An energy storage system-based power dispatch method

Abstract

The present application relates to the technical field of electric energy scheduling, and particularly relates to an electric energy scheduling method based on an energy storage system.The method comprises the following steps: deploying a distributed sensing array inside a battery module to collect multi-source heterogeneous data, and performing standardization processing to generate a standardized energy storage dataset; performing tensor completion on the standardized energy storage dataset to generate energy storage tensor completion data; performing time-space dimension analysis on the energy storage tensor completion data, and performing cross-modal feature fusion to generate an energy storage cross-modal feature space;Therefore, through the structured processing of multi-dimensional fused data and the dynamic feedback optimization of the digital twin model, the present application solves the problems of lagging prediction and poor model adaptability in traditional cold-chain warehouse environments, and improves the accuracy of temperature prediction and the intelligent level of warehouse management.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electric energy scheduling, and in particular to an electric energy scheduling method based on an energy storage system. BACKGROUND

[0002] Existing energy storage electric energy scheduling technology mainly relies on static strategies or simplified models, and fails to effectively consider multi-dimensional factors such as battery aging, frequency response and load dynamic changes, resulting in insufficient scheduling accuracy and adaptability, and being difficult to cope with high-frequency load fluctuations and abnormal power grid disturbances. Data collection is mostly focused on a single dimension, such as voltage, current or SOC, ignoring comprehensive analysis of multi-modal data such as temperature and polarization loss, and using traditional filtering and interpolation methods to handle data missing and noise, lacking deep modeling of data continuity and spatial correlation, limiting the refinement of scheduling. Although digital twin technology has been proposed, it is mostly limited to visualization and limited prediction, lacking dynamic optimization and adaptive updating capabilities, and being difficult to cope with long-term wear and tear and changes in charging and discharging paths. In terms of optimization strategies, existing methods usually single-mindedly focus on energy saving or stability, lack of coordinated control of energy saving and frequency response, and fail to build an effective game model, easily causing imbalance between system performance and energy saving benefits. SUMMARY

[0003] Therefore, it is necessary to provide an electric energy scheduling method based on an energy storage system to solve at least one of the above technical problems.

[0004] To achieve the above-mentioned purpose, an electric energy scheduling method based on an energy storage system, the method comprising the following steps:

[0005] Step S1: deploying a distributed sensing array inside a battery module to collect multi-source heterogeneous data, and performing protocol conversion to obtain an initial multi-modal data set; performing noise spectrum adjustment filtering threshold on the initial multi-modal data set, and performing long-term drift compensation to generate a standardized energy storage data set;

[0006] Step S2: performing tensor completion on the standardized energy storage data set to generate energy storage tensor completion data; performing time-space dimension analysis on the energy storage tensor completion data, and performing cross-modal feature fusion to generate an energy storage cross-modal feature space;

[0007] Step S3: constructing a battery health state digital twin based on the energy storage cross-modal feature space to obtain an energy storage system digital twin model; generating a game scheduling strategy based on the energy storage system digital twin model to obtain an energy storage system electric energy scheduling strategy;

[0008] Step S4: obtaining real-time energy storage monitoring data; performing process deviation calculation based on the real-time energy storage monitoring data and the energy scheduling strategy of the energy storage system; if the energy storage deviation data is greater than a preset threshold, generating SOC attenuation data; returning the SOC attenuation data to the energy storage system digital twin model for network parameter optimization, and performing full-process scheduling analysis of the energy storage system to obtain full-process scheduling analysis data; generating a scheduling report based on the full-process scheduling analysis data to obtain an energy storage system energy scheduling full-process report.

[0009] The application has the advantages that through the deployment of the multi-source sensor array, real-time collection of multi-dimensional heterogeneous data in the cold chain storage environment is realized, and accurate synchronization of different data sources in the time dimension is ensured based on the timestamp alignment technology, forming high timeliness and integrity of the cold chain storage timestamp alignment data. Subsequently, through three-dimensional space mapping and construction of the feature tensor matrix, the spatial distribution information of temperature, humidity and other environmental parameters is integrated to generate a fusion feature matrix, which can reflect the multi-dimensional dynamic characteristics of the storage environment and provide a structured data basis for subsequent analysis. On this basis, a preset long short-term memory (LSTM) model is used to predict the temperature distribution of the fusion feature matrix, capture the time sequence characteristics and spatial correlation of temperature changes, and generate storage temperature prediction distribution data. The data is further combined with the fusion feature matrix to construct a cold chain causal knowledge graph, revealing the causal relationship between environmental variables, and based on the graph, a digital twin model is constructed to realize multi-dimensional simulation and dynamic simulation of the storage system state. The digital twin model then undergoes error residual iterative analysis, the system identifies the deviation between the model prediction and the actual monitoring, generates difference analysis data, automatically adjusts and compensates the parameters through the feedback loop, optimizes and fine-tunes the model, and forms a more accurate storage optimization digital twin model. Finally, the optimized model drives the construction of the logistics and storage management report, systematically presenting key indicators such as temperature prediction accuracy, environmental stability and management efficiency, supporting scientific decision-making and precise scheduling. Overall, this method forms a closed loop through data collection, synchronization, fusion, prediction and feedback optimization, enhancing the environmental perception ability and prediction accuracy of the cold chain storage system, and realizing the dynamic adaptability and continuous optimization of the model. Therefore, through structured processing of multi-dimensional fusion data and dynamic feedback optimization of the digital twin model, the application solves the problems of traditional cold chain storage environment prediction lag and poor model adaptability, improving the accuracy of temperature prediction and the intelligent level of storage management.

[0010] Preferably, step S1 comprises the following steps:

[0011] Step S11: deploying voltage sensors and temperature sensors inside the battery module to collect single cell temperature data and terminal voltage data at a frequency of 10 Hz, wherein the accuracy of the voltage sensor is 0.1%, and the accuracy of the temperature sensor is 0.5℃;

[0012] Step S12: Deploy a high-frequency current sensor with a bandwidth of 20 kHz on the DC side of the energy storage converter to collect the charge and discharge current waveform at a frequency of 1 kHz, and generate the energy storage charge and discharge waveform data;

[0013] Step S13: Deploy a synchronous phase measurement unit at the grid access point to collect the grid frequency, voltage phase angle and harmonic distortion rate at a frequency of 50 Hz, and generate the grid phase raw data;

[0014] Step S14: Protocol conversion of single temperature data, terminal voltage data, energy storage charge and discharge waveform data and grid phase raw data through a multi-protocol adapter to generate an initial multi-modal data set;

[0015] Step S15: Noise spectrum adjustment filtering threshold is performed on the initial multi-modal data set, and long-term drift compensation is performed to generate a standardized energy storage data set.

[0016] The application realizes high-frequency (10Hz) dynamic acquisition of the operating state of the single battery by deploying high-precision voltage and temperature sensors inside the battery module, ensuring that the small features of temperature gradient change and voltage fluctuation in each time slice are accurately captured, effectively supporting cell thermal management and aging assessment. At the same time, a high-frequency current sensor with a bandwidth of 20 kHz is arranged on the DC side of the energy storage converter, combined with a data sampling frequency of 1 kHz, to obtain the fine-grained waveform characteristics of the current during charging and discharging, which can accurately reflect the dynamic response and power disturbance mode caused by load changes, forming high-time energy storage charge and discharge waveform data. In addition, a synchronous phase measurement unit is introduced at the grid access point to synchronously collect key measurement parameters such as grid frequency, voltage phase angle and harmonic distortion rate at a frequency of 50 Hz, which not only provides data support for the stability of the grid operation, but also provides a data basis for subsequent power quality analysis and frequency response optimization. The three types of data with different sources and frequencies are converted by a multi-protocol adapter to solve the compatibility problem between different data formats and communication protocols, and an initial multi-modal data set is generated. Subsequently, in view of the high-frequency noise and low-frequency signal drift problems existing in the acquisition process, the system introduces a spectrum domain filtering algorithm to separate the high-pass and low-pass components of the signal by setting a threshold, filter out non-structural noise such as electromagnetic interference and thermal disturbance, and combine time domain analysis for long-term drift compensation, so that all data remain physically consistent in amplitude and reference value, and finally form a standardized energy storage data set with unified standards and high data reliability. The data set provides a multi-source fusion, high-dimensional accuracy and spatiotemporal consistent key basis for subsequent model training, state recognition and energy scheduling optimization.

[0017] Preferably, the tensor completion of the standardized energy storage data set in step S2 comprises:

[0018] The standardized energy storage dataset is mapped to a three-dimensional space for timestamp sequencing to obtain energy storage three-dimensional mapping standardized data;

[0019] The energy storage three-dimensional mapping standardized data is repaired for temperature and current data by using low-rank Tucker decomposition to obtain standard tensor repair data;

[0020] The standard tensor repair data is constructed into a three-dimensional feature tensor of device-time-feature, and missing value detection is performed to obtain energy storage tensor completion data.

[0021] The application maps the standardized energy storage dataset to a three-dimensional space, and completes the time sequence ordering of multidimensional data by using the timestamp as the index, establishes a spatial structured data framework based on the device number, collection time and physical characteristics (such as voltage, temperature, current, etc.), and thus constructs the energy storage three-dimensional mapping standardized data in a unified coordinate, effectively retaining the dynamic change law of various sensor data in the time dimension. Subsequently, the low-rank Tucker tensor decomposition technology is used to repair the temperature and current subspace in the three-dimensional structure, which is based on the tensor rank constraint principle, and the original data is represented as the product of the core tensor and multiple factor matrices, and the redundancy between sensors and the low-rank data are extracted by minimizing the tensor kernel norm or the reconstruction error, so as to fill in the abnormal missing area caused by sensor fluctuation, collection delay or external interference, and generate standard tensor repair data with structural continuity and statistical consistency. Then, the system reconstructs the three-dimensional feature tensor based on the tensor data according to the logic of device-time-feature, further ensures the normalization of data in the device dimension, the equidistance of data in the time dimension, and the stability of data in the feature dimension, and lays a unified data foundation for subsequent processing. On the basis of this tensor, the system introduces a multi-scale missing value detection mechanism, combines a sliding window and context semantic analysis, analyzes the null values, pseudo-zero values and abnormal values at different positions, and performs interpolation completion processing according to the self-consistency rules of the tensor space structure, and finally generates energy storage tensor completion data that meets the requirements in terms of physical consistency, mathematical distinguishability and time logic, and provides structural integrity and semantic explicit data support for high-quality twin modeling, state diagnosis and prediction control.

[0022] Preferably, the time-space dimension analysis of the energy storage tensor completion data and the cross-modal feature fusion of step S2 include:

[0023] The energy storage tensor completion data is subjected to time dimension data extraction to generate battery SOC data;

[0024] The energy storage tensor completion data is subjected to spatial dimension data extraction to generate power grid frequency fluctuation rate data;

[0025] The battery SOC data and the grid frequency fluctuation rate data are time axis aligned to generate a synchronized energy storage dataset;

[0026] The battery SOC data is subjected to long-term dependence relation analysis to obtain battery dependence relation analysis data; the battery dependence relation analysis data is subjected to charge-discharge behavior feature extraction to obtain a time step feature vector;

[0027] The grid frequency fluctuation rate data is subjected to unit topology impedance calculation to obtain grid topology impedance calculation data; the grid topology impedance calculation data is subjected to charge-discharge path electric energy loss relation analysis and spatial distribution feature extraction to obtain a spatial dimension feature matrix;

[0028] The time step feature vector and the spatial dimension feature matrix are subjected to tensor splicing to obtain an energy storage feature splicing matrix; the energy storage feature splicing matrix is subjected to synchronization processing by using the synchronized energy storage dataset to obtain an energy storage scheduling feature fusion matrix;

[0029] Time-space initial weight distribution is performed based on the energy storage scheduling feature fusion matrix, and a feature space is constructed to obtain an energy storage cross-modal feature space.

[0030] The application extracts SOC data representing the state of battery remaining capacity in the time dimension to depict the charging and discharging evolution process of the battery at each time step; meanwhile, frequency fluctuation rate data representing the dynamic characteristics of the power grid system are extracted in the spatial dimension to capture the supply and demand disturbance characteristics of different spatial nodes. By aligning the SOC data and the power grid frequency fluctuation rate data on the time axis, a synchronized energy storage dataset is constructed, thereby ensuring the analysis consistency of different source data under the same time reference. Then, long-term dependence relationship modeling is performed on the SOC data, the influence mechanism of the historical state of the battery on the future charging and discharging behavior is extracted by analyzing the nonlinear correlation across time steps in the time series, and a time step feature vector is constructed to reflect the dynamic evolution trend on the time axis; the unit topology impedance calculation is performed on the power grid frequency fluctuation rate data, the electrical connection characteristic model of each node in the power grid is constructed, the energy transmission loss of different nodes in the charging and discharging process is calculated, and the spatial distribution characteristics are further extracted to generate a spatial dimension feature matrix. Subsequently, the time step feature vector and the spatial dimension feature matrix are spliced in the tensor structure to form an energy storage feature splicing matrix, which maintains the independence of the physical meanings of various features while ensuring the overall semantic consistency. Then, the synchronized energy storage dataset is used to reconstruct the multi-modal data under the unified time reference after the splicing of the matrix, and an energy storage scheduling feature fusion matrix is generated. On this basis, the time-space bidirectional initial weight distribution is performed, the multi-scale feature weight field is constructed according to the historical correlation strength and the spatial node load response capability, and the energy storage cross-modal feature space is constructed. The feature space fully integrates the evolution trend in the time dimension and the energy transmission mechanism in the spatial dimension, and has strong robustness, cross-domain expressiveness and scheduling explainability.

[0031] Preferably, step S3 comprises the following steps:

[0032] Step S31: performing multi-scale energy storage analysis based on the energy storage cross-modal feature space, and constructing a battery health state digital twin to obtain an energy storage system digital twin model;

[0033] Step S32: designing a double-layer optimization framework based on the energy storage system digital twin model to obtain an energy storage double-layer optimization framework, wherein the energy storage double-layer optimization framework comprises an upper-layer consumption reduction framework and a lower-layer energy consumption stability framework;

[0034] Step S33: generating a game scheduling strategy based on the upper-layer consumption reduction framework and the lower-layer energy consumption stability framework to obtain an energy storage system energy scheduling strategy.

[0035] The application builds a multi-scale analysis driven energy storage system digital twin model based on the data expression basis of energy storage cross-modal feature space, realizes the whole process coupling modeling and hierarchical control mechanism from data fusion to scheduling optimization. At the data level, first, the cross-modal feature space is taken as the input, multi-source information such as time step features, spatial node features and energy consumption loss features are integrated, and multi-scale energy storage analysis method is used to model the state evolution at different time granularity (such as hour level, day level) and spatial resolution (such as node level, regional level), to form the joint expression of battery state of charge, temperature gradient, impedance change trend and other health state parameters, and then to build a battery health state digital twin with high fitting degree and traceability, and to map the prediction atlas of the future evolution path of the battery from the current state. Subsequently, taking the digital twin as the core, a double-layer optimization framework of the energy storage system is designed: the upper layer of the consumption reduction framework focuses on the data of electricity price, load prediction and battery energy efficiency, and executes the 24-hour rolling optimization of the charge and discharge plan to realize the cost minimization or benefit maximization; the lower layer of the energy consumption stability framework implements the frequency domain and time domain response strategies for the grid frequency disturbance, current sudden change or charge and discharge fluctuation data according to the dynamic feedback results of the digital twin, to ensure the operation stability of the energy storage system and the grid system in the dynamic coupling state. Finally, based on the data input and output response parameters in the double-layer optimization framework, through the game modeling method of the consumption reduction weight factor and the stability weight factor, the energy scheduling strategy of the energy storage system is jointly constructed, so that the scheduling instruction meets the real-time frequency regulation demand of the grid side while considering the consumption reduction operation target.

[0036] Preferably, step S31 comprises the following steps:

[0037] Step S311: Ohmic polarization / diffusion polarization loss calculation based on energy storage cross-modal feature space, to generate electrochemical scale analysis data;

[0038] Step S312: Thermodynamic analysis for capacity attenuation prediction based on energy storage cross-modal feature space, to generate thermodynamic scale analysis data;

[0039] Step S313: Electrode stress crack detection and crack diffusion rate calculation based on energy storage cross-modal feature space, to generate mechanical scale analysis data;

[0040] Step S314: Parameter initialization of electrochemical scale analysis data, thermodynamic scale analysis data and mechanical scale analysis data, and digital twin construction, to generate an energy storage system digital twin model.

[0041] The application improves the state visualization ability and dynamic prediction accuracy of the energy storage system under complex operating conditions by deeply analyzing the data of the energy storage cross-modal feature space and constructing a unified analysis framework under multiple physical scales. At the data level, first, the three-dimensional tensor feature representation after fusion is used to model the electrochemical behavior at the microscopic scale of multiple modal signals such as time, voltage, current, temperature and impedance. With the voltage-current response curve under non-steady state, the ohmic polarization and concentration polarization contribution are separated, the polarization loss function is established, and the electrochemical scale analysis data is generated. Secondly, in the thermodynamic dimension, combined with the joint distribution characteristics of temperature-current under long-term operation, the thermal gradient tensor is used to map the thermal accumulation degree in different SOC intervals, predict the capacity attenuation trend and output the thermodynamic scale analysis data. Further, in the mechanical dimension, the stress response analysis is performed on the relevant tensor of current fluctuation and temperature change under high-frequency sampling, the micro-cracks in the electrode material caused by cyclic load are detected, the crack propagation rate is quantified through crack contour fitting and diffusion path modeling method, and the mechanical scale analysis data is generated. The above three types of scale analysis data are used as the initial parameter set, which is normalized and spliced into the energy storage system digital twin model, so that the model not only has the ability to reflect the current multi-physical state of the energy storage unit, but also has the ability to deduce the future operation risk. In the process of continuously updating the state data, the twin model can accurately describe the charging and discharging strategy, the remaining life of the battery and the precursors of faults according to the evolution trend of different physical mechanisms.

[0042] Preferably, step S33 comprises the following steps:

[0043] The upper layer of the energy consumption reduction framework and the lower layer of the energy consumption stability framework are used to generate a game scheduling strategy to obtain an energy storage system energy scheduling strategy, wherein the energy storage system energy scheduling strategy includes an energy consumption reduction optimization strategy and a frequency response optimization strategy.

[0044] Obtain the price data; perform load prediction extraction on the upper layer of the energy consumption reduction framework to obtain energy storage load prediction data; perform 24-hour linear charging and discharging planning based on the load prediction data and the price data to obtain the energy consumption reduction optimization strategy.

[0045] Real-time power grid frequency data is obtained by performing real-time power grid frequency extraction on the lower layer of the energy consumption stability framework; when the real-time power grid frequency data exceeds 0.2 Hz, a storage correction instruction is triggered to adjust to obtain the frequency response optimization strategy.

[0046] This invention, within an upper-level energy conservation framework, uses historical and real-time electricity price data to model the correlation between the system and the historical operating trajectory of energy storage load. By performing high-dimensional time series prediction of load characteristics, it extracts predicted energy storage load data for the next 24 hours. This load prediction process combines short-cycle fluctuations and long-cycle trend factors to establish a time-step distribution feature vector, which is then cross-linearly modeled with electricity price data. This ultimately generates a continuous 24-hour linear charging and discharging behavior sequence, constituting an energy conservation optimization strategy. This allows the system to prioritize charging during low-price periods and discharging during high-price periods during intraday scheduling, thereby minimizing energy costs. Secondly, within a lower-level energy consumption stability framework, instantaneous changes in grid frequency are extracted using synchronous phase measurement data with a real-time sampling frequency of 50Hz. This constructs a grid frequency time series. When an abnormal event with a frequency deviation exceeding 0.2Hz occurs in this series, a frequency response mechanism is immediately triggered, invoking pre-set energy storage correction commands to rapidly adjust battery charging and discharging power. This correction strategy relies on regression analysis of energy storage response data under historical frequency disturbances to form a response boundary model, enabling rapid absorption and mitigation of frequency disturbances. Finally, the two strategies are fused using tensor-level weights to generate an energy dispatch strategy for the energy storage system that combines optimal energy reduction with frequency regulation capabilities. This strategy retains predictability across multiple time scales while possessing dynamic adaptability with a second-level response.

[0047] Preferably, step S4 includes the following steps:

[0048] Step S41: Obtain real-time energy storage monitoring data;

[0049] Step S42: Calculate the execution process deviation based on real-time energy storage monitoring data and energy storage system power dispatch strategy. If the energy storage deviation data is greater than the preset threshold, generate SOC attenuation data.

[0050] Step S43: The SOC attenuation data is transmitted back to the digital twin model of the energy storage system for network parameter optimization, generating an optimized digital twin model of the energy storage system;

[0051] Step S44: Based on the optimized digital twin model of the energy storage system, perform full-process scheduling analysis of the energy storage system and generate a report to obtain a full-process report on the power scheduling of the energy storage system.

[0052] The application realizes high-precision modeling of the operating state of the energy storage system and enhances the traceability of the whole scheduling process by introducing an execution process deviation feedback mechanism and a dynamic optimization process of the digital twin model, thereby improving the operation transparency and prediction and control capability of the system from the data level. First, the system obtains real-time energy storage monitoring data through a sensor network deployed at each key node of the energy storage subsystem. The data types include high-frequency operation indicators such as battery module SOC, terminal voltage, instantaneous power, and charge-discharge instruction response delay. After these data enter the data processing link, they are dynamically compared with the currently executed energy storage system power scheduling strategy to construct a time alignment mapping of the expected response sequence and the actual feedback sequence, thereby performing execution process deviation calculation. Through statistical sliding window analysis of the deviation data, the fluctuation mean and relative deviation rate are extracted. When the deviation exceeds the preset threshold (such as 5% or SOC fluctuation exceeding 2%), the system regards the deviation as a sign reflecting the internal degradation trend of the energy storage system and derives the SOC attenuation data accordingly. This attenuation data is not processed as an isolated indicator, but is fed back to the energy storage system digital twin model to participate in the update optimization of the twin network parameters. Through the BP iterative algorithm, the activation function weight and state transition function related to SOC prediction in the twin model are corrected, thereby generating an optimized digital twin model that is closer to the actual state. This model not only retains the physical modeling capability of prior knowledge such as battery thermal behavior and electrochemical response, but also integrates the dynamic feedback characteristics of operation data, realizing the self-evolution process of the twin. Finally, the system generates an energy storage system power scheduling whole process report based on the optimized digital twin model. The report can cover the full structural topology of the energy storage system in the spatial dimension, trace the evolution path of the execution deviation in the time dimension, and make quantitative predictions of the future operation trend within a certain period, providing high-precision reference for operation and maintenance personnel.

[0053] Preferably, step S42 comprises the following steps:

[0054] Step S421: Calculate the deviation value based on real-time energy storage monitoring data and energy storage system power scheduling strategy to generate energy storage deviation value data;

[0055] Step S422: Calculate the SOC deviation based on real-time energy storage monitoring data and energy storage system power scheduling strategy to obtain energy storage SOC deviation data;

[0056] Step S423: If the energy storage deviation value data is greater than 5% and the energy storage SOC deviation data is greater than 2%, generate a compensation coefficient for error propagation path backtracking and obtain SOC attenuation data.

[0057] The application effectively improves the accuracy and response efficiency of state anomaly identification and SOC (State of Charge) attenuation trend tracking of the energy storage system by introducing a multiple deviation quantification mechanism and error propagation path backtracking analysis. First, the system uses the comparison relationship between real-time energy storage monitoring data and the current scheduling strategy to quantitatively calculate the overall scheduling deviation and the battery SOC deviation. Among them, the energy storage deviation value data is obtained by sampling and normalizing the difference between the real-time charging and discharging power and the target power, constructing a time series offset function to reflect the actual response deviation of the scheduling system at the execution level. At the same time, the SOC deviation data is based on the point-to-point error between the SOC prediction curve and the actual observation curve, combined with the volatility and difference gradient change in the current time window, to extract a dynamic SOC residual feature vector. This dual deviation index system can realize parallel perception of scheduling deviation and battery aging trend. Further, when the energy storage deviation value data exceeds 5% and the SOC deviation data exceeds 2%, a composite threshold condition, the system starts the error propagation path backtracking mechanism to track the cause of the deviation, including multi-stage control instructions, charging and discharging rate adjustment history, battery pack thermal behavior trend, and power grid instantaneous frequency disturbance. Through principal component analysis and error sensitivity evaluation, the key nodes and main influence variables of the deviation are mined. On this basis, the system constructs an error compensation coefficient matrix based on path backtracking, combines the historical attenuation law to correct and fit the current deviation, and finally generates SOC attenuation data with physical rationality and data-driven accuracy. The SOC attenuation data not only has the advantages of high precision and fast response, but also can be used as feedback input to optimize the digital twin parameters, supporting subsequent state assessment and prediction analysis.

[0058] In the present specification, a power scheduling system based on an energy storage system is provided for performing the above-mentioned power scheduling method based on an energy storage system, which comprises:

[0059] A multi-source heterogeneous perception standardization module is configured to deploy a distributed sensing array inside a battery module to collect multi-source heterogeneous data, and to perform protocol conversion to obtain an initial multi-modal data set; the initial multi-modal data set is filtered by a noise spectrum adjustment threshold and long-term drift compensation to generate a standardized energy storage data set;

[0060] A cross-modal feature fusion module is configured to perform tensor completion on the standardized energy storage data set to generate energy storage tensor completion data; the energy storage tensor completion data is analyzed in time and space dimensions, and cross-modal feature fusion is performed to generate an energy storage cross-modal feature space;

[0061] The digital twin modeling and game scheduling module is used for constructing a battery health state digital twin based on a cross-modal feature space of the energy storage, obtaining a digital twin model of the energy storage system, and generating a game scheduling strategy for the digital twin model of the energy storage system to obtain an electric energy scheduling strategy for the energy storage system.

[0062] The real-time deviation analysis and closed-loop optimization module is used for obtaining real-time energy storage monitoring data, performing deviation calculation based on the real-time energy storage monitoring data and the electric energy scheduling strategy for the energy storage system, generating SOC attenuation data if the energy storage deviation data is greater than a preset threshold, returning the SOC attenuation data to the digital twin model of the energy storage system for network parameter optimization, performing full-process scheduling analysis of the energy storage system, obtaining full-process scheduling analysis data, and generating a scheduling report based on the full-process scheduling analysis data to obtain an electric energy scheduling report for the whole process of the energy storage system.

[0063] The beneficial effects of the present application are that through the deployment of a multi-source sensor array, real-time acquisition of multi-dimensional heterogeneous data in a cold chain warehouse environment is realized, and accurate synchronization of different data sources in the time dimension is ensured based on timestamp alignment technology, forming high timeliness and integrity of cold chain warehouse timestamp alignment data. Subsequently, through three-dimensional space mapping and construction of feature tensor matrix, the spatial distribution information of temperature, humidity and other environmental parameters is integrated to generate a fusion feature matrix, which can reflect the multi-dimensional dynamic characteristics of the warehouse environment and provide a structured data basis for subsequent analysis. On this basis, a preset long short-term memory network (LSTM) model is used to predict the temperature distribution of the fusion feature matrix, capturing the time series characteristics and spatial correlation of temperature changes, and generating warehouse temperature prediction distribution data. This data is further combined with the fusion feature matrix to construct a cold chain causal knowledge graph, revealing the causal relationship between environmental variables, and based on the graph, a digital twin model is constructed to realize multi-dimensional simulation and dynamic simulation of the warehouse system state. The digital twin model then undergoes error residual iterative analysis, the system identifies the deviation between model prediction and actual monitoring, generates difference analysis data, and automatically adjusts compensation parameters through a feedback loop to complete the optimization and fine-tuning of the model, forming a more accurate warehouse optimization digital twin model. Finally, the optimized model drives the construction of a logistics warehouse management report, systematically presenting key indicators such as temperature prediction accuracy, environmental stability and management efficiency, supporting scientific decision-making and precise scheduling. Overall, this method enhances the environmental perception ability and prediction accuracy of the cold chain warehouse system through a closed loop of data acquisition, synchronization, fusion, prediction and feedback optimization, achieving dynamic adaptability and continuous optimization of the model. BRIEF DESCRIPTION OF DRAWINGS

[0064] Figure 1 It is a step flowchart of an electric energy scheduling method based on an energy storage system.

[0065] Figure 2 To Figure 1 Detailed implementation step flow diagram of step S4 in the method;

[0066] The implementation, functional features and advantages of the present application will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION

[0067] The technical method of the present application will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0068] In addition, the accompanying drawings are only schematic illustrations of the present application and are not necessarily drawn to scale. The same reference numerals in the drawings represent the same or similar parts, and thus repeated description thereof will be omitted. Some of the block diagrams shown in the drawings are functional entities, which do not necessarily have to correspond to physically or logically independent entities. The functional entities can be implemented in the form of software, or in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0069] It should be understood that although the terms "first", "second" and the like can be used herein to describe various elements, these elements should not be limited by these terms. These terms are only used to distinguish one element from another. For example, without departing from the scope of the exemplary embodiments, a first element can be called a second element, and similarly a second element can be called a first element. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0070] To achieve the above-mentioned purpose, please refer to Figures 1 to 2 A power scheduling method based on an energy storage system, the method comprising the following steps:

[0071] Step S1: Deploying a distributed sensing array inside a battery module to collect multi-source heterogeneous data, and performing protocol conversion to obtain an initial multi-modal data set; performing noise spectrum adjustment filtering threshold on the initial multi-modal data set, and performing long-term drift compensation to generate a standardized energy storage data set;

[0072] Step S2: Performing tensor completion on the standardized energy storage data set to generate energy storage tensor completion data; performing time-space dimension analysis on the energy storage tensor completion data, and performing cross-modal feature fusion to generate an energy storage cross-modal feature space;

[0073] Step S3: Constructing a battery health state digital twin based on the energy storage cross-modal feature space to obtain an energy storage system digital twin model; performing game scheduling strategy generation on the energy storage system digital twin model to obtain an energy storage system power scheduling strategy;

[0074] Step S4: Obtaining real-time energy storage monitoring data; performing execution process deviation calculation based on the real-time energy storage monitoring data and the energy storage system power scheduling strategy, and if the energy storage deviation data is greater than a preset threshold, generating SOC attenuation data; returning the SOC attenuation data to the energy storage system digital twin model for network parameter optimization, and performing energy storage system full-process scheduling analysis to obtain full-process scheduling analysis data; generating a scheduling report based on the full-process scheduling analysis data to obtain an energy storage system power scheduling full-process report.

[0075] In the embodiment of the present application, reference Figure 1 As shown in the figure, it is a step flow diagram of an energy scheduling method based on an energy storage system, and in the present example, the energy scheduling method based on the energy storage system comprises the following steps:

[0076] Step S1: Deploying a distributed sensor array inside a battery module to collect multi-source heterogeneous data, and performing protocol conversion to obtain an initial multi-modal data set; performing noise spectrum adjustment filtering threshold on the initial multi-modal data set, and performing long-term drift compensation to generate a standardized energy storage data set;

[0077] In the embodiment of the present application, key physical quantity monitoring points are selected in the battery module, and multiple types of sensing units such as temperature, voltage, current, stress and humidity are arranged to realize synchronous acquisition of multi-source physical signals in the battery operation process. These sensor nodes are connected to the edge collection terminal through wired or wireless communication mode to form a distributed data collection network, and data synchronization and fusion are performed through time stamp and sampling frequency. Since the collected data has obvious heterogeneity in physical source, dimension unit, time scale and spatial distribution, data cleaning is required to eliminate missing values, abnormal values and invalid sampling points, and then numerical scale standardization and feature mode unification are performed based on normalization, standard deviation scaling and wavelet transform, to ensure the comparability and processability of different types of data. On this basis, the standardized data of all sensor nodes under the same time window are structured and organized to construct a multi-dimensional tensor form data expression structure, taking time series as the main axis and spatial nodes and sensing types as the auxiliary axis, to form a standardized energy storage data set with integrity, time sequence and structured features. The data set not only retains the correlation between multi-source physical information, but also has scalability, providing a unified and standardized data basis for subsequent tensor completion, modal fusion and twin construction.

[0078] Step S2: tensor completion is performed on the standardized energy storage dataset to generate energy storage tensor completion data; time-space dimension analysis is performed on the energy storage tensor completion data, and cross-modal feature fusion is performed to generate an energy storage cross-modal feature space;

[0079] In the embodiment of the present application, the standardized energy storage dataset generally appears as a sparse tensor with a time-space-modal three-dimensional structure. Some observed values in the tensor are missing due to sensor instantaneous failure, communication delay or environmental interference. To improve data integrity, the tensor structure needs to be completed first. The completion process can use a tensor decomposition method based on low-rank constraint, such as CANDECOMP / PARAFAC decomposition (CP decomposition) or Tucker decomposition, to separate the core tensor and factor matrix from the original sparse tensor, and to estimate the missing data by minimizing the reconstruction error; or a neural completion method based on deep learning can be introduced, such as constructing a time convolution network (TCN) or a space-time graph convolution network (ST-GCN), to perform nonlinear interpolation and reconstruction using the continuity features of data in the time and space dimensions. The completed tensor data serves as a complete multi-dimensional data structure, providing a basis for multi-dimensional joint analysis. On this basis, time series trend extraction is further performed through a sliding time window technique, and spatial dependence analysis is performed in combination with the spatial topology of the sensor nodes, such as using dynamic time warping (DTW) to align the time series of different nodes, or using a graph neural network to model the spatial correlation. After completing the time-space joint modeling, in combination with the characteristics of different physical quantities such as voltage, current, temperature and displacement on the modal axis, the feature vectors of each modal are mapped to a unified embedding space through principal component analysis (PCA), canonical correlation analysis (CCA) or an attention mechanism controlled multi-modal fusion network, a high-dimensional interaction matrix is constructed, and the collaborative representation and fusion of cross-modal features are realized, finally forming an energy storage cross-modal feature space with rich semantic structure and strong expression ability, which is used to support subsequent digital twin construction and intelligent scheduling strategy generation.

[0080] Step S3: constructing a battery health state digital twin based on the energy storage cross-modal feature space to obtain an energy storage system digital twin model; performing game scheduling strategy generation on the energy storage system digital twin model to obtain an energy storage system energy scheduling strategy;

[0081] In the embodiments of the present application, the holographic mapping and game- type optimization control of the energy storage system operating state are realized through the structured modeling of multi-dimensional heterogeneous characteristics and the state prediction mechanism. From the data level, the energy storage cross-modal characteristic space contains high-dimensional characteristic tensors generated by tensor completion and space-time fusion, covering multi-source modal indicators such as voltage, current, temperature, strain, and insulation state, and maintaining continuity and consistency in time and space dimensions. For this high-dimensional characteristic space, a feature-driven digital twin body oriented to health state representation needs to be first constructed, usually using a deep neural network architecture to model the characteristic space, including but not limited to multilayer perceptron (MLP), convolutional neural network (CNN), or a hybrid structure combining graph neural network (GNN) and recurrent neural network (RNN), to enable the model to recognize potential failure modes, degradation processes, and life trends inside the battery module. During model training, historical operating state labels or expert annotated data are used as supervisory signals to minimize the error function between predicted state and true state, and gradually complete the convergence of digital twin model parameters. On this basis, the health state digital twin body is used as an interactive bridge between the system virtual body and the physical body, and a reinforcement learning or game theory optimization strategy (such as the policy gradient method based on Markov decision process, Q-learning algorithm, or Stackelberg game model) is introduced, and according to the health state evolution trend predicted by the twin model, the load distribution, power regulation, and balance control of each energy storage unit are generated, and an executable power scheduling action set is output. The strategy generation process takes maximizing resource utilization, minimizing system degradation, and optimizing response time as the objective function, and performs multi-round strategy game solving in the feasible solution space, and feeds back the results to the physical system to guide the allocation and execution of energy storage tasks, forming a closed-loop scheduling mechanism combining virtual and real.

[0082] Step S4: acquiring real-time energy storage monitoring data; performing execution process deviation calculation based on the real-time energy storage monitoring data and the energy storage system power scheduling strategy, if the energy storage deviation data is greater than a preset threshold, generating SOC attenuation data; returning the SOC attenuation data to the energy storage system digital twin model for network parameter optimization, and performing energy storage system full-process scheduling analysis to obtain full-process scheduling analysis data; generating a scheduling report based on the full-process scheduling analysis data to obtain an energy storage system power scheduling whole-process report.

[0083] In the embodiments of the present application, the edge computing node or embedded monitoring unit collects multi-source index data streams in real time during system operation, including but not limited to battery monomer voltage, current, temperature, state of charge (SOC), insulation resistance and energy flow distribution. The data is transmitted to the local or cloud time series database through high-frequency sampling, and after time alignment, abnormal value elimination and unit normalization, the real-time energy storage monitoring data after cleaning is formed. Subsequently, the real-time data will be input into the energy storage system electric energy scheduling strategy module generated by the game, and by comparing the deviation between the actual operation result and the strategy expected output, the execution process deviation is calculated using evaluation functions such as mean square error (MSE), dynamic time warping (DTW) or Bayesian error boundary. If the deviation exceeds the preset dynamic tolerance threshold interval, the deviation will be further analyzed to identify the root cause of performance degradation, and the SOC (State of Charge) degradation data is quantified with battery capacity attenuation, internal resistance rise or thermal runaway trend as an index. Such SOC degradation data not only has time correlation, but also shows high-dimensional sparse sequence related to historical state, environmental factors and usage patterns, which needs to be modeled through fusion attention mechanism or variational autoencoder (VAE) structure to enhance its description ability of system state degradation path. Subsequently, the SOC degradation data is fed back to the energy storage system digital twin model, and by adjusting the neural network parameters, state transition probability matrix or dynamic discriminator weight in it, the adaptive update of the twin to the environment and degradation changes is realized. Finally, the monitoring data, deviation evaluation results, strategy adjustment records and optimized state parameters in the whole process are integrated to generate an energy storage system electric energy scheduling whole process report through a unified information integration module. The report not only has time sequence and traceability, but also can be used as data basis for subsequent strategy verification and health assessment.

[0084] Preferably, step S1 comprises the following steps:

[0085] Step S11: Deploy voltage sensors and temperature sensors inside the battery module to collect monomer temperature data and terminal voltage data at a frequency of 10 Hz, wherein the accuracy of the voltage sensor is 0.1%, and the accuracy of the temperature sensor is 0.5℃;

[0086] Step S12: Deploy a high-frequency current sensor with a bandwidth of 20 kHz on the DC side of the energy storage converter to collect charge and discharge current waveforms at a frequency of 1 kHz, and generate energy storage charge and discharge waveform data;

[0087] Step S13: Deploy a synchronous phase measurement unit at the grid access point to collect grid frequency, voltage phase angle and harmonic distortion rate at a frequency of 50 Hz, and generate grid phase raw data;

[0088] Step S14: Convert the individual cell temperature data, terminal voltage data, energy storage charging and discharging waveform data, and grid phase raw data through a multi-protocol adapter to generate an initial multi-mode dataset;

[0089] Step S15: Adjust the noise spectrum filtering threshold of the initial multimodal dataset and perform long-term drift compensation to generate a standardized energy storage dataset.

[0090] In this embodiment of the invention, a voltage sensor with an accuracy of 0.1% and a temperature sensor with an accuracy of 0.5°C are deployed within the battery module. Individual cell terminal voltage and temperature data are collected at a frequency of 10Hz. The resulting signals form a low-frequency sampling state trajectory on the time axis, primarily used to evaluate the balance and thermal stability between cells. Secondly, a high-frequency current sensor with a bandwidth of 20kHz is deployed on the DC side of the energy storage converter. Charging and discharging current waveforms are collected at a frequency of 1kHz, capturing dynamic behaviors such as high-frequency disturbances and electromagnetic transient responses. This data is presented in time-series form and combined with methods such as Fourier transform or wavelet packet decomposition for time-frequency domain feature analysis. Simultaneously, a synchronous phasor measurement unit (PMU) is installed at the grid connection point to synchronously collect the grid frequency, voltage phase angle, and harmonic distortion rate at a frequency of 50Hz, forming a grid time-series vector containing phasor information, used to describe the dynamic interaction between the system and the grid. The three types of data sources mentioned above have different sampling frequencies, data dimensions, and encoding formats. A multi-protocol adapter is needed to uniformly encapsulate and convert the data carried by communication protocols such as Modbus, CAN, and IEC 61850. Negotiation and handshake, and data stream packaging are completed at the physical and network layers to ultimately generate an initial multimodal dataset with unified timestamps and a consistent structure. After completing protocol compatibility processing, to ensure the stability and comparability of the data in subsequent modeling processes, noise spectrum analysis is performed on the initial multimodal dataset. Based on the power spectral density (PSD) calculation results, a filtering threshold is set, and adaptive filters or bandpass filters are used to reduce high-frequency interference signals. Simultaneously, to address slow variable shifts caused by factors such as temperature drift and electrochemical aging during long-term use, drift correction methods such as least squares fitting of drift trends, incremental moving average compensation, or exponentially weighted moving average (EWMA) are used to correct signal drift. Finally, the above processing generates a standardized energy storage dataset with uniform scale, low noise, and high stability, providing a consistent basic input for subsequent data modeling and state estimation.

[0091] Preferably, step S2, tensor completion of the standardized energy storage dataset, includes:

[0092] The standardized energy storage dataset is mapped to a three-dimensional space and sorted by timestamp to obtain standardized three-dimensional mapping data for energy storage.

[0093] The temperature and current data of the standardized data of the three-dimensional mapping of the energy storage are repaired by using a low-rank Tucker decomposition to obtain standard tensor repair data.

[0094] The standard tensor repair data is constructed into a three-dimensional feature tensor of device-time-feature, and missing value detection is performed to obtain energy storage tensor completion data.

[0095] In the embodiment of the application, the multi-source energy storage data set after standardization is reorganized based on a unified timestamp, and three-dimensional mapping operation is performed according to three dimensions of "device number-sampling time-measured feature", thereby constructing a three-dimensional sparse tensor taking an energy storage unit as a spatial index, taking a physical time sequence as an evolution index, and taking temperature, voltage, current and other state parameters as a feature index. In the three-dimensional structure, there are problems of inconsistent sampling frequency and data missing in different dimensions, especially under high-frequency measurement, temperature and current signals are easily affected by sensor failure or communication delay to appear null. In order to mathematically complete the data gap, the Tucker low-rank tensor decomposition method is selected, the three-dimensional tensor is first decomposed into a core tensor and multiple mode matrices through high-order singular value decomposition (HOSVD), and then rank constraint optimization is performed in the tensor subspace, and the missing area is numerically estimated by minimizing the tensor reconstruction error. This process relies on the low-rank structure assumption of the observed data, thereby realizing the global collaborative repair of missing values in the temperature and current dimensions. After completing the tensor decomposition and numerical filling, the reconstructed tensor is reorganized into the form of device-time-feature to form a three-dimensional feature tensor with complete structure for subsequent modeling analysis. Then, a missing value detection mechanism is introduced into the tensor data, and the initial missing area and the repair area are marked by a mask matrix, and the integrity of the tensor, the repair error and the consistency of the statistical distribution are checked, and the abnormal points with large error are removed or the repair algorithm parameters are further refined, so as to obtain the final energy storage tensor completion data, thereby providing a data basis for feature fusion and state modeling.

[0096] Preferably, the step S2 of performing time-space dimension analysis on the energy storage tensor completion data and performing cross-modal feature fusion comprises:

[0097] The energy storage tensor completion data is subjected to time dimension data extraction to generate battery SOC data;

[0098] The energy storage tensor completion data is subjected to spatial dimension data extraction to generate power grid frequency fluctuation rate data;

[0099] The battery SOC data and the power grid frequency fluctuation rate data are time-axis aligned to generate a synchronized energy storage data set;

[0100] The battery SOC data is subjected to long-term dependence analysis to obtain battery dependence analysis data; and the battery dependence analysis data is subjected to charge-discharge behavior feature extraction to obtain a time step feature vector;

[0101] The grid frequency fluctuation rate data is subjected to unit topology impedance calculation to obtain grid topology impedance calculation data; the grid topology impedance calculation data is subjected to charge-discharge path electric energy loss relationship analysis, and spatial distribution feature extraction is performed to obtain a spatial dimension feature matrix;

[0102] The time step feature vector and the spatial dimension feature matrix are subjected to tensor splicing to obtain a storage feature splicing matrix; the storage feature splicing matrix is subjected to synchronization processing by using a synchronized storage dataset to obtain a storage scheduling feature fusion matrix;

[0103] The storage scheduling feature fusion matrix is subjected to time-space initial weight distribution, and a feature space is constructed to obtain a storage cross-modal feature space.

[0104] In the embodiment of the present application, the key state variables are extracted from the completed energy storage three-dimensional tensor along the time axis direction to construct the battery SOC (State of Charge) data sequence in time sequence mode to capture the dynamic change trend of the battery state of charge; then the frequency fluctuation data at the corresponding positions on the grid side and the device side are extracted along the spatial axis direction, the fluctuation rate is calculated based on the continuous time slice, and the spatial distributed frequency response matrix is constructed to form the grid frequency fluctuation rate data. After the preliminary extraction of the two modal data, the interpolation alignment, window synchronization and other methods are used to map them to the unified time axis to generate the synchronized energy storage data set for supporting the subsequent joint modeling. Then, the time-dependent relationship modeling is performed for the battery SOC data, the local time-dependent features are extracted by using the sliding window mechanism, and further the autocorrelation structure or long-term memory structure is constructed to capture the long-range correlation in the time evolution of the battery state by statistical analysis and frequency domain transformation (such as Fourier or wavelet transformation) to obtain the battery-dependent relationship analysis data. On this basis, the rate, amplitude change, peak-valley turning and other behavior characteristics in the charging and discharging process are extracted to construct the time step feature vector with time recursion. At the same time, for the grid frequency fluctuation rate data, the grid unit impedance matrix is constructed according to the electrical topological structure and the node position relationship, the energy loss of the charging and discharging path between nodes is analyzed by the impedance solving algorithm to form the grid topological impedance calculation data, and the spatial distribution characteristics are extracted by the high-dimensional feature coding or spatial grouping strategy to form the spatial dimension feature matrix. Then the time step feature vector and the spatial dimension feature matrix are subjected to tensor splicing operation to construct the energy storage feature splicing matrix with unified dimension, and the feature splicing matrix is subjected to standardization, normalization and time stamp synchronization processing by using the synchronized energy storage data set to generate the energy storage scheduling feature fusion matrix with alignment consistency and modal complementarity. Finally, the initial weight distribution of the time dimension and the spatial dimension is performed on the fusion matrix, the joint feature expression space is constructed by using the attention mechanism, position embedding or weight regularization strategy to form the energy storage cross-modal feature space which can be used for subsequent modeling.

[0105] Preferably, step S3 comprises the following steps:

[0106] Step S31: performing multi-scale energy storage analysis based on the energy storage cross-modal feature space, and constructing a battery health state digital twin to obtain an energy storage system digital twin model;

[0107] Step S32: designing a double-layer optimization framework based on the energy storage system digital twin model to obtain an energy storage double-layer optimization framework, wherein the energy storage double-layer optimization framework comprises an upper-layer consumption reduction framework and a lower-layer energy consumption stability framework;

[0108] Step S33: generating a game scheduling strategy based on the upper-layer consumption reduction framework and the lower-layer energy consumption stability framework to obtain an energy storage system power scheduling strategy.

[0109] In the embodiments of the present application, on the basis of the energy storage cross-modal feature space, the multi-scale space-time decomposition method is used to analyze the operation data of the energy storage system in layers, including the battery charge change pattern in the short time scale and the aging trend identification in the long time scale, and the state evolution path is constructed by combining the joint distribution characteristics of key variables such as temperature, current and voltage. By mapping these multi-scale time series into dynamic Bayesian networks or time sequence coding structures based on graph convolution, the dynamic characterization of the battery health state can be realized, and then the digital twin model reflecting the performance degradation rule and abnormal behavior response of the battery is trained to form the digital twin model of the energy storage system. Subsequently, based on the digital twin model, a double-layer optimization framework is constructed, in which the upper layer consumption reduction framework takes the maximum revenue or minimum cost as the objective function, and based on the predicted market price, load demand and energy storage state distribution, a mixed integer linear programming or Lagrangian relaxation optimization model is constructed; the lower layer energy consumption stability framework focuses on the energy balance and system dynamic constraints in the real-time operation process, and a second-order state space model or robust optimization structure is constructed to ensure that the stability conditions such as voltage, current and frequency are met in the actual dispatching process. The upper and lower layers establish a coupling feedback channel through an iterative updating mechanism, that is, the upper layer provides a candidate solution space for the dispatching strategy, and the lower layer returns the stability evaluation index and feeds back the correction boundary conditions, forming a closed-loop optimization iteration system. Finally, the non-cooperative game or Stackelberg game modeling method is introduced on the double-layer optimization framework to consider the resource competition relationship between the battery module, the grid node and the external load and other multi-agents, and through the construction of a Nash equilibrium solving model or an evolutionary game strategy learning algorithm, the dynamic balance of the dispatching strategy among different agents is realized, so as to generate an energy scheduling strategy for the energy storage system facing the global efficiency and local stability. This process highly depends on the time consistency, spatial integrity and modal coordination of the energy storage data, ensuring the rationality and effectiveness of the game dispatching mechanism at the data level.

[0110] Preferably, step S31 comprises the following steps:

[0111] Step S311: Calculate the Ohmic polarization / diffusion polarization loss based on the energy storage cross-modal feature space to generate electrochemical scale analysis data;

[0112] Step S312: Perform thermodynamic analysis for capacity attenuation prediction based on the energy storage cross-modal feature space to generate thermodynamic scale analysis data;

[0113] Step S313: Perform electrode stress crack detection and crack diffusion rate calculation based on the energy storage cross-modal feature space to generate mechanical scale analysis data;

[0114] Step S314: parameter initialization is performed on the electrochemical scale analysis data, the thermodynamic scale analysis data and the mechanical scale analysis data, and a digital twin is constructed to generate a digital twin model of the energy storage system.

[0115] In the embodiment of the present application, on the electrochemical scale, key variables such as instantaneous voltage, current density, internal resistance estimate and temperature field change are extracted from the cross-modal feature space, and the energy loss caused by Ohmic polarization and concentration polarization is quantified according to the extended Butler-Volmer equation framework and the RC submodule response function in the equivalent circuit model of the battery. By introducing impedance spectrum mapping and multi-node voltage response differential analysis method, hidden variables such as electrolyte conductivity and electrode interface reaction kinetics parameters can be further separated to form electrochemical scale analysis data. Secondly, in the thermodynamic scale analysis, the temperature gradient distribution, current waveform and energy input and output time series continuously collected in the cross-modal feature space are used to construct a non-steady-state heat conduction model, and the battery material heat capacity and thermal conductivity parameters are combined to perform finite difference time step promotion to derive the battery temperature rise trend and local overheating probability density function. The key thermal driving mechanism causing capacity attenuation is predicted through the energy conservation model to generate thermodynamic scale analysis data. Then, in the mechanical scale analysis, with the aid of high-dimensional feature patterns such as abnormal growth of internal resistance, voltage step change and compression of charge and discharge cycles in the cross-modal feature space, a crack detection network is trained combined with XRD parameters or finite element simulation results to calculate the maximum principal stress and shear stress distribution of the electrode sheet under periodic load. According to the Paris law or the crack growth empirical model, the crack propagation rate is estimated to construct the mechanical stress driven crack evolution data set. Finally, the analysis data of the electrochemical, thermodynamic and mechanical scales are initialized and fused through the method of high-dimensional tensor parameter alignment and physical constraint embedding to generate a cross-scale consistent input feature tensor, and an energy storage system digital twin model with a structure-behavior-performance three-layer mapping mechanism is constructed to provide multi-scale coupling support for subsequent scheduling simulation and performance prediction.

[0116] Especially important is that step S32 includes the following:

[0117] Step S32: based on the digital twin model of the energy storage system, a double-layer optimization framework is designed to obtain an energy storage double-layer optimization framework, wherein the energy storage double-layer optimization framework includes an upper-layer consumption reduction framework and a lower-layer energy consumption stability framework.

[0118] The formula of the upper-layer consumption reduction framework is:

[0119]

[0120] wherein, is a time index; is the total number of scheduling periods; The electricity purchase / sale unit price of the power grid for the t period; The power exchanged with the power grid (kW) for the t period, positive for electricity purchase, negative for electricity sale; The battery aging cost coefficient; The battery capacity attenuation amount for the t period; The rated capacity of the battery;

[0121] The formula for the lower energy consumption stability is:

[0122]

[0123] Wherein, The actual frequency of the power grid for the t period; The state of charge of the battery for the t period; The SOC reference value; The SOC deviation penalty coefficient.

[0124] In the embodiment of the application, the time sequence consumption target and real-time stability constraint are cooperatively controlled through hierarchical processing. The specific technical means is embodied as follows: based on the battery capacity attenuation prediction data output by the digital twin model (regenerated by the LSTM network from the real-time collected temperature , current and historical charge and discharge cycle data) and the time sequence data of the power exchanged with the power grid (originating from the 15-minute granularity electricity price of the SCADA system and the day-ahead load prediction data), the upper consumption reduction framework quantifies the battery loss as a consumption reduction cost item (wherein is an aging cost conversion coefficient, which is dynamically updated by battery full life cycle cost allocation), and forms a linear weighted objective function with the electricity purchase and sale cost , and adopts branch and bound to solve the 96-period mixed integer programming problem; the lower energy consumption stability framework real-time accesses the millisecond-level power grid frequency data stream collected by the PMU and the second-level SOC data reported by the BMS, calculates the L2 norm of the frequency deviation and the SOC tracking error (wherein is dynamically set as a safety window of 40%-70% by the upper optimization result), introduces a regularization coefficient to balance the weights of the two (when the frequency fluctuation exceeds the threshold, it is automatically reduced to relax the SOC constraint), finally forms a quadratic programming problem and uses the interior point method to solve it online, and outputs the power correction instruction to the energy storage converter execution layer, while the actual running data is fed back to the digital twin to realize parameter closed-loop calibration.

[0125] Preferably, the step S33 comprises the following steps:

[0126] The game scheduling strategy is generated based on the upper layer consumption reduction framework and the lower layer energy consumption stability framework, and the energy scheduling strategy of the energy storage system is obtained, wherein the energy scheduling strategy of the energy storage system comprises a consumption reduction optimization strategy and a frequency response optimization strategy.

[0127] The price data is obtained, the load prediction extraction is performed on the upper layer consumption reduction framework, and the energy storage load prediction data is obtained; the 24-hour linear charging and discharging plan is performed based on the load prediction data and the price data, and the consumption reduction optimization strategy is obtained.

[0128] The real-time power grid frequency data is obtained by performing the real-time power grid frequency extraction on the lower layer energy consumption stability framework; when the real-time power grid frequency data exceeds 0.2 Hz, the energy storage correction instruction is triggered to adjust, and the frequency response optimization strategy is obtained.

[0129] In the embodiment of the present application, in the upper layer consumption reduction framework, the system receives historical power consumption load data and meteorological prediction factors (such as temperature, humidity, holiday index, etc.), and constructs an energy storage load prediction model by using an integrated time series regression method (such as an LSTM model or a SARIMA model based on a sliding window), to output a 24-hour time period energy demand curve, thereby forming energy storage load prediction data. At the same time, the system introduces the price time series data provided by the electricity market, and performs trend analysis and price fluctuation interval identification on the price time series data. Based on the two types of data, the objective function is set as "minimizing the consumption cost", and under the premise of meeting the capacity and power constraints of the energy storage device, the minimum SOC limit value and the life degradation constraint, a 24-hour linear programming model is constructed, the charging and discharging instructions of each hour are set, and the consumption reduction optimization strategy is obtained by solving the linear programming model. Secondly, in the lower layer energy consumption stability framework, the system accesses the frequency data stream generated by the power grid synchronous phasor measurement device (PMU) in real time, the sampling frequency is 10 points per second, the median filter and trend elimination processing are performed first, and then the frequency change rate (ROCOF) is differentiated and analyzed and the threshold is judged. When the instantaneous disturbance of the frequency deviation from the center value 50 Hz exceeding ±0.2 Hz is detected, the frequency response mechanism is triggered, and the current consumption reduction optimization strategy is offset and corrected according to the frequency deviation size and direction. This correction dynamically adjusts the charging and discharging power based on the pre-defined proportional-integral adjustment factor, and the correction command is issued within the strategy control frame (such as within 1 minute), so as to generate the frequency response optimization strategy. Finally, the consumption reduction optimization strategy and the frequency response optimization strategy are jointly scheduled through the game coordination mechanism to ensure that the system realizes dynamic balance between consumption reduction and power grid stability, generates a complete energy scheduling strategy of the energy storage system, and sends the energy scheduling strategy to the execution control unit.

[0130] As an example of the present application, reference is made to Figure 2As shown, the step S4 includes:

[0131] Step S41: acquiring real-time energy storage monitoring data;

[0132] Step S42: performing process deviation calculation based on real-time energy storage monitoring data and energy scheduling strategy of the energy storage system, and if the energy storage deviation data is greater than a preset threshold, generating SOC attenuation data;

[0133] Step S43: returning the SOC attenuation data to the energy storage system digital twin model for network parameter optimization, and generating an optimized digital twin model of the energy storage system;

[0134] Step S44: performing energy storage system full-process scheduling analysis based on the optimized digital twin model of the energy storage system, and generating a report to obtain an energy scheduling full-process report of the energy storage system.

[0135] In the embodiment of the present application, a plurality of types of sensor arrays deployed in the battery module, the electric energy conversion device and the grid side are used to acquire real-time energy storage monitoring data at a high frequency, including parameters such as single-cell SOC, voltage, current, temperature, charging and discharging state, energy storage device operating state, grid frequency fluctuation, and the like, and the data is synchronized by using a unified timestamp. On this basis, the system calls the currently activated energy scheduling strategy of the energy storage system, compares the expected control signal output by the strategy with the actually collected energy storage behavior data step by step, constructs an execution process deviation measurement function based on indicators such as mean square error and mean absolute error, and quantifies the difference between the strategy execution and the device response. If the deviation index exceeds a preset dynamic threshold (the threshold is automatically generated by the fluctuation confidence interval in the running period), a state attenuation analysis module is started, an SOC attenuation rate curve is extracted from the deviation accumulation trend, and SOC attenuation data is generated by surface fitting and parameter regression method. Then, the SOC attenuation data is transmitted as feedback input to the state estimation sub-network in the digital twin model of the energy storage system, and the network weight and input mapping parameter are iteratively reconstructed by using a weighted update mechanism (such as Bayesian optimization or stochastic gradient update), so as to form an optimized digital twin model of the energy storage system. After the optimization of the digital twin model is completed, the system calls the full-process scheduling simulation module, compares the current system load prediction, electricity price data and frequency disturbance level, re-plans the charging and discharging path and checks the stability, and generates multi-period operation trajectories and control instructions by using a multi-objective scheduling solution algorithm. Finally, the strategy execution result, deviation correction record, SOC evolution trend, scheduling load response and topology change are combined in the form of text and graphics to construct an energy scheduling full-process report of the energy storage system, and to provide basic data support for subsequent operation evaluation and strategy fine-tuning.

[0136] Preferably, the step S42 includes the following steps:

[0137] Step S421: Calculate the deviation value based on the real-time energy storage monitoring data and the energy scheduling strategy of the energy storage system, and generate energy storage deviation value data;

[0138] Step S422: Calculate the SOC deviation based on the real-time energy storage monitoring data and the energy scheduling strategy of the energy storage system, and obtain energy storage SOC deviation data;

[0139] Step S423: If the energy storage deviation value data is greater than 5% and the energy storage SOC deviation data is greater than 2%, generate a compensation coefficient for error propagation path backtracking, and obtain SOC attenuation data.

[0140] In the embodiment of the application, by comparing the real-time energy storage monitoring data with the current scheduling strategy of the energy storage system step by step, the difference between the actual response value and the expected scheduling value of the key control variables such as voltage, current and power is extracted, a state variable difference sequence is constructed, and a comprehensive energy storage deviation value data is calculated based on a time weighted residual function. Subsequently, the system further independently models the SOC state, analyzes the cumulative error in the SOC estimation path through a two-way integral model or a Kalman filter, compares the theoretical SOC trajectory calculated with the actual SOC data collected, and generates energy storage SOC deviation data. After obtaining the deviation data, the system sets a threshold judgment mechanism. If the comprehensive energy storage deviation value exceeds 5% and the SOC deviation exceeds 2%, the error propagation path analysis module is activated, and an error path atlas based on causal inference and residual transmission mechanism is constructed. The atlas analyzes the influence intensity of the deviation source node on the SOC output result using a multi-scale dynamic Bayesian network or a graph convolution mechanism, generates error transmission channels across variable dimensions, and extracts stable influence paths combined with the statistical behavior of the historical running interval. On this basis, a compensation coefficient generation model is constructed. The model performs hierarchical weighting according to the current path propagation weight and the deviation amplitude, outputs SOC correction factors in multiple dimensions, and finally superimposes the original SOC estimation value to form corrected SOC attenuation data for subsequent digital twin parameter updating and scheduling strategy adjustment. This method forms a complete deviation diagnosis and correction chain from data difference, path identification, causal modeling to compensation generation, improving the state observability and response stability of the energy storage system in a dynamic environment.

[0141] Especially important is step S43:

[0142] Step S431: Return the SOC attenuation data to the energy storage system digital twin model for loss parameter optimization, and obtain the model loss function;

[0143] Step S432: Re-train the PINN grid parameters of the energy storage system digital twin model based on the model loss function, and obtain the model grid parameter data;

[0144] Step S433: feature space reconstruction is performed on the model grid parameter data to obtain an optimized digital twin model of the energy storage system.

[0145] In the embodiment of the present application, by introducing the backhaul mechanism of SOC attenuation data, combining the structural advantages of Physics-Informed Neural Networks (PINN), the dynamic update and structural reconstruction of the digital twin model of the energy storage system at the data level are realized. Specifically, step S431 first compares the error between the SOC attenuation data generated by the bias analysis and error backtracking link and the current output value of the model as the target output, and constructs a composite loss function containing time series residual, non-linear energy loss function and battery aging influence factor. This loss function not only retains the fitting accuracy requirement of the data-driven layer, but also integrates the physical constraint relationship existing in the battery thermal-electric-mechanical coupling mechanism, thereby providing a structural optimization target for subsequent network update. Step S432 re-trains the grid parameters in the PINN model based on the constructed loss function, which not only includes the backpropagation update of the weights and biases in the traditional neural network, but also covers the re-encoding of the physical constraint boundary conditions in the space-time discrete area of the PINN model and the re-optimization of its gradient response function. During the training process, the system uses the SOC attenuation trend curve as the main supervision signal to guide the neural network to update the feature mapping relationship related to battery health in a specific grid area, especially focusing on the error distribution characteristics of the current density gradient, voltage stability distribution and temperature evolution path in the multi-scale grid cells, and completing high-precision weight adjustment and structure fitting with adaptive gradient descent and L-BFGS optimizers. Step S433 further reconstructs the high-dimensional feature space of the updated model grid parameter data, reconstructs the mutual dependence and non-linear correlation between features, and outputs an optimized digital twin model of the energy storage system with current system state awareness. In this reconstruction process, Tensor PCA and Laplace Eigen Mapping are used to extract significant factors in the high-order structure of the model to ensure that the updated twin model has a high prediction generalization ability while maintaining physical consistency, thereby realizing the iterative evolution of the model under the data-physical dual driving condition.

[0146] In the present specification, a power scheduling system based on an energy storage system is provided for performing the above-mentioned power scheduling method based on an energy storage system, which comprises:

[0147] A multi-source heterogeneous perception standardization module is configured to deploy a distributed sensing array inside a battery module to collect multi-source heterogeneous data, perform protocol conversion, and obtain an initial multi-modal data set; perform noise spectrum adjustment filtering threshold on the initial multi-modal data set, and perform long-term drift compensation to generate a standardized energy storage data set;

[0148] The cross-modal feature fusion module is configured to perform tensor completion on the standardized energy storage dataset to generate energy storage tensor completion data, perform time-space dimension analysis on the energy storage tensor completion data, and perform cross-modal feature fusion to generate an energy storage cross-modal feature space.

[0149] The digital twin modeling and game scheduling module is configured to construct a battery health state digital twin based on the energy storage cross-modal feature space to obtain an energy storage system digital twin model, and perform game scheduling strategy generation on the energy storage system digital twin model to obtain an energy storage system electric energy scheduling strategy.

[0150] The real-time deviation analysis and closed-loop optimization module is configured to obtain real-time energy storage monitoring data, perform execution process deviation calculation based on the real-time energy storage monitoring data and the energy storage system electric energy scheduling strategy, generate SOC attenuation data if the energy storage deviation data is greater than a preset threshold, perform network parameter optimization on the SOC attenuation data by returning the SOC attenuation data to the energy storage system digital twin model, perform energy storage system full-process scheduling analysis to obtain full-process scheduling analysis data, and perform scheduling report generation based on the full-process scheduling analysis data to obtain an energy storage system electric energy scheduling whole-process report.

[0151] Therefore, from any viewpoint, the embodiments should be considered as being exemplary and not limiting, the scope of the present application being defined by the appended claims and not by the above description, and it is intended to embrace all variations falling within the meaning and the scope of the equivalent elements of the claims.

[0152] The above description is merely that of a specific implementation of the present application, which enables those skilled in the art to understand or implement the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to these embodiments shown herein, but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for electrical energy dispatching based on energy storage system, characterized in that, The method comprises the following steps: Step S1: deploying a distributed sensing array inside a battery module to collect multi-source heterogeneous data, performing protocol conversion, and obtaining an initial multi-modal data set; performing noise spectrum adjustment filtering threshold on the initial multi-modal data set, and performing long-term drift compensation to generate a standardized energy storage data set; Step S2: performing tensor completion on the standardized energy storage data set to generate energy storage tensor completion data; performing time-space dimension analysis on the energy storage tensor completion data, and performing cross-modal feature fusion to generate an energy storage cross-modal feature space; Step S3: constructing a battery health state digital twin based on the energy storage cross-modal feature space to obtain an energy storage system digital twin model; The energy storage system digital twin model is subjected to game scheduling strategy generation to obtain an energy storage system power scheduling strategy; Step S3 comprises the following steps: Step S31: performing multi-scale energy storage analysis based on the energy storage cross-modal feature space, and constructing a battery health state digital twin to obtain an energy storage system digital twin model; Step S31 comprises the following steps: Step S311: performing Ohm polarization / diffusion polarization loss calculation based on the energy storage cross-modal feature space to generate electrochemical scale analysis data; Step S312: performing thermodynamic analysis for capacity attenuation prediction based on the energy storage cross-modal feature space to generate thermodynamic scale analysis data; Step S313: performing electrode stress crack detection based on the energy storage cross-modal feature space, and performing crack diffusion rate calculation to generate mechanical scale analysis data; Step S314: performing parameter initialization on the electrochemical scale analysis data, the thermodynamic scale analysis data, and the mechanical scale analysis data, and constructing a digital twin to generate an energy storage system digital twin model; Step S32: designing a double-layer optimization framework based on the energy storage system digital twin model to obtain an energy storage double-layer optimization framework, wherein the energy storage double-layer optimization framework comprises an upper-layer consumption reduction framework and a lower-layer energy consumption stability framework; Step S33: generating a game scheduling strategy based on the upper-layer consumption reduction framework and the lower-layer energy consumption stability framework to obtain an energy storage system power scheduling strategy; Step S33 comprises: Generating a game scheduling strategy based on the upper-layer consumption reduction framework and the lower-layer energy consumption stability framework to obtain an energy storage system power scheduling strategy, wherein the energy storage system power scheduling strategy comprises a consumption reduction optimization strategy and a frequency response optimization strategy; Obtaining electricity price data; performing load prediction extraction on the upper-layer consumption reduction framework to obtain energy storage load prediction data; performing 24-hour linear charging and discharging planning based on the load prediction data and the electricity price data to obtain the consumption reduction optimization strategy; Performing real-time power grid frequency extraction on the lower-layer energy consumption stability framework to obtain real-time power grid frequency data; when the real-time power grid frequency data exceeds 0.2 Hz, triggering an energy storage correction instruction for adjustment to obtain the frequency response optimization strategy; Step S4: Obtain real-time energy storage monitoring data; perform process deviation calculation based on the real-time energy storage monitoring data and the energy scheduling strategy of the energy storage system; if the energy storage deviation data is greater than a preset threshold, generate SOC attenuation data; return the SOC attenuation data to the digital twin model of the energy storage system for network parameter optimization, and perform full-process scheduling analysis of the energy storage system to obtain full-process scheduling analysis data; generate a scheduling report based on the full-process scheduling analysis data to obtain a full-process report of the energy scheduling of the energy storage system.

2. The method of claim 1, wherein, Step S1 includes the following steps: Step S11: Deploy voltage sensors and temperature sensors inside the battery module to collect single cell temperature data and terminal voltage data at a frequency of 10Hz, wherein the accuracy of the voltage sensor is 0.1% and the accuracy of the temperature sensor is 0.5℃; Step S12: Deploy a high-frequency current sensor with a bandwidth of 20kHz on the DC side of the energy storage converter to collect charge and discharge current waveforms at a frequency of 1kHz, and generate energy storage charge and discharge waveform data; Step S13: Deploy a synchronous phase measurement unit at the grid access point to collect grid frequency, voltage phase angle and harmonic distortion rate at a frequency of 50Hz, and generate grid phase raw data; Step S14: Convert the single cell temperature data, terminal voltage data, energy storage charge and discharge waveform data and grid phase raw data into initial multi-modal data sets through a multi-protocol adapter; Step S15: Adjust the noise spectrum filtering threshold of the initial multi-modal data set, and perform long-term drift compensation to generate a standardized energy storage data set.

3. The method of claim 1, wherein, The tensor completion of the standardized energy storage data set in step S2 includes: Map the standardized energy storage data set to a three-dimensional space for timestamp sorting to obtain energy storage three-dimensional mapping standardized data; Repair the temperature and current data of the energy storage three-dimensional mapping standardized data using low-rank Tucker decomposition to obtain standard tensor repair data; Construct a three-dimensional feature tensor of device x time x feature from the standard tensor repair data, and perform missing value detection to obtain energy storage tensor completion data.

4. The electrical energy dispatch method based on energy storage system according to claim 1, wherein, The time-space dimension analysis and cross-modal feature fusion of the energy storage tensor completion data in step S2 include: Extract the battery SOC data from the energy storage tensor completion data in the time dimension; Extract the grid frequency fluctuation rate data from the energy storage tensor completion data in the spatial dimension; Align the battery SOC data and the grid frequency fluctuation rate data on the time axis to generate a synchronized energy storage data set; Perform long-term dependency relationship analysis on the battery SOC data to obtain battery dependency relationship analysis data; extract the charge and discharge behavior features of the battery dependency relationship analysis data to obtain a time step feature vector; Perform unit topology impedance calculation on the grid frequency fluctuation rate data to obtain grid topology impedance calculation data; perform charge and discharge path energy loss relationship analysis on the grid topology impedance calculation data, and extract spatial distribution features to obtain a spatial dimension feature matrix; The time step feature vector and the spatial dimension feature matrix are tensor spliced to obtain an energy storage feature splicing matrix; the energy storage feature splicing matrix is synchronized by using a synchronized energy storage dataset to obtain an energy storage scheduling feature fusion matrix; Based on the energy storage scheduling feature fusion matrix, time-space initial weight distribution is performed, and a feature space is constructed to obtain an energy storage cross-modal feature space.

5. The electrical energy dispatch method based on energy storage system according to claim 1, wherein, Step S4 includes the following steps: Step S41: acquiring real-time energy storage monitoring data; Step S42: performing execution process deviation calculation based on the real-time energy storage monitoring data and the energy storage system electric energy scheduling strategy, and if the energy storage deviation data is greater than a preset threshold, generating SOC attenuation data; Step S43: returning the SOC attenuation data to the energy storage system digital twin model for network parameter optimization to generate an optimized energy storage system digital twin model; Step S44: performing energy storage system full-process scheduling analysis based on the optimized energy storage system digital twin model, and generating a report to obtain an energy storage system electric energy scheduling full-process report.

6. The electrical energy dispatch method based on energy storage system according to claim 1, wherein, Step S42 includes the following steps: Step S421: performing deviation value calculation based on the real-time energy storage monitoring data and the energy storage system electric energy scheduling strategy to generate energy storage deviation value data; Step S422: performing SOC deviation calculation based on the real-time energy storage monitoring data and the energy storage system electric energy scheduling strategy to obtain energy storage SOC deviation data; Step S423: if the energy storage deviation value data is greater than 5% and the energy storage SOC deviation data is greater than 2%, a compensation coefficient is generated for error propagation path backtracking, and SOC attenuation data is obtained.

7. An electrical energy dispatch system based on an energy storage system, characterized by, A system for performing the energy storage system-based electric energy scheduling method of claim 1, the energy storage system-based electric energy scheduling system comprising: A multi-source heterogeneous perception standardization module for deploying a distributed sensing array inside a battery module to collect multi-source heterogeneous data, performing protocol conversion, and obtaining an initial multi-modal dataset; performing noise spectrum adjustment filtering thresholding, and long-term drift compensation on the initial multi-modal dataset to generate a standardized energy storage dataset; A cross-modal feature fusion module for performing tensor completion on the standardized energy storage dataset to generate energy storage tensor completion data; performing time-space dimension analysis on the energy storage tensor completion data, and performing cross-modal feature fusion to generate an energy storage cross-modal feature space; A digital twin modeling and game scheduling module for constructing a battery health state digital twin based on the energy storage cross-modal feature space to obtain an energy storage system digital twin model; performing game scheduling strategy generation on the energy storage system digital twin model to obtain an energy storage system electric energy scheduling strategy; A real-time deviation analysis and closed-loop optimization module for acquiring real-time energy storage monitoring data; performing execution process deviation calculation based on the real-time energy storage monitoring data and the energy storage system electric energy scheduling strategy, and if the energy storage deviation data is greater than a preset threshold, generating SOC attenuation data; returning the SOC attenuation data to the energy storage system digital twin model for network parameter optimization, and performing energy storage system full-process scheduling analysis to obtain full-process scheduling analysis data; performing scheduling report generation based on the full-process scheduling analysis data to obtain an energy storage system electric energy scheduling full-process report.

Images