power generation time series data is input into a pre-built time series feature extraction model, and the predicted training high-level time series features are obtained through forward propagation calculation. Based on the high-level time-series training features, a fusion operation is performed with the meteorological data of the training area during the prediction time period to obtain the training fusion features; The training fusion features are input into the pre-built power generation prediction model to obtain the predicted power generation value of the training area for the prediction period. Calculate the evaluation index between the predicted power generation value and the actual power generation label of the training area; based on the evaluation index, execute the backpropagation algorithm to calculate the gradient of all parameters of the time series feature extraction model and the power generation prediction model; A gradient clipping operation is performed on the gradient based on a preset gradient threshold to obtain the clipped gradient; Based on the adaptive moment estimation optimizer, the model parameters are iteratively trained and updated according to the pruned gradient; wherein, during the parameter update process, the learning rate of the adaptive moment estimation optimizer is dynamically adjusted according to a preset learning rate scheduling strategy; The process continues until the preset iteration termination condition is met, resulting in a trained time-series feature extraction model and a trained power generation prediction model.

[0013] Furthermore, the adaptive moment estimation optimizer includes an Adam optimizer or an AdamW optimizer; The evaluation metrics include root mean square error and mean absolute percentage error.

[0014] On the other hand, the present invention discloses a regional power generation prediction system, applicable to the above-mentioned regional power generation prediction method, comprising: The data acquisition module is used to acquire preprocessed historical power generation time series data of the area to be predicted, as well as regional meteorological data for the prediction period. The feature extraction module is used to input the historical power generation time series data into a trained time series feature extraction model based on a self-attention mechanism to extract high-level time series features. The feature stitching module is used to perform a fusion operation with the regional meteorological data of the predicted time period based on the advanced temporal features to obtain fused features; The power generation prediction module is used to input the fused features into the trained power generation prediction model to obtain the regional power generation prediction value for the prediction period.

[0015] Compared with the prior art, the beneficial effects achieved by the present invention are as follows: The regional power generation forecasting method and system of this invention firstly employs a time-series feature extraction model based on a self-attention mechanism, overcoming the inherent defects of traditional models in terms of long-term dependence. This model can extract more informative and discriminative high-level time-series features from historical data. Secondly, it integrates the deeply extracted high-level time-series features with regional meteorological data for the forecast period, fully considering meteorological conditions that have a decisive impact on power generation. This achieves cross-modal complementarity between internal operational data and external environmental data, avoiding the problem of inaccurate predictions during sudden weather changes caused by relying solely on historical data, and significantly improving the accuracy and robustness of the forecast results. Finally, the power generation forecasting model achieves accurate and reliable predictions of future regional power generation, especially improving the robustness of predictions under conditions of large fluctuations in meteorological conditions. This provides key technical support for the safe and stable operation of the power system and has good prospects for industrial application. Attached Figure Description

[0016] Figure 1 This is a flowchart of the regional power generation prediction method provided in Embodiment 1 of the present invention.

[0017] Figure 2 This is a prediction result diagram of the regional power generation prediction method provided in Embodiment 1 of the present invention; Figure 3 This is a prediction result diagram of the convolutional neural network-long short-term memory network hybrid model provided in Embodiment 1 of the present invention; Figure 4 This is a prediction result diagram of the time-series feature extraction model based on self-attention mechanism provided in Embodiment 1 of the present invention; Figure 5 This is a prediction result diagram of the gradient boosting decision tree ensemble model provided in Embodiment 1 of the present invention. Detailed Implementation

[0018] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention. Example 1

[0019] This embodiment 1 provides a method for predicting regional power generation, such as Figure 1 As shown, it includes the following steps: Obtain preprocessed historical power generation time series data for the area to be predicted, as well as regional meteorological data for the prediction period; Historical power generation time series data are input into a trained time series feature extraction model based on a self-attention mechanism to extract high-level time series features; Based on the advanced time series features, a fusion operation is performed with the regional meteorological data for the prediction time period to obtain the fused features; The fused features are input into the trained power generation prediction model to obtain the regional power generation prediction value for the prediction period.

[0020] The design concept of this invention is as follows: First, a time-series feature extraction model based on a self-attention mechanism is adopted to overcome the inherent defects of traditional models in terms of long-term dependence, enabling the extraction of more informative and discriminative high-level time-series features from historical data. Second, the deeply extracted high-level time-series features are fused with regional meteorological data for the prediction period, fully considering meteorological conditions that have a decisive impact on power generation. This achieves cross-modal complementarity between internal operational data and external environmental data, avoiding the problem of inaccurate predictions during sudden weather changes caused by relying solely on historical data, and significantly improving the accuracy and robustness of the prediction results. Finally, the power generation prediction model achieves accurate and reliable prediction of future regional power generation, especially improving the prediction robustness under conditions of large fluctuations in meteorological conditions, providing key technical support for the safe and stable operation of the power system, and has good prospects for industrial application.

[0021] The specific steps are as follows: Step 1: Obtain the preprocessed historical power generation time series data of the area to be predicted, as well as the regional meteorological data for the prediction period.

[0022] Regional meteorological data for the forecast period include ambient temperature, horizontal wind speed, wind direction angle, and total horizontal solar irradiance at the surface. Specifically, the steps for obtaining preprocessed historical power generation time-series data for the region to be predicted are as follows: 1.1 Obtain the original historical power generation time series data of the area to be predicted.

[0023] The original historical power generation time series data includes hydropower generation data, new energy power generation data, imported power data, total regional power generation data, and corresponding time series labels for preset historical time periods.

[0024] In this embodiment, the preset historical time period is selected as the 14 days prior to the predicted time period.

[0025] 1.2 Identify missing values ​​in the original historical power generation time series data and perform imputation operations to obtain historical power generation time series data after missing value processing.

[0026] The filling operation includes time-series-based linear interpolation, previous valid value filling, or mean filling. Preferably, in this embodiment, time-series-based linear interpolation is used.

[0027] 1.3 Identify outliers in the historical power generation time series data after missing value processing and perform correction operations to obtain the historical power generation time series data after outlier processing.

[0028] Correction operations include mean replacement or boundary value truncation. Preferably, this embodiment uses boundary value truncation, which truncates data using boundary values ​​within the valid data range defined by the Raida criterion.

[0029] 1.4. Based on the historical power generation time series data after outlier processing, a standardization operation is performed to obtain preprocessed historical power generation time series data.

[0030] Standardization operations include normalization or Z-score standardization. Preferably, this embodiment uses Z-score standardization to ensure that all data points are of the same order of magnitude. Step 2: Input the historical power generation time series data into the trained time series feature extraction model based on the self-attention mechanism to extract high-level time series features.

[0031] Specifically, time-series feature extraction models based on self-attention mechanisms include: The data input unit is used to receive preprocessed historical power generation time-series data; The location encoding unit is used to embed and encode the preprocessed historical power generation time series data according to the time series label, forming a feature vector sequence containing time series information. The feature encoding unit is used to obtain high-level temporal features that incorporate contextual information from all time steps by computing the global attention weight matrix of the sequence in parallel based on the feature vector sequence containing temporal information. The data output unit is used to output advanced time-series features.

[0032] In this embodiment, the temporal feature extraction model has a model dimension of 32, and the feature encoding unit contains only one encoder layer. This encoder layer employs a multi-head self-attention mechanism, with two attention heads. Within the feature encoding unit, the feature vector sequence is first aggregated into a single, globally representative 32-dimensional feature vector through global average pooling. This vector is then passed through a fully connected layer and mapped to a fixed 16-dimensional vector, serving as the final output high-level temporal feature.

[0033] Step 3: Based on the advanced time series features, perform a fusion operation with the regional meteorological data for the prediction time period to obtain the fused features.

[0034] Fusion operations include vector concatenation or feature concatenation to form multidimensional fused features.

[0035] Step 4: Input the fused features into the trained power generation prediction model to obtain the predicted power generation value for the region during the prediction period.

[0036] The power generation prediction model adopts a gradient boosting decision tree ensemble model; The gradient boosting decision tree ensemble model includes multiple decision tree base predictors and an additive ensemble module. The additive ensemble module is communicatively connected to the multiple decision tree base predictors and is used to accumulate the preliminary prediction scores output by the multiple decision tree base predictors to obtain the final regional power generation prediction value.

[0037] Furthermore, the training steps for the time-series feature extraction model and the power generation prediction model are as follows: Obtain the original dataset, which includes preprocessed training historical power generation time series data of the region to be predicted, meteorological data of the training region for the prediction period, and corresponding actual power generation labels.

[0038] The original dataset is split chronologically, with the first 90% used as the training set and the last 10% as the test set, thus strictly maintaining the continuity of the time series.

[0039] Based on the training set, the pre-built time-series feature extraction model and power generation prediction model are jointly trained end-to-end to obtain the trained time-series feature extraction model and the trained power generation prediction model.

[0040] The end-to-end joint training steps are as follows: The training historical power generation time series data is input into a pre-built time series feature extraction model, and the predicted high-level time series features are obtained through forward propagation calculation. Based on the high-level time-series features trained, they are fused with meteorological data from the training area during the prediction period to obtain the fused features trained. The training fusion features are input into the pre-built power generation prediction model to obtain the predicted power generation value of the training area for the prediction period. Calculate the evaluation index between the predicted power generation value and the actual power generation label in the training area; based on the evaluation index, execute the backpropagation algorithm to calculate the gradient of all parameters of the time series feature extraction model and the power generation prediction model; A gradient clipping operation is performed on the gradient based on a preset gradient threshold to obtain the clipped gradient. Based on the adaptive moment estimation optimizer, the model parameters are iteratively trained and updated according to the pruned gradient; during the parameter update process, the learning rate of the adaptive moment estimation optimizer is dynamically adjusted according to the preset learning rate scheduling strategy. The process continues until the preset iteration termination condition is met, resulting in a trained time-series feature extraction model and a trained power generation prediction model.

[0041] Furthermore, adaptive moment estimation optimizers include the Adam optimizer or the AdamW optimizer; The evaluation metrics include root mean square error and mean absolute percentage error.

[0042] In this embodiment, the preferred adaptive moment estimation optimizer is the Adam optimizer, and the evaluation metric is the root mean square error.

[0043] It should be noted that this invention employs an end-to-end joint training strategy, training all training models within the same framework. During training, the root mean square error loss function, Adam optimizer, gradient pruning, and learning rate scheduling are used. An early stopping mechanism is also implemented to prevent overfitting. Each training iteration can simultaneously process 16 time-series samples and use historical data from the past 14 days to predict the current target value.

[0044] In training the power generation prediction model, important variables such as hydropower generation and external power input are set with conservative parameters of low complexity, while the rest are set with relaxed parameters of medium complexity. L1 / L2 regularization is used to prevent overfitting.

[0045] To further verify the experimental effectiveness of the method of this invention, a multi-model method will be used for comparative verification on more than 10 target variables, including key indicators such as new energy power generation and total provincial power generation. The evaluation indicators used are root mean square error and mean absolute percentage error. Experimental results are as follows: Figures 2-5 As shown in Table 1, it is clear that the regional power generation prediction method proposed in this invention outperforms the comparative model method in both indicators, effectively demonstrating its superiority.

[0046] Table 1. Comparison of errors in regional total power generation under different prediction model methods Example 2

[0047] This embodiment 2 provides a regional power generation prediction system, applicable to the regional power generation prediction method of embodiment 1 above, including: The data acquisition module is used to acquire preprocessed historical power generation time series data of the area to be predicted, as well as regional meteorological data for the prediction period. The feature extraction module is used to input historical power generation time series data into a trained time series feature extraction model based on a self-attention mechanism to extract high-level time series features. The feature stitching module is used to fuse high-level time-series features with regional meteorological data for the predicted time period to obtain fused features; The power generation prediction module is used to input the fused features into the trained power generation prediction model to obtain the regional power generation prediction value for the prediction period.

[0048] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0049] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0050] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0051] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0052] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for predicting regional power generation, characterized in that, Includes the following steps: Obtain preprocessed historical power generation time series data for the area to be predicted, as well as regional meteorological data for the prediction period; The historical power generation time series data is input into a trained time series feature extraction model based on a self-attention mechanism to extract high-level time series features; Based on the advanced time-series features, a fusion operation is performed with the regional meteorological data for the predicted time period to obtain fused features; The fused features are input into the trained power generation prediction model to obtain the regional power generation prediction value for the prediction period.

2. The regional power generation prediction method according to claim 1, characterized in that, The steps for obtaining the preprocessed historical power generation time-series data of the region to be predicted are as follows: Obtain the original historical power generation time series data for the region to be predicted; Identify missing values ​​in the original historical power generation time series data and perform imputation operations to obtain historical power generation time series data after missing value processing; Identify outliers in the historical power generation time series data after missing value processing and perform correction operations to obtain historical power generation time series data after outlier processing. Based on the historical power generation time series data after outlier processing, a standardization operation is performed to obtain preprocessed historical power generation time series data.

3. The regional power generation prediction method according to claim 2, characterized in that, The original historical power generation time series data includes hydropower generation data, new energy power generation data, external power input data, regional total power generation data, and corresponding time series labels for preset historical time periods.

4. The regional power generation prediction method according to claim 2, characterized in that, The filling operation includes linear interpolation based on time series, filling with the previous valid value, or filling with the mean. The correction operations include mean replacement or boundary value truncation. The standardization process includes normalization or Z-Score standardization.

5. The regional power generation prediction method according to claim 1, characterized in that, The time-series feature extraction model based on the self-attention mechanism includes: The data input unit is used to receive preprocessed historical power generation time-series data; The location encoding unit is used to embed and encode the preprocessed historical power generation time series data according to time series labels to form a feature vector sequence containing time series information. The feature encoding unit is used to obtain high-level temporal features that integrate contextual information of all time steps by parallel computing the attention weight matrix of the global sequence based on the feature vector sequence containing temporal information. The data output unit is used to output the advanced timing features.

6. The regional power generation prediction method according to claim 1, characterized in that, The power generation prediction model adopts a gradient boosting decision tree ensemble model. The gradient boosting decision tree ensemble model includes multiple decision tree base predictors and an additive ensemble module. The additive ensemble module is communicatively connected to the multiple decision tree base predictors and is used to accumulate the preliminary prediction scores output by the multiple decision tree base predictors to obtain the final regional power generation prediction value.

7. The regional power generation prediction method according to claim 1, characterized in that, The regional meteorological data for the predicted time period includes ambient temperature, horizontal wind speed, wind direction angle, and total horizontal solar irradiance on the ground surface. The fusion operation includes vector concatenation or feature concatenation to form multidimensional fused features.

8. The regional power generation prediction method according to claim 1, characterized in that, The training steps for the time-series feature extraction model and the power generation prediction model are as follows: Obtain a training set, which includes preprocessed training historical power generation time series data of the area to be predicted, meteorological data of the training area during the prediction period, and corresponding actual power generation labels. Based on the training set, the pre-constructed time-series feature extraction model and power generation prediction model are jointly trained end-to-end to obtain the trained time-series feature extraction model and the trained power generation prediction model. The end-to-end joint training steps are as follows: The training historical power generation time series data is input into a pre-built time series feature extraction model, and the predicted training high-level time series features are obtained through forward propagation calculation. Based on the high-level time-series training features, a fusion operation is performed with the meteorological data of the training area during the prediction time period to obtain the training fusion features; The training fusion features are input into the pre-built power generation prediction model to obtain the predicted power generation value of the training area for the prediction period. Calculate the evaluation index between the predicted power generation value and the actual power generation label of the training area; based on the evaluation index, execute the backpropagation algorithm to calculate the gradient of all parameters of the time series feature extraction model and the power generation prediction model; A gradient clipping operation is performed on the gradient based on a preset gradient threshold to obtain the clipped gradient; Based on the adaptive moment estimation optimizer, the model parameters are iteratively trained and updated according to the pruned gradient; wherein, during the parameter update process, the learning rate of the adaptive moment estimation optimizer is dynamically adjusted according to a preset learning rate scheduling strategy; The process continues until the preset iteration termination condition is met, resulting in a trained time-series feature extraction model and a trained power generation prediction model.

9. The regional power generation prediction method according to claim 8, characterized in that, The adaptive moment estimation optimizer includes the Adam optimizer or the AdamW optimizer; The evaluation metrics include root mean square error and mean absolute percentage error.

10. A regional power generation prediction system, applicable to the regional power generation prediction method according to any one of claims 1-9, characterized in that, include: The data acquisition module is used to acquire preprocessed historical power generation time series data of the area to be predicted, as well as regional meteorological data for the prediction period. The feature extraction module is used to input the historical power generation time series data into a trained time series feature extraction model based on a self-attention mechanism to extract high-level time series features. The feature stitching module is used to perform a fusion operation with the regional meteorological data of the predicted time period based on the advanced temporal features to obtain fused features; The power generation prediction module is used to input the fused features into the trained power generation prediction model to obtain the regional power generation prediction value for the prediction period.

Images