A power grid scheduling method and device based on multi-scale price prediction, equipment and medium

Abstract

The application relates to the technical field of power grid dispatching, and discloses a power grid dispatching method and device based on multi-scale price prediction, equipment and a medium, time scale feature decomposition is performed on the price time sequence data of a plurality of power regions belonging to a target power grid, feature trend components and feature residual components are obtained, multi-scale trend features are extracted according to the feature trend components, trend correlation features are extracted between the plurality of power regions according to the feature residual components, future price data of a target region is predicted according to the multi-scale trend features and the trend correlation features, and an optimized dispatching scheme is obtained through optimized dispatching decision-making according to the future price data. The method has the beneficial effect that the power resource distribution of the target power grid can be dispatched in a targeted manner, and the safety and stability of the target power grid are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power grid dispatching, and in particular to a power grid dispatching method and device based on multi-scale electricity price prediction, equipment and medium. BACKGROUND

[0002] The balance of power resources between different power regions in a power grid system is crucial. On the one hand, the distribution of power resources affects the power supply stability of each power region and determines whether the power load of each power region can work normally. On the other hand, the imbalance of power resources causes frequency fluctuations between power regions, affecting the safety of the power grid system. Therefore, a suitable power grid dispatching method is needed to dispatch the power grid system in a timely manner to ensure the safe and stable operation of the power grid system. In actual situations, the power resource distribution relationship between multiple regions is relatively complex and cannot be directly calculated and obtained through conventional data, which further leads to the inability to make dispatching decisions for the power grid system through intuitive data, causing delays in dispatching actions and affecting the safety and stability of the power grid system.

[0003] Therefore, there is an urgent need to provide a technical solution that can make dispatching decisions for the power grid system based on intuitive and accurate data. SUMMARY

[0004] The present application provides a power grid dispatching method and device based on multi-scale electricity price prediction, equipment and medium, which makes electricity price prediction for any target region based on the electricity price trend of each power region and the trend correlation between the power regions, so as to serve as data basis for power grid dispatching decisions, thereby enabling targeted dispatching of the power resource distribution of the target power grid and improving the safety and stability of the target power grid.

[0005] In order to achieve the above-mentioned purpose, the main technical solution adopted by the present application includes:

[0006] In a first aspect, the present application provides a power grid dispatching method based on multi-scale electricity price prediction, which comprises:

[0007] Time-scale feature decomposition is performed on the electricity price time series data of multiple power regions to obtain characteristic trend components and characteristic residual components of the electricity price time series data; wherein the multiple power regions belong to a target power grid;

[0008] Multi-scale trend features are extracted from the characteristic trend components of the multiple power regions to obtain multi-scale trend features of the multiple power regions; and trend correlation features between the multiple power regions are extracted from the characteristic residual components to obtain trend correlation features between the multiple power regions;

[0009] According to the multi-scale trend feature and the trend correlation feature, a target region in the plurality of power regions is subjected to electricity price change prediction, to obtain future electricity price data of the target region;

[0010] According to the future electricity price data, an optimized scheduling decision is made, to obtain an optimized scheduling scheme for multi-region scheduling optimization of the target power grid.

[0011] The power grid scheduling method based on multi-scale electricity price prediction provided in the embodiments of the present application decomposes the electricity price time series data of each power region into a feature trend component and a feature residual component according to the time scale feature, extracts trend features from the feature trend component at multiple time scales, to obtain multi-scale trend features, extracts correlation features between the feature residual components among the plurality of power regions, to obtain trend correlation features, predicts the electricity price change of a target region according to the multi-scale trend features and the trend correlation features, and uses the prediction results as data basis to make an optimized scheduling decision for a target power grid, to obtain an optimized scheduling scheme for multi-region scheduling optimization of the target power grid. Compared with related technologies, the present application obtains the electricity price change mode features of each power region at different time scales and the trend correlation features between the plurality of power regions, to predict the electricity price change of any target region, thereby comprehensively characterizing the resource allocation relationship in the power region from multiple dimensions, improving the accuracy and authenticity of the obtained future electricity price data, and further performing multi-region scheduling optimization of the target power grid based on the future electricity price data, to realize targeted scheduling of the power resource allocation of the target power grid, and improve the safety and stability of the target power grid.

[0012] Optionally, the multi-scale trend feature extraction according to the feature trend component of each of the plurality of power regions comprises:

[0013] For any power region in the plurality of power regions, trend features are extracted from the feature trend component of the power region, to obtain scale trend features of the power region at multiple time scales;

[0014] According to the feature trend component of each of the plurality of power regions, weight generation is performed on the plurality of time scales respectively, to obtain time scale weights;

[0015] According to the time scale weights, multi-scale feature fusion is performed on the scale trend features of all power regions at corresponding time scales, to obtain the multi-scale trend features.

[0016] Optionally, the trend correlation feature extraction according to the feature residual component among the plurality of power regions comprises:

[0017] performing time-frequency feature conversion on the feature residual components of the plurality of power regions respectively, to obtain frequency domain feature representations of the plurality of power regions respectively;

[0018] performing correlation relationship learning according to a plurality of frequency components in the frequency domain feature representations, to obtain frequency domain correlation features between the plurality of frequency components;

[0019] performing time domain restoration on the frequency domain correlation features, to obtain trend correlation features between the plurality of power regions.

[0020] Optionally, the time scale feature decomposition on the electricity price time series data of the plurality of power regions respectively, to obtain feature trend components and feature residual components of the electricity price time series data, comprises:

[0021] for any one of the plurality of power regions, performing feature encoding on the electricity price time series data of the any one power region, to obtain time series encoding features of the any one power region;

[0022] based on a preset sliding window, performing trend feature extraction on the time series encoding features in a time sequence direction, to obtain the feature trend components;

[0023] performing residual feature extraction on the time series encoding features according to the feature trend components, to obtain the feature residual components.

[0024] Optionally, the electricity price time series data of the plurality of power regions respectively correspond to time series encoding features, and there are regional correlation features between the electricity price time series data of the plurality of power regions; and the electricity price change prediction on a target region of the plurality of power regions according to the multi-scale trend features and the trend correlation features, to obtain future electricity price data of the target region, comprises:

[0025] performing dynamic feature fusion on the multi-scale trend features and the trend correlation features according to the time series encoding features and the regional correlation features, to obtain trend fusion features;

[0026] performing feature integrity fusion on the trend fusion features and the time series encoding features, to obtain comprehensive trend features between the plurality of power regions;

[0027] performing the electricity price change prediction on the target region according to the comprehensive trend features, to obtain the future electricity price data of the target region.

[0028] Optionally, the electricity price change prediction on the target region according to the comprehensive trend features, to obtain the future electricity price data of the target region, comprises:

[0029] input the comprehensive trend feature into a multi-head prediction model, and predict the price data distribution of the target region by using the multi-head prediction model to obtain a plurality of price data quantile points of the target region within a preset time; wherein the multi-head prediction model is obtained by training according to a quantile loss function;

[0030] According to the plurality of price data quantile points, the price data of the target region within a preset time is predicted to obtain the future price data.

[0031] Optionally, the optimization scheduling decision is made according to the future price data to obtain an optimized scheduling scheme, comprising:

[0032] According to the topological relationship between the plurality of power regions, an initial population is randomly constructed; wherein each particle in the initial population corresponds to an initial scheduling scheme of the target power grid;

[0033] Based on the scheduling constraint condition and the scheduling objective function, the fitness of each particle in the initial population is evaluated to obtain the optimal fitness of the initial population; wherein the scheduling constraint condition is obtained according to the future price data;

[0034] When the optimal fitness does not meet the preset optimization requirement, the initial population is updated to obtain an evolved population; wherein each particle in the evolved population corresponds to an evolved scheduling scheme of the target power grid;

[0035] The evolved population is used as the initial population, and the evolved scheduling scheme is used as the initial scheduling scheme, and the fitness evaluation and population update process are repeated until the preset termination condition is met;

[0036] The optimal fitness corresponds to the scheduling scheme to obtain the optimized scheduling scheme.

[0037] In a second aspect, the embodiments of the present application provide a power grid scheduling device based on multi-scale price prediction, comprising:

[0038] A time scale decomposition module is configured to decompose the price time series data of a plurality of power regions into feature trend components and feature residual components; wherein the plurality of power regions belong to a target power grid;

[0039] A component feature extraction module is configured to extract multi-scale trend features from the feature trend components of the plurality of power regions to obtain the multi-scale trend features of the plurality of power regions, and extract trend correlation features between the plurality of power regions from the feature residual components to obtain the trend correlation features between the plurality of power regions.

[0040] a price change prediction module, configured to perform price change prediction on a target region in the plurality of power regions according to the multi-scale trend feature and the trend correlation feature, to obtain future price data of the target region;

[0041] an optimal scheduling decision module, configured to perform optimal scheduling decision according to the future price data, to obtain an optimal scheduling scheme for multi-region scheduling optimization of the target power grid.

[0042] In a third aspect, an embodiment of the present application provides a computer device, including a memory and a processor, which are in communication connection with each other, and the memory stores computer instructions, and the processor executes the computer instructions to perform the method in any of the above embodiments.

[0043] In a fourth aspect, an embodiment of the present application provides a computer readable storage medium, which stores computer instructions, and the computer instructions are used to make a computer perform the method in any of the above embodiments.

[0044] In a fifth aspect, an embodiment of the present application provides a computer program product, which includes computer instructions, and the computer instructions are used to make a computer perform the method in any of the above embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0045] In order to more clearly illustrate the technical solutions in the specific embodiments or prior art of the present application, the drawings needed in the specific embodiments or prior art description will be briefly introduced as follows. Obviously, the drawings in the following description are some embodiments of the present application, and those skilled in the art can also obtain other drawings according to these drawings without creative labor.

[0046] Figure 1 a step diagram of the power grid scheduling method based on multi-scale price prediction provided by the embodiment of the present application;

[0047] Figure 2 a step diagram of the multi-scale trend feature extraction in the embodiment of the present application;

[0048] Figure 3 a network architecture schematic diagram of the multi-scale trend feature extraction in the embodiment of the present application;

[0049] Figure 4 an architecture schematic diagram of the Mamba expert network in the embodiment of the present application;

[0050] Figure 5 a step diagram of the trend correlation feature extraction in the embodiment of the present application;

[0051] Figure 6 A network architecture diagram for trend correlation feature extraction in the embodiment of the present application;

[0052] Figure 7 A step diagram for time scale feature decomposition in the embodiment of the present application;

[0053] Figure 8 A network architecture diagram for parallel Mamba network in the embodiment of the present application;

[0054] Figure 9 A step diagram for electricity price change prediction in the embodiment of the present application;

[0055] Figure 10 A step diagram for electricity price change prediction in the embodiment of the present application;

[0056] Figure 11 A step diagram for optimization scheduling decision in the embodiment of the present application;

[0057] Figure 12 A module diagram of the power grid scheduling device based on multi-scale electricity price prediction provided in the embodiment of the present application;

[0058] Figure 13 A structural schematic diagram of a computer device provided in the embodiment of the present application. DETAILED DESCRIPTION

[0059] To make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some but not all of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0060] The balance of power resources between different regions in the power grid system is crucial. On the one hand, the distribution of power resources affects the stability of power supply in each region and determines whether the power load in each region can work normally. On the other hand, the imbalance of power resources will cause frequency fluctuations between regions, affecting the safety of the power grid system. Therefore, it is necessary to timely schedule the power grid system through a suitable power grid scheduling method to ensure the safe and stable operation of the power grid system. In actual situations, the power resource distribution relationship between multiple regions is relatively complex and difficult to directly calculate and obtain through conventional data, which further leads to the inability to make scheduling decisions for the power grid system through intuitive data, causing delay in scheduling actions and affecting the safety and stability of the power grid system. Therefore, it is urgent to provide a technical solution that can make scheduling decisions for the power grid system based on intuitive and accurate data.

[0061] Based on the above problems, the application provides a power grid scheduling method and device based on multi-scale price prediction, medium, and equipment. Time scale feature decomposition is performed on the price time series data of each power region belonging to the target power grid to obtain feature trend components and feature residual components. Multi-scale trend features are extracted according to the feature trend components to obtain multi-scale trend features. Trend correlation features are extracted between the power regions according to the feature residual components to obtain trend correlation features. The future price data of the target region is predicted according to the multi-scale trend features and the trend correlation features. The optimized scheduling scheme is obtained by making an optimized scheduling decision according to the future price data.

[0062] The power grid scheduling method based on multi-scale price prediction provided in the application decomposes the price time series data of each power region into feature trend components and feature residual components according to time scale features. Trend features are extracted from the feature trend components at multiple time scales to obtain multi-scale trend features. Correlation features are extracted from the feature residual components between the power regions to obtain trend correlation features. The multi-scale trend features and the trend correlation features are used to predict the price change of the target region. The optimized scheduling scheme for multi-region scheduling optimization of the target power grid is obtained by making an optimized scheduling decision based on the prediction results.

[0063] Compared with related technologies, the application obtains the price change mode features of each power region at different time scales and the trend correlation features between the power regions, which are used to predict the price change of any target region. Therefore, the resource allocation relationship in the power region can be comprehensively characterized in multiple dimensions, the accuracy and authenticity of the obtained future price data are improved, and the multi-region scheduling optimization of the target power grid is performed based on the future price data, which realizes the targeted scheduling of the power resource allocation of the target power grid and improves the safety and stability of the target power grid.

[0064] According to an embodiment of the application, a power grid scheduling method based on multi-scale price prediction is provided. It should be noted that the steps shown in the flowchart can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, the steps shown or described herein can be executed in a different order in some cases.

[0065] Referring to Figure 1 The method includes the following steps.

[0066] S100. Time-scale feature decomposition is performed on the electricity price time series data of each of the plurality of power regions to obtain a characteristic trend component and a characteristic residual component of the electricity price time series data, wherein the plurality of power regions belong to the target power grid.

[0067] S200. Multi-scale trend feature extraction is performed according to the characteristic trend component of each of the plurality of power regions to obtain multi-scale trend features of the plurality of power regions, and trend correlation feature extraction is performed between the plurality of power regions according to the characteristic residual component to obtain trend correlation features between the plurality of power regions.

[0068] S300. Electricity price change prediction is performed on a target region in the plurality of power regions according to the multi-scale trend features and the trend correlation features to obtain future electricity price data of the target region.

[0069] S400. An optimized scheduling decision is made according to the future electricity price data to obtain an optimized scheduling scheme for multi-region scheduling optimization of the target power grid.

[0070] Specifically, the target power grid includes a plurality of power regions, and each power region corresponds to respective electricity price data. According to a preset sample time window, the electricity price change data of each power region in the preset sample time window is extracted to obtain respective electricity price time series data of each power region. After obtaining the electricity price time series data, the electricity price time series data is standardized to eliminate the dimensional differences between different electricity price time series data. Exemplarily, the standardization method can be Z-Score standardization.

[0071] Further, data feature extraction is performed on the electricity price time series data to obtain data features of the electricity price time series data. It should be noted that the electricity price time series data of each power region is simultaneously affected by a plurality of influencing factors, each influencing factor has a different influence mode on the electricity price time series data, and thus the electricity price time series data changes based on a plurality of influence modes when it changes over time, showing a complex overall change. Considering the above reasons, the data features of the electricity price time series data of all power regions are time-scale decomposed to extract the change mode of the electricity price time series data, to obtain a characteristic trend component of the electricity price time series data that changes periodically over time, and a characteristic residual component that changes randomly over time. It can be understood that by time-scale decomposing the data features of the electricity price time series data, the plurality of change modes of the electricity price time series data can be modeled in detail, and thus the influencing factors and the corresponding influence modes of each change mode can be mastered, providing a clear and accurate data basis for electricity price prediction and optimized scheduling decision.

[0072] Further, the characteristic trend component contains the periodic change pattern of each power region at different time scales, and is a coupling characteristic of multiple periodic change patterns. In this case, the change pattern of the price time series data at different time scales cannot be analyzed separately, and pattern confusion is likely to occur. Therefore, the embodiment extracts trend characteristics of the characteristic trend component at multiple time scales to obtain the periodic change pattern of each power region at different time scales, and obtains a multi-scale trend characteristic according to the periodic change pattern of each power region, thereby separating the change pattern of the price time series data at different time scales. It can be understood that the multi-scale trend characteristic can be used to analyze the change pattern at each time scale, thereby enhancing the analysis capability of the dynamic change of the price time series data, and further improving the accuracy of the price change prediction, thereby providing an accurate data basis for the subsequent optimization scheduling decision process.

[0073] Further, for the characteristic residual component, the embodiment extracts trend correlation characteristics of the characteristic residual component between multiple power regions to obtain the mutual relationship between the price data of different power regions, and obtains the trend correlation characteristics. It can be understood that through the trend correlation characteristics, on the one hand, the correlation between the non-periodic change patterns of different power regions can be analyzed to determine the different influences of the sudden events on different power regions, and on the other hand, the mutual influence relationship between different power regions is also related to the periodic change pattern of each power region, thereby providing correlation data basis for the analysis of the periodic change pattern.

[0074] On the basis of obtaining the multi-scale trend characteristics and the trend correlation characteristics, the two are fused to fuse the change pattern of each power region and the pattern correlation between different power regions, and obtain the corresponding feature fusion result. According to the feature fusion result, the price change of a target region in the multiple power regions is predicted to obtain the price change of the target region in a preset future period, and to obtain the future price data of the target region. It can be understood that the feature fusion result of the multi-scale trend characteristics and the trend correlation characteristics represents the overall change pattern of the price data in the target power grid, thereby being capable of performing targeted price change prediction on any target region in the multiple power regions. Exemplarily, the target region can be a power region in the multiple power regions that needs to be predicted, and the price data of which will be changed by one or more influence factors. The number of target regions can be one or more.

[0075] Further, on the basis of the future electricity price data of the target region, an optimal scheduling decision is made for the target power grid to obtain an optimal scheduling scheme for multi-region scheduling optimization of the target power grid, so as to balance and allocate the power resources in the target power grid among the power regions and ensure smooth operation of each power region. It can be understood that the future electricity price data of the target region represents the resource supply and demand relationship of the target region at the corresponding future time, and when the future electricity price data decreases, it means that the power resources of the target region are abundant, and on the basis of maintaining normal work, the power resources of the target region can be dispatched to other power regions to supplement the resource gap of other power regions. When the future electricity price data increases, it means that the power resources of the target region are tight, and there may be a resource gap, which needs to dispatch resources from other power regions to meet the resource demand. Through the future electricity price data, the resource supply and demand relationship of each power region in the target power grid can be comprehensively mastered, and real-time and accurate response of the power generation side operation strategy to the change of the power supply and demand relationship of the power grid is realized, and the power supply stability of the power grid is improved. Therefore, the power resource allocation of the target power grid can be dispatched in a targeted manner, and the safety and stability of the target power grid are improved.

[0076] The power grid scheduling method based on multi-scale electricity price prediction provided by the embodiment decomposes the electricity price time series data of each power region into a characteristic trend component and a characteristic residual component according to the time scale characteristics, extracts trend characteristics of the characteristic trend component at multiple time scales to obtain multi-scale trend characteristics, extracts correlation characteristics between the characteristic residual components of the power regions to obtain trend correlation characteristics, and predicts the electricity price change of the target region according to the multi-scale trend characteristics and the trend correlation characteristics, and uses the prediction result as data basis to make an optimal scheduling decision for the target power grid to obtain an optimal scheduling scheme for multi-region scheduling optimization of the target power grid.

[0077] Compared with the related art, the electricity price change mode characteristics of each power region at different time scales and the trend correlation characteristics between the power regions are obtained to predict the electricity price change of any target region, so that the resource allocation relationship in the power region can be fully characterized in multiple dimensions, the accuracy and authenticity of the obtained future electricity price data are improved, and then the multi-region scheduling optimization of the target power grid is performed on the basis of the future electricity price data, the power resource allocation of the target power grid is dispatched in a targeted manner, and the safety and stability of the target power grid are improved.

[0078] Referring to Figure 2 As an embodiment of the present application, multi-scale trend characteristics are extracted from the characteristic trend components of the multiple power regions to obtain multi-scale trend characteristics of the multiple power regions, including:

[0079] S210. For any power region among multiple power regions, extract the trend features of the characteristic trend components of any power region to obtain the scale trend features of any power region at multiple time scales.

[0080] S220. Based on the characteristic trend components of each of the multiple power regions, weights are generated for each of the multiple time scales to obtain the time scale weights.

[0081] S230. Based on the time scale weight, multi-scale feature fusion is performed on the scale trend characteristics of each power region at the corresponding time scale to obtain multi-scale trend characteristics.

[0082] Specifically, for any given power region, trend features are extracted from its characteristic trend components at multiple time scales to obtain the scale trend features of that power region at multiple time scales. For example, referring to... Figure 3 As shown, trend feature extraction can be performed using a Mamba expert network. The architecture of the Mamba expert network can be referenced. Figure 4 As shown, different Mamba expert networks differ in their dimensionality settings for selective state-space models, enabling them to capture trend features at different time scales. The number of Mamba expert networks is related to the number of selected time scales. Figure 3 Taking three examples, the different Mamba expert networks operate in parallel. The characteristic trend components of any power region are projected onto the state space through a linear mapping, and the projection results are input into the corresponding Mamba expert network. This allows multiple Mamba expert networks to encode the characteristic trend components of any power region, resulting in scale trend features at multiple time scales.

[0083] Furthermore, after obtaining the scale trend features, weights are generated for the scale trend features at each time scale to obtain time scale weights corresponding to the time scale. Then, based on the time scale weights, feature fusion is performed on the characteristic trend components of multiple power regions at each time scale to obtain multi-scale trend features. For example, referring to… Figure 3 As shown, the scale trend features output by each Mamba expert network are input into a gating network G. The gating network G generates time-scale weights for each Mamba expert network, and the outputs of all Mamba expert networks are weighted and fused according to these time-scale weights to obtain multi-scale trend features. The multi-scale trend features can be expressed as:

[0084]

[0085] in, It exhibits multi-scale trend characteristics; is the time scale weight corresponding to the i-th Mamba expert network; is the feature trend component; is the scale trend feature output by the i-th Mamba expert network; is the total number of Mamba expert networks.

[0086] Referring to Figure 5 As an embodiment of the present application, according to the feature residual component, trend correlation feature extraction is performed between multiple power regions to obtain trend correlation features between the multiple power regions, including:

[0087] S240. Time-frequency feature conversion is performed on the feature residual component of each of the multiple power regions to obtain a frequency domain feature representation of each of the multiple power regions.

[0088] S250. Correlation relationship learning is performed according to multiple frequency components in the frequency domain feature representation to obtain frequency domain correlation features between the multiple frequency components.

[0089] S260. Time domain restoration is performed on the frequency domain correlation features to obtain trend correlation features between the multiple power regions.

[0090] Referring to Figure 6 For all power regions, fast Fourier transform is respectively performed on the feature residual components of each of the power regions to convert the feature residual components to the frequency domain to obtain a frequency domain feature representation of each of the multiple power regions. It can be understood that the frequency domain feature representation includes multiple frequency components, each of which corresponds to a specific time scale. The energy of the frequency component corresponds to the influence of the influencing factor on the electricity price data at the time scale. The higher the energy of the frequency component, the stronger the influence of the influencing factor on the electricity price data at the time scale, and vice versa. Exemplarily, the frequency domain feature representation can be represented as:

[0091]

[0092] wherein, is the frequency domain feature representation of the k-th frequency; is the feature residual component of the n-th time point; is the total length of the feature residual component; is the time domain index; is the frequency domain index.

[0093] Further, the frequency domain feature representation is input into a feedforward neural network, which comprises a first linear transformation layer, a nonlinear activation function layer, a Dropout layer and a second linear transformation layer connected in sequence. The first linear transformation layer performs dimension expansion on the frequency domain feature representation to enhance the feature expression capability of the frequency domain feature representation. The frequency domain feature representation after dimension expansion is input into the nonlinear activation function layer, which performs nonlinear transformation on the frequency domain feature representation through a GELU (Gaussian Error Linear Unit) activation function, and prevents overfitting through the Dropout layer. After processing, the frequency domain feature representation processed by multiple layers is input into the second linear transformation layer, which performs dimension compression on the frequency domain feature representation to compress the dimension of the frequency domain feature representation to the dimension before input into the feedforward neural network.

[0094] It should be noted that the feedforward neural network enhances the feature expression capability of the frequency domain feature representation through the linear transformation layer, and performs nonlinear transformation on the frequency domain feature representation through the nonlinear activation function layer to nonlinearly learn the mutual relationship between different frequency components in the frequency domain feature representation, so as to obtain the trend correlation between the power regions in the frequency domain. Exemplarily, the mutual relationship between the frequency components can be a harmonic relationship or an energy distribution mode, etc.

[0095] Further, the frequency domain correlation feature is inverse fast Fourier transformed to convert the frequency domain correlation feature back to the time domain to obtain the correlation feature in the time domain as the trend correlation feature between the multiple power regions. Exemplarily, the trend correlation feature can be represented as:

[0096]

[0097] wherein, is the trend correlation feature; is the frequency domain correlation feature on the kth frequency.

[0098] Referring to FIG. 1, Figure 7 As an embodiment of the present application, the electricity price time series data of the multiple power regions are subjected to time scale feature decomposition to obtain the characteristic trend component and the characteristic residual component of the electricity price time series data, including:

[0099] S110. For any power region in the multiple power regions, the electricity price time series data of the power region are subjected to feature encoding to obtain the time series encoding feature of the power region.

[0100] S120. Based on a preset sliding window, the trend feature of the time series encoding feature is extracted in the time sequence direction to obtain the characteristic trend component.

[0101] S130. The residual feature of the time series encoding feature is extracted according to the characteristic trend component to obtain the characteristic residual component.

[0102] Specifically, before the time-scale feature decomposition is obtained, the respective electricity price time series data of all electricity regions are respectively feature encoded to obtain the data features of the respective electricity price time series data of each electricity region. Referring to Figure 8 As shown, the feature encoding process can be implemented through a parallel Mamba network, which includes two parallel Mamba models that perform opposite encoding processes on the input in the time series direction. The electricity price time series data is input into the parallel Mamba network, and the electricity price time series data is linearly mapped through a linear layer to perform high-dimensional feature embedding on the electricity price time series data to obtain a time series feature tensor. The time series feature tensor is input into the parallel Mamba models, respectively, wherein the time series feature tensor is forward feature encoded in the Mamba model, and the process can be represented as:

[0103]

[0104] wherein, is the forward encoded feature; is the time series feature tensor; denotes the forward feature encoding operation. Similarly, the time series feature tensor is reversely feature encoded in the reverse Mamba model, and the process can be represented as:

[0105]

[0106] wherein, is the reverse encoded feature; is the time series feature tensor after time series reversal; denotes the reverse feature encoding operation.

[0107] Further, the forward encoded feature and the reverse encoded feature output by the parallel Mamba model are feature fused to obtain a parallel encoded feature. Exemplarily, the way of feature fusion of the forward encoded feature and the reverse encoded feature can be additive fusion, and the form can be represented as:

[0108]

[0109] wherein, is the parallel encoded feature. On this basis, the obtained parallel encoded feature is fused with the original time series feature tensor, and the fusion result is normalized to obtain a time series encoded feature. Exemplarily, the time series encoded feature can be represented as:

[0110]

[0111] wherein, The time series encoding feature.

[0112] Further, based on the preset sliding window, a trend feature of the time series encoding feature is extracted by sliding the preset sliding window in the time sequence direction, to obtain a feature trend component. In some embodiments, the trend feature extraction can be implemented by a learnable moving average filter, which can be a one-dimensional convolution layer. By moving the one-dimensional convolution layer in the time sequence direction, a moving average calculation is performed on the time series encoding feature to obtain the feature trend component. Exemplarily, the feature trend component can be represented as:

[0113]

[0114] wherein, is the time series encoding feature of the vth power region.

[0115] Further, after obtaining the feature trend component, a feature residual component is obtained by performing residual feature extraction on the time series encoding feature according to the feature trend component. Exemplarily, the feature residual component can be represented as:

[0116]

[0117] wherein, is the feature residual component.

[0118] Referring to Figure 9 As an embodiment of the present application, the time series data of the electricity price of each of the plurality of power regions corresponds to a time series encoding feature, and the time series data of the electricity price of each of the plurality of power regions has a region association feature. According to the multi-scale trend feature and the trend association feature, the future electricity price data of the target region is obtained by predicting the change of the electricity price of the target region, including:

[0119] S310. According to the time series encoding feature and the region association feature, the multi-scale trend feature and the trend association feature are dynamically fused to obtain a trend fusion feature.

[0120] S320. The trend fusion feature and the time series encoding feature are completely fused to obtain a comprehensive trend feature between the plurality of power regions.

[0121] S330. According to the comprehensive trend feature, the change of the electricity price of the target region is predicted to obtain the future electricity price data of the target region.

[0122] Specifically, the multi-scale trend features and the trend correlation features are input into a gating network, the combination of the multi-scale trend features and the trend correlation features is dynamically learned by the gating network, and after learning, the multi-scale trend features and the trend correlation features are weighted and distributed, so that the multi-scale trend features and the trend correlation features are dynamically fused according to the distributed gating weights, and the trend fusion features are obtained.

[0123] In some embodiments, the gating weights can be obtained according to the time series encoding features, the time series encoding features are taken as the context information of the multi-scale trend features and the trend correlation features, and the dynamic weight generation is performed by the gating network according to the time series encoding features. Exemplarily, the gating network can include two linear convolution layers and a Sigmoid activation function, and the gating weights can be represented as:

[0124]

[0125] wherein, is the gating weight. The trend fusion features obtained according to the gating weights can be represented as:

[0126]

[0127] wherein, is the trend fusion feature; represents element-level multiplication.

[0128] Further, after obtaining the trend fusion features, the trend fusion features are connected with the residual of the time series encoding features, the trend fusion features and the time series encoding features are completely fused, and the comprehensive trend features between the plurality of power regions are obtained to ensure the feature information integrity of the time series data of the electricity price. Exemplarily, the comprehensive trend features can be represented as:

[0129]

[0130] wherein, is the comprehensive trend feature; and LayerNorm represents a layer normalization operation. It can be understood that the target region can be predicted according to the comprehensive trend features, and the future electricity price data of the target region is obtained to generate an optimized scheduling scheme for multi-region scheduling optimization of the target power grid.

[0131] Referring to FIG. 1, Figure 10 As an embodiment of the present application, the target region is predicted according to the comprehensive trend features, and the future electricity price data of the target region is obtained, including:

[0132] S332. Input the comprehensive trend feature into the multi-head prediction model, and predict the price data distribution of the target region by the multi-head prediction model to obtain a plurality of price data quantile points of the target region within the preset time; wherein the multi-head prediction model is obtained by training according to the quantile loss function.

[0133] S334. According to the plurality of price data quantile points, the price data of the target region within the preset time is predicted to obtain future price data.

[0134] Specifically, the multi-head prediction model includes a multi-head linear layer and a linear projection layer, wherein each head of the multi-head linear layer corresponds to a price data quantile point. For example, when the number of heads in the multi-head linear layer is three, each head corresponds to a 10% quantile point, a 50% quantile point and a 90% quantile point, wherein the 10% quantile point corresponds to a pessimistic prediction, the 50% quantile point corresponds to a median prediction, and the 90% quantile point corresponds to an optimistic prediction.

[0135] Further, the multi-head prediction model is obtained by training according to the quantile loss function, and the overall conditional distribution of the price data is learned by the multi-head prediction model driven by the quantile loss function. For example, the quantile loss function can be expressed as:

[0136]

[0137] wherein, is the quantile loss function; is the true value of the price data; is the predicted value of the price data. Based on the quantile loss function, the loss of each quantile point is optimized to adjust the parameters of the linear projection layer, master the mapping relationship between the quantile point and the price data, and obtain the trained multi-head prediction model.

[0138] Further, the comprehensive trend feature is input into the multi-head prediction model to predict the price data distribution of the target region, and a plurality of price data quantile points of the target region within the preset time are obtained. It can be understood that the plurality of price data quantile points collectively form a quantile tensor, which provides rich probabilistic prediction information for price prediction, such as the prediction interval between the 10% quantile point and the 90% quantile point, and realizes advanced risk management and robust decision-making.

[0139] Further, the comprehensive trend feature is projected to the preset time according to the plurality of price data quantile points by the linear projection layer to obtain the future price data of the target region within the preset time. For example, the future price data can be expressed as:

[0140]

[0141] wherein, For the future electricity price data before transposition.

[0142] Referring to Figure 11 As an embodiment of the present application, the optimized scheduling decision is made according to the future electricity price data, and the optimized scheduling scheme is obtained, which includes:

[0143] S410. According to the topological relationship between the plurality of power regions, an initial population is randomly constructed; wherein each particle in the initial population respectively corresponds to an initial scheduling scheme of the target power grid.

[0144] S420. Based on the scheduling constraint condition and the scheduling target function, the fitness of each particle in the initial population is evaluated respectively, and the optimal fitness of the initial population is obtained; wherein the scheduling constraint condition is obtained according to the future electricity price data.

[0145] S430. When the optimal fitness does not meet the preset optimization requirement, the initial population is updated to obtain an evolved population; wherein each particle in the evolved population respectively corresponds to an evolved scheduling scheme of the target power grid.

[0146] S440. The evolved population is taken as the initial population, and the evolved scheduling scheme is taken as the initial scheduling scheme, and the fitness evaluation and population update process are repeated until the preset termination condition is met.

[0147] S450. The optimized scheduling scheme is obtained according to the scheduling scheme corresponding to the optimal fitness.

[0148] Specifically, according to the topological relationship between the plurality of power regions including the target region, an initial population is randomly constructed, and each particle in the initial population respectively corresponds to an initial scheduling scheme of the target power grid. For example, each particle in the initial population can respectively correspond to a random parameter within a reasonable scheduling range, such as the power output value of the generator and the load demand, etc.

[0149] Further, based on the scheduling constraint condition and the scheduling target function, the fitness of each particle in the initial population is evaluated respectively, and the optimal fitness of the initial population is obtained. Wherein, the scheduling constraint condition is obtained according to the future electricity price data, for example, the scheduling constraint condition can limit the power purchase source of the grid load or the intermediate merchant according to the future electricity price data, encourage the grid load or the intermediate merchant to purchase electricity from the power region with lower electricity price, and suppress the grid load or the intermediate merchant to purchase electricity from the power region with higher electricity price. In addition, the scheduling constraint condition obtained according to the future electricity price data can also provide operation constraints for the energy storage system in the target power grid to determine the charging and discharging time of the energy storage system and reduce the waste of power resources.

[0150] In some embodiments, the scheduling objective function for fitness evaluation can be any one or a combination of multiple of the following objectives: minimizing power production cost, minimizing pollutant emission, maximizing utilization of renewable energy, and minimizing power grid load fluctuation. After obtaining the optimal fitness, the optimal fitness is evaluated according to a preset optimization requirement to obtain the fitness evaluation result of the current iteration round. When the optimal fitness does not meet the preset optimization requirement, the initial population is updated, and the position and speed of each particle in the initial population are updated to obtain an evolved population. It can be understood that each particle in the evolved population has a different position and speed from each particle in the initial population, and each particle corresponds to an evolved scheduling scheme of the target power grid, which is closer to the objective represented by the scheduling objective function than the initial scheduling scheme.

[0151] Further, the evolved population is taken as the initial population, and the evolved scheduling scheme is taken as the initial scheduling scheme. The above process of fitness evaluation and population update of the initial population is repeated to perform multiple iteration optimizations on the scheduling scheme corresponding to each particle until a preset termination condition is met, and an optimized scheduling scheme is obtained according to the scheduling scheme corresponding to the optimal fitness. It can be understood that the preset termination condition can be that the number of iteration rounds reaches a preset upper limit, or the variation amplitude of the scheduling objective function in multiple iteration rounds is less than a preset lower limit.

[0152] Correspondingly, with reference to Figure 12 The embodiment of the present application provides a power grid scheduling device based on multi-scale price prediction, which comprises:

[0153] A time scale decomposition module 1210 is configured to perform time scale feature decomposition on the price time series data of each of the plurality of power regions to obtain a characteristic trend component and a characteristic residual component of the price time series data; wherein the plurality of power regions belong to the target power grid.

[0154] A component feature extraction module 1220 is configured to perform multi-scale trend feature extraction on the characteristic trend component of each of the plurality of power regions to obtain multi-scale trend features of the plurality of power regions; and perform trend correlation feature extraction on the characteristic residual component between the plurality of power regions to obtain trend correlation features between the plurality of power regions.

[0155] A price change prediction module 1230 is configured to perform price change prediction on a target region in the plurality of power regions according to the multi-scale trend features and the trend correlation features to obtain future price data of the target region.

[0156] An optimized scheduling decision module 1240 is configured to perform optimized scheduling decision according to the future price data to obtain an optimized scheduling scheme for multi-region scheduling optimization of the target power grid.

[0157] In some optional embodiments, the component feature extraction module 1220 includes:

[0158] a trend feature extraction unit, configured to perform trend feature extraction on the trend component of each of the plurality of power regions, to obtain scale trend features of each of the plurality of power regions at a plurality of time scales.

[0159] a scale weight generation unit, configured to perform weight generation on the plurality of time scales according to the trend component of each of the plurality of power regions, to obtain time scale weights.

[0160] a multi-scale feature fusion unit, configured to perform multi-scale feature fusion on the scale trend features of all of the plurality of power regions at corresponding time scales according to the time scale weights, to obtain multi-scale trend features.

[0161] In some optional embodiments, the component feature extraction module 1220 further includes:

[0162] a time-frequency feature conversion unit, configured to perform time-frequency feature conversion on the residual component of each of the plurality of power regions, to obtain a frequency domain feature representation of each of the plurality of power regions.

[0163] a correlation relationship learning unit, configured to perform correlation relationship learning on a plurality of frequency components in the frequency domain feature representation, to obtain a frequency domain correlation feature between the plurality of frequency components.

[0164] a time domain restoration unit, configured to perform time domain restoration on the frequency domain correlation feature, to obtain a trend correlation feature between the plurality of power regions.

[0165] In some optional embodiments, the time scale decomposition module 1210 includes:

[0166] a feature encoding unit, configured to perform feature encoding on the electricity price time series data of each of the plurality of power regions, to obtain time series encoded features of each of the plurality of power regions.

[0167] a trend feature extraction unit, configured to perform trend feature extraction on the time series encoded features in a time sequence direction based on a preset sliding window, to obtain a trend component.

[0168] a residual feature extraction unit, configured to perform residual feature extraction on the time series encoded features according to the trend component, to obtain a residual component.

[0169] In some optional embodiments, the electricity price change prediction module 1230 includes:

[0170] The dynamic feature fusion unit is configured to perform dynamic feature fusion on the multi-scale trend features and the trend correlation features according to the time-series encoding features and the region correlation features, to obtain trend fusion features.

[0171] The feature integrity fusion unit is configured to perform feature integrity fusion on the trend fusion features and the time-series encoding features, to obtain comprehensive trend features between the plurality of power regions.

[0172] The electricity price change prediction unit is configured to perform electricity price change prediction on the target region according to the comprehensive trend features, to obtain future electricity price data of the target region.

[0173] In some optional embodiments, the electricity price change prediction unit includes:

[0174] The quantile prediction sub-unit is configured to input the comprehensive trend features into a multi-head prediction model, and to perform prediction on the electricity price data distribution of the target region by the multi-head prediction model, to obtain a plurality of electricity price data quantiles of the target region within a preset time; wherein the multi-head prediction model is obtained by training according to a quantile loss function.

[0175] The electricity price data prediction sub-unit is configured to perform prediction on the electricity price data of the target region within the preset time according to the plurality of electricity price data quantiles, to obtain the future electricity price data.

[0176] In some optional embodiments, the optimization scheduling decision module 1240 includes:

[0177] The initial population construction unit is configured to construct an initial population randomly according to the topological relationship between the plurality of power regions; wherein each particle in the initial population corresponds to an initial scheduling scheme of the target power grid.

[0178] The fitness evaluation unit is configured to perform fitness evaluation on each particle in the initial population based on a scheduling constraint condition and a scheduling objective function, to obtain an optimal fitness of the initial population; wherein the scheduling constraint condition is obtained according to the future electricity price data.

[0179] The population updating unit is configured to perform population updating on the initial population when the optimal fitness does not satisfy a preset optimization requirement, to obtain an evolved population; wherein each particle in the evolved population corresponds to an evolved scheduling scheme of the target power grid.

[0180] The process repeating unit is configured to repeat the fitness evaluation and the population updating process by taking the evolved population as the initial population and taking the evolved scheduling scheme as the initial scheduling scheme, until a preset termination condition is satisfied.

[0181] The scheme determination unit is configured to obtain an optimization scheduling scheme according to a scheduling scheme corresponding to the optimal fitness.

[0182] Further function description of each module and unit is the same as the corresponding embodiment described above, and will not be repeated here.

[0183] The power grid scheduling device based on multi-scale price prediction in the embodiment is presented in the form of functional units. The units refer to ASIC (Application Specific Integrated Circuit) circuits, processors and memories that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0184] Please refer to Figure 13 , Figure 13 is a structural schematic diagram of a computer device provided by an embodiment of the present application. As shown in the figure, the computer device includes one or more processors 10, a memory 20, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. Various components are communicatively connected to each other by different buses, and can be installed on a common mainboard or in other ways as needed. The processor can process instructions executed within the computer device, including instructions stored in the memory or on the 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 optional embodiments, multiple processors and / or multiple buses can be used with multiple memories and multiple memory, if necessary. Also, multiple computer devices can be connected, each providing part of the necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Figure 13 In the embodiment, the processor 10 is taken as an example.

[0185] The processor 10 can be a central processor, a network processor, or a combination thereof. The processor 10 can further include a hardware chip. The hardware chip can be an application specific integrated circuit, a programmable logic device, or a combination thereof. The programmable logic device can be a complex programmable logic device, a field programmable logic gate array, a generic array logic, or any combination thereof.

[0186] The memory 20 stores instructions executable by the at least one processor 10, so that the at least one processor 10 executes the method shown in the above embodiment.

[0187] The memory 20 can include a program storage area and a data storage area. The program storage area can store an operating system, application programs required by at least one function, and the like. The data storage area can store data created according to the use of the computer device, and the like. In addition, the memory 20 can include a high-speed random access memory, and can also include a non-transitory memory such as at least one magnetic disk storage device, a flash memory device, or other non-transitory solid-state memory device. In some alternative embodiments, the memory 20 can optionally include a memory disposed remotely from the processor 10, which can be connected to the computer device through a network. Examples of the network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof.

[0188] The memory 20 can include a volatile memory such as a random access memory, and can also include a non-volatile memory such as a flash memory, a hard disk, or a solid state disk. The memory 20 can also include a combination of the above-mentioned kinds of memories.

[0189] The computer device also includes a communication interface 30 for communication of the computer device with other devices or communication networks.

[0190] The embodiments of the present application also provide a computer readable storage medium. The above-mentioned method according to the embodiments of the present application can be implemented in hardware, firmware, or recorded in a storage medium, or implemented as computer code originally stored in a remote storage medium or non-transitory machine readable storage medium and downloaded to a local storage medium, so that the method described herein can be processed by such software on a storage medium using a general-purpose computer, a special-purpose processor, or programmable or special-purpose hardware. The storage medium can be a magnetic disk, an optical disk, a read-only memory, a random access memory, a flash memory, a hard disk, or a solid state disk, etc. Further, the storage medium can also include a combination of the above-mentioned kinds of memories. It can be understood that the computer, the processor, the microprocessor controller, or the programmable hardware includes a storage component that can store or receive software or computer code, which, when accessed and executed by the computer, the processor, or the hardware, implements the method shown in the above-mentioned embodiments.

[0191] The embodiments of the present application provide a computer program product, which includes computer instructions stored in a computer readable storage medium. The processor of the computer device reads the computer instructions from the computer readable storage medium, and the processor executes the computer instructions to make the computer device execute the method of any of the embodiments of the present application.

[0192] Although the embodiments of the present application are described in conjunction with the accompanying drawings, various modifications and changes can be suggested to one skilled in the art, and it is intended that the present application encompass such modifications and changes as fall within the scope of the appended claims.

[0193] The systems, apparatuses, modules or units illustrated by the above embodiments can be specifically implemented by a computer chip or entity, or by a product with certain functions. A typical implementation device is a computer. Specifically, the computer can be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0194] For the sake of brevity, the above description has been described in terms of functional modules. Of course, the modules are not required to be implemented in software and / or hardware. They can be implemented in software and / or hardware, for example.

[0195] Those skilled in the art will understand that the embodiments of the present application can be provided as methods, systems or computer program products. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk memory, CD-ROM, optical memory and the like) containing computer-usable program code.

[0196] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts 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, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device produce the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in a flow or multiple flows and / or blocks Figure 1 The apparatus that implements the functions specified in a flow or multiple flows and / or blocks

[0197] These computer program instructions can also be stored in a computer-readable memory that can direct the computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory produce the manufactured product including the instruction apparatus that implements the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in a flow or multiple flows and / or blocks Figure 1the function specified in the one or more blocks.

[0198] These computer program instructions can also be loaded into a computer or other programmable data processing devices, so that a series of operation steps are performed on the computer or other programmable devices to generate a computer implemented process, thus the instructions executed on the computer or other programmable devices provide a process for implementing the functions described in the flowcharts Figure 1 one or more flowcharts and / or blocks Figure 1 the steps of the function specified in the one or more blocks.

[0199] It should also be noted that the terms "comprising", "comprises", "including", "includes" or any other variations thereof are intended to cover a non-exclusive inclusion, such that processes, methods, articles, or devices that comprise a list of elements do not include only those elements but can also include other elements not expressly listed or inherent to such processes, methods, articles, or devices. Without further limitation, an element preceded by "comprises a" does not, without more constraints, foreclose the existence of additional identical elements in the process, method, article, or device that includes the recited element.

[0200] Each of the embodiments in the present specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the difference from other embodiments. In particular, for the system embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the part of the method embodiments.

[0201] The above only describes the embodiments of the present application and does not limit the present application. Those skilled in the art can make various modifications and changes to the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the scope of the claims of the present application.

[0202] Although the embodiments of the present application are described in conjunction with the drawings, various modifications and changes can be made by those skilled in the art without departing from the spirit and scope of the present application, and such modifications and changes shall fall within the scope defined by the appended claims.

Claims

1. A power grid dispatching method based on multi-scale electricity price forecasting, characterized in that, The method includes: The time-scale feature decomposition of electricity price time-series data for multiple power regions is performed to obtain the characteristic trend component and characteristic residual component of the electricity price time-series data; wherein, the multiple power regions belong to the target power grid; Multi-scale trend features are extracted based on the characteristic trend components of each of the multiple power regions to obtain the multi-scale trend features of the multiple power regions; Fast Fourier Transform is performed on the characteristic residual components of each of the multiple power regions to obtain the frequency domain feature representations of each of the multiple power regions; The frequency domain feature representations are input into a feedforward neural network, and the feedforward neural network performs dimensional expansion, nonlinear transformation, and dimensional compression on the frequency domain feature representations to obtain frequency domain correlation features; Inverse Fast Fourier Transform is performed on the frequency domain correlation features to obtain the time domain correlation features, which serve as the trend correlation features between the multiple power regions; Based on the multi-scale trend characteristics and the trend correlation characteristics, electricity price changes are predicted for target areas in the multiple power regions to obtain future electricity price data for the target areas. Based on the topological relationships between the multiple power regions, an initial population is randomly constructed. Each particle in the initial population corresponds to an initial scheduling scheme for the target power grid, including random parameters within a reasonable scheduling range, such as generator power output and load power demand. Based on scheduling constraints and a scheduling objective function, the fitness of each particle in the initial population is evaluated to obtain the optimal fitness of the initial population. The scheduling constraints are obtained based on future electricity price data. When the optimal fitness does not meet preset optimization requirements, the initial population is updated to obtain an evolved population. Each particle in the evolved population corresponds to an evolved scheduling scheme for the target power grid. Using the evolved population as the initial population and the evolved scheduling scheme as the initial scheduling scheme, the fitness evaluation and population update process is repeated until a preset termination condition is met. An optimized scheduling scheme is obtained based on the scheduling scheme corresponding to the optimal fitness to optimize multi-region scheduling of the target power grid.

2. The method according to claim 1, characterized in that, The step of extracting multi-scale trend features based on the characteristic trend components of each of the multiple power regions to obtain the multi-scale trend features of the multiple power regions includes: For any one of the multiple power regions, trend features are extracted from the characteristic trend components of the power region to obtain the scale trend features of the power region at multiple time scales. Based on the characteristic trend components of each of the multiple power regions, weights are generated for each of the multiple time scales to obtain time scale weights; Based on the time scale weights, the scale trend characteristics of each power region at the corresponding time scale are fused using multi-scale features to obtain the multi-scale trend characteristics.

3. The method according to claim 1, characterized in that, The process of performing time-scale feature decomposition on the electricity price time-series data for multiple power regions to obtain the characteristic trend component and characteristic residual component of the electricity price time-series data includes: For any one of the multiple power regions, the electricity price time series data of any one power region is feature-encoded to obtain the time series coding feature of any one power region; Based on a preset sliding window, trend features are extracted from the temporal coding features in the time sequence direction to obtain the feature trend components. The residual features of the temporal coding features are extracted based on the characteristic trend components to obtain the characteristic residual components.

4. The method according to claim 1, characterized in that, The electricity price time-series data of each of the multiple power regions correspond to time-series coding features, and there are regional correlation features among the electricity price time-series data of the multiple power regions; the step of predicting electricity price changes for a target region among the multiple power regions based on the multi-scale trend features and the trend correlation features to obtain future electricity price data for the target region includes: Based on the temporal coding features and the regional correlation features, the multi-scale trend features and the trend correlation features are dynamically fused to obtain trend fusion features; The trend fusion feature and the time-series coding feature are fully fused to obtain the comprehensive trend feature among the multiple power regions; Based on the comprehensive trend characteristics, electricity price changes in the target area are predicted to obtain future electricity price data for the target area.

5. The method according to claim 4, characterized in that, The step of predicting electricity price changes in the target area based on the comprehensive trend characteristics to obtain future electricity price data for the target area includes: The comprehensive trend features are input into the multi-bulk prediction model, which then predicts the distribution of electricity price data in the target area to obtain multiple electricity price data quantiles in the target area within a preset time period. The multi-bulk prediction model is trained based on a quantile loss function. Based on the multiple electricity price data quantiles, the electricity price data of the target area within a preset time period is predicted to obtain the future electricity price data.

6. A power grid dispatching device based on multi-scale electricity price forecasting, characterized in that, The device includes: The time-scale decomposition module is used to perform time-scale feature decomposition on the electricity price time-series data of multiple power regions to obtain the characteristic trend component and characteristic residual component of the electricity price time-series data; wherein, the multiple power regions belong to the target power grid; The component feature extraction module is used to extract multi-scale trend features based on the characteristic trend components of each of the multiple power regions, thereby obtaining multi-scale trend features of the multiple power regions; to perform fast Fourier transform on the characteristic residual components of each of the multiple power regions, thereby obtaining frequency domain feature representations of each of the multiple power regions; to input the frequency domain feature representations into a feedforward neural network, and to perform dimensional expansion, nonlinear transformation, and dimensional compression on the frequency domain feature representations through the feedforward neural network, thereby obtaining frequency domain correlation features; and to perform inverse fast Fourier transform on the frequency domain correlation features, thereby obtaining time domain correlation features, which serve as trend correlation features between the multiple power regions. The electricity price change prediction module is used to predict the electricity price change of a target area in the multiple power regions based on the multi-scale trend features and the trend correlation features, so as to obtain the future electricity price data of the target area. An optimized scheduling decision module is used to randomly construct an initial population based on the topological relationships between the multiple power regions. Each particle in the initial population corresponds to an initial scheduling scheme for the target power grid, including random parameters within a reasonable scheduling range, such as generator power output and load power demand. Based on scheduling constraints and a scheduling objective function, the fitness of each particle in the initial population is evaluated to obtain the optimal fitness of the initial population. The scheduling constraints are obtained based on future electricity price data. When the optimal fitness does not meet preset optimization requirements, the initial population is updated to obtain an evolved population. Each particle in the evolved population corresponds to an evolved scheduling scheme for the target power grid. The evolved population is used as the initial population, and the evolved scheduling scheme is used as the initial scheduling scheme. The fitness evaluation and population update process is repeated until a preset termination condition is met. An optimized scheduling scheme is obtained based on the scheduling scheme corresponding to the optimal fitness to optimize multi-region scheduling of the target power grid.

7. 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 5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing a computer to perform the method of any one of claims 1 to 5.

Images