Method and System for Generating Operation Strategy of User-Side Energy Storage

Abstract

The present application discloses a method and system for generating a user-side energy storage operation strategy, which relates to the field of energy storage strategy generation. First, it performs time series encoding on historical electricity prices to extract periodic patterns. At the same time, it converts parameters such as temperature and light in weather forecasts and holiday event information into structured vectors, and realizes semantic alignment in a high-dimensional hidden space to form a feature combination body containing external interference factors. Then, it adopts an internal-external collaborative response coding mechanism to jointly encode the electricity price time series features and external factors for chain interaction. Finally, the generated electricity price prediction result comprehensively considers the dual effects of external factors on load and market supply and demand, so as to support the energy storage system to execute precise charge and discharge strategies in the peak and valley intervals of electricity prices, effectively overcoming the problem of strategy lag caused by data fragmentation in traditional methods.

Description

Technical Field

[0001] This application relates to the field of energy storage strategy generation, and more specifically, to a method and system for generating an energy storage operation strategy on the user side. Background Art

[0002] With the deepening of the electricity market mechanism and the increasing penetration of distributed energy sources, energy storage on the user side has become a key facility for reducing electricity costs and enhancing the energy self-balancing ability. The core lies in dynamically adjusting the charge and discharge behavior of energy storage to minimize electricity costs and maximize economic benefits. As the electricity market gradually opens up, time-of-use electricity price mechanisms and demand charge models are widely implemented, and the economic benefits of user-side energy storage systems highly depend on the accurate prediction of peak and valley electricity price periods and load peaks.

[0003] Since the electricity consumption behavior of users is easily affected by external events such as sudden weather changes and holiday activities, the load curve shows non-stationary fluctuation characteristics; at the same time, the electricity market is driven by factors such as supply and demand relationships and fluctuations in new energy output, and electricity price signals often contain complex non-linear dynamic characteristics.

[0004] Traditional prediction methods usually independently analyze electricity price or load data using a single-dimensional time series model. They neither perform multi-modal feature fusion on structured parameters such as temperature and sunlight in weather forecasts and unstructured information of special events, nor establish an association mapping between the dynamic nature of the external environment and historical electricity price fluctuation patterns. For example, when a cold snap strikes, the traditional ARIMA model may underestimate the time period when peak electricity prices appear due to its lack of ability to predict the sudden increase in air-conditioning loads; and if the lighting load superposition caused by family gatherings during holidays is not included in the prediction system, it will cause the energy storage system to discharge prematurely, missing higher price difference benefits. More critically, in the prior art, external factors such as weather and events and the time series characteristics of electricity prices are often in separate analysis dimensions, making it difficult to capture the relationship between a sudden rise in temperature and the advance of the electricity price peak period, and also unable to analyze the potential impact of load mutations triggered by special events on the market supply and demand balance. This makes the prediction error of traditional methods increase significantly when facing multi-dimensional interference factors, ultimately leading to the deviation of the energy storage operation strategy from the optimal economic path.

[0005] Therefore, an optimized solution for generating an energy storage operation strategy on the user side is desired. Summary of the Invention

[0006] To solve the above technical problems, the present application is proposed. According to one aspect of the present application, there is provided a method for generating a user-side energy storage operation strategy, which includes: obtaining market historical electricity price data; obtaining short-term weather forecast data and short-term event data; performing time series encoding on the market historical electricity price data to obtain an electricity price time series pattern feature encoding vector; performing structured joint encoding on the short-term weather forecast data and the short-term event data to obtain an external factor joint structured encoding vector; performing time series prediction based on short-term electricity price on the electricity price time series pattern feature encoding vector and the external factor joint structured encoding vector to obtain a short-term electricity price prediction result, including: performing inner-outer collaborative response encoding with chain local interaction on the electricity price time series pattern feature encoding vector and the external factor joint structured encoding vector to obtain an electricity price time series pattern feature modulation encoding vector; performing feature decoding on the electricity price time series pattern feature modulation encoding vector to obtain the short-term electricity price prediction result; and determining a user-side energy storage operation strategy based on the short-term electricity price prediction result.

[0007] According to another aspect of the present application, there is provided a system for generating a user-side energy storage operation strategy, which includes: a market historical electricity price data acquisition module for obtaining market historical electricity price data; a short-term data acquisition module for obtaining short-term weather forecast data and short-term event data; an electricity price time series encoding module for performing time series encoding on the market historical electricity price data to obtain an electricity price time series pattern feature encoding vector; an external factor joint module for performing structured joint encoding on the short-term weather forecast data and the short-term event data to obtain an external factor joint structured encoding vector; a short-term electricity price prediction module for performing time series prediction based on short-term electricity price on the electricity price time series pattern feature encoding vector and the external factor joint structured encoding vector to obtain a short-term electricity price prediction result, wherein the short-term electricity price prediction module is configured to: perform inner-outer collaborative response encoding with chain local interaction on the electricity price time series pattern feature encoding vector and the external factor joint structured encoding vector to obtain an electricity price time series pattern feature modulation encoding vector; perform feature decoding on the electricity price time series pattern feature modulation encoding vector to obtain the short-term electricity price prediction result; and a user-side energy storage operation strategy generation module for determining a user-side energy storage operation strategy based on the short-term electricity price prediction result.

[0008] Compared with the prior art, a method and system for generating a user-side energy storage operation strategy provided by the present application first perform temporal encoding on historical electricity prices to extract periodic patterns. At the same time, parameters such as temperature and sunlight in weather forecasts and holiday event information are converted into structured vectors, and semantic alignment is achieved in a high-dimensional latent space to form a feature combination body containing external interference factors. Then, an internal-external collaborative response encoding mechanism is adopted to enable chained interaction between the electricity price temporal features and external factors through joint encoding. Finally, the generated electricity price prediction result comprehensively considers the dual effects of external factors on load and market supply and demand, thereby supporting the energy storage system to execute precise charge and discharge strategies in the peak-valley intervals of electricity prices and effectively overcoming the problem of strategy lag caused by data fragmentation in traditional methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] The above and other objects, features, and advantages of the present application will become more apparent by describing the embodiments of the present application in more detail with reference to the accompanying drawings. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. They are used to explain the present application together with the embodiments of the present application and do not constitute a limitation to the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0010] Figure 1 It is a flowchart of the method for generating a user-side energy storage operation strategy according to an embodiment of the present application.

[0011] Figure 2 It is a flowchart of step S140 in the method for generating a user-side energy storage operation strategy according to an embodiment of the present application.

[0012] Figure 3 It is a flowchart of step S150 in the method for generating a user-side energy storage operation strategy according to an embodiment of the present application.

[0013] Figure 4 It is a flowchart of step S151 in the method for generating a user-side energy storage operation strategy according to an embodiment of the present application.

[0014] Figure 5 It is a block diagram of the system for generating a user-side energy storage operation strategy according to an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0015] Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although some embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are for illustrative purposes only and are not used to limit the protection scope of the present disclosure.

[0016] Therefore, in view of the problems in the above-mentioned background art, the technical concept of this application is to establish the correlation between external environmental interference and electricity price fluctuation patterns through multi-modal data collaborative encoding and dynamic feature modulation mechanism. Specifically, first, perform time-series encoding on historical electricity prices to extract periodic patterns, and at the same time convert the temperature, light parameters of weather forecasts and event information such as holidays into structured vectors respectively; achieve semantic alignment of meteorological data and event data through high-dimensional latent space mapping to form a feature combination incorporating external interference factors. Subsequently, adopt an internal-external collaborative response encoding mechanism to jointly encode the electricity price time-series features and external factors for chain interaction. For example, capture the correlation strength between cold wave forecast data and historical electricity price peak sections, and dynamically adjust the electricity price prediction weights for future periods. This mechanism enables the model to analyze the conduction effect of the change in supply-demand relationship caused by the sharp increase in air-conditioning load on electricity prices without explicitly inputting load data - indirectly representing the impact of load mutations on the market through implicit correlation modeling between temperature parameters and electricity price fluctuations. The finally generated electricity price prediction results essentially integrate the dual effects of external factors on load and market supply-demand, thereby supporting the energy storage system to execute precise charge-discharge strategies during peak and valley electricity price intervals and overcoming the problem of strategy lag caused by data fragmentation in traditional methods.

[0017] Figure 1 FIG. is a flowchart of a method for generating a user-side energy storage operation strategy according to an embodiment of the present application. As Figure 1 shown, the method for generating a user-side energy storage operation strategy according to an embodiment of the present application includes: S110, obtaining market historical electricity price data; S120, obtaining short-term weather forecast data and short-term event data; S130, performing time-series encoding on the market historical electricity price data to obtain an electricity price time-series pattern feature encoding vector; S140, performing structured joint encoding on the short-term weather forecast data and the short-term event data to obtain an external factor joint structured encoding vector; S150, performing time-series prediction based on short-term electricity prices on the electricity price time-series pattern feature encoding vector and the external factor joint structured encoding vector to obtain a short-term electricity price prediction result; S160, determining a user-side energy storage operation strategy based on the short-term electricity price prediction result.

[0018] In step S110, market historical electricity price data is obtained. It should be understood that market historical electricity price data refers to the recorded data of the change of electricity prices over time in the power market in the past period, including electricity price information in different time periods, different seasons, different electricity consumption scenarios, etc. In particular, the electricity prices in the power market have certain periodicity and regularity. By analyzing market historical electricity price data, the change patterns of electricity prices on different time scales such as daily, weekly, monthly, and even annually can be found. For example, the occurrence time and price level of peak-valley electricity prices, as well as seasonal electricity price fluctuation characteristics, etc. These rules are crucial for predicting future electricity price trends.

[0019] In the specific implementation process, when obtaining historical market electricity price data, the power market data service platform is called through the interface to batch collect multi-period electricity price data such as intraday electricity price curve, day-ahead electricity price curve, and real-time settlement price. After collection, these data are uniformly time-series standardized to eliminate the time alignment error of data from different sources, ensure that the obtained market historical electricity price data is consistent and comparable in time dimension, and provide accurate and standardized basic data for subsequent time series coding and other processing of market historical electricity price data, so that subsequent analysis can be implemented based on a unified time benchmark, avoiding the interference of data source time series offset on the accuracy of strategy generation.

[0020] In step S120, short-term weather forecast data and short-term event data are obtained. Accordingly, short-term weather forecast data refers to forecast information on weather conditions in the near future (usually the next few days), including factors such as temperature, humidity, and light intensity. Short-term event data refers to calendar information and event markers related to a specific date, including date types (such as weekdays, weekends), holiday markers, and special event markers (such as large-scale events, emergencies, etc.). Since users' electricity consumption behavior is easily affected by weather changes and special events, the electricity price in the power market will also be driven by factors such as supply and demand relationships and fluctuations in new energy output. By obtaining short-term weather forecast data and short-term event data, these external factors can be taken into consideration to more comprehensively analyze and predict the changing trend of electricity prices.

[0021] In the specific implementation process, when obtaining short-term weather forecast data, real-time capture is carried out by connecting to the National Meteorological Administration, regional meteorological centers or third-party meteorological data API interfaces. At the technical implementation level, the open API interface of the meteorological data network can be called to obtain various types of meteorological data such as ground, high altitude, and satellite, or the meteorological data unified service interface (MUSIC) can be used to realize site data retrieval, grid data analysis and file product download, supporting multiple data formats such as XML / JSON and programming languages such as C# and Java. Third-party meteorological data services such as Xinzhi Weather and QWeather provide high-precision grid-level forecasts, and minute-level updated meteorological actual and forecast data can be obtained through the RestfulAPI interface to meet the refined requirements of short-term energy storage strategies for meteorological parameters. The acquisition of short-term event data requires the combination of multi-source information, including official holiday arrangements issued by the government, event announcements on social media platforms, and user-defined event lists. User-defined events establish standardized data formats and are integrated into the system through interfaces or manual entry.

[0022] In step S130, the market historical electricity price data is time-series encoded to obtain an electricity price time-series pattern feature encoding vector. Specifically, in an embodiment of the present application, the market historical electricity price data is time-series encoded to obtain an electricity price time-series pattern feature encoding vector, including: performing time-series encoding based on causal expansion convolution on the market historical electricity price data to obtain the electricity price time-series pattern feature encoding vector. It should be understood that considering that the historical electricity price data not only contains periodic laws (such as daily peaks and valleys, weekly cycles), but also implicitly contains non-stationary time series characteristics triggered by events such as sudden changes in supply and demand relationships and fluctuations in new energy output. Although traditional time series encoding methods (such as sliding averages and Fourier transforms) can extract explicit periodic patterns, it is difficult to capture causal dependencies across time scales in electricity price sequences. For example, a surge in load caused by a cold wave may cause the peak of the next day's electricity price to be advanced by several hours. This delayed causal effect needs to be identified through long-term time series modeling. However, ordinary convolutional neural networks are limited by local receptive fields and cannot effectively model potential correlations across dozens of hours or even days in electricity price fluctuations, resulting in key features (such as market inertia after extreme events) being ignored, which directly affects the accuracy of subsequent electricity price forecasts. Based on this, in this application, the market historical electricity price data is time-series encoded based on causal expansion convolution to obtain the electricity price time series pattern feature encoding vector. Specifically, the time series encoding based on causal expansion convolution expands the receptive field layer by layer under the premise of strictly maintaining time causality by superimposing multiple layers of expansion convolution kernels. For example, the first layer of convolution captures hourly fluctuations (such as intraday peaks and valleys), the second layer is expanded to the daily cycle pattern through the expansion coefficient, and the third layer further associates the weekly cycle trend. The features output by each layer depend only on the current and historical time data to avoid future information leakage. Through this hierarchical feature extraction, short-term disturbances (such as a sudden rise in electricity prices at noon), medium-term cycles (such as differences between weekdays and weekends), and long-term event associations (such as the continued high price trend after extreme weather) in electricity price data are encoded as multi-dimensional electricity price time series pattern feature vectors.

[0023] In step S140, the short-term weather forecast data and the short-term event data are structured jointly encoded to obtain an external factor joint structured coding vector. Figure 2 Flow chart of step S140 in the method for generating a user-side energy storage operation strategy according to an embodiment of the present application. Specifically, in the embodiment of the present application, Figure 2As shown, in step S140, performing structured joint encoding on the short-term weather forecast data and the short-term event data to obtain an external factor joint structured encoding vector, including: S141, performing structured encoding on the short-term weather forecast data and the short-term event data to obtain a short-term weather forecast data structured encoding vector and a short-term event data structured encoding vector; S142, mapping the short-term weather forecast data structured encoding vector and the short-term event data structured encoding vector to the same high-dimensional latent space to obtain an aligned short-term weather forecast data structured encoding vector and an aligned short-term event data structured encoding vector; S143, combining the aligned short-term weather forecast data structured encoding vector and the aligned short-term event data structured encoding vector to obtain the external factor joint structured encoding vector.

[0024] Specifically, in step S141, performing structured encoding on the short-term weather forecast data and the short-term event data to obtain a short-term weather forecast data structured encoding vector and a short-term event data structured encoding vector. Correspondingly, considering that short-term weather forecasts (such as temperature, light intensity) and special events (such as holidays, family gatherings) are essentially multi-modal heterogeneous data: weather forecasts contain continuous structured parameters (such as temperature values), while event data are mostly discrete unstructured information (such as the "National Day holiday" label). If these raw data are directly input into the model, there will be a semantic gap problem - it is difficult to establish an effective association between the time series changes in temperature and the sudden characteristics of event labels within the same analysis framework. For example, there are significant differences in the data forms between the continuous low-temperature curve of a cold wave forecast and the single-point marker of a "family gathering" event. Without unified representation, the model cannot capture the synergistic impact of the sudden drop in temperature and the superimposed lighting load during the gathering period on electricity prices. Based on this, in the technical solution of this application, performing structured encoding on the short-term weather forecast data and the short-term event data to obtain a short-term weather forecast data structured encoding vector and a short-term event data structured encoding vector. Specifically, for weather forecast data, parameters strongly related to the electricity load, such as temperature and light, are extracted, and their time series (such as hourly temperature for the next 24 hours) are converted into structured vectors with physical meanings, for example, by using one-dimensional convolution to capture temporal information. For event data, a semantic embedding method is used to map discrete event labels (such as "cold wave warning", "Spring Festival holiday") to low-dimensional dense vectors.

[0025] Specifically, in step S142, the short-term weather forecast data structured encoding vector and the short-term event data structured encoding vector are mapped to the same high-dimensional latent space to obtain an aligned short-term weather forecast data structured encoding vector and an aligned short-term event data structured encoding vector. It should be understood that considering that the operation strategy of user-side energy storage highly depends on the coupling relationship between the external environment dynamics and electricity price fluctuations, there are essential differences in the data forms between weather forecast data (such as temperature curves, light intensity) and event data (such as holiday labels, special events): the former is a continuous time series, and the latter is mostly discrete semantic labels. If the two are directly concatenated or processed independently, cross-modal semantic associations cannot be established. For example, the continuous low temperature during a cold wave and the "family gathering" event belong to different dimensions in the original data space, and it is difficult for the model to recognize the possible resonance effect of air conditioner and lighting loads caused by their superposition. This semantic fragmentation of heterogeneous data will lead to a fragmented analysis of the impact of external factors on electricity prices and cannot accurately capture the sudden change of the supply-demand relationship under the synergistic effect of multiple factors. Based on this, in the technical solution of this application, the short-term weather forecast data structured encoding vector and the short-term event data structured encoding vector are mapped to the same high-dimensional latent space to obtain an aligned short-term weather forecast data structured encoding vector and an aligned short-term event data structured encoding vector. Specifically, a non-linear transformation matrix is used to perform spatial transformation on meteorological feature vectors such as temperature and light, and event embedding vectors respectively, so that they have comparable vector distributions in the hidden layer. For example, after the temperature drop trend vector of a cold wave forecast is mapped, it may show a neighboring distribution with the "extreme weather warning" event vector in the high-dimensional space, while the "holiday" event vector has a topological association with the electricity price peak period vector in the same historical period. In actual calculations, the parameters of the non-linear transformation matrix can be continuously optimized through model training. Its core idea is similar to finding a suitable "coordinate transformation" method for different modal data in the high-dimensional space, so that they can be compared and fused under the same measurement system.

[0026] Specifically, a non-linear transformation matrix is used to perform spatial transformation on meteorological feature vectors and event embedding vectors. Taking the cold wave forecast as an example, its temperature drop trend vector and the "extreme weather warning" event vector are significantly different in the original data space, but through the action of the non-linear transformation matrix, they can show a neighboring distribution in the hidden layer. This means that in the high-dimensional space, they are closer in position, which is convenient for the model to identify the potential connection between the two. Similarly, after the "holiday" event vector is transformed, it can have a topological association with the electricity price peak period vector in the same historical period. This association can be understood as in the high-dimensional space, their relative positions and connection relationships reflect the internal connection between the actual data.

[0027] In actual calculations, the parameters of the non - linear transformation matrix are continuously optimized through model training. The model training process is like a process of continuous exploration and adjustment. Based on a large amount of historical data, it searches for the most suitable matrix parameters, enabling different - modality data to be compared and fused under the same metric system. For example, during multiple trainings, the model will continuously adjust the non - linear transformation matrix according to the actual data feedback of the impact of temperature drops and "extreme weather warnings" on electricity prices, making the distributions of these two vectors in the high - dimensional space more reasonable and better reflecting their synergy in influencing electricity prices. In this way, the structured - encoded vectors of short - term weather forecast data and the structured - encoded vectors of short - term event data are mapped to the same high - dimensional latent space, obtaining comparable and related aligned vectors, laying a solid foundation for subsequent analysis of the comprehensive impact of external factors on electricity prices.

[0028] Specifically, in step S143, the aligned structured - encoded vector of short - term weather forecast data and the aligned structured - encoded vector of short - term event data are combined to obtain the jointly structured - encoded vector of external factors. Correspondingly, considering that the generation of the user - side energy storage strategy needs to comprehensively consider the synergistic effects of multiple external factors, but after the weather forecast data and event data are aligned in the high - dimensional space, they are still independently represented and cannot directly reflect the joint impact mechanism of the two on electricity price fluctuations. If the method only analyzes the weather or events in isolation, it will ignore the interaction between the two in the time - space dimension (such as load resonance when extreme weather and special events occur simultaneously), resulting in the model being unable to accurately predict the intensity and duration of electricity price peaks and valleys. Therefore, in this application, the aligned structured - encoded vector of short - term weather forecast data and the aligned structured - encoded vector of short - term event data are combined to obtain the jointly structured - encoded vector of external factors. Specifically, in an example of this application, an attention - driven feature cross - mechanism is adopted to dynamically calculate the correlation weights between the weather vector and the event vector. For example, after the event vector of "cold wave warning" and the temperature - drop curve vector in the next 24 hours are aligned in the latent space, by calculating their attention scores, the correlation intensity of the overlapping area between the temperature - drop period and the event validity period is identified. Subsequently, based on the attention weights, the aligned vectors are weighted concatenated or tensor - fused to generate a jointly encoded vector. This process not only retains the independent semantics of the original aligned vectors (such as temperature trends, event types), but also captures the synergistic influence pattern of the two in a specific time - space range through the cross - terms (such as the probability of a sharp increase in load caused by the superposition of the first day of a cold wave and a holiday).

[0029] In step S150, a time - series prediction based on short - term electricity prices is performed on the time - series pattern feature - encoded vector of electricity prices and the jointly structured - encoded vector of external factors to obtain a short - term electricity price prediction result. Figure 3 It is a flowchart of step S150 in the method for generating a user - side energy storage operation strategy according to an embodiment of the present application. Specifically, asFigure 3 As shown, in step S150, performing time series prediction based on short-term electricity prices on the time series pattern feature encoding vector of electricity prices and the external factor joint structured encoding vector, including: S151, performing inner-outer collaborative response encoding with chain local interaction on the time series pattern feature encoding vector of electricity prices and the external factor joint structured encoding vector to obtain a time series pattern feature modulation encoding vector of electricity prices; S152, performing feature decoding on the time series pattern feature modulation encoding vector of electricity prices to obtain the short-term electricity price prediction result.

[0030] Specifically, in step S151, performing inner-outer collaborative response encoding with chain local interaction on the time series pattern feature encoding vector of electricity prices and the external factor joint structured encoding vector to obtain a time series pattern feature modulation encoding vector of electricity prices. Further, considering that the generation of the user-side energy storage operation strategy depends on the deep coupling of the electricity price time series characteristics and external environmental disturbances, but there are significant limitations in the traditional methods for dealing with the relationship between the two: on the one hand, there are essential differences in the data forms and action mechanisms between the electricity price time series patterns (such as daily peak-valley cycles, weekly cycle trends) and external factors (such as sudden temperature changes, special events), and simple feature splicing or linear superposition cannot capture the dynamic interaction effects between the two; on the other hand, the impact of external factors on electricity prices is often local and delayed. For example, the sharp increase in air-conditioning load caused by a cold snap may trigger a jump in electricity prices at a specific time period (such as at night), and it is difficult for traditional global interaction models to finely model such local spatio-temporal correlations. Due to the lack of fine-grained modeling of the interaction mechanism of multi-source heterogeneous data in the prior art, the energy storage strategy cannot dynamically respond to the combined effects of environmental disturbances and market behaviors. Therefore, in this application, inner-outer collaborative response encoding with chain local interaction is performed on the time series pattern feature encoding vector of electricity prices and the external factor joint structured encoding vector to obtain a time series pattern feature modulation encoding vector of electricity prices.

[0031] Specifically, first, one-dimensional convolution processing is performed on the time-series pattern feature encoding vector of electricity price and the joint encoding vector of external factors respectively to extract local implicit features (such as hourly fluctuation segments in the electricity price sequence and temperature drop trend segments in weather data). Subsequently, through single-feature interaction, the local features of electricity price and the local features of external factors are matched group by group to generate interaction vectors, capturing the association between the two within specific spatio-temporal segments (such as the co-variation pattern of electricity price fluctuations at night on the first day of a cold snap and the temperature curve). Then, a chain perception weight mechanism is introduced to evaluate the importance of each local interaction. For example, identify the key impact of local interactions during the mid-stage of a cold snap superimposed on holidays on electricity price prediction, and accordingly perform weighted modulation on the interaction response vectors. Finally, use forward LSTM to reason about the set of weighted local interactions, integrating cross-time-slice interaction information in the time series dimension (such as the conduction path of the cumulative effect of daily load during the duration of a cold snap on electricity price), and generate a time-series pattern feature modulation encoding vector of electricity price.

[0032] Figure 4 The flowchart of step S151 in the method for generating the user-side energy storage operation strategy according to the embodiment of the present application. Specifically, as Figure 4 shown, in step S151, the time-series pattern feature encoding vector of electricity price and the joint structured encoding vector of external factors are subjected to in-out collaborative response encoding of chain local interaction to obtain a time-series pattern feature modulation encoding vector of electricity price, including: S1511, performing local implicit feature single interaction on the time-series pattern feature encoding vector of electricity price and the joint structured encoding vector of external factors respectively to obtain a set of electricity price-external factor joint local implicit feature interaction encoding vectors; S1512, performing chain perception weighted modulation on the set of electricity price-external factor joint local implicit feature interaction encoding vectors to obtain a set of electricity price-external factor joint local implicit feature interaction modulation encoding vectors; S1513, inputting the set of electricity price-external factor joint local implicit feature interaction modulation encoding vectors into a forward LSTM model to obtain the time-series pattern feature modulation encoding vector of electricity price.

[0033] Specifically, in the embodiment of the present application, step S1511, performing local implicit feature single interaction on the time-series pattern feature encoding vector of electricity price and the joint structured encoding vector of external factors respectively to obtain a set of electricity price-external factor joint local implicit feature interaction encoding vectors, includes: performing local implicit feature one-dimensional convolution encoding on the time-series pattern feature encoding vector of electricity price and the joint structured encoding vector of external factors respectively to obtain a set of local time-series pattern implicit feature encoding vectors of electricity price and a set of joint local structured implicit encoding vectors of external factors. This process can be expressed by the formula: ; where is the time - series pattern feature encoding vector of electricity price, is the external factor joint structured encoding vector, is the one - dimensional convolutional encoding of local implicit features, is the length of the one - dimensional convolutional kernel, and are respectively the 1st, 2nd, th, and th electricity price local time - series pattern implicit feature encoding vectors in the set of electricity price local time - series pattern implicit feature encoding vectors, and are respectively the 1st, 2nd, th, and th external factor joint local structured implicit encoding vectors in the set of external factor joint local structured implicit encoding vectors, is and is the number of vectors in and and they have the same length.

[0034] Perform single - entity feature interaction on each group of corresponding electricity price local time - series pattern implicit feature encoding vectors and external factor joint local structured implicit encoding vectors in the set of electricity price local time - series pattern implicit feature encoding vectors and the set of external factor joint local structured implicit encoding vectors respectively to obtain the set of electricity price - external factor joint local implicit feature interaction encoding vectors. This process can be expressed by the formula: ; where, is the th electricity price local time - series pattern implicit feature encoding vector in the set of electricity price local time - series pattern implicit feature encoding vectors, is the th external factor joint local structured implicit encoding vector in the set of external factor joint local structured implicit encoding vectors, is dot - multiplication by position, is addition by position, is subtraction by position, is the concatenation operation, is the th local implicit feature interaction weight matrix in the set of local implicit feature interaction weight matrices, is the th local implicit feature interaction bias vector in the set of local implicit feature interaction bias vectors, is the th electricity price - external factor joint local implicit feature interaction encoding vector in the set of electricity price - external factor joint local implicit feature interaction encoding vectors.

[0035] It should be understood that since the optimization of the user-side energy storage strategy needs to accurately capture the local coupling effects of electricity price fluctuations and external environmental disturbances, when the electricity price time series characteristics and the external factor joint coding vector are used as overall features, the implicit local dynamic associations are often masked by global statistical features. For example, a sudden spike in a certain period of time in the electricity price sequence may only be related to the surge in air-conditioning load caused by a sudden drop in temperature in the next few hours, while the impact of the "cold wave warning + holidays" event in the external factor joint coding vector may be concentrated in a specific date range. If the overall features are directly calculated using a global interaction model, it is impossible to focus on these local key fragments, resulting in a delayed or misjudged response of the model to emergencies. Based on this, the present application performs one-dimensional convolution coding of local implicit features on the electricity price time series pattern feature coding vector and the external factor joint structured coding vector respectively to efficiently extract local patterns in serialized features (such as the load accumulation effect during continuous high temperature periods) based on convolution operations, and obtains a set of implicit feature coding vectors of local time series patterns of electricity prices and a set of local structured implicit coding vectors of joint external factors.

[0036] Accordingly, it is considered that the interaction effect between the two is often local and heterogeneous. For example, the sudden drop in temperature caused by a cold wave may cause a surge in air-conditioning load during the night period, thereby pushing up the electricity price during that period, while the same cold wave event may have a weaker impact on electricity prices during the day due to sufficient sunlight. Traditional global interaction methods (such as fully connected layers) ignore this kind of local temporal and spatial correlation, and interact the electricity price series with external factors as a whole, resulting in the model being unable to focus on the feature coupling of key time periods, weakening the ability to respond to sudden disturbances. In addition, the mechanism of action between electricity price fluctuations and external factors may be nonlinear or conditionally dependent. For example, the synergistic effect of the "high temperature + commercial promotion" event in the evening period is much higher than that in other periods, and such complex patterns need to be captured through fine-grained interactions. To this end, the present application performs monomer feature interactions on each group of corresponding electricity price local time series pattern implicit feature encoding vectors and external factors joint local structured implicit encoding vectors to obtain a set of electricity price-external factor joint local implicit feature interaction encoding vectors. In particular, the electricity price-external factor joint local interaction encoding vector generated by this process not only retains the independent semantics of the original local features, but also encodes the synergistic effect pattern of the two in a specific spatiotemporal segment.

[0037] Specifically, in the embodiment of the present application, step S1512 of performing chained perception weighted modulation on the set of electricity price-external factor joint local implicit feature interaction coding vectors to obtain a set of electricity price-external factor joint local implicit feature interaction modulation coding vectors includes: S1512-1, determining the chained perception modulation weights of each electricity price-external factor joint local implicit feature interaction coding vector based on the feature distribution characteristics of each electricity price-external factor joint local implicit feature interaction coding vector in the set of electricity price-external factor joint local implicit feature interaction coding vectors to obtain a set of electricity price-external factor joint local implicit feature interaction chained perception modulation weights; S1512-2, performing weighted modulation on the set of electricity price-external factor joint local implicit feature interaction coding vectors based on the set of electricity price-external factor joint local implicit feature interaction chained perception modulation weights to obtain the set of electricity price-external factor joint local implicit feature interaction modulation coding vectors.

[0038] Specifically, in step S1512-1, based on the feature distribution characteristics of each electricity price-external factor joint local implicit feature interaction coding vector in the set of electricity price-external factor joint local implicit feature interaction coding vectors, the chained perception modulation weights of each electricity price-external factor joint local implicit feature interaction coding vector are determined to obtain a set of electricity price-external factor joint local implicit feature interaction chained perception modulation weights. This process can be expressed by the formula: ; where is the th electricity price-external factor joint local implicit feature interaction coding vector in the set of electricity price-external factor joint local implicit feature interaction coding vectors, is the th eigenvalue in is the square of the Euclidean norm of the calculation vector, is the number of eigenvalues in is the is the th electricity price-external factor joint local implicit feature interaction chained perception modulation weight in the set of electricity price-external factor joint local implicit feature interaction chained perception modulation weights.

[0039] It should be understood that there are significant differences in the contributions of different interaction features to electricity price forecasting. For example, the local interaction between the sharp temperature drop at night on the first day of a cold snap and the sharp rise in electricity price may be more valuable for decision-making than the interaction during the subsequent gentle cooling stage; while the synergy effect of the "high temperature + commercial promotion" event in the evening may be much higher than that in other periods. When equal treatment is given to all local interactions, noise segments and key signals are integrated with equal weights, resulting in the model being unable to focus on the spatio-temporal coupling effect that truly affects the market supply-demand balance. In addition, there may be a chain-dependent relationship between local interactions (such as the load accumulation effect caused by continuous high temperatures needs to be transmitted across time periods), and it is necessary to explicitly model the correlation strength of interaction units in the time dimension. Therefore, in this application, based on the characteristic distribution characteristics of the feature distribution of each electricity price-external factor joint local implicit feature interaction coding vector, the chain-aware modulation weights of the respective electricity price-external factor joint local implicit feature interaction coding vectors are determined, so that the finally generated set of chain-aware modulation weights of the electricity price-external factor joint local implicit feature interaction quantifies the relative importance of each interaction unit in subsequent reasoning.

[0040] More specifically, in the embodiment of this application, step S1512-2, based on the set of chain-aware modulation weights of the electricity price-external factor joint local implicit feature interaction, performing weighted modulation on the set of electricity price-external factor joint local implicit feature interaction coding vectors to obtain the set of electricity price-external factor joint local implicit feature interaction modulation coding vectors, includes: performing semantic space reconstruction and feature enhancement on the set of chain-aware modulation weights of the electricity price-external factor joint local implicit feature interaction for dynamic specification regulation to obtain the set of chain-aware enhanced modulation weights of the electricity price-external factor joint local implicit feature interaction, and this process can be represented by the formula: ; where is the th electricity price-external factor joint feature linear interaction energy operator in the set of electricity price-external factor joint feature linear interaction energy operators, is the th electricity price-external factor joint feature difference measurement operator in the set of electricity price-external factor joint feature difference measurement operators, is the th electricity price-external factor joint feature non-linear coupling strength operator in the set of electricity price-external factor joint feature non-linear coupling strength operators, is the value of the logarithmic function with the natural constant as the base, is the th electricity price-external factor joint feature periodic local compensation factor in the set of electricity price-external factor joint feature periodic local compensation factors, is the th and are and corresponding weight coefficients, is the th

[0041] Based on the set of the ; wherein, and are the 1st, 2nd, th th is the set of the

[0042] Specifically, when calculating each etc. These operations essentially correspond to different interaction specifications, and then form a differential spatial constraint association mechanism in the interaction space.

[0043] To enhance the dynamic specification regulation gain of the and is regarded as a linear transformation effect (i.e., the gradient direction is consistent with the spatial interaction direction), while represents a non - linear oscillation effect (i.e., the gradient direction is orthogonal to the spatial interaction direction). For the vector statistics of different effects (such as ), the non - linear oscillation effect will cause a local periodic modulation effect in the direction of the linear local feature representation, so it is necessary to calculate its periodic local compensation factor: ; wherein, As a linear local feature representation, the growth of its logarithmic size will enhance the non-linear oscillation effect contribution. At the same time, the non-linear oscillation effect will also cause a phase transition of the linear local feature representation, resulting in periodic phase co-modulation: .

[0044] Finally, based on and weighted sum of, the weight of the interactive chain perception modulation of the electricity price - external factor joint local implicit feature is corrected .

[0045] By distinguishing the spatial constraint paradigm of the interaction specification in the interaction space, this method effectively enhances the collaborative correlation between different spatial constraint association mechanisms, thus significantly improving the calculation accuracy of the weight of the interactive chain perception modulation of the electricity price - external factor joint local implicit feature.

[0046] After that, each weight of the interactive chain perception modulation of the electricity price - external factor joint local implicit feature is weighted modulated with the corresponding interactive coding vector of the electricity price - external factor joint local implicit feature to ensure that the feature activation intensity of the high-weight interactive vector is significantly higher than that of the low-weight unit, obtaining a set of interactive modulation coding vectors of the electricity price - external factor joint local implicit feature. That is, the set of weighted modulated interactive coding vectors highlights the decision value of key spatio-temporal segments, enabling the model inference to focus on high-impact coupling events. For example, after the modulation of the high-weight interactive vector on the first night of the cold wave, the associated feature of the sudden temperature drop and electricity price jump carried by it is strengthened, promoting the model to accurately predict that this period is the core window of the charge and discharge strategy; while the low-weight interaction at noon on holidays is weakened by the modulation, avoiding misleading the strategy generation.

[0047] Specifically, in step S1513, the set of the interactive modulation coding vectors of the electricity price - external factor joint local implicit feature is input into the forward LSTM model to obtain the feature modulation coding vector of the electricity price time series pattern, and this process can be expressed by the formula: ; where is the set of the interactive modulation coding vectors of the electricity price - external factor joint local implicit feature, is the forward LSTM coding, is the feature modulation coding vector of the electricity price time series pattern.

[0048] Finally, input the set of the joint local implicit feature interaction modulation coding vectors of electricity price - external factors into the forward LSTM model to obtain the electricity price time - series pattern feature modulation coding vector. Specifically, input the vector set into the LSTM in chronological order, and use its gating mechanism (forget gate, input gate, output gate) to gradually integrate the time - series context information. For example, when processing the interaction vector on the first day of a cold snap, the LSTM memory unit stores the associated features of the temperature drop and electricity price jump during this period; when processing the interaction vector on the next day, the model combines historical memory and current input to identify the trend of intensified supply - demand tension caused by load accumulation, and dynamically adjusts the electricity price prediction logic for subsequent periods. Through this cyclic iteration process, the scattered local interaction information (such as the hourly load increase during high - temperature periods) is encoded into a coherent global feature representation, capturing the cross - period causal relationships (such as the gradual shift of the electricity price peak period to the previous day due to consecutive high temperatures), so that the generated electricity price time - series pattern feature modulation coding vector contains global market dynamics.

[0049] Specifically, in step S152, perform feature decoding on the electricity price time - series pattern feature modulation coding vector to obtain the short - term electricity price prediction result. Specifically, in the technical solution of this application, performing feature decoding on the electricity price time - series pattern feature modulation coding vector to obtain the short - term electricity price prediction result includes: using a short - term predictor based on a decoder to perform feature decoding on the electricity price time - series pattern feature modulation coding vector to obtain the short - term electricity price prediction result. In particular, the short - term electricity price prediction result refers to the predicted value of the electricity price in the short future. It should be understood that the electricity price time - series pattern feature modulation coding vector obtained through the previous encoding process is a processed and transformed high - dimensional feature vector, which contains the comprehensive information of the electricity price time - series pattern and external factors. However, this vector form cannot be directly used to represent the actual electricity price prediction result. The role of the decoder is to convert this encoded feature vector into an understandable and applicable short - term electricity price prediction data form, making it correspond to the actual electricity price value. It is worth mentioning that the decoder usually has the ability to restore specific data from the abstract feature representation. When dealing with the electricity price prediction problem, it can, based on the various feature information contained in the coding vector, through a series of decoding operations, convert this information into the predicted value of the electricity price in the short future period. This conversion process from features to specific values is realized based on the structure and training mechanism of the decoder, and can effectively utilize the information in the coding vector to generate meaningful prediction results. In this way, through feature decoding of the electricity price time - series pattern feature modulation coding vector, the various feature information extracted and fused before is converted into specific short - term electricity price prediction values, thereby providing an expectation of the electricity price change in the short future for electricity market participants, power system operators, etc., and helping them make reasonable decisions, such as adjusting the power generation plan and formulating electricity - using strategies.

[0050] In specific implementation, the short-term predictor based on the decoder usually includes a series of neural network structures, which have the ability to restore specific data from the abstract feature representation. When dealing with the electricity price prediction problem, the decoder performs decoding operations according to various feature information contained in the encoded vector. For example, the encoded vector may contain the periodic law information of historical electricity prices, such as daily peak-valley, weekly cycle and other patterns, as well as the influence information of external factors such as sudden temperature changes and holidays on electricity prices. The decoder will analyze and process this information and convert it into the predicted value of the electricity price in the short term in the future.

[0051] During the decoding process, the decoder will use its internal parameters and training mechanism to correspond the feature information in the encoded vector with the actual electricity price value. The training mechanism is obtained by training based on a large amount of historical data. By continuously adjusting the parameters of the decoder, the decoder can more accurately extract useful information from the encoded vector and convert it into an accurate electricity price prediction. For example, during the training process, a large amount of historical data containing different external factors and electricity price changes will be input, allowing the decoder to learn the internal relationship between these data, so that when facing a new encoded vector, it can accurately predict the short-term electricity price.

[0052] In step S160, based on the short-term electricity price prediction result, determine the operation strategy of the user-side energy storage. It should be understood that the charging and discharging behavior of the user-side energy storage will be directly affected by the electricity price. The fluctuation of the electricity price provides an arbitrage space for the energy storage system. By charging at a low electricity price and discharging at a high electricity price, the energy storage system can help users reduce the electricity cost. Therefore, the accurate short-term electricity price prediction result is the key basis for formulating a reasonable energy storage operation strategy. That is to say, by reasonably arranging the charging and discharging of the energy storage according to the short-term electricity price prediction result, users can avoid consuming a large amount of electricity during high electricity price periods, but use the electric energy stored by the energy storage system during low electricity price periods, thereby reducing the electricity bill expenditure and achieving the maximization of economic benefits.

[0053] In an embodiment of the present application, the implementation process of determining the operation strategy of the user-side energy storage based on the short-term electricity price prediction result is as follows: First, deeply analyze the short-term electricity price prediction result. Identify the peak and trough periods of the predicted electricity price, as well as the specific values of the electricity price in each period. At the same time, combined with the user's historical electricity consumption data, understand the electricity consumption demand characteristics in different periods. For example, industrial users have a large and relatively stable electricity consumption demand during the production period, commercial users have a high electricity consumption demand during business hours, and residential users have obvious electricity consumption peaks during morning and evening rush hours.

[0054] Next, based on the peak-valley periods of electricity prices and the electricity consumption demands of users, a preliminary charge-discharge strategy is formulated. During the low-price periods of electricity, the energy storage system should conduct charging operations as much as possible. Take a residential user as an example. If it is predicted that the electricity price is low in the early morning period and the electricity consumption demand of the user is extremely small during this period, the energy storage system can fully charge during this period to store electrical energy. For industrial users, if the low-price period of electricity coincides with the production intermission period, this period can also be used to charge the energy storage system. During the high-price periods of electricity, the energy storage system needs to discharge to meet the electricity consumption demands of users. Suppose it is predicted that the electricity price is high in the evening period. At this time, the residential user turns on various electrical appliances and the electricity consumption demand increases greatly. The energy storage system then starts to discharge, reducing the amount of electricity directly purchased by the user from the power grid, thereby reducing the electricity consumption cost.

[0055] In the process of formulating the charge-discharge strategy, the characteristics of the energy storage system itself also need to be considered. Attention should be paid to the charge-discharge efficiency of the energy storage system because there will be certain energy losses during the charge-discharge process. If the charge-discharge efficiency is low, it is necessary to balance the charge amount and the discharge amount when formulating the strategy to avoid excessive charge-discharge resulting in energy waste. At the same time, the capacity limit of the energy storage system should be considered. If the capacity of the energy storage system is small, it may not be able to store enough electrical energy during the low-price period of electricity for use during the high-price period. At this time, it is necessary to reasonably arrange the charging time and the charge amount to give priority to ensuring the electricity consumption demands during critical periods. For example, commercial users have a large demand for electricity during the business peak period. The energy storage system should give priority to ensuring that there is enough electricity supply during this period and reasonably plan the charging in combination with this demand.

[0056] In addition, it is also necessary to monitor in real time and dynamically adjust the energy storage operation strategy. The power market and the electricity consumption situation of users may change at any time. For example, sudden weather changes may cause a sudden increase in the electricity consumption demand of users, or changes in the supply-demand relationship in the power market may cause fluctuations in electricity prices. By monitoring the electricity price and the electricity consumption data of users in real time, once it is found that there is a deviation between the actual situation and the predicted result, the energy storage operation strategy should be adjusted in a timely manner. If the originally predicted high-price period of electricity arrives earlier and the electricity consumption demand increases earlier, the energy storage system should start discharging earlier to meet the electricity consumption demands of users and avoid purchasing a large amount of electricity from the power grid during the high-price period.

[0057] Finally, apply the formulated energy storage operation strategy to the actual user-side energy storage system and continuously track and evaluate the implementation effect of the strategy. By comparing the electricity consumption cost and economic benefits before and after implementing the strategy, continuously optimize and improve the energy storage operation strategy. If it is found that the strategy implemented in a certain stage has poor effect in reducing the electricity consumption cost, it is necessary to analyze the reasons. It may be that the prediction result is not accurate enough, or some factors are not considered sufficiently in the process of formulating the strategy, and then make targeted improvements so that the energy storage system can better adapt to different electricity consumption scenarios and market changes and achieve long-term stable economic benefits.

[0058] In another embodiment of the present application, the implementation process of determining the user-side energy storage operation strategy based on the short-term electricity price prediction result is as follows: First, identify the peak and valley periods based on the short-term electricity price prediction result, formulate a charging strategy during the low electricity price period, and preferentially use low-cost electric energy to fill the energy storage system. At this time, it is necessary to ensure that the state of charge (SOC) of the energy storage does not exceed the capacity limit and the charging power does not exceed the rated power of the equipment. During the peak electricity price period, execute the discharging strategy, and meet part or all of the load demand through energy storage discharging to reduce the expenditure on purchasing high-cost electricity. At the same time, ensure that the SOC after discharging is not lower than the safety threshold to cope with subsequent load fluctuations.

[0059] In addition, for load mutations caused by external factors, when formulating the strategy, the short-term electricity price prediction result is combined with the load prediction output. When it is predicted that the load demand in a specific period exceeds the normal level and the electricity price is at a high level, the energy storage is preferentially dispatched to discharge to match the load peak, avoiding a decline in revenue due to too high grid electricity purchase costs. If external factors cause the peak period of the electricity price to move forward or extend, the charging and discharging time windows are dynamically adjusted, the charging is completed in advance, and the discharging period is extended to ensure the response accuracy of the energy storage system within the electricity price fluctuation range.

[0060] In summary, the method for generating the user-side energy storage operation strategy based on the embodiments of the present application is elucidated. First, the historical electricity prices are encoded in time series to extract periodic patterns. At the same time, parameters such as temperature and light in the weather forecast and holiday event information are converted into structured vectors, and semantic alignment is achieved in a high-dimensional latent space to form a feature combination body containing external interference factors. Then, an internal-external collaborative response coding mechanism is adopted to jointly encode the electricity price time series features and external factors for chain interaction. Finally, the generated electricity price prediction result comprehensively considers the dual effects of external factors on load and market supply and demand, thereby supporting the energy storage system to execute precise charging and discharging strategies in the peak and valley intervals of the electricity price, effectively overcoming the problem of strategy lag caused by data fragmentation in traditional methods.

[0061] Figure 5 It is a block diagram of the system for generating the user-side energy storage operation strategy according to the embodiments of the present application. As Figure 5As shown, the system 100 for generating the user-side energy storage operation strategy according to the embodiment of the present application includes: a market historical electricity price data acquisition module 110 for acquiring market historical electricity price data; a short-term data acquisition module 120 for acquiring short-term weather forecast data and short-term event data; a electricity price time series encoding module 130 for performing time series encoding on the market historical electricity price data to obtain an electricity price time series pattern feature encoding vector; an external factor joint module 140 for performing structured joint encoding on the short-term weather forecast data and the short-term event data to obtain an external factor joint structured encoding vector; a short-term electricity price prediction module 150 for performing time series prediction based on short-term electricity prices on the electricity price time series pattern feature encoding vector and the external factor joint structured encoding vector to obtain a short-term electricity price prediction result; and a user-side energy storage operation strategy generation module 160 for determining a user-side energy storage operation strategy based on the short-term electricity price prediction result. Among them, the short-term electricity price prediction module is used to: perform inner-outer collaborative response encoding with chain local interaction on the electricity price time series pattern feature encoding vector and the external factor joint structured encoding vector to obtain an electricity price time series pattern feature modulation encoding vector; and perform feature decoding on the electricity price time series pattern feature modulation encoding vector to obtain the short-term electricity price prediction result.

[0062] Specifically, first, it consists of a market historical electricity price data acquisition module and a short-term data acquisition module. The market historical electricity price data acquisition module docks with the power market data platform, collects multi-period electricity price data and performs unified time series standardization processing to form a continuous electricity price data set. The short-term data acquisition module, through the meteorological API interface and the event information system, real-time grabs weather forecast parameters such as temperature and light, and event markers such as holidays and special activities. The two types of data are strictly aligned based on the unified time axis, providing input support with the same source time dimension for subsequent processing.

[0063] Secondly, the electricity price time series encoding module uses causal dilated convolution technology to perform time series encoding on historical electricity price data, and layer by layer extracts time series pattern features such as daily peak-valley, weekly cycle, and long-range event correlation to generate an electricity price time series pattern feature encoding vector. The external factor joint module synchronously processes weather forecast and event data. First, it separately converts continuous meteorological parameters and discrete event labels into independent structured vectors, then realizes semantic alignment through high-dimensional latent space mapping, and finally uses an attention-driven feature cross mechanism to generate an external factor joint structured encoding vector that integrates the synergistic effects of weather changes and event impacts.

[0064] Then, as the core of data fusion, the short-term electricity price prediction module receives the combined structured encoding vectors of the electricity price time series pattern feature encoding vector and external factors, and performs chain interaction processing through the internal-external collaborative response encoding mechanism: First, it extracts local hidden features through one-dimensional convolution and performs group-by-group interaction, then highlights the influence of key spatio-temporal segments through chain-aware weighted modulation, and finally uses the forward LSTM model to integrate cross-period correlations to generate an electricity price time series pattern feature modulation encoding vector containing external factor modulation. This vector is converted into a short-term electricity price prediction result by a decoder, which directly drives the formulation of the energy storage strategy.

[0065] Finally, the user-side energy storage operation strategy generation module constructs an optimization model with the goal of minimizing the electricity consumption cost based on the prediction results, and determines the charge-discharge power and energy scheduling plan for each period in combination with the physical constraints of the energy storage system and the user load characteristics. This module monitors the actual electricity price and load response in the market in real time. When the prediction deviation or load mutation exceeds the threshold, it triggers the short-term electricity price prediction module to run again through the feedback link, updates the prediction results based on the latest data, and adjusts the strategy. This closed-loop mechanism prompts the electricity price time series encoding and external factor joint module to dynamically optimize the feature extraction parameters, ensuring that the prediction results fit the actual market environment, and forming a full-process adaptive system of "data acquisition - feature processing - prediction generation - strategy formulation - real-time feedback".

[0066] Here, those skilled in the art can understand that the specific operations of each step in the above-mentioned user-side energy storage operation strategy generation system have been introduced in detail in the description of the Figures 1 to 4 user-side energy storage operation strategy generation method, and therefore, the repeated description will be omitted.

[0067] Specifically, in another example of this application, a user-side energy storage optimization and control system based on multi-factor collaboration is also proposed. Specifically, it includes: 1. The optimal operation strategy of the user-side energy storage that integrates multiple factors. Considering the mutual influence of energy storage configuration and operation, starting from analyzing the energy storage configuration constraint conditions, energy transfer formula, and charge-discharge constraint conditions, with the goal of minimizing the comprehensive economic cost and voltage deviation of energy storage configuration operation, combined with the constraint conditions for the safe and stable operation of the distribution network, an optimization configuration model of distributed energy storage in the active distribution network is established. Since the optimization of this model is a complex dynamic multi-objective optimization problem and traditional methods have defects, the teaching-learning-based optimization (TLBO) algorithm is used to improve the search mechanism of the particle swarm optimization (PSO) algorithm, and the multi-objective HTL-MOPSO algorithm with enhanced performance is explored in combination with the leader particle selection strategy. The energy storage configuration model is transformed into a two-layer hierarchical structure multi-objective optimization model. The outer layer determines the energy storage configuration plan, and the inner layer optimizes the energy storage operation strategy. The inner and outer layers interact with each other, and the multi-objective HTL-MOPSO algorithm is used for optimization and solution to obtain the optimal capacity configuration of the energy storage and the corresponding optimal operation strategy.

[0068] 2. Exploration and Benefit Evaluation of Potential Energy Storage Users Considering Economy and Distribution Network Stability. Starting from the two dimensions of the diverse electricity consumption demand characteristics of users and the operating characteristics on the grid side, record the diverse electricity consumption demands on the user side (such as user load capacity and composition, fee reduction demand indicators, etc.) and the operating characteristics on the grid side (such as voltage fluctuations, frequency fluctuations, and power quality, etc.) on a daily and monthly basis, and establish a feature library for exploring target users of energy storage services on the user side. Based on the analytic hierarchy process and the method of determining weights with the maximum deviation, determine the subjective weights and objective weights of each level of features and indicators respectively, and introduce the concept of connection numbers to determine their combined weights. Construct the GRA-TOPSIS quantification model, the entropy weight method model, and the fuzzy analytic hierarchy process respectively, comprehensively consider multiple indicators and factors for quantifying and grading user demands and grid-side operating characteristics, introduce a non-linear information aggregation method, add an index grading vector and the optimal and worst index value vectors to the decision matrix, realize the classification of user characteristics, and accurately identify energy storage users. Intuitively display the quantitative results of user classification in a visual way and convert them into a qualitative form. Combining the energy storage operation strategy and capacity configuration, calculate the economic benefits from three aspects: arbitraging the peak-valley electricity price difference using electrochemical energy storage, the environmental benefits brought by energy storage devices, and subsidy income. Quantify the distribution network stability benefits from the improvement of power quality (severity of transient and steady-state power quality, economic losses, etc.) before and after installing energy storage devices. Design a visual software platform for classifying energy storage users, record relevant data to calculate various benefits, comprehensively evaluate and determine whether users are suitable for installing distributed energy storage, and help promote the user-side energy storage market.

[0069] 3. User-side energy storage optimizes the power quality of the distribution network and the grid development plan. Based on the mining and benefit evaluation of existing potential energy storage users, the K-Means clustering algorithm is used to divide the multi-dimensional features into operational categories, the Monte Carlo algorithm is introduced to build a time series data prediction model, and the multi-layer weighted long short-term memory model (LSTM) is used to predict time series data, accurately capture long-term and short-term dependencies, evaluate the constraints of the system supply and demand balance, obtain the supply and demand balance conditions and the regional power grid capacity and the user-side configuration transformer capacity limit data, mathematically model the capacity limit data as a constraint, jointly establish the main optimization goals of the optimal energy storage device configuration planning, establish a multi-objective optimization mathematical model for the configuration planning of energy storage devices in the region, and use the dual optimization method in convex optimization theory to solve the model, and obtain the supply and demand balance conditions that can reasonably utilize the upper and lower limits of the user-side and regional equipment performance. On the basis of the above, determine the grid development planning strategy required to cope with the dynamic changes in user demand, mathematically model the key equipment in the grid (such as transformers, harmonic removal devices, etc.), obtain various performance optimization parameters as the main goal, and use the previously obtained supply and demand balance conditions as constraints to establish an optimization model for achieving the optimal distribution network supply and demand stability. The optimization model is simplified by relaxing the objective function and constraints, and is converted into a two-layer optimization problem of optimal energy storage capacity configuration and grid power supply stability. The outer layer optimizes the supply and demand balance indicators and power supply stability of the regional distribution network to determine the overall configuration of the energy storage device; the lower layer optimizes the supply and demand balance conditions of the multidimensional feature labels of the electricity users under the premise of the energy storage device configuration plan in the given optimal area, and takes grid stability as the main optimization goal. The particle swarm algorithm or alternating direction multiplier algorithm is used to solve the inner and outer optimization problems according to the multidimensional feature labels of users in different regions and the electrical effects of user-side energy storage on the grid. Finally, the grid operation is simulated to determine the accuracy.

[0070] 4. User-side energy storage status monitoring and energy storage benefit evaluation system. Design a health monitoring module and a net income calculation module based on the status evaluation criteria, record the charging and discharging conditions of the user-side energy storage device on a daily and monthly basis, draw a power curve, calculate the net income according to the charging and discharging data combined with the income model, design the health status evaluation criteria for the energy storage device, and develop a monitoring module for real-time monitoring of the health status of the energy storage device, so as to monitor the health status of the energy storage device in real time, and maximize the comprehensive income of users through effective charging and discharging strategies. Based on the measurement data, develop a system to monitor the voltage at the energy storage connection point, the active power in the line, the reactive power flow, and the available margin of the energy storage itself, develop an intelligent sensing detection module to monitor the operating status of the connection point in real time, use a time series neural network to predict future state variables, and when there is a large deviation between the prediction result and the set target, send a warning signal to the superior dispatching center and assist in the timely adjustment of the energy storage operation strategy, so as to solve the complexity of the connection point status monitoring and the reliability of the state prediction, and realize the monitoring of the impact of the user-side energy storage on the grid voltage and power after connecting to the grid, as well as the prediction and warning of the future changes in the grid operating status.

[0071] In summary, it is intended that the above detailed description be regarded as illustrative rather than restrictive, and it should be understood that the above embodiments should be construed as only for illustrating the present invention and not for limiting the protection scope of the present invention.

Claims

1. A method for generating an operation strategy of user-side energy storage, characterized in that, Including: Obtaining historical market electricity price data; Obtaining short-term weather forecast data and short-term event data; Performing time series encoding on the historical market electricity price data to obtain an electricity price time series pattern feature encoding vector; Performing structured joint encoding on the short-term weather forecast data and the short-term event data to obtain an external factor joint structured encoding vector; Performing time series prediction based on short-term electricity prices on the electricity price time series pattern feature encoding vector and the external factor joint structured encoding vector to obtain a short-term electricity price prediction result, including: performing chain local interaction inner-outer collaborative response encoding on the electricity price time series pattern feature encoding vector and the external factor joint structured encoding vector to obtain an electricity price time series pattern feature modulation encoding vector; performing feature decoding on the electricity price time series pattern feature modulation encoding vector to obtain the short-term electricity price prediction result; Determining the user-side energy storage operation strategy based on the short-term electricity price prediction result; Among them, performing chain local interaction inner-outer collaborative response encoding on the electricity price time series pattern feature encoding vector and the external factor joint structured encoding vector to obtain an electricity price time series pattern feature modulation encoding vector includes: respectively performing local implicit feature monomer interaction on the electricity price time series pattern feature encoding vector and the external factor joint structured encoding vector to obtain a set of electricity price-external factor joint local implicit feature interaction encoding vectors; performing chain perception weighted modulation on the set of electricity price-external factor joint local implicit feature interaction encoding vectors to obtain a set of electricity price-external factor joint local implicit feature interaction modulation encoding vectors; inputting the set of electricity price-external factor joint local implicit feature interaction modulation encoding vectors into a forward LSTM model to obtain the electricity price time series pattern feature modulation encoding vector.

2. The method for generating the user-side energy storage operation strategy according to claim 1, wherein Performing time series encoding on the historical market electricity price data to obtain an electricity price time series pattern feature encoding vector includes: performing time series encoding based on causal dilated convolution on the historical market electricity price data to obtain the electricity price time series pattern feature encoding vector.

3. The method for generating the user-side energy storage operation strategy according to claim 1, wherein Performing structured joint encoding on the short-term weather forecast data and the short-term event data to obtain an external factor joint structured encoding vector includes: Performing structured encoding on the short-term weather forecast data and the short-term event data to obtain a short-term weather forecast data structured encoding vector and a short-term event data structured encoding vector; Mapping the short-term weather forecast data structured encoding vector and the short-term event data structured encoding vector to the same high-dimensional implicit space to obtain an aligned short-term weather forecast data structured encoding vector and an aligned short-term event data structured encoding vector; Combining the aligned short-term weather forecast data structured encoding vector and the aligned short-term event data structured encoding vector to obtain the external factor joint structured encoding vector.

4. The method for generating the user-side energy storage operation strategy according to claim 1, characterized in that Respectively performing local implicit feature monomer interaction on the electricity price time series pattern feature encoding vector and the external factor joint structured encoding vector to obtain a set of electricity price-external factor joint local implicit feature interaction encoding vectors includes: Perform local implicit feature one-dimensional convolutional encoding on the electricity price time series pattern feature encoding vector and the external factor joint structured encoding vector respectively to obtain a set of electricity price local time series pattern implicit feature encoding vectors and a set of external factor joint local structured implicit encoding vectors; Perform single-entity feature interaction on each corresponding electricity price local time series pattern implicit feature encoding vector and external factor joint local structured implicit encoding vector in the set of electricity price local time series pattern implicit feature encoding vectors and the set of external factor joint local structured implicit encoding vectors to obtain a set of electricity price-external factor joint local implicit feature interaction encoding vectors.

5. The method for generating the user-side energy storage operation strategy according to claim 4, wherein Perform chained perception weighted modulation on the set of electricity price-external factor joint local implicit feature interaction encoding vectors to obtain a set of electricity price-external factor joint local implicit feature interaction modulation encoding vectors, including: Based on the feature distribution characteristics of each electricity price-external factor joint local implicit feature interaction encoding vector in the set of electricity price-external factor joint local implicit feature interaction encoding vectors, determine the chained perception modulation weights of the respective electricity price-external factor joint local implicit feature interaction encoding vectors to obtain a set of electricity price-external factor joint local implicit feature interaction chained perception modulation weights; Based on the set of electricity price-external factor joint local implicit feature interaction chained perception modulation weights, perform weighted modulation on the set of electricity price-external factor joint local implicit feature interaction encoding vectors to obtain the set of electricity price-external factor joint local implicit feature interaction modulation encoding vectors.

6. The method for generating the user-side energy storage operation strategy according to claim 5, wherein Based on the set of electricity price-external factor joint local implicit feature interaction chained perception modulation weights, perform weighted modulation on the set of electricity price-external factor joint local implicit feature interaction encoding vectors to obtain a set of electricity price-external factor joint local implicit feature interaction modulation encoding vectors, including: Perform semantic space reconstruction and feature enhancement with dynamic specification control on the set of electricity price-external factor joint local implicit feature interaction chained perception modulation weights to obtain a set of electricity price-external factor joint local implicit feature interaction chained perception enhanced modulation weights; Based on the set of electricity price-external factor joint local implicit feature interaction chained perception enhanced modulation weights, perform weighted modulation on the set of electricity price-external factor joint local implicit feature interaction encoding vectors to obtain a set of electricity price-external factor joint local implicit feature interaction modulation encoding vectors.

7. The method for generating the user-side energy storage operation strategy according to claim 6, wherein Perform feature decoding on the electricity price time series pattern modulation encoding vector to obtain the short-term electricity price prediction result, including: using a short-term predictor based on a decoder to perform feature decoding on the electricity price time series pattern modulation encoding vector to obtain the short-term electricity price prediction result.

8. A generation system for the operation strategy of user-side energy storage, characterized in that, Including: A market historical electricity price data acquisition module for acquiring market historical electricity price data; A short-term data acquisition module for acquiring short-term weather forecast data and short-term event data; An electricity price time series encoding module for performing time series encoding on the market historical electricity price data to obtain an electricity price time series pattern feature encoding vector; External factor joint module, which is used to perform structured joint encoding on the short-term weather forecast data and the short-term event data to obtain an external factor joint structured encoding vector; Short-term electricity price prediction module, which is used to perform time series prediction based on the short-term electricity price on the electricity price time series pattern feature encoding vector and the external factor joint structured encoding vector, wherein the short-term electricity price prediction module is used to: perform chain local interaction inner-outer collaborative response encoding on the electricity price time series pattern feature encoding vector and the external factor joint structured encoding vector to obtain an electricity price time series pattern feature modulation encoding vector; perform feature decoding on the electricity price time series pattern feature modulation encoding vector to obtain the short-term electricity price prediction result; User-side energy storage operation strategy generation module, which is used to determine the user-side energy storage operation strategy based on the short-term electricity price prediction result; Among them, performing chain local interaction inner-outer collaborative response encoding on the electricity price time series pattern feature encoding vector and the external factor joint structured encoding vector to obtain an electricity price time series pattern feature modulation encoding vector includes: respectively performing local implicit feature monomer interaction on the electricity price time series pattern feature encoding vector and the external factor joint structured encoding vector to obtain a set of electricity price-external factor joint local implicit feature interaction encoding vectors; performing chain perception weighted modulation on the set of electricity price-external factor joint local implicit feature interaction encoding vectors to obtain a set of electricity price-external factor joint local implicit feature interaction modulation encoding vectors; inputting the set of electricity price-external factor joint local implicit feature interaction modulation encoding vectors into a forward LSTM model to obtain the electricity price time series pattern feature modulation encoding vector.

Images