Methods, devices, electronic equipment and storage media for predicting day-ahead electricity prices

Abstract

This invention provides a method, apparatus, electronic device, and storage medium for predicting day-ahead electricity prices, belonging to the field of electricity price prediction technology. The method includes: acquiring the province-wide dispatch load, renewable energy output, mandatory start / stop unit capacity, and tie-line planning information for any given time point in the power grid node where the power plant is located; inputting these information into three electricity price prediction models to obtain predicted values ​​for the three day-ahead electricity prices; and finally, calculating the weighted allocation ratio of the predicted values ​​based on a weighted allocation model to obtain the predicted day-ahead electricity price. This invention, by associating and fusing multiple electricity price prediction models and dynamically allocating weights, enables the fused model to accurately predict day-ahead electricity prices in the spot market, while avoiding the problems of lack of connection and singular weight allocation methods in existing model fusion processes. This provides a reliable basis for power generation companies to make day-ahead trading decisions.

Description

Technical Field

[0001] This invention relates to the field of electricity price forecasting technology, specifically to a day-ahead electricity price forecasting method, a day-ahead electricity price forecasting device, an electronic device, and a readable storage medium. Background Technology

[0002] As market-oriented reforms in my country's power industry deepen, market conditions are becoming increasingly diverse, and market competition is intensifying. Currently, some provinces have opened spot electricity trading markets, requiring power generation companies to submit day-ahead market bids the day before the transaction to secure their day-ahead power generation and increase revenue. Therefore, power generation companies need a method to quickly and accurately predict day-ahead electricity prices in the spot market. This will enable them to make precise judgments, formulate reasonable trading strategies, enhance their competitiveness, and maximize profits. Accurately predicting day-ahead electricity prices is a challenging but crucial task.

[0003] In recent years, the informatization level of various power generation enterprises has gradually improved, and the data has become more complete. Moreover, through long-term observation, it can be found that the day-ahead node electricity price in the spot electricity trading market is related to factors such as the central dispatch load, renewable energy output, the capacity of units that must be started and stopped, and the tie line plan, and follows certain patterns. Therefore, it is possible to use historical data of factors such as central dispatch load, renewable energy output, the capacity of units that must be started and stopped, and tie line plan, and to reasonably use machine learning algorithms to build an electricity price prediction model. This has become a common method. Furthermore, with the widespread application of the model fusion approach, an electricity price prediction fusion model has also been developed.

[0004] However, existing electricity price forecasting fusion models mostly allocate weights based on the algorithm's performance in the sample rather than through dynamic weight allocation; and during model fusion, the models are independent of each other with little correlation. These two points, from different perspectives, contribute to the inaccuracy of day-ahead electricity prices predicted by existing electricity price forecasting fusion models, making it impossible to provide a reliable basis for power generation companies to make day-ahead trading decisions. Summary of the Invention

[0005] The purpose of this invention is to provide a method, apparatus, electronic device, and storage medium for predicting day-ahead electricity prices in the spot electricity trading market, in order to solve the problem that the day-ahead electricity prices predicted by existing electricity price prediction fusion models are not accurate enough, making it impossible to provide a reliable basis for power generation companies to make day-ahead trading decisions.

[0006] To achieve the above objectives, embodiments of the present invention provide a day-ahead nodal electricity price forecasting method, comprising:

[0007] Obtain information on the province's overall dispatch load, new energy output, mandatory start-up and mandatory shutdown unit capacity, and tie line plans at any given time point in the power grid node where the power plant is located;

[0008] The information on the centrally dispatched load, new energy output, capacity of units that must be started or stopped, and tie line plans is input into the first electricity price prediction model to obtain the predicted value of the first day-ahead node electricity price.

[0009] Input the centrally dispatched load, new energy output, capacity of units that must be turned on or off, tie line planning information, and the predicted value of the first day-ahead node electricity price into the second electricity price prediction model to obtain the predicted value of the second day-ahead node electricity price.

[0010] The centrally dispatched load, new energy output, capacity of units that must be started or stopped, tie line planning information, and the predicted value of the electricity price at the second day's node are input into the third electricity price prediction model to obtain the predicted value of the electricity price at the third day's node.

[0011] Based on the weighted allocation model, the weighted allocation ratios of the predicted values ​​of the first, second, and third day-ahead electricity prices are calculated, and the weighted values ​​are then used to obtain the predicted day-ahead electricity prices.

[0012] Optionally, before the step of inputting the centrally dispatched load, renewable energy output, mandatory start-up and mandatory shutdown unit capacity, and tie-line plan information into the first electricity price prediction model to obtain the predicted value of the first day-ahead nodal electricity price, the method further includes:

[0013] The information on the centrally dispatched load, new energy output, capacity of units that must be started or stopped, and tie line plans is preprocessed; wherein, the preprocessing includes noise reduction processing and normalization processing.

[0014] Optionally, the step of calculating the weighted distribution ratio of the predicted values ​​of the first, second, and third day-ahead electricity prices based on the weighted allocation model, and then weighting them to obtain the predicted day-ahead electricity price, includes:

[0015] The information on the centrally dispatched load, new energy output, mandatory start-up and mandatory shutdown unit capacity, and tie line plan is input into the weight allocation model to obtain the weight allocation combination; wherein, the weight allocation combination includes the weight value of the predicted value of the first day-ahead node electricity price, the weight value of the predicted value of the second day-ahead node electricity price, and the weight value of the predicted value of the third day-ahead node electricity price.

[0016] Based on the weighted combination, the weighted average of the predicted values ​​of the first, second, and third day-ahead node electricity prices is calculated to obtain the predicted value of the day-ahead node electricity price.

[0017] Optionally, the first electricity price prediction model is obtained in the following way:

[0018] Obtain sample data; wherein, the sample data includes the province's overall dispatch load, new energy output, mandatory start-up and mandatory shutdown unit capacity, tie line planning information and day-ahead node electricity price at any historical point in time of the power grid node where the power plant is located;

[0019] The pre-constructed random forest model is trained using the province's overall dispatch load, new energy output, mandatory start-up and shutdown unit capacity, and tie line plan information at any historical point in time. The first electricity price prediction model is obtained when the electricity price predicted by the pre-constructed random forest model is equal to the province's day-ahead node electricity price at any historical point in time.

[0020] Optionally, the second electricity price prediction model is obtained in the following way:

[0021] Obtain sample data; wherein, the sample data includes the province's overall dispatch load, new energy output, mandatory start-up and mandatory shutdown unit capacity, tie line planning information and day-ahead node electricity price at any historical point in time of the power grid node where the power plant is located;

[0022] The pre-built extreme gradient boosting model is trained using the province's overall dispatch load, new energy output, mandatory start-up and shutdown unit capacity, tie line planning information, and the electricity price predicted by the first electricity price prediction model at any historical time point. When the electricity price predicted by the pre-built extreme gradient boosting model is equal to the province's day-ahead node electricity price at any historical time point, the second electricity price prediction model is obtained.

[0023] Optionally, the third electricity price prediction model is obtained in the following way:

[0024] Obtain sample data; wherein, the sample data includes the province's overall dispatch load, new energy output, mandatory start-up and mandatory shutdown unit capacity, tie line planning information and day-ahead node electricity price at any historical point in time of the power grid node where the power plant is located;

[0025] The pre-constructed multiple linear regression model is trained using the province's overall dispatch load, new energy output, mandatory start-up and shutdown unit capacity, tie line planning information, and the electricity price predicted by the second electricity price prediction model at any historical time point. The third electricity price prediction model is obtained when the electricity price predicted by the pre-constructed multiple linear regression model is equal to the province's day-ahead node electricity price at any historical time point.

[0026] Optionally, the weight allocation model is obtained in the following way:

[0027] Obtain sample data; wherein, the sample data includes the province's overall dispatch load, new energy output, mandatory start-up and mandatory shutdown unit capacity, tie line planning information and day-ahead node electricity price at any historical point in time of the power grid node where the power plant is located;

[0028] According to the preset weight calculation accuracy, the first electricity price prediction model, the second electricity price prediction model and the third electricity price prediction model are weighted and assigned to obtain multiple weight assignment combinations;

[0029] According to the multiple weight allocation combinations, the weighted average of the electricity price predicted by the first electricity price prediction model, the electricity price predicted by the second electricity price prediction model, and the electricity price predicted by the third electricity price prediction model is calculated to obtain multiple predicted electricity prices;

[0030] The predicted electricity price with the smallest difference from the provincial day-ahead node electricity price at any historical point in time is selected from the multiple predicted electricity prices, and the weight allocation combination corresponding to the predicted electricity price with the smallest difference is taken as the optimal weight allocation combination.

[0031] The pre-constructed multiple linear regression model is trained using the province's overall dispatch load, new energy output, mandatory start-up and shutdown unit capacity, and tie line planning information at any historical point in time. The weight allocation model is obtained when the weight allocation combination predicted by the pre-constructed multiple linear regression model is equal to the optimal weight allocation combination.

[0032] In a second aspect of the present invention, a day-ahead nodal electricity price forecasting device is provided, comprising:

[0033] The data acquisition module is used to acquire the province's overall dispatch load, new energy output, mandatory start-up and mandatory shutdown unit capacity, and tie line planning information at any point in time at the power grid node where the power plant is located.

[0034] The first calculation module is used to input the unified dispatch load, new energy output, capacity of units that must be turned on or off, and tie line plan information into the first electricity price prediction model to obtain the predicted value of the first day-ahead node electricity price.

[0035] The second calculation module is used to input the centrally dispatched load, new energy output, capacity of units that must be turned on or off, tie line planning information and the predicted value of the first day-ahead node electricity price into the second electricity price prediction model to obtain the predicted value of the second day-ahead node electricity price.

[0036] The third calculation module is used to input the centrally dispatched load, new energy output, capacity of units that must be turned on or off, tie line planning information, and the predicted value of the electricity price at the second day's node into the third electricity price prediction model to obtain the predicted value of the electricity price at the third day's node.

[0037] The comprehensive calculation module is used to calculate the weight allocation ratio of the predicted values ​​of the first, second, and third day-ahead electricity prices based on the weight allocation model, and then obtain the predicted value of the day-ahead electricity price after weighting.

[0038] In a third aspect of the present invention, an electronic device is provided, comprising: a processor and a memory, the memory storing machine-readable instructions executable by the processor, wherein the machine-readable instructions, when executed by the processor, perform the day-ahead electricity price forecasting method described above.

[0039] In a fourth aspect of the present invention, a computer-readable storage medium is provided, on which instructions are stored for causing a machine to perform the day-ahead electricity price forecasting method as described above.

[0040] In this embodiment of the invention, by acquiring the province-wide dispatch load, new energy output, mandatory start-up and mandatory shutdown unit capacity, and tie-line plan information at any point in time at the power grid node where the power plant is located, and then inputting the first, second, and third electricity price prediction models, the predicted values ​​of the first, second, and third day-ahead node electricity prices are obtained. Based on the weight allocation model, the weights of the predicted values ​​of the first, second, and third day-ahead node electricity prices are calculated, and the final day-ahead node electricity price prediction value is obtained after weighted averaging. This realizes the correlation and fusion of multiple electricity price prediction models and the dynamic allocation of weights, so that the fused model can not only accurately predict the day-ahead node electricity price in the spot market, but also avoids the problems of lack of connection and single weight allocation method in the model fusion process of the prior art, thus providing a reliable basis for power generation companies to make day-ahead trading decisions.

[0041] Other features and advantages of the embodiments of the present invention will be described in detail in the following detailed description section. Attached Figure Description

[0042] The accompanying drawings are provided to further illustrate embodiments of the present invention and form part of the specification. They are used together with the following detailed description to explain the embodiments of the present invention, but do not constitute a limitation thereof. In the drawings:

[0043] Figure 1 This is a schematic diagram illustrating the construction principle of the composite electricity price prediction model provided in this invention.

[0044] Figure 2 This is a schematic diagram of the first process of the day-ahead node electricity price forecasting method provided in an embodiment of the present invention;

[0045] Figure 3 This is a schematic diagram of the second process of the day-ahead node electricity price forecasting method provided in an embodiment of the present invention;

[0046] Figure 4 This is a flowchart illustrating the training process of the first electricity price prediction model provided in an embodiment of the present invention;

[0047] Figure 5 This is a flowchart illustrating the training process of the second electricity price prediction model provided in an embodiment of the present invention;

[0048] Figure 6 This is a flowchart illustrating the training process of the third electricity price prediction model provided in an embodiment of the present invention;

[0049] Figure 7 This is a flowchart illustrating the training process of the weight allocation model provided in an embodiment of the present invention;

[0050] Figure 8 This is a schematic diagram of the day-ahead electricity price forecasting device provided in an embodiment of the present invention. Detailed Implementation

[0051] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit the scope of the present invention.

[0052] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs; the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit this application.

[0053] In the description of the embodiments of this application, technical terms such as "first" and "second" are used only to distinguish different objects and should not be construed as indicating or implying relative importance or implicitly specifying the number, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly defined.

[0054] To facilitate understanding of the inventive concept of this invention, the following detailed description is provided:

[0055] The concept of this invention is based on model fusion and dynamic weight allocation to construct a composite electricity price prediction model. The overall construction method of the composite electricity price prediction model is as follows: Figure 1 As shown, during model application, the input conditions for the model are the province-wide dispatch load, renewable energy output, mandatory start-up and shutdown unit capacity, and tie-line plan information at any given time point in the power grid node where the power plant is located. The model outputs the final predicted day-ahead node electricity price. During the model training phase, the input conditions for the model are the province-wide dispatch load, renewable energy output, mandatory start-up and shutdown unit capacity, and tie-line plan information at any given historical time point in the power grid node where the power plant is located. The output condition for the model is the province-wide day-ahead node electricity price at any given historical time point. This is used to train the model.

[0056] Please refer to Figure 2 , Figure 2 This is a schematic diagram of the first process of the day-ahead nodal electricity price forecasting method provided in an embodiment of the present invention. The method includes the following steps:

[0057] S100 obtains information on the province's unified dispatch load, new energy output, mandatory start-up and mandatory shutdown unit capacity, and tie line planning at any point in time at the power grid node where the power plant is located;

[0058] Day-ahead trading refers to a type of trading in the spot electricity market, used to trade the electricity volume of the current day one day in advance.

[0059] It should be noted that in provinces and regions where the spot electricity trading market is open, information on the overall dispatch load, renewable energy output, mandatory start / stop unit capacity, and tie line plans for each node is publicly disclosed through the respective provincial electricity trading platforms. However, the information on overall dispatch load, renewable energy output, mandatory start / stop unit capacity, and tie line plans is time-of-use information, meaning that the day is divided into multiple time points. In other words, the trading platform publishes the overall dispatch load, renewable energy output, mandatory start / stop unit capacity, and tie line plans for the entire province at multiple time points on the same day. The information on the overall dispatch load, renewable energy output, mandatory start / stop unit capacity, and tie line plans for the entire province is different at each time point.

[0060] A node refers to a node in the power grid where a power plant is located.

[0061] A time point refers to dividing the day preceding a trading day into multiple time periods proportionally, with the starting point of each time period being the time point corresponding to the day-ahead electricity price. These proportional divisions include, but are not limited to, 24, 48, and 96 minutes. For example: 1440 minutes (24h) ÷ 24 = 60 minutes, meaning each 60-minute unit contains time points including: 0:00, 1:00, 2:00, ..., 22:00 and 23:00; 1440 minutes (24h) ÷ 48 = 30 minutes, meaning each 30-minute unit contains time points including: 0:00, 0:30, 1:00, 1:30, ..., 22:30, 23:00 and 23:30; 1440 minutes (24h) ÷ 96 = 15 minutes, meaning each 15-minute unit contains time points including: 0:00, 0:15, 0:30, 1:00, ..., 23:15, 23:30 and 23:45.

[0062] Based on the above, this embodiment provides a day-ahead node price forecasting method, specifically a day-ahead node price forecasting method for any point in time at the power grid node where the power plant is located, applied to the spot electricity trading market to assist power generation companies in making day-ahead trading decisions.

[0063] It should be noted that the subsequent training and prediction of the model are carried out at different time points, that is, the electricity price is predicted at multiple time points for the power grid node where the power plant is located. After predicting the electricity price at all time points, the predicted value of the day-ahead node electricity price is finally obtained.

[0064] The load under unified dispatch refers to the load of the entire province at a certain moment, for example: 37669MW.

[0065] New energy power output refers to the total power output of all new energy power plants in the province at a certain moment, such as 5656MW.

[0066] The mandatory operating and mandatory shut-off capacity refers to the total capacity of all mandatory operating and mandatory shut-off units in the province at a certain moment, such as a mandatory operating capacity of 11,770 MW and a mandatory shut-off capacity of 3,424 MW.

[0067] Inter-provincial transaction link planning information refers to the total planned transaction volume of inter-provincial transaction links at a certain moment, such as 1000MW.

[0068] S200 inputs the information on the centrally dispatched load, new energy output, capacity of units that must be started or stopped, and tie line planning into the first electricity price prediction model to obtain the predicted value of the first day-ahead node electricity price;

[0069] The first electricity price prediction model is a model obtained after training the random forest model. The training of the first electricity price prediction model will be explained in detail later, and will not be elaborated here.

[0070] S300 inputs the centrally dispatched load, renewable energy output, capacity of units that must be started or stopped, tie line planning information, and the predicted value of the first day-ahead node price into the second electricity price prediction model to obtain the predicted value of the second day-ahead node price.

[0071] It should be noted that, in order to address the issue of model independence, the predicted day-ahead electricity price output by the first electricity price prediction model is used as the input to the second electricity price prediction model. This establishes a correlation between the models, thereby improving the overall prediction accuracy. Similarly, using the predicted day-ahead electricity price output by the second electricity price prediction model as the input to the third electricity price prediction model is also for the purpose of establishing a correlation between the models; this will not be elaborated upon further.

[0072] The second electricity price prediction model is obtained by training the extreme gradient boosting model. The training of the second electricity price prediction model will be explained in detail later, and will not be elaborated here.

[0073] S400 inputs the centrally dispatched load, new energy output, capacity of units that must be started or stopped, tie line planning information, and the predicted value of the second-day-ahead node price into the third electricity price prediction model to obtain the predicted value of the third-day-ahead node price.

[0074] The third electricity price prediction model is a model obtained by training a multiple linear regression model. The training of the third electricity price prediction model will be explained in detail later, and will not be elaborated here.

[0075] S500, based on a weighted allocation model, calculates the weighted allocation ratio of the predicted values ​​of the first, second, and third day-ahead electricity prices, and then calculates the predicted values ​​of the day-ahead electricity prices after weighting.

[0076] It should be noted that, in order to address the problem that the weight allocation of each model in the existing technology is often based solely on the algorithm's performance in the samples, resulting in unsatisfactory prediction results, a weight allocation model is used to dynamically allocate weights to the first, second, and third electricity price prediction models, thereby improving the accuracy of the final day-ahead electricity price prediction value.

[0077] The predicted day-ahead electricity price refers to the price of electricity traded one day in advance for the current day's electricity volume.

[0078] The weight allocation model is a model obtained by training a multiple linear regression model. The training of the weight allocation model will be explained in detail later and will not be elaborated here.

[0079] It is worth mentioning that, as mentioned above, this embodiment predicts the day-ahead electricity price for any given point in time at the power grid node where the power plant is located. However, predicting the electricity price for each different point in time requires inputting the corresponding central dispatch load, renewable energy output, mandatory on / off unit capacity, and tie line plan information for each different point in time into three separate electricity price prediction models trained using these same data. A weight allocation model is then used to assign weights to these models to obtain the predicted electricity price values ​​for each point in time. This is different from inputting the same first, second, and third electricity price prediction models and weight allocation model into the same model to predict the electricity price values ​​for different points in time. In other words, while the algorithms for the first, second, and third electricity price prediction models and the weight allocation model are the same for different points in time, they each need to be trained separately.

[0080] Specifically, the province-wide dispatch load, renewable energy output, mandatory start-up and shutdown unit capacity, and tie-line plan information for any given time point in the power grid node where the power plant is located are input into the first electricity price prediction model to obtain the predicted value for the first day-ahead node. Then, the obtained dispatch load, renewable energy output, mandatory start-up and shutdown unit capacity, tie-line plan information, and the predicted value for the first day-ahead node are input into the second electricity price prediction model to obtain the predicted value for the electricity price for the second day-ahead node. Then, the obtained dispatch load, renewable energy output, mandatory start-up and shutdown unit capacity, tie-line plan information, and the predicted value for the electricity price for the second day-ahead node are input into the third electricity price prediction model to obtain the predicted value for the electricity price for the third day-ahead node. Then, a weight allocation model is used to assign weights to the first, second, and third electricity price prediction models respectively. Finally, the weighted average of the predicted values ​​for the first, second, and third day-ahead node electricity prices is calculated based on the weights assigned to the three electricity price prediction models to obtain the predicted value for the day-ahead node electricity price.

[0081] In this embodiment, by acquiring the province-wide dispatch load, renewable energy output, mandatory start-up and mandatory shutdown unit capacity, and tie-line plan information at any given time point in the power grid node where the power plant is located, and then inputting the first, second, and third electricity price prediction models, the predicted values ​​of the first, second, and third day-ahead node electricity prices are obtained. Based on the weight allocation model, the weights of the predicted values ​​of the first, second, and third day-ahead node electricity prices are calculated, and the final day-ahead node electricity price prediction value is obtained after weighted averaging. This realizes the correlation and fusion of multiple electricity price prediction models and the dynamic allocation of weights, so that the fused model can not only accurately predict the day-ahead node electricity price in the spot market, but also avoids the problems of lack of connection and single weight allocation method in the model fusion process of the prior art, thus providing a reliable basis for power generation companies to make day-ahead trading decisions.

[0082] Optionally, the above step S200 may be preceded by:

[0083] Preprocessing is performed on information regarding centrally dispatched load, renewable energy output, capacity of units that must be started or stopped, and tie line plans; the preprocessing includes noise reduction and normalization.

[0084] In the field of algorithm models, the data cleaning process is also known as "feature engineering". Common methods include selecting features based on correlation, denoising based on the degree of dispersion, and data normalization. The purpose is to improve the data quality, make more reasonable use of each feature to predict more accurate results, or fit a model with higher accuracy. Therefore, this embodiment uses noise reduction and normalization to preprocess the data.

[0085] Specifically, the least squares method was used to fit the data on the centrally dispatched load, renewable energy output, mandatory start-up and mandatory shutdown unit capacity, and tie line planning information. A deviation threshold h was set for each data type, and points where the deviation between the sample value and the fitted value exceeded the threshold h were treated as noise and deleted. The threshold h was determined based on the centrally dispatched load, renewable energy output, mandatory start-up and mandatory shutdown unit capacity, and tie line planning information.

[0086] Specifically, the noise-reduced information on centrally dispatched load, renewable energy output, mandatory start-up and shutdown unit capacity, and tie-line planning is normalized by dividing each data point by its maximum value. This normalization process flattens the value ranges of various data points, keeping them within the same range and avoiding parameter clustering issues caused by large differences in value ranges between data points, thus improving the model's final prediction performance.

[0087] In this embodiment, by performing noise reduction and normalization processing on the acquired information on the province's overall dispatch load, new energy output, mandatory start-up and shutdown unit capacity, and tie line plans, the quality of the data is guaranteed, thereby providing favorable input conditions for multiple electricity price prediction models, resulting in higher accuracy of the output results of multiple electricity price prediction models.

[0088] Optionally, step S500 may further include steps S510-S520.

[0089] Please refer to Figure 3 , Figure 3 This is a schematic diagram of the second process of the day-ahead node electricity price forecasting method provided in the embodiments of the present invention.

[0090] S510 inputs the information on the load under unified dispatch, the output of new energy sources, the capacity of units that must be started or stopped, and the tie line plan into the weight allocation model to obtain the weight allocation combination; wherein, the weight allocation combination includes the weight value of the predicted value of the first day-ahead node electricity price, the weight value of the predicted value of the second day-ahead node electricity price, and the weight value of the predicted value of the third day-ahead node electricity price.

[0091] Understandably, the weights range from 0 to 1 (inclusive), and the sum of the weights is 1. That is, the sum of the weights of the predicted electricity price at the first day's node, the predicted electricity price at the second day's node, and the predicted electricity price at the third day's node is 1, and the three constitute the weight allocation combination.

[0092] To make it easier to understand, the following examples are provided:

[0093] For example, assuming the weight calculation precision is designed to be 0.001, then the weight allocation combination can be {0.001, 0.002, 0.997}, where 0.001 represents the weight value of the predicted electricity price at the first day's node, 0.002 represents the weight value of the predicted electricity price at the second day's node, and 0.997 represents the weight value of the predicted electricity price at the third day's node.

[0094] For example, assuming the weight calculation precision is designed to be 0.002, then the weight allocation combination can be {0.002, 0.006, 0.992}, where 0.002 represents the weight value of the predicted electricity price at the first day's node, 0.006 represents the weight value of the predicted electricity price at the second day's node, and 0.992 represents the weight value of the predicted electricity price at the third day's node.

[0095] If the precision of the weight calculation is set to other values, the same principle applies; this will not be elaborated upon here.

[0096] S520, based on the weighted combination, calculates the weighted average of the predicted values ​​of the first, second, and third day-ahead node electricity prices to obtain the predicted day-ahead node electricity price.

[0097] Specifically, let P1 be the predicted value of the day-ahead electricity price for the first day, P2 for the second day, and P3 for the third day. Let W1 be the weight assigned to the predicted value of the day-ahead electricity price for the first day, W2 for the second day, and W3 for the third day. Then, the predicted value of the day-ahead electricity price can be calculated as Pz using the following formula: PZ = P1xW1 + P2xW2 + P3xW3.

[0098] To make it easier to understand, the following examples are provided:

[0099] Assuming the predicted daytime electricity price is 2 yuan / kWh for the first day, 1.5 yuan / kWh for the second day, and 3 yuan / kWh for the third day, and the weighting of the predicted daytime electricity price is 0.025 for the first day, 0.05 for the second day, and 0.025 for the third day, then the predicted daytime electricity price is: 2 x 0.25 + 1.5 x 0.5 + 3 x 0.25 = 2 yuan / kWh.

[0100] In this embodiment, by inputting the province-wide dispatch load, renewable energy output, mandatory-operation / mandatory-shutdown unit capacity, and tie-line plan information at any given time point in the power grid node where the power plant is located into the weighted allocation model, the weighted values ​​of the predicted day-ahead node electricity price for the first, second, and third days are obtained. Then, a weighted average is calculated and summed to obtain the predicted day-ahead node electricity price. Since the weighted allocation model outputs the weighted combination based on real-time acquired dispatch load, renewable energy output, mandatory-operation / mandatory-shutdown unit capacity, and tie-line plan information, the weighted values ​​of each electricity price prediction model are no longer fixed but dynamically allocated according to changes in the acquired dispatch load, renewable energy output, mandatory-operation / mandatory-shutdown unit capacity, and tie-line plan information. This results in higher accuracy for the final predicted day-ahead node electricity price.

[0101] Optionally, Figure 4 This is a flowchart illustrating the training process of the first electricity price prediction model provided in an embodiment of the present invention. The training process of the first electricity price prediction model includes the following steps:

[0102] S210, Obtain sample data; the sample data includes the province's overall dispatch load, new energy output, mandatory start-up and mandatory shutdown unit capacity, tie line planning information and day-ahead node electricity price at any point in time in the power grid node where the power plant is located;

[0103] The power trading platform publishes the day-ahead clearing results for each node daily. Therefore, historical data on the province's overall dispatch load, renewable energy output, mandatory start-up and shutdown unit capacity, tie-line planning information, and historical day-ahead node electricity prices can be downloaded and collected through the power trading platform. The historical data on the province's overall dispatch load, renewable energy output, mandatory start-up and shutdown unit capacity, and tie-line planning information serve as input conditions for model training, while the historical day-ahead node electricity prices serve as output conditions for model training.

[0104] In one embodiment, before inputting historical provincial load data, renewable energy output, mandatory start / stop unit capacity, and tie-line planning information into the random forest model, the sample data can be denoised and normalized to improve the quality of the input data and thus enhance the model's training accuracy. Similarly, the input data for the extreme gradient boosting model and the multiple linear regression model can also undergo denoising and normalization during training, which will not be elaborated upon further.

[0105] S220 uses the province's overall dispatch load, new energy output, mandatory start-up and shutdown unit capacity, and tie line planning information at any historical point in time to train a pre-built random forest model. When the electricity price predicted by the pre-built random forest model is equal to the province's day-ahead node electricity price at any historical point in time, the first electricity price prediction model is obtained.

[0106] Random Forest (RF) is a widely used and mature algorithm that is implemented by ensemble multiple decision trees, where each decision tree is independent of the others. Extensive experiments have shown that the Random Forest algorithm can effectively predict results from structured data samples.

[0107] Specifically, the system uses four basic features—the province's overall dispatch load, renewable energy output, mandatory start-up and shutdown capacity of generating units, and tie-line planning information—as inputs into the random forest model for multiple training iterations at any historical point in time when the power plant is located. Each time, the predicted electricity price is compared with the province's day-ahead node electricity price at any given time to reduce model error. The training continues until the predicted electricity price equals the province's day-ahead node electricity price at any given time, at which point the model completes training and the first electricity price prediction model is obtained.

[0108] In this embodiment, the random forest model is trained using the province-wide load adjustment, renewable energy output, mandatory start-up and shutdown unit capacity, and tie line planning information at any historical point in time of the power grid node where the power plant is located as input conditions, and the province-wide day-ahead node electricity price at any historical point in time as output conditions, to obtain the first electricity price prediction model. This allows the first electricity price prediction model to be used for preliminary prediction of day-ahead node electricity prices in practical applications.

[0109] Optionally, Figure 5 This is a flowchart illustrating the training process of the second electricity price prediction model provided in an embodiment of the present invention. The training process of the second electricity price prediction model includes the following steps:

[0110] S310, Obtain sample data; the sample data includes the province's overall dispatch load, new energy output, mandatory start-up and mandatory shutdown unit capacity, tie line planning information and day-ahead node electricity price at any point in time in the power grid node where the power plant is located;

[0111] S320 uses the province's overall dispatch load, renewable energy output, mandatory start-up and shutdown unit capacity, tie line planning information, and the electricity price predicted by the first electricity price prediction model at any historical point in time to train a pre-built extreme gradient boosting model. When the electricity price predicted by the pre-built extreme gradient boosting model is equal to the province's day-ahead node electricity price at any historical point in time, a second electricity price prediction model is obtained.

[0112] Extreme Gradient Boosting (XGBoost) is a special type of gradient boosting algorithm. While it also uses decision trees as the basic unit, unlike random forests, each decision tree in XGBoost is fitted based on the previous decision tree; essentially, each tree fits the residual of the previous tree. Furthermore, by introducing the complexity of decision trees and using Taylor expansion to approximate the residuals, the algorithm effectively improves the model's accuracy and generalization ability.

[0113] Specifically, the four basic features are the province-wide dispatch load, renewable energy output, mandatory start-up and shutdown capacity of generating units, and tie line planning information at any historical point in time when the power plant is located. The electricity price predicted by the first electricity price prediction model is then introduced as the fifth feature. The model is then fed into the extreme gradient boosting model for multiple training sessions. Each time, the predicted electricity price is compared with the province-wide day-ahead node electricity price at any historical point in time to reduce the model error. The training continues until the predicted electricity price equals the province-wide day-ahead node electricity price at any historical point in time. At this point, the model is considered complete and the second electricity price prediction model is obtained.

[0114] In this embodiment, the extreme gradient boosting model is trained using the province-wide dispatch load, renewable energy output, mandatory start-up and shutdown unit capacity, tie line planning information, and the predicted electricity price output from the first electricity price prediction at any historical point in time in the power grid node where the power plant is located as input conditions, and the province-wide day-ahead node electricity price at any historical point in time as output conditions. This yields a second electricity price prediction model, enabling the use of the second electricity price prediction model for preliminary prediction of day-ahead node electricity prices in practical applications. Furthermore, by combining the predicted electricity price output from the first electricity price prediction model as input conditions, the correlation between the first and second electricity price prediction models is effectively established, resulting in more accurate predictions after the fusion of the two models.

[0115] Optionally, Figure 6 This is a flowchart illustrating the training process of the third electricity price prediction model provided in an embodiment of the present invention. The training process of the third electricity price prediction model includes the following steps:

[0116] S410, Obtain sample data; the sample data includes the province's overall dispatch load, new energy output, mandatory start-up and mandatory shutdown unit capacity, tie line planning information and day-ahead node electricity price at any point in time in the power grid node where the power plant is located;

[0117] S420 uses the province's unified dispatch load, new energy output, mandatory start-up and shutdown unit capacity, tie line planning information, and the electricity price predicted by the second electricity price prediction model at any historical point in time to train a pre-built multiple linear regression model. When the electricity price predicted by the pre-built multiple linear regression model is equal to the province's day-ahead node electricity price at any historical point in time, a third electricity price prediction model is obtained.

[0118] The Lasso (Least Absolute Shrinkage and Selection Operator) model assumes the existence of an approximate linear relationship that allows each feature to regress linearly. Lasso typically uses the mean squared error as the loss function. Its implementation is very simple, the regression speed is fast, and it exhibits strong fitting effects when the number of features is relatively small. It is a widely used algorithmic model.

[0119] Specifically, the four basic features are the province-wide dispatch load, renewable energy output, mandatory start-up and shutdown capacity of generating units, and tie-line planning information at any historical point in time when the power plant is located. The electricity price predicted by the second electricity price prediction model is then introduced as the fifth feature. The model is then fed into the extreme gradient boosting model for multiple training sessions. Each time, the predicted electricity price is compared with the province-wide day-ahead node electricity price at any historical point in time to reduce the model's error. The training continues until the predicted electricity price equals the province-wide day-ahead node electricity price at any historical point in time. At this point, the model has completed training and the third electricity price prediction model is obtained.

[0120] In this embodiment, the province's overall dispatch load, renewable energy output, mandatory start-up and shutdown unit capacity, tie line planning information, and the predicted electricity price output from the first electricity price prediction are used as input conditions at any historical point in time in the power grid node where the power plant is located. The province's day-ahead node electricity price at any historical point in time is used as the output condition to train the multiple linear regression model, resulting in a third electricity price prediction model. This allows the third electricity price prediction model to be used for preliminary prediction of day-ahead node electricity prices in practical applications. Furthermore, by combining the predicted electricity price output from the second electricity price prediction model as the input condition of the model, the correlation between the second and third electricity price prediction models is effectively established, making the prediction results after the fusion of the two models more accurate.

[0121] Optionally, Figure 7 This is a flowchart illustrating the training process of the weight allocation model provided in an embodiment of the present invention. The training process of the weight allocation model includes the following steps:

[0122] S511, Obtain sample data; the sample data includes the province's overall dispatch load, new energy output, mandatory start-up and mandatory shutdown unit capacity, tie line planning information and day-ahead node electricity price at any point in time in the power grid node where the power plant is located;

[0123] S512, according to the preset weight calculation accuracy, assign weights to the first electricity price prediction model, the second electricity price prediction model and the third electricity price prediction model to obtain multiple weight allocation combinations;

[0124] In this embodiment, the calculation precision of the preset weight can be designed to be 0.001, but those skilled in the art can design other values ​​for the calculation precision of the weight according to actual application needs, etc., without making specific limitations.

[0125] Specifically, if the set of multiple weight allocation combinations is W z Each weight allocation combination is labeled W. i The weight label of the electricity price predicted by the first electricity price prediction model is W[1], the weight label of the electricity price predicted by the second electricity price prediction model is W[2], and the weight label of the electricity price predicted by the third electricity price prediction model is W[3]. Then it can be expressed as W z ={{W1[1], W1[2], W1[3]}, {W2[1], W2[2], W2[3]}, {W3[1], W3[2], W3[3]}, ........., {W n [1], W n [2], W n [3]}}, the number of specific weight allocation combinations is determined by the calculation accuracy of the weights.

[0126] For example, if the weight calculation precision is 0.001, then the number of acceptable weight allocation combinations is 501501, such as W. Z = {{0.001,0.001,0.998}, {0.001,0.002,0.997}, and so on.

[0127] S513, according to multiple weight allocation combinations, calculate the weighted average of the electricity price predicted by the first electricity price prediction model, the electricity price predicted by the second electricity price prediction model, and the electricity price predicted by the third electricity price prediction model, to obtain multiple predicted electricity prices;

[0128] Specifically, if the various weight allocation combinations are labeled as {W1[1], W1[2], W1[3]}, {W2[1], W2[2], W2[3]}, {W3[1], W3[2], W3[3]}, ..., {W n [1], W n [2], W n [3]}, the electricity price calculated under the weighted grouping combination is labeled P. ZLet the electricity price predicted by the first electricity price prediction model be labeled P1, the electricity price predicted by the second electricity price prediction model be labeled P2, and the electricity price predicted by the third electricity price prediction model be labeled P3. Then, it can be expressed as P... z ={{P1*W1[1]+P2*W1[2]+P3*W1[3]}, {P1*W2[1]+P2*W2[2]+P3*W2[3]}, ........., {P1*W n [1]+P2*W n [2]+P3*W n [3]}}.

[0129] To make it easier to understand, the following examples are provided:

[0130] Based on the above, we know that the weight allocation combination is 501501 groups, such as {0.001, 0.001, 0.998} and {0.001, 0.002, 0.997}. Assuming the first electricity price prediction model predicts a price of 2, the second model predicts a price of 3, and the third model predicts a price of 1.5, then the predicted electricity price is: P z = {0.001*2+0.001*3+0.998*1.5, 0.001*2+0.002*3+0.997*1.5.....} etc., 501501 possible results.

[0131] S514: Select the predicted electricity price with the smallest difference from the provincial day-ahead node electricity price at any historical point in time from multiple predicted electricity prices, and take the weight allocation combination corresponding to the predicted electricity price with the smallest difference as the optimal weight allocation combination.

[0132] Specifically, by programming, the final electricity price corresponding to each weight allocation combination can be traversed to find the optimal combination, denoted as {W}. 1,max W 2,max W 3,max Similarly, the optimal weight combination for each point in time over multiple days can be calculated. Assuming there are N days of historical data, an Nx3 array can be obtained, which serves as the target sample for the weight allocation model.

[0133] For example, the weighted combination for one day of data is: {{0.001, 0.001, 0.998}, {0.001, 0.002, 0.997}, ...}; the weighted combination for N days of data is: {{0.001, 0.001, 0.998}, {0.001, 0.002, 0.997}, ...

[0134] {0.001, 0.001, 0.998}, {0.001, 0.002, 0.997}, ...

[0135] {0.001, 0.001, 0.998}, {0.001, 0.002, 0.997}, ...

[0136] {0.001, 0.001, 0.998}, {0.001, 0.002, 0.997}, ...... ... ... ...

[0140] {0.001, 0.001, 0.998}, {0.001, 0.002, 0.997}, ...}

[0141] To make it easier to understand, the following examples are provided:

[0142] In one embodiment, the optimal weight combination of the three electricity price prediction models is determined by programming calculation. The pseudocode for the calculation process is as follows:

[0143] E min =E max #Note: Minimum deviation requires a large initial value.

[0144] W min =null #Note: Declare a combination that minimizes the deviation.

[0145] For(W i )in W Z #Note: Iterate through the set W of all possible weight assignment combinations Z

[0146] W i [1] #Note: Weights of the random forest model

[0147] W i [2] #Note: Weights in the extreme gradient boosting model

[0148] W i [3] #Note: Weights in the multiple linear regression model

[0149] P Z =P1*W i [1]+P2*W i [2]+P3*W i [3]#Note: P1, P2, and P3 are the day-ahead nodal electricity prices predicted by the three models, respectively. z To calculate the electricity price under this weighted combination

[0150] E = |P Z -P R|#Note: E represents the deviation between the predicted result and the actual value under this weight combination; P R The actual value of electricity price

[0151] If(E <E min ): #Note: The bubble sort method finds the weight combination with the smallest deviation.

[0152] E min =E

[0153] W min =W i

[0154] W obtained through the above method min The optimal weight allocation combination is represented by a 1x3 array. Similarly, the optimal weight combination schemes at other time points can be calculated to form the optimal weight set, which serves as the target sample for the weight allocation model.

[0155] S515 uses the province's overall dispatch load, new energy output, mandatory start-up and shutdown unit capacity, and tie line planning information at any historical point in time to train a pre-constructed multiple linear regression model. When the weight allocation combination predicted by the pre-constructed multiple linear regression model equals the optimal weight allocation combination, the weight allocation model is obtained.

[0156] It should be noted that for samples with few features, the multiple linear regression model is a simple and effective algorithm to implement. Therefore, the weight allocation model is trained using the multiple linear regression model.

[0157] Specifically, the total load of the entire province, the output of new energy sources, the capacity of units that must be started and stopped, and the interconnection line plan information at any point in time in the history of the power grid node where the power plant is located are taken as four basic features and input into the multiple linear regression model for multiple training. Each time, the predicted weight allocation combination is compared with the optimal weight allocation combination to reduce the model error. The model is completed when the weight allocation combination predicted by the training is equal to the optimal weight allocation combination, and the weight allocation model is obtained.

[0158] In this embodiment, by using the province-wide unified dispatch load, renewable energy output, mandatory start-up and shutdown unit capacity, and tie line plan information at any historical point in time of the power grid node where the power plant is located as input conditions, and using the optimal weight allocation combination as output conditions, a multiple linear regression model is trained to obtain a weight allocation model. This allows for the dynamic allocation of weight values ​​for the first, second, and third electricity price prediction models by inputting real-time unified dispatch load, renewable energy output, mandatory start-up and shutdown unit capacity, and tie line plan information during practical applications. This avoids the problem of a single weight allocation mode in the prior art, thereby improving the accuracy of the final day-ahead node electricity price prediction value.

[0159] In one or more of the above embodiments, in addition to using random forest prediction model, extreme gradient boosting model and multiple linear regression model for fusion training, other statistical models can also be used for fusion training. The fusion concept and weight allocation concept of other statistical models are the same as those described above.

[0160] Based on the same inventive concept, embodiments of the present invention also provide a day-ahead node electricity price prediction device 200 for spot electricity trading markets.

[0161] Please refer to Figure 8 , Figure 8 This is a schematic diagram of the day-ahead node electricity price forecasting device for the spot electricity trading market provided in an embodiment of the present invention.

[0162] The data acquisition module 210 is used to acquire the province's unified dispatch load, new energy output, mandatory start-up and mandatory shutdown unit capacity, and tie line planning information at any point in time at the power grid node where the power plant is located.

[0163] The first calculation module 220 is used to input the unified dispatch load, new energy output, capacity of units that must be turned on and off and tie line planning information into the first electricity price prediction model to obtain the predicted value of the first day-ahead node electricity price;

[0164] The second calculation module 230 is used to input the centrally dispatched load, new energy output, capacity of units that must be turned on or off, tie line planning information and the predicted value of the first day-ahead node electricity price into the second electricity price prediction model to obtain the predicted value of the second day-ahead node electricity price.

[0165] The third calculation module 240 is used to input the centrally dispatched load, new energy output, capacity of units that must be turned on or off, tie line planning information, and the predicted value of the electricity price at the second day before the node into the third electricity price prediction model to obtain the predicted value of the electricity price at the third day before the node.

[0166] The comprehensive calculation module 250 is used to calculate the weight allocation ratio of the predicted values ​​of the first, second, and third day-ahead electricity prices based on the weight allocation model, and then obtain the predicted values ​​of the day-ahead electricity prices after weighting.

[0167] It should be understood that this device corresponds to the above-described embodiment of the pre-market electricity price forecasting method and is capable of performing the various steps involved in the above-described method embodiment. The specific functions of this device can be found in the description above, and detailed descriptions are omitted here to avoid repetition. The device includes at least one software functional module that can be stored in memory or embedded in the device's operating system (OS) in the form of software or firmware.

[0168] Based on the same inventive concept, embodiments of the present invention provide an electronic device, which includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0169] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0170] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0171] Based on the same inventive concept, embodiments of the present invention also provide a computer-readable storage medium storing instructions that, when executed by a processor, are adapted to execute a program having the following method steps: obtaining provincial dispatch load, renewable energy output, mandatory start / stop unit capacity, and tie-line plan information at any given time point in the power grid node where the power plant is located; inputting the dispatch load, renewable energy output, mandatory start / stop unit capacity, and tie-line plan information into a first electricity price prediction model to obtain the predicted value of the node electricity price for the first day; inputting the dispatch load, renewable energy output, mandatory start / stop unit capacity, and tie-line plan information into a first electricity price prediction model to obtain the predicted value of the node electricity price for the first day; inputting the dispatch load, renewable energy output, and mandatory start / stop unit capacity ... The capacity of units that must be shut down, the planned information of tie lines, and the predicted value of the first day-ahead node price are input into the second electricity price prediction model to obtain the predicted value of the second day-ahead node price. The central dispatch load, renewable energy output, capacity of units that must be turned on and off, the planned information of tie lines, and the predicted value of the second day-ahead node price are input into the third electricity price prediction model to obtain the predicted value of the third day-ahead node price. Based on the weight allocation model, the weight allocation ratio of the predicted values ​​of the first day-ahead node price, the second day-ahead node price, and the third day-ahead node price is calculated, and the weighted average is used to obtain the predicted value of the day-ahead node price.

[0172] In one embodiment, the above-mentioned method for predicting day-ahead electricity prices in the spot electricity trading market further includes: preprocessing information on centrally dispatched loads, renewable energy output, capacity of units that must be started or stopped, and tie-line plans; wherein, the preprocessing includes: noise reduction processing and normalization processing.

[0173] In one embodiment, the above-mentioned method for predicting day-ahead node electricity prices in the spot electricity trading market further includes: inputting information on centrally dispatched load, renewable energy output, capacity of units that must be started or stopped, and tie-line planning into a weighted allocation model to obtain a weighted allocation combination; wherein, the weighted allocation combination includes the weighted values ​​of the predicted values ​​of the first day-ahead node electricity price, the second day-ahead node electricity price, and the third day-ahead node electricity price; based on the weighted allocation combination, calculating the weighted average of the predicted values ​​of the first day-ahead node electricity price, the second day-ahead node electricity price, and the third day-ahead node electricity price to obtain the predicted value of the day-ahead node electricity price.

[0174] In one embodiment, the above-mentioned method for predicting the day-ahead node electricity price in the spot electricity trading market further includes: acquiring sample data; wherein the sample data includes the province-wide dispatch load, renewable energy output, mandatory-operation and mandatory-outage unit capacity, tie-line planning information, and day-ahead node electricity price at any historical point in time for the power plant's grid node; training a pre-constructed random forest model using the province-wide dispatch load, renewable energy output, mandatory-operation and mandatory-outage unit capacity, and tie-line planning information at any historical point in time; and obtaining a first electricity price prediction model when the electricity price predicted by the pre-constructed random forest model is equal to the province-wide day-ahead node electricity price at any historical point in time.

[0175] In one embodiment, the above-mentioned method for predicting the day-ahead node electricity price in the spot electricity trading market further includes: acquiring sample data; wherein, the sample data includes the province-wide dispatch load, renewable energy output, mandatory-operation and mandatory-outage unit capacity, tie-line planning information, and day-ahead node electricity price at any historical point in time for the power plant's grid node; training a pre-constructed extreme gradient boosting model using the province-wide dispatch load, renewable energy output, mandatory-operation and mandatory-outage unit capacity, tie-line planning information, and the electricity price predicted by the first electricity price prediction model at any historical point in time; and obtaining a second electricity price prediction model when the electricity price predicted by the pre-constructed extreme gradient boosting model is equal to the province-wide day-ahead node electricity price at any historical point in time.

[0176] In one embodiment, the above-mentioned method for predicting day-ahead node electricity prices in the spot electricity trading market further includes: acquiring sample data; wherein the sample data includes the province-wide dispatch load, renewable energy output, mandatory-operation and mandatory-outage unit capacity, tie-line planning information, and day-ahead node electricity price at any historical point in time for the power plant's grid node; using the province-wide dispatch load, renewable energy output, mandatory-operation and mandatory-outage unit capacity, tie-line planning information, and electricity price predicted by the second electricity price prediction model at any historical point in time to train a pre-constructed multiple linear regression model, and obtaining a third electricity price prediction model when the electricity price predicted by the pre-constructed multiple linear regression model is equal to the province-wide day-ahead node electricity price at any historical point in time.

[0177] In one embodiment, the method for predicting the day-ahead electricity price in the spot electricity trading market further includes: acquiring sample data; wherein the sample data includes the province-wide dispatch load, renewable energy output, mandatory start-up and mandatory shutdown unit capacity, tie-line planning information, and day-ahead electricity price at any historical point in time at the power grid node where the power plant is located; assigning weights to the first electricity price prediction model, the second electricity price prediction model, and the third electricity price prediction model according to preset weight calculation accuracy, obtaining multiple weight allocation combinations; and calculating the electricity price predicted by the first electricity price prediction model and the electricity price predicted by the second electricity price prediction model according to the multiple weight allocation combinations. The weighted average of the electricity price and the electricity price predicted by the third prediction electricity price model is used to obtain multiple predicted electricity prices. From these multiple predicted electricity prices, the predicted electricity price with the smallest difference from the provincial day-ahead node electricity price at any historical point in time is selected, and the weight allocation combination corresponding to the predicted electricity price with the smallest difference is taken as the optimal weight allocation combination. The pre-constructed multiple linear regression model is trained using the provincial unified dispatch load, renewable energy output, mandatory start and stop unit capacity, and tie line plan information at any historical point in time. When the weight allocation combination predicted by the pre-constructed multiple linear regression model is equal to the optimal weight allocation combination, the weight allocation model is obtained.

[0178] 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.

[0179] 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.

[0180] 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.

[0181] 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.

[0182] It should also be noted that the various specific technical features described in the above embodiments can be combined in any suitable manner without contradiction. To avoid unnecessary repetition, the embodiments of the present invention will not describe the various possible combinations separately.

[0183] In addition, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0184] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0185] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A day-ahead nodal electricity price forecasting method, characterized in that, include: Obtain information on the province's overall dispatch load, new energy output, mandatory start-up and mandatory shutdown unit capacity, and tie line plans at any given time point in the power grid node where the power plant is located; The information on the centrally dispatched load, new energy output, capacity of units that must be started or stopped, and tie line plans is input into the first electricity price prediction model to obtain the predicted value of the first day-ahead node electricity price. Input the centrally dispatched load, new energy output, capacity of units that must be turned on or off, tie line planning information, and the predicted value of the first day-ahead node electricity price into the second electricity price prediction model to obtain the predicted value of the second day-ahead node electricity price. The centrally dispatched load, new energy output, capacity of units that must be started or stopped, tie line planning information, and the predicted value of the electricity price at the second day's node are input into the third electricity price prediction model to obtain the predicted value of the electricity price at the third day's node. Based on the weighted allocation model, the weighted allocation ratios of the predicted values ​​of the first, second, and third day-ahead electricity prices are calculated, and the weighted values ​​are then used to obtain the predicted values ​​of the day-ahead electricity prices. Training method for the first electricity price prediction model: The first electricity price prediction model is obtained by training a pre-built random forest model with the province's overall dispatch load, new energy output, mandatory start-up and mandatory shutdown unit capacity and tie line plan information at any point in history. The electricity price predicted by the pre-built random forest model is equal to the province's day-ahead node electricity price at any point in history. The training method for the second electricity price prediction model: The pre-built extreme gradient boosting model is trained using the province's unified dispatch load, new energy output, mandatory start-up and mandatory shutdown unit capacity, tie line planning information, and the electricity price predicted by the first electricity price prediction model at any historical point in time. When the electricity price predicted by the pre-built extreme gradient boosting model is equal to the province's day-ahead node electricity price at any historical point in time, the second electricity price prediction model is obtained. The training method for the third electricity price prediction model: The pre-constructed multiple linear regression model is trained using the province's unified dispatch load, new energy output, mandatory start-up and mandatory shutdown unit capacity, tie line planning information, and the electricity price predicted by the second electricity price prediction model at any historical point in time. The third electricity price prediction model is obtained when the electricity price predicted by the pre-constructed multiple linear regression model is equal to the province's day-ahead node electricity price at any historical point in time. The training method for the weight allocation model is as follows: According to the preset weight calculation accuracy, the first electricity price prediction model, the second electricity price prediction model and the third electricity price prediction model are weighted and assigned to obtain multiple weight assignment combinations; According to the multiple weight allocation combinations, the weighted average of the electricity price predicted by the first electricity price prediction model, the electricity price predicted by the second electricity price prediction model, and the electricity price predicted by the third electricity price prediction model is calculated to obtain multiple predicted electricity prices; The predicted electricity price with the smallest difference from the provincial day-ahead node electricity price at any historical point in time is selected from the multiple predicted electricity prices, and the weight allocation combination corresponding to the predicted electricity price with the smallest difference is taken as the optimal weight allocation combination. The pre-constructed multiple linear regression model is trained using the province's overall dispatch load, new energy output, mandatory start-up and shutdown unit capacity, and tie line planning information at any historical point in time. The weight allocation model is obtained when the weight allocation combination predicted by the pre-constructed multiple linear regression model is equal to the optimal weight allocation combination.

2. The day-ahead nodal electricity price forecasting method according to claim 1, characterized in that, Before the step of inputting the centrally dispatched load, renewable energy output, mandatory start-up and mandatory shutdown unit capacity, and tie-line plan information into the first electricity price prediction model to obtain the predicted value of the first day-ahead nodal electricity price, the method further includes: The information on the centrally dispatched load, new energy output, capacity of units that must be started or stopped, and tie line plans is preprocessed; wherein, the preprocessing includes noise reduction processing and normalization processing.

3. The day-ahead nodal electricity price forecasting method according to claim 1, characterized in that, The weighted allocation model calculates the weighted allocation ratios of the predicted day-ahead electricity prices for the first, second, and third days, and then weights them to obtain the predicted day-ahead electricity prices, including: The information on the centrally dispatched load, new energy output, mandatory start-up and mandatory shutdown unit capacity, and tie line plan is input into the weight allocation model to obtain the weight allocation combination; wherein, the weight allocation combination includes the weight value of the predicted value of the first day-ahead node electricity price, the weight value of the predicted value of the second day-ahead node electricity price, and the weight value of the predicted value of the third day-ahead node electricity price. Based on the weighted combination, the weighted average of the predicted values ​​of the first, second, and third day-ahead node electricity prices is calculated to obtain the predicted value of the day-ahead node electricity price.

4. The day-ahead nodal electricity price forecasting method according to claim 1, characterized in that, The training method for the first electricity price prediction model also includes: Obtain sample data; wherein, the sample data includes the province-wide dispatch load, renewable energy output, mandatory start-up and mandatory shutdown unit capacity, tie line planning information and day-ahead node electricity price at any historical point in time of the power grid node where the power plant is located.

5. The day-ahead nodal electricity price forecasting method according to claim 1, characterized in that, The training method for the second electricity price prediction model also includes: Obtain sample data; wherein, the sample data includes the province-wide dispatch load, renewable energy output, mandatory start-up and mandatory shutdown unit capacity, tie line planning information and day-ahead node electricity price at any historical point in time of the power grid node where the power plant is located.

6. The day-ahead nodal electricity price forecasting method according to claim 1, characterized in that, The training method for the third electricity price prediction model also includes: Obtain sample data; wherein, the sample data includes the province-wide dispatch load, renewable energy output, mandatory start-up and mandatory shutdown unit capacity, tie line planning information and day-ahead node electricity price at any historical point in time of the power grid node where the power plant is located.

7. The day-ahead nodal electricity price forecasting method according to claim 3, characterized in that, Training methods for weight allocation models also include: Obtain sample data; wherein, the sample data includes the province-wide dispatch load, renewable energy output, mandatory start-up and mandatory shutdown unit capacity, tie line planning information and day-ahead node electricity price at any historical point in time of the power grid node where the power plant is located.

8. A day-ahead nodal electricity price forecasting device, characterized in that, include: The data acquisition module is used to acquire the province's overall dispatch load, new energy output, mandatory start-up and mandatory shutdown unit capacity, and tie line planning information at any point in time at the power grid node where the power plant is located. The first calculation module is used to input the unified dispatch load, new energy output, capacity of units that must be turned on or off, and tie line plan information into the first electricity price prediction model to obtain the predicted value of the first day-ahead node electricity price. The second calculation module is used to input the centrally dispatched load, new energy output, capacity of units that must be turned on or off, tie line planning information and the predicted value of the first day-ahead node electricity price into the second electricity price prediction model to obtain the predicted value of the second day-ahead node electricity price. The third calculation module is used to input the centrally dispatched load, new energy output, capacity of units that must be turned on or off, tie line planning information, and the predicted value of the electricity price at the second day's node into the third electricity price prediction model to obtain the predicted value of the electricity price at the third day's node. The comprehensive calculation module is used to calculate the weight allocation ratio of the predicted values ​​of the first, second, and third day-ahead electricity prices based on the weight allocation model, and then obtain the predicted values ​​of the day-ahead electricity prices after weighting. Training method for the first electricity price prediction model: The first electricity price prediction model is obtained by training a pre-built random forest model with the province's overall dispatch load, new energy output, mandatory start-up and mandatory shutdown unit capacity and tie line plan information at any point in history. The electricity price predicted by the pre-built random forest model is equal to the province's day-ahead node electricity price at any point in history. The training method for the second electricity price prediction model: The pre-built extreme gradient boosting model is trained using the province's unified dispatch load, new energy output, mandatory start-up and mandatory shutdown unit capacity, tie line planning information, and the electricity price predicted by the first electricity price prediction model at any historical point in time. When the electricity price predicted by the pre-built extreme gradient boosting model is equal to the province's day-ahead node electricity price at any historical point in time, the second electricity price prediction model is obtained. The training method for the third electricity price prediction model: The pre-constructed multiple linear regression model is trained using the province's unified dispatch load, new energy output, mandatory start-up and mandatory shutdown unit capacity, tie line planning information, and the electricity price predicted by the second electricity price prediction model at any historical point in time. The third electricity price prediction model is obtained when the electricity price predicted by the pre-constructed multiple linear regression model is equal to the province's day-ahead node electricity price at any historical point in time. The training method for the weight allocation model is as follows: According to the preset weight calculation accuracy, the first electricity price prediction model, the second electricity price prediction model and the third electricity price prediction model are weighted and assigned to obtain multiple weight assignment combinations; According to the multiple weight allocation combinations, the weighted average of the electricity price predicted by the first electricity price prediction model, the electricity price predicted by the second electricity price prediction model, and the electricity price predicted by the third electricity price prediction model is calculated to obtain multiple predicted electricity prices; The predicted electricity price with the smallest difference from the provincial day-ahead node electricity price at any historical point in time is selected from the multiple predicted electricity prices, and the weight allocation combination corresponding to the predicted electricity price with the smallest difference is taken as the optimal weight allocation combination. The pre-constructed multiple linear regression model is trained using the province's overall dispatch load, new energy output, mandatory start-up and shutdown unit capacity, and tie line planning information at any historical point in time. The weight allocation model is obtained when the weight allocation combination predicted by the pre-constructed multiple linear regression model is equal to the optimal weight allocation combination.

9. An electronic device, characterized in that, include: A processor and a memory, the memory storing machine-readable instructions executable by the processor, which, when executed by the processor, perform the day-ahead node electricity price forecasting method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions for causing a machine to perform the day-ahead electricity price forecasting method according to any one of claims 1-7.

Images