Power generation strategy adjustment method and device based on multi-modal electricity price prediction result, equipment and medium

Abstract

The invention relates to the technical field of power grid dispatching, and discloses a power generation strategy adjustment method based on a multi-modal electricity price prediction result, and the method comprises the steps: obtaining current electricity price time sequence data, current text event data and current visual event data of a current time period; the current electricity price time sequence data correspond to enhanced electricity price time sequence features, the current text event data correspond to enhanced text event features, and the current visual event data correspond to enhanced visual event features; performing cross-modal event feature fusion on the enhanced text event feature and the enhanced visual event feature to obtain a multi-modal event feature; performing feature modulation fusion on the multi-modal event feature and the enhanced electricity price time sequence feature to obtain a multi-modal fusion time sequence feature so as to perform electricity price prediction to obtain predicted electricity price data; and generating a power generation strategy according to the predicted electricity price data to obtain a target power generation strategy. According to the invention, real-time accurate response of the power generation side operation strategy to the power grid supply-demand relationship change is realized, and the power supply stability of the power grid is improved.

Description

Technical Field

[0001] This application relates to the field of power grid dispatching technology, and in particular to a method, apparatus, equipment and medium for adjusting power generation strategies based on multimodal electricity price prediction results. Background Technology

[0002] With the deepening of social development and energy transition, the penetration rate of new energy power generation continues to increase, making it an important power source in the power grid and contributing significantly to ensuring stable power supply. However, new energy power generation is highly dependent on environmental conditions and exhibits significant volatility, leading to instability in the electricity input to the grid from new energy sources. To address these issues, modern power grids have gradually incorporated new energy power generation systems into adjustable strategies to adapt to dispatching needs under actual conditions. In reality, when the power supply and demand relationship in the grid changes, the power generation side often cannot accurately perceive these changes, resulting in an inability to adjust its power generation strategies in a timely manner. This can lead to energy shortages or redundancies in the grid, affecting the stability of power supply.

[0003] In related technologies, the ability to perceive the power supply and demand relationship when adjusting power generation strategies on the power generation side still needs to be improved. Summary of the Invention

[0004] This application provides a method, apparatus, equipment, and medium for adjusting power generation strategies based on multimodal electricity price prediction results. It fuses and predicts electricity price data for future time periods based on multimodal data in the current time period, and generates corresponding power generation strategies based on the predicted electricity price data to adjust the operation of the power generation side. This enables the power generation side operation strategy to respond in real time and accurately to changes in the power grid supply and demand relationship, thereby improving the power supply stability of the power grid.

[0005] To achieve the above objectives, the main technical solutions adopted in this application include: In a first aspect, embodiments of this application provide a method for adjusting power generation strategies based on multimodal electricity price prediction results, the method comprising: Acquire the current electricity price time-series data, current text event data, and current visual event data for the current time period; wherein, the current electricity price time-series data corresponds to enhanced electricity price time-series features enhanced by historical data, the current text event data corresponds to enhanced text event features enhanced by historical data, and the current visual event data corresponds to enhanced visual event features enhanced by historical data. Cross-modal event feature fusion is performed on the enhanced text event features and the enhanced visual event features to obtain multimodal event features; The multimodal event features and the enhanced electricity price time series features are fused by feature modulation to obtain multimodal fused time series features, and electricity price is predicted based on the multimodal fused time series features to obtain predicted electricity price data for future time periods. A power generation strategy is generated based on the predicted electricity price data, and a target power generation strategy for the future time period is obtained for adjusting the operation of the power generation side.

[0006] The generation strategy adjustment method based on multimodal electricity price prediction results proposed in this application acquires multimodal data such as electricity price time-series data, text event data, and visual event data within the current time period, along with their corresponding enhanced data features. It performs cross-modal fusion of these enhanced data features to obtain multimodal fused time-series features, and then predicts the electricity price data for future time periods based on these features. On this basis, a generation strategy is generated according to the predicted electricity price data, and the generation side's operation is adjusted. Compared with related technologies, this application, based on multimodal data such as electricity price time-series data, text event data, and visual event data, performs cross-modal fusion of the data features of each multimodal data as the data foundation for electricity price prediction. This enhances the perception of the actual power grid environment, improves the accuracy of electricity price prediction results, and thus improves the matching degree between the target generation strategy and the actual power grid environment. It achieves real-time and accurate response of the generation side's operation strategy to changes in the power grid's supply and demand relationship, thereby improving the power grid's power supply stability.

[0007] Optionally, the step of fusing the enhanced text event features and the enhanced visual event features across modalities to obtain multimodal event features includes: The enhanced text event features are used as cross-modal query vectors, and the enhanced visual event features are used as cross-modal key vectors and cross-modal value vectors. Attention weights are calculated for the enhanced text event features and the enhanced visual event features to obtain the cross-modal attention weights corresponding to the enhanced text event features and the enhanced visual event features respectively. Based on the cross-modal attention weights, attention fusion is performed on the enhanced text event features and the enhanced visual event features to obtain the multimodal event features.

[0008] Optionally, the step of performing feature modulation fusion on the multimodal event features and the enhanced electricity price time-series features to obtain multimodal fused time-series features includes: The enhanced electricity price time series features are modulated using the multimodal event features to obtain event-modulated time series features; The event modulation timing features and the enhanced electricity price timing features are fused to obtain the multimodal fused timing features.

[0009] Optionally, the current electricity price time-series data corresponds to multiple historical electricity price time-series data that meet preset similarity requirements; the enhanced electricity price time-series features are obtained through the following methods: Feature extraction is performed on the multiple historical electricity price time series data, and feature aggregation is performed on the data features of the multiple historical electricity price time series data to obtain unified historical time series features; The unified historical time-series features and the current electricity price time-series data are fused to obtain fused electricity price time-series features; The enhanced electricity price time series features are obtained by performing self-attention deep fusion on the fused electricity price time series features.

[0010] Optionally, the current electricity price time-series data corresponds to multiple historical electricity price time-series data that meet preset similarity requirements. Each of the multiple historical electricity price time-series data corresponds to a historical time period, and each historical time period corresponds to historical text event data and historical visual event data. The enhanced text event features are obtained through the following methods: The enhanced text event features are obtained by concatenating the data features of the historical text event data and the data features of the current text event data. The enhanced visual event features are obtained through the following method: The enhanced visual event features are obtained by sequentially concatenating the data features of the historical visual event data and the data features of the current visual event data.

[0011] Optionally, the step of predicting electricity prices based on the multimodal fusion time-series features to obtain predicted electricity price data for future time periods includes: The multimodal fused temporal features are input into the prediction head module; wherein, the prediction head module is trained using mean squared error as the loss function; The prediction head module maps the multimodal fusion time-series features to a preset future prediction step size to obtain the predicted electricity price data.

[0012] Optionally, generating a power generation strategy based on the predicted electricity price data to obtain the target power generation strategy for the future time period includes: The predicted electricity price data is input into the power generation decision model, which then outputs multiple candidate power generation strategies for the future time period. The power generation decision model is a pre-trained reinforcement learning network model. Based on the current power generation strategy for the current time period, the adjustment loss is calculated for the multiple candidate power generation strategies to obtain the adjustment loss cost of each candidate power generation strategy. The target power generation strategy is obtained by screening the multiple candidate power generation strategies based on the adjusted loss cost.

[0013] Secondly, embodiments of this application provide a power generation strategy adjustment device based on multimodal electricity price prediction results, the device comprising: The electricity price prediction multimodal data acquisition module is used to acquire current electricity price time series data, current text event data, and current visual event data for the current time period; wherein, the current electricity price time series data corresponds to enhanced electricity price time series features enhanced by historical data, the current text event data corresponds to enhanced text event features enhanced by historical data, and the current visual event data corresponds to enhanced visual event features enhanced by historical data. The electricity price prediction cross-modal feature fusion module is used to perform cross-modal event feature fusion on the enhanced text event features and the enhanced visual event features to obtain multimodal event features; The feature fusion electricity price prediction module is used to perform feature modulation fusion on the multimodal event features and the enhanced electricity price time series features to obtain multimodal fusion time series features, and to predict electricity prices based on the multimodal fusion time series features to obtain predicted electricity price data for future time periods. The power generation strategy adjustment module is used to generate a power generation strategy based on the predicted electricity price data, and obtain the target power generation strategy for the future time period, so as to adjust the operation of the power generation side.

[0014] Thirdly, embodiments of this application provide a computer device, including: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the method described in any of the above embodiments.

[0015] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer instructions, which are used to cause a computer to perform the method described in any one of the above embodiments.

[0016] Fifthly, embodiments of this application provide a computer program product, including computer instructions, which are used to cause a computer to perform the method described in any of the above embodiments. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the specific embodiments of this application or the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1 A flowchart illustrating the steps of a power generation strategy adjustment method based on multimodal electricity price prediction results provided in an embodiment of this application; Figure 2 This is a flowchart illustrating the multimodal fusion temporal features obtained in the embodiments of this application; Figure 3 This is a flowchart illustrating the steps of cross-modal event feature fusion in an embodiment of this application. Figure 4 This is a flowchart illustrating the steps of feature modulation fusion in an embodiment of this application; Figure 5 This is a flowchart illustrating the steps involved in obtaining enhanced electricity price time-series characteristics in an embodiment of this application. Figure 6 This is a flowchart illustrating the steps for obtaining enhanced text event features and enhanced visual event features in the embodiments of this application; Figure 7 This is a flowchart illustrating the steps of electricity price forecasting in an embodiment of this application; Figure 8 This is a flowchart illustrating the steps involved in generating a power generation strategy in an embodiment of this application. Figure 9 A block diagram of a power generation strategy adjustment device based on multimodal electricity price prediction results provided in an embodiment of this application; Figure 10 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0020] With the deepening of social development and energy transition, the penetration rate of new energy power generation continues to increase, becoming an important power source in the power grid and making a key contribution to ensuring stable power supply. New energy power generation is highly dependent on environmental conditions and exhibits significant volatility, leading to instability in the electricity input to the grid from new energy power generation. To address these issues, modern power grids have gradually incorporated new energy power generation systems into adjustable strategies to adapt to dispatching needs under actual conditions. However, in reality, when the power supply and demand relationship in the power grid changes, the power generation side often cannot accurately perceive these changes, resulting in an inability to adjust its power generation strategies in a timely manner. This can lead to energy shortages or redundancies in the power grid, affecting the stability of power supply. In related technologies, the ability to perceive power supply and demand relationships when adjusting power generation strategies still needs improvement.

[0021] To address the aforementioned issues, this application provides a method for adjusting power generation strategies based on multimodal electricity price prediction results. The method involves acquiring current electricity price time-series data, current text event data, and current visual event data for the current time period. The current electricity price time-series data corresponds to enhanced electricity price time-series features, the current text event data corresponds to enhanced text event features, and the current visual event data corresponds to enhanced visual event features. Cross-modal event feature fusion is performed on the enhanced text event features and enhanced visual event features to obtain multimodal event features. Feature modulation fusion is then performed on the multimodal event features and enhanced electricity price time-series features to obtain multimodal fused time-series features, which are used for electricity price prediction to obtain predicted electricity price data. A power generation strategy is generated based on the predicted electricity price data to obtain the target power generation strategy.

[0022] The generation strategy adjustment method based on multimodal electricity price prediction results provided in this application acquires multimodal data such as electricity price time series data, text event data, and visual event data within the current time period, as well as their respective corresponding enhanced data features; performs cross-modal fusion of multiple enhanced data features to obtain multimodal fused time series features, and makes predictions based on the multimodal fused time series features to obtain predicted electricity price data for future time periods; on this basis, a generation strategy is generated based on the predicted electricity price data to adjust the operation of the generation side.

[0023] Compared with related technologies, this application uses multimodal data such as electricity price time-series data, text event data, and visual event data to perform cross-modal fusion of the data features of each multimodal data as the data foundation for electricity price prediction. This enhances the perception of the actual power grid environment, improves the accuracy of electricity price prediction results, and thus improves the matching degree between the target power generation strategy and the actual power grid environment. It also enables the power generation side operation strategy to respond in real time and accurately to changes in the power grid supply and demand relationship, thereby improving the power supply stability of the power grid.

[0024] According to an embodiment of this application, a method for adjusting power generation strategy based on multimodal electricity price prediction results is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0025] Reference Figure 1 As shown in this embodiment, a method for adjusting power generation strategies based on multimodal electricity price prediction results is provided. The method includes: S100. Obtain the current electricity price time series data, current text event data, and current visual event data for the current time period; wherein, the current electricity price time series data corresponds to enhanced electricity price time series features enhanced by historical data, the current text event data corresponds to enhanced text event features enhanced by historical data, and the current visual event data corresponds to enhanced visual event features enhanced by historical data.

[0026] S200. Perform cross-modal event feature fusion on enhanced text event features and enhanced visual event features to obtain multimodal event features.

[0027] S300. Feature modulation and fusion are performed on multimodal event features and enhanced electricity price time series features to obtain multimodal fused time series features, and electricity price prediction is performed based on multimodal fused time series features to obtain predicted electricity price data for future time periods.

[0028] S400. Generate a power generation strategy based on the predicted electricity price data to obtain the target power generation strategy for the future time period, which can be used to adjust the operation of the power generation side.

[0029] The current electricity price time-series data can be the electricity price change data of the target electricity market in which the target power grid participates within the current time period. After acquisition, the current electricity price time-series data is also normalized to eliminate the influence caused by different units of measurement between different data. The data features of the current electricity price time-series data can be obtained in the following way: the current electricity price time-series data is segmented according to a preset step size to obtain multiple overlapping blocks; each overlapping block is mapped to a high-dimensional feature space, and position encoding is superimposed on the mapping result to obtain a block embedding sequence, which serves as the data feature of the current electricity price time-series data.

[0030] Current text event data can be text data related to the target electricity market within the current time period, which is time-aligned with the current electricity price time series data, representing external events affecting the target electricity market within the current time period. Current text event data can include, but is not limited to, data from energy-related news portals, government policy announcements, electricity market transaction reports, and meteorological disaster warnings. The data features of current text event data can be obtained by feature encoding the current text event data using a pre-trained language model. Current visual event data can be visual image data related to the target electricity market within the current time period, which is time-aligned with the current electricity price time series data, representing external events affecting the target electricity market within the current time period. Current visual event data can include, but is not limited to, data from meteorological satellite cloud images, wind radar images, and regional electricity load heat maps. The data features of current visual event data can be obtained by feature encoding the current visual event data using a pre-trained visual model.

[0031] The current electricity price time-series data corresponds to historical electricity price time-series data in one or more historical time periods preceding the current time period. This historical electricity price time-series data can be time-series data with similar data characteristics to the current electricity price time-series data and can serve as historical prior information for the current electricity price time-series data. The historical time periods corresponding to the historical electricity price time-series data also correspond to historical text event data and historical visual event data. Both historical text event data and historical visual event data are time-aligned with the historical electricity price time-series data, and the number of each type of historical text event data and historical visual event data is the same as the number of historical electricity price time-series data.

[0032] In some embodiments, all electricity price data prior to the current time period are stored in an enhanced memory. The electricity price data corresponds to a time period, and based on the time period, there are text data and visual data, stored in the form of multimodal data triples. The process of obtaining historical electricity price time-series data may include: querying the enhanced memory based on the data features of the current electricity price time-series data; calculating the cosine similarity between the data features of the current electricity price time-series data and the data features of the electricity price data in the enhanced memory; sorting the data according to the cosine similarity calculation results; and selecting the multiple electricity price data with the highest cosine similarity as historical electricity price time-series data.

[0033] It is understandable that enhancing electricity price time-series features can be achieved by using historical electricity price time-series data features and applying historical prior enhancement to the current electricity price time-series data features. Similarly, enhancing text event features can be achieved by using historical text event data features and applying historical prior enhancement to the current text event data features. Likewise, enhancing visual event features can be achieved by using historical visual event data features and applying historical prior enhancement to the current visual event data features.

[0034] Reference Figure 2 As shown, the current electricity price time series data, current text event data, and current visual event data for the current time period are obtained. Similar historical electricity price time series data are obtained based on the current electricity price time series data. Historical text event data and historical visual event data are determined based on the historical time periods corresponding to the historical electricity price time series data. Then, based on the historical electricity price time series data, historical text event data, and historical visual event data, corresponding feature enhancements are performed on the current electricity price time series data, current text event data, and current visual event data, respectively, to obtain enhanced electricity price time series features, enhanced text event features, and enhanced visual event features.

[0035] Furthermore, cross-modal event feature fusion is performed on the enhanced text event features and enhanced visual event features to obtain multimodal event features, which serve as a unified enhanced representation of external events. It is understood that both enhanced text event features and enhanced visual event features are in vector form, and the specific methods of cross-modal event feature fusion can be cross-modal attention fusion, early fusion, late fusion, or hybrid fusion, etc.

[0036] It should be noted that in this embodiment, text event data can describe external events related to the target electricity market from a semantic information perspective, while visual event data can describe external events related to the target electricity market from a visualization perspective, providing actual spatial information for external events. By integrating the data characteristics of text event data and visual event data, they can complement each other, jointly describing external events from multiple dimensions, enriching the event information of external events, and providing a sufficient and comprehensive data foundation for subsequent electricity price forecasting and generation strategy generation.

[0037] Furthermore, the objects of cross-modal event feature fusion are enhanced text event features and enhanced visual event features enhanced by historical data. Therefore, the resulting multimodal event features also incorporate external event features from similar situations in the historical time period, thereby effectively strengthening the multimodal event features of the current time period by utilizing historical experience.

[0038] Furthermore, based on multimodal event characteristics and enhanced electricity price time-series characteristics, several external events with the greatest impact on electricity price changes in the target electricity market are identified. These external events are then fused with their corresponding multimodal event characteristics and enhanced electricity price time-series characteristics to obtain multimodal fused time-series characteristics. These multimodal fused time-series characteristics serve as the data foundation for electricity price prediction. Based on these characteristics, electricity price changes in the target electricity market over future time periods are predicted, yielding predicted electricity price data for those periods. It is understood that this embodiment enhances the correlation between external events and electricity price changes by modulating and fusing multimodal event characteristics and enhanced electricity price time-series characteristics. This allows for analysis and prediction from both internal influencing factors of the target electricity market and external influencing factors brought about by external events during subsequent electricity price prediction, improving the accuracy of predicted electricity price data and providing a reliable data foundation for subsequent generation strategy generation.

[0039] Furthermore, based on the predicted electricity price data, generation strategies are generated for each power generation device in the target power grid, yielding a target generation strategy for each device in the future time period. Operational adjustments are then made to each device according to its target generation strategy, enabling it to increase generation when electricity demand increases to meet grid load requirements and decrease generation when demand decreases to avoid electricity waste. It is understandable that generating generation strategies based on predicted electricity price data obtained from multimodal fusion time-series features achieves real-time and accurate responses of generation-side operational strategies to changes in grid supply and demand, improving the matching degree between the target generation strategy and the actual grid environment. This effectively enhances the adjustment effect of generation strategies and ultimately improves the power supply stability of the grid.

[0040] The generation strategy adjustment method based on multimodal electricity price prediction results provided in this embodiment acquires multimodal data such as electricity price time series data, text event data, and visual event data within the current time period, as well as their corresponding enhanced data features; performs cross-modal fusion of multiple enhanced data features to obtain multimodal fused time series features, and makes predictions based on the multimodal fused time series features to obtain predicted electricity price data for future time periods; on this basis, a generation strategy is generated based on the predicted electricity price data to adjust the operation of the generation side.

[0041] Compared with related technologies, this application uses multimodal data such as electricity price time-series data, text event data, and visual event data to perform cross-modal fusion of the data features of each multimodal data as the data foundation for electricity price prediction. This enhances the perception of the actual power grid environment, improves the accuracy of electricity price prediction results, and thus improves the matching degree between the target power generation strategy and the actual power grid environment. It also enables the power generation side operation strategy to respond in real time and accurately to changes in the power grid supply and demand relationship, thereby improving the power supply stability of the power grid.

[0042] Reference Figure 3 As shown, in one embodiment of this application, cross-modal event feature fusion is performed on enhanced text event features and enhanced visual event features to obtain multimodal event features, including: S210. Using the enhanced text event features as the cross-modal query vector and the enhanced visual event features as the cross-modal key vector and cross-modal value vector, calculate the attention weights for the enhanced text event features and the enhanced visual event features to obtain the cross-modal attention weights corresponding to the enhanced text event features and the enhanced visual event features respectively.

[0043] S220. Based on the cross-modal attention weights, attention fusion is performed on the enhanced text event features and the enhanced visual event features to obtain multimodal event features.

[0044] Specifically, enhanced text event features are used as cross-modal query vectors, and enhanced visual event features are used as cross-modal key vectors and cross-modal value vectors. This allows for matching visual event features corresponding to the same external event within the enhanced visual event features, based on the enhanced text event features. Attention weights are calculated for the enhanced text and enhanced visual event features based on the degree of matching, resulting in cross-modal attention weights. Attention fusion is then performed on the enhanced text and enhanced visual event features based on these cross-modal attention weights. This allows for visual feature enhancement of the enhanced text event features using enhanced visual event features, providing a comprehensive description of external events through multimodal features and improving the feature refinement of the multimodal event features.

[0045] For example, the process of cross-modal event feature fusion can be represented as:

[0046] in, Features of multimodal events; To enhance text event features; To enhance visual event features; , and These are the learnable weight matrices.

[0047] Reference Figure 4 As shown, in one embodiment of this application, feature modulation fusion is performed on multimodal event features and enhanced electricity price time-series features to obtain multimodal fused time-series features, including: S310. The event-modulated time-series characteristics of the enhanced electricity price are modulated using multimodal event characteristics to obtain event-modulated time-series characteristics.

[0048] S320. Perform feature fusion on the event modulation timing features and the enhanced electricity price timing features to obtain multimodal fused timing features.

[0049] Specifically, the enhanced electricity price time-series feature is used as the query vector, and the multimodal event feature is used as the key and value vectors. Based on the internal influencing factors of the target electricity market represented by the enhanced electricity price time-series feature, the external influencing factors represented by the multimodal event feature are matched to determine the correlation between the enhanced electricity price time-series feature and the multimodal event feature, thus understanding how external events affect the target electricity market. Based on the above correlation, the enhanced electricity price time-series feature is modulated using the multimodal event feature to obtain event-modulated time-series features that enhance the correlation between external events and electricity price changes. For example, the event-modulated time-series feature can be represented as:

[0050] in, For event modulation timing characteristics; To enhance the time-series characteristics of electricity prices; Features of multimodal events; , and These are the learnable weight matrices.

[0051] Furthermore, the price changes in the target electricity market are influenced by both internal factors and external events, and the degree of influence of both changes over time. To balance the magnitude of the influence between internal and external factors and ensure that the multimodal fused time-series features reflect the actual situation in the current time period, this embodiment fuses the event modulation time-series features and the enhanced electricity price time-series features to obtain multimodal fused time-series features that comprehensively include the features of both internal and external factors.

[0052] In some embodiments, the feature fusion method for event modulation time-series features and enhanced electricity price time-series features can be weighted fusion, where the feature weights can be obtained through a gating network. This gating network can include a linear layer and a sigmoid activation function. The event modulation time-series features and enhanced electricity price time-series features are input into the gating network, and the gating network performs feature learning on the network input, thereby outputting the respective feature weights of the event modulation time-series features and the enhanced electricity price time-series features. For example, the feature weights can be expressed as:

[0053] in, The feature weights have a range of (0,1); and Let these be the weights and biases of the gated network; This represents the Sigmoid activation function; This indicates a splicing operation.

[0054] Based on feature weights, feature fusion is performed on event modulation time-series features and enhanced electricity price time-series features to obtain multimodal fused time-series features, which can be expressed as follows:

[0055] in, Multimodal fusion of temporal features; This represents element-wise multiplication. It can be understood that feature weights... When the value approaches 1, it indicates that the enhanced electricity price time series features account for a higher proportion in the multimodal fusion time series features, and the subsequent electricity price forecasting process relies more on the internal influencing factors of the target electricity market; feature weights When the value approaches 0, it indicates that the event modulation time series features account for a higher proportion in the multimodal fusion time series features, and the subsequent electricity price prediction process is more dependent on the external influencing factors of external events.

[0056] Reference Figure 5 As shown, in one embodiment of this application, the current electricity price time series data corresponds to multiple historical electricity price time series data that meet preset similarity requirements; the enhanced electricity price time series features are obtained through the following method: S110. Extract features from multiple historical electricity price time series data, and aggregate the data features of multiple historical electricity price time series data to obtain unified historical time series features.

[0057] S120. Perform data feature fusion on the unified historical time series characteristics and the current electricity price time series data to obtain the fused electricity price time series characteristics.

[0058] S130. Perform self-attention deep fusion on the time-series features of the integrated electricity price to obtain enhanced time-series features of the electricity price.

[0059] Specifically, based on the current electricity price time-series data, similarity matching is performed on all electricity price data prior to the current time period to obtain multiple historical electricity price time-series data that meet preset similarity requirements, which serve as historical prior information for the current electricity price time-series data. For example, the preset similarity requirement can be that the similarity to the current electricity price time-series data is higher than a preset threshold. Feature extraction is performed on each of the multiple historical electricity price time-series data to obtain the data features of each historical electricity price time-series data, and feature aggregation is performed on all data features to obtain a unified historical time-series feature. For example, the specific method of feature aggregation can be additive aggregation, which can be expressed as:

[0060] in, To unify the characteristics of historical timeline; This represents the k-th historical electricity price time series data; This represents the total number of historical electricity price time-series data.

[0061] Furthermore, based on feature aggregation of historical electricity price time-series data, data feature fusion is performed on the unified historical time-series features and the current electricity price time-series data to obtain fused electricity price time-series features. This unified historical time-series features serve as historical prior information to enhance the actual scenario of the current time period represented by the current electricity price time-series data, thereby improving the ability of the electricity price time-series features to represent the internal influencing factors of the target electricity market. For example, the specific method of data feature fusion can adopt an additive fusion strategy, which can be expressed as follows:

[0062] in, To integrate the time-series characteristics of electricity prices; This refers to the data characteristics of the current electricity price time series data.

[0063] Furthermore, the fused electricity price time-series features are simultaneously used as query vector, key vector, and value vector. Self-attention deep fusion is then performed on the fused electricity price time-series features to extract the correlation between current information and historical prior information, revealing the complex dependencies between them. Based on these relationships, global deep fusion is performed on the current information and historical prior information in the fused electricity price time-series features to enhance their context-awareness, resulting in enhanced electricity price time-series features. For example, the self-attention deep fusion process can be represented as follows:

[0064] in, This indicates multi-head attention operation.

[0065] Reference Figure 6 As shown in one embodiment of this application, the current electricity price time series data corresponds to multiple historical electricity price time series data that meet preset similarity requirements. Each of the multiple historical electricity price time series data corresponds to a historical time period, and each historical time period corresponds to historical text event data and historical visual event data. Enhanced text event features are obtained through the following method: S140. The data features of historical text event data and the data features of current text event data are sequentially concatenated to obtain enhanced text event features.

[0066] Enhanced visual event features are obtained through the following methods: S150. The data features of historical visual event data and the data features of current visual event data are sequentially concatenated to obtain enhanced visual event features.

[0067] Specifically, based on the current electricity price time-series data, similarity matching is performed on all electricity price data prior to the current time period to obtain multiple historical electricity price time-series data that meet preset similarity requirements. The time periods corresponding to the historical electricity price time-series data are taken as historical time periods, and historical text event data and historical visual event data corresponding to the historical time periods are obtained. Enhanced text event features can be obtained by using the data features of historical text event data to perform historical prior enhancement on the data features of the current text event data. For example, the specific method of historical prior enhancement can be to concatenate the two sequences, in which case the enhanced text event features can be represented as:

[0068] in, Features of the current text event; Features of historical text events; This indicates a feature splicing operation.

[0069] Similarly, enhancing visual event features can be achieved by using the data features of historical visual event data to augment the data features of the current visual event data with historical priors. For example, the specific method of historical prior augmentation can be to concatenate the two sequences, in which case the enhanced visual event features can be represented as:

[0070] in, Features of the current visual event; Features of historical visual events.

[0071] It is understandable that by performing historical prior enhancement on the data features of the current text event data and the current visual event data respectively, not only is the ability of the enhanced text event features and the enhanced visual event features to represent the external event features within the current time period strengthened, but the generalization patterns in the historical data can also be learned in the subsequent electricity price forecasting process, thereby improving the accuracy of electricity price forecasting, and thus providing an accurate and reliable data foundation for the generation of power generation strategies, effectively improving the adjustment effect of power generation strategies.

[0072] Reference Figure 7 As shown in the figure, as one embodiment of this application, electricity price prediction is performed based on multimodal fusion time-series features to obtain predicted electricity price data for future time periods, including: S330. Input the multimodal fusion temporal features into the prediction head module; wherein, the prediction head module is trained using mean squared error as the loss function.

[0073] S340. The multimodal fusion time-series features are mapped to a preset future prediction step size through the prediction head module to obtain the predicted electricity price data.

[0074] Specifically, during the training phase of the prediction head module, the prediction head module is trained using a prediction training dataset to obtain the training prediction values ​​output by the prediction head module. The prediction training dataset can include multimodal fusion time-series features prior to the current time period and corresponding real electricity price data. The difference between the training prediction values ​​and the real values ​​in the prediction training dataset is calculated, and the parameters of the prediction head module are optimized based on the difference and the loss function, so that the prediction results output by the prediction head module gradually approach the real values, ultimately completing the training. For example, the prediction head module can be composed of a multilayer perceptron, and the loss function of the prediction head module during training can be the mean squared error, which can be expressed as:

[0075] in, The loss function; This represents the total batch size for training predictions; To preset the future prediction step size; For training prediction values; This is the actual value.

[0076] Furthermore, the multimodal fusion time-series features are input into the prediction head module. Based on the preset future prediction step size, the prediction head module maps the multimodal fusion time-series features to the future time period to obtain the predicted electricity price data for the future time period.

[0077] Reference Figure 8 As shown in the embodiment of this application, a power generation strategy is generated based on predicted electricity price data to obtain a target power generation strategy for a future time period, including: S410. Input the predicted electricity price data into the power generation decision model, and output multiple candidate power generation strategies for future time periods through the power generation decision model; wherein, the power generation decision model is a pre-trained reinforcement learning network model.

[0078] S420. Based on the current power generation strategy for the current time period, calculate the adjustment loss for multiple candidate power generation strategies to obtain the adjustment loss cost for each candidate power generation strategy.

[0079] S430. Based on the adjustment loss cost, multiple candidate power generation strategies are screened to obtain the target power generation strategy.

[0080] Specifically, during the training phase of the power generation decision model, an algorithm based on the Actor-Critic framework is used to train the model. Preset electricity price data is used as input to the model. In each training round, the model generates a power generation strategy based on the preset price data, resulting in a trained power generation strategy. A reward calculation is performed on the trained strategy based on a preset reward function to evaluate the model's decision-making behavior under the preset price data, obtaining the cumulative reward value corresponding to the trained strategy. The model is then updated based on the cumulative reward value, enabling it to generate better power generation strategies. It is understood that the training rounds can be repeated multiple times until a preset termination condition is met. For example, the preset reward function can be derived based on the load satisfaction of the target grid, the wasted electricity of each power generation device, and the safety of the power generation devices. The preset termination condition can be that the cumulative reward value of the trained power generation strategy changes less than a preset threshold across multiple training rounds, or that the number of training rounds reaches its maximum limit.

[0081] Furthermore, based on the trained power generation decision-making model, multiple candidate power generation strategies are generated according to the predicted electricity price data for future time periods. For each power generation device in the target grid, adjustment losses are calculated for each candidate power generation strategy based on its current power generation strategy in the current time period. The cost of adjusting from the current power generation strategy to the candidate power generation strategy is calculated, resulting in the adjustment loss cost for each candidate power generation strategy. Based on this, the candidate power generation strategies are screened according to the adjustment loss cost, and the candidate power generation strategy with the lowest adjustment loss cost is selected as the target power generation strategy. The operation of each power generation device on the power generation side is then adjusted according to the target power generation strategy, thereby achieving real-time and accurate response of the power generation side operation strategy to changes in the power grid supply and demand relationship, and improving the power supply stability of the grid.

[0082] Accordingly, please refer to Figure 9 This application provides a power generation strategy adjustment device based on multimodal electricity price prediction results. The device includes: The electricity price prediction multimodal data acquisition module 910 is used to acquire the current electricity price time series data, current text event data, and current visual event data for the current time period. Among them, the current electricity price time series data corresponds to the enhanced electricity price time series features enhanced by historical data, the current text event data corresponds to the enhanced text event features enhanced by historical data, and the current visual event data corresponds to the enhanced visual event features enhanced by historical data.

[0083] The electricity price prediction cross-modal feature fusion module 920 is used to perform cross-modal event feature fusion on enhanced text event features and enhanced visual event features to obtain multimodal event features.

[0084] The feature fusion electricity price prediction module 930 is used to perform feature modulation fusion on multimodal event features and enhanced electricity price time series features to obtain multimodal fused time series features, and to predict electricity prices based on the multimodal fused time series features to obtain predicted electricity price data for future time periods.

[0085] The power generation strategy adjustment module 940 is used to generate a power generation strategy based on the predicted electricity price data, and obtain the target power generation strategy for the future time period, so as to make operational adjustments on the power generation side.

[0086] In some optional implementations, the electricity price forecasting cross-modal feature fusion module 920 includes: The attention weight calculation unit is used to calculate the attention weights of the enhanced text event features and the enhanced visual event features as cross-modal query vectors and cross-modal key vectors and cross-modal value vectors, respectively, to obtain the cross-modal attention weights corresponding to the enhanced text event features and the enhanced visual event features.

[0087] The cross-modal attention fusion unit is used to perform attention fusion on enhanced text event features and enhanced visual event features according to cross-modal attention weights to obtain multimodal event features.

[0088] In some optional implementations, the feature fusion electricity price prediction module 930 includes: The event feature modulation unit is used to modulate the enhanced electricity price time series features using multimodal event features to obtain event-modulated time series features.

[0089] The feature fusion unit is used to fuse event modulation time-series features and enhanced electricity price time-series features to obtain multimodal fused time-series features.

[0090] In some optional implementations, the electricity price forecasting multimodal data acquisition module 910 includes: The feature extraction and aggregation unit is used to extract features from multiple historical electricity price time series data and aggregate the data features of multiple historical electricity price time series data to obtain unified historical time series features.

[0091] The data feature fusion unit is used to fuse the data features of unified historical time series features and current electricity price time series data to obtain fused electricity price time series features.

[0092] The self-attention deep fusion unit is used to perform self-attention deep fusion on the fused electricity price time series features to obtain enhanced electricity price time series features.

[0093] In some optional implementations, the electricity price forecasting multimodal data acquisition module 910 further includes: The text feature sequence splicing unit is used to splice the data features of historical text event data and the data features of current text event data to obtain enhanced text event features.

[0094] The visual feature sequence splicing unit is used to splice the data features of historical visual event data and the data features of current visual event data to obtain enhanced visual event features.

[0095] In some optional implementations, the feature fusion electricity price prediction module 930 further includes: The prediction head module input unit is used to input multimodal fused temporal features into the prediction head module; wherein, the prediction head module is trained using mean squared error as the loss function.

[0096] The time-series feature mapping unit is used to map multimodal fusion time-series features to a preset future prediction step size through the prediction head module to obtain predicted electricity price data.

[0097] In some optional implementations, the power generation strategy adjustment module 940 includes: The candidate strategy output unit is used to input the predicted electricity price data into the power generation decision model, and output multiple candidate power generation strategies for future time periods through the power generation decision model; wherein, the power generation decision model is a pre-trained reinforcement learning network model.

[0098] The adjustment loss calculation unit is used to calculate the adjustment loss of multiple candidate power generation strategies based on the current power generation strategy in the current time period, and obtain the adjustment loss cost of each candidate power generation strategy.

[0099] The power generation strategy screening unit is used to screen multiple candidate power generation strategies based on the adjustment loss cost to obtain the target power generation strategy.

[0100] Further functional descriptions of the above modules and units are the same as those in the corresponding embodiments described above, and will not be repeated here.

[0101] In this embodiment, the power generation strategy adjustment device based on multimodal electricity price prediction results is presented in the form of a functional unit. Here, a unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0102] Please see Figure 10 , Figure 10This is a schematic diagram of a computer device according to an embodiment of this application. As shown in the figure, the computer device includes one or more processors 10, a memory 20, and interfaces for connecting the various components, including high-speed interfaces and low-speed interfaces. The various components communicate with each other using different buses and can be mounted on a common motherboard or otherwise installed as needed. The processors can process instructions executed within the computer device, including instructions stored in or on memory to display graphical information of a GUI on an external input / output device (such as a display device coupled to the interface). In some alternative embodiments, multiple processors and / or multiple buses can be used with multiple memories and multiple memory modules, if desired. Similarly, multiple computer devices can be connected, each providing some of the necessary operations (e.g., as a server array, a group of blade servers, or a multiprocessor system). Figure 10 Take a processor 10 as an example.

[0103] Processor 10 may be a central processing unit, a network processor, or a combination thereof. Processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The programmable logic device may be a complex programmable logic device (CAMP), a field-programmable gate array (FPGA), a general-purpose array logic (GDA), or any combination thereof.

[0104] The memory 20 stores instructions executable by at least one processor 10 to cause the at least one processor 10 to perform the method shown in the above embodiments.

[0105] The memory 20 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the computer device. Furthermore, the memory 20 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some alternative embodiments, the memory 20 may optionally include memory remotely located relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0106] The memory 20 may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as flash memory, hard disk or solid-state drive; the memory 20 may also include a combination of the above types of memory.

[0107] The computer device also includes a communication interface 30 for communicating with other devices or communication networks.

[0108] This application also provides a computer-readable storage medium. The methods described in this application can be implemented in hardware or firmware, or implemented as recordable on a storage medium, or implemented as computer code downloaded over a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and subsequently stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the methods shown in the above embodiments are implemented.

[0109] This application provides a computer program product including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the method of any embodiment of this application.

[0110] Although embodiments of this application have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of this application, and all such modifications and variations fall within the scope defined by the appended claims.

[0111] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, a computer can be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smartphone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or any combination of these devices.

[0112] For ease of description, the above devices are described separately by function as various units. Of course, in implementing this application, the functions of each unit can be implemented in one or more software and / or hardware.

[0113] 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.

[0114] 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.

[0115] 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.

[0116] 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.

[0117] 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 a process, method, article, or apparatus. Without further limitation, 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 said element.

[0118] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0119] The above description is merely an embodiment of this application and is 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.

[0120] Although embodiments of this application have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of this application, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method for adjusting power generation strategies based on multimodal electricity price forecasting results, characterized in that, The method includes: Acquire the current electricity price time-series data, current text event data, and current visual event data for the current time period; wherein, the current electricity price time-series data corresponds to enhanced electricity price time-series features enhanced by historical data, the current text event data corresponds to enhanced text event features enhanced by historical data, and the current visual event data corresponds to enhanced visual event features enhanced by historical data. Cross-modal event feature fusion is performed on the enhanced text event features and the enhanced visual event features to obtain multimodal event features; The multimodal event features and the enhanced electricity price time series features are fused by feature modulation to obtain multimodal fused time series features, and electricity price is predicted based on the multimodal fused time series features to obtain predicted electricity price data for future time periods. A power generation strategy is generated based on the predicted electricity price data, and a target power generation strategy for the future time period is obtained for adjusting the operation of the power generation side.

2. The method according to claim 1, characterized in that, The process of fusing the enhanced text event features and the enhanced visual event features across modalities to obtain multimodal event features includes: The enhanced text event features are used as cross-modal query vectors, and the enhanced visual event features are used as cross-modal key vectors and cross-modal value vectors. Attention weights are calculated for the enhanced text event features and the enhanced visual event features to obtain the cross-modal attention weights corresponding to the enhanced text event features and the enhanced visual event features respectively. Based on the cross-modal attention weights, attention fusion is performed on the enhanced text event features and the enhanced visual event features to obtain the multimodal event features.

3. The method according to claim 1, characterized in that, The step of performing feature modulation and fusion on the multimodal event features and the enhanced electricity price time-series features to obtain multimodal fused time-series features includes: The enhanced electricity price time series features are modulated using the multimodal event features to obtain event-modulated time series features; The event modulation timing features and the enhanced electricity price timing features are fused to obtain the multimodal fused timing features.

4. The method according to claim 1, characterized in that, The current electricity price time series data corresponds to multiple historical electricity price time series data that meet preset similarity requirements; The enhanced electricity price time-series characteristics are obtained in the following manner: Feature extraction is performed on the multiple historical electricity price time series data, and feature aggregation is performed on the data features of the multiple historical electricity price time series data to obtain unified historical time series features; The unified historical time-series features and the current electricity price time-series data are fused to obtain fused electricity price time-series features; The enhanced electricity price time series features are obtained by performing self-attention deep fusion on the fused electricity price time series features.

5. The method according to claim 1, characterized in that, The current electricity price time series data corresponds to multiple historical electricity price time series data that meet preset similarity requirements. The multiple historical electricity price time series data correspond to historical time periods, and the historical time periods correspond to historical text event data and historical visual event data. The enhanced text event features are obtained in the following way: The enhanced text event features are obtained by concatenating the data features of the historical text event data and the data features of the current text event data. The enhanced visual event features are obtained in the following manner: The enhanced visual event features are obtained by sequentially concatenating the data features of the historical visual event data and the data features of the current visual event data.

6. The method according to any one of claims 1 to 5, characterized in that, The step of predicting electricity prices based on the multimodal fusion time-series features to obtain predicted electricity price data for future time periods includes: The multimodal fused temporal features are input into the prediction head module; wherein, the prediction head module is trained using mean squared error as the loss function; The prediction head module maps the multimodal fusion time-series features to a preset future prediction step size to obtain the predicted electricity price data.

7. The method according to any one of claims 1 to 5, characterized in that, The step of generating a power generation strategy based on the predicted electricity price data to obtain the target power generation strategy for the future time period includes: The predicted electricity price data is input into the power generation decision model, which then outputs multiple candidate power generation strategies for the future time period. The power generation decision model is a pre-trained reinforcement learning network model. Based on the current power generation strategy for the current time period, the adjustment loss is calculated for the multiple candidate power generation strategies to obtain the adjustment loss cost of each candidate power generation strategy. The target power generation strategy is obtained by screening the multiple candidate power generation strategies based on the adjusted loss cost.

8. A power generation strategy adjustment device based on multimodal electricity price prediction results, characterized in that, The device includes: The electricity price prediction multimodal data acquisition module is used to acquire current electricity price time series data, current text event data, and current visual event data for the current time period; wherein, the current electricity price time series data corresponds to enhanced electricity price time series features enhanced by historical data, the current text event data corresponds to enhanced text event features enhanced by historical data, and the current visual event data corresponds to enhanced visual event features enhanced by historical data. The electricity price prediction cross-modal feature fusion module is used to perform cross-modal event feature fusion on the enhanced text event features and the enhanced visual event features to obtain multimodal event features; The feature fusion electricity price prediction module is used to perform feature modulation fusion on the multimodal event features and the enhanced electricity price time series features to obtain multimodal fusion time series features, and to predict electricity prices based on the multimodal fusion time series features to obtain predicted electricity price data for future time periods. The power generation strategy adjustment module is used to generate a power generation strategy based on the predicted electricity price data, and obtain the target power generation strategy for the future time period, so as to adjust the operation of the power generation side.

9. A computer device, characterized in that, include: A memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, the processor executing the computer instructions to perform the method of any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to perform the method of any one of claims 1 to 7.

Images