Industrial and commercial energy storage system energy resource scheduling method and system

Abstract

The invention relates to the technical field of power systems and energy management, discloses an industrial and commercial energy storage system energy resource scheduling method and system, and aims to solve the problems that in the prior art, economy optimization and battery health management are separated, degradation model simplification causes life damage, and the adaptability to load and electricity price uncertainty is insufficient. Specifically, multi-source heterogeneous state data are fused; constructing and calibrating a battery digital twinborn model coupled with electrochemistry, a thermal field and an aging mechanism; and executing a three-layer cooperative strategy of long-term degradation track planning, medium-term random scheduling optimization and short-term model prediction control, enabling the battery degradation cost to be generated in an economic scheduling target, and achieving high-robustness real-time control through rolling optimization. The system comprises a data acquisition and fusion unit, a digital twinning module, a probability prediction module, a hierarchical scheduling engine and an instruction execution unit. According to the invention, the unification of full-life-cycle value maximization and operation reliability is realized.

Description

Technical Field

[0001] This invention belongs to the field of power system and energy management technology, specifically relating to an energy resource scheduling method and system for industrial and commercial energy storage systems. Background Technology

[0002] With the advancement of the "dual-carbon" strategy and the deepening of power market reform, commercial and industrial energy storage systems, as a key component of virtual power plants (VPPs), are playing an increasingly important role in peak shaving, demand response, and grid frequency regulation. Currently, energy storage systems generally adopt a centralized dispatch architecture, with a central control unit directly issuing power commands to various energy storage devices. However, the actual deployed commercial and industrial energy storage resources exhibit high heterogeneity, encompassing multiple technologies such as lithium batteries, lead-acid batteries, and supercapacitors, which differ significantly in response speed, energy density, cycle life, and dynamic characteristics. This physical diversity makes it difficult for traditional dispatch methods to achieve unified modeling and coordinated control, often requiring customized communication protocols and control logic for each type of device, greatly increasing system integration complexity and operation and maintenance costs.

[0003] The core challenge of energy resource scheduling lies in achieving efficient coordination of multiple types of energy storage units without relying on the details of underlying equipment. Existing scheduling strategies typically rely on fixed-weight allocation or simple capacity ratios for power decomposition, neglecting the differences in dynamic response capabilities of different energy storage media over time. For example, lithium batteries have millisecond-level response capabilities, making them suitable for high-frequency power regulation; lead-acid batteries, while having large capacity, have a slow response and are more suitable for steady-state loads; while supercapacitors can complete instantaneous power throughput in microseconds, used to suppress short-term fluctuations.

[0004] If scheduling instructions do not take into account such dynamic characteristics, it is easy to cause resource mismatch. Fast-response resources are inefficiently used to support basic loads, and slow-response equipment is forced to participate in high-frequency regulation. This not only reduces the overall system efficiency, but may also accelerate equipment aging and even cause safety risks.

[0005] Therefore, there is an urgent need for an energy resource scheduling method and system for energy storage systems that can deeply integrate the dynamic characteristics of equipment, support real-time adaptive mapping, and have strong fault tolerance. Summary of the Invention

[0006] The purpose of this invention is to provide an energy resource scheduling method and system for industrial and commercial energy storage systems, in order to solve the technical problems in the prior art where energy storage scheduling models separate economic optimization from battery health management, adopt simplified battery degradation models that lead to scheduling strategies that damage long-term lifespan, and have insufficient adaptability to uncertainties in load, electricity price and new energy output, resulting in a significant deviation between actual operating benefits and theoretical optimal values.

[0007] To achieve the above objectives, according to one aspect of the present invention, a method for energy resource scheduling in an industrial and commercial energy storage system is provided, comprising the following steps: Real-time data collection and fusion are used to form a multi-source heterogeneous state dataset, which includes: energy storage system operation data, grid interaction data, renewable energy generation data, and user-side load data. The energy storage system operation data includes the voltage, current, surface temperature, internal temperature, and state of charge and health estimated by the battery management system for each battery cell in the energy storage unit cluster. The grid interaction data includes electricity metering data at the grid connection point, electricity price signals, and grid frequency. The renewable energy generation data includes the output power of the photovoltaic array and environmental irradiance. The user-side load data includes the total electricity load power curve. A battery multiphysics digital twin model mapped to a physical energy storage system is constructed and updated in real time. The battery multiphysics digital twin model is coupled with an electrochemical mechanism model, a three-dimensional thermal field distribution model, and a multi-mechanism coupled battery aging model. Through a nonlinear state observer, using the real-time collected energy storage system operation data, the internal state variables of the battery multiphysics digital twin model are calibrated and estimated online, and high-precision battery internal electrochemical state, temperature field distribution state, and health state are output. Based on historical multi-source heterogeneous state datasets, a probabilistic joint prediction model is driven to generate probability distribution prediction data of user-side load, renewable energy power generation, and time-of-use electricity price within a future scheduling cycle. The probability distribution prediction data is represented by a series of prediction trajectories under quantiles. A hierarchical collaborative energy resource scheduling optimization strategy is implemented, which includes a long-term degradation trajectory planning layer, a medium-term stochastic scheduling optimization layer, and a short-term model prediction control layer. The long-term degradation trajectory planning layer is based on the full life cycle asset value model of the energy storage system and sets an upper limit for the battery health state degradation rate over a long period, such as one month, as a hard constraint on the medium-term stochastic scheduling optimization layer. The mid-term stochastic scheduling optimization layer receives the upper limit of the health state decay rate set by the long-term degradation trajectory planning layer, and constructs a stochastic optimization problem with the objective of maximizing the expected value of comprehensive economic benefits within the scheduling cycle based on the probability distribution prediction data generated by the probabilistic joint prediction model. The comprehensive economic benefits are the difference between the grid interaction benefits and the battery degradation costs. The battery degradation costs are obtained by performing forward simulations of the alternative scheduling strategies using the battery multiphysics digital twin model. Solving the stochastic optimization problem generates a reference trajectory for the baseline charge and discharge power for the next scheduling cycle. The short-term model prediction control layer takes the reference charge-discharge power trajectory as the target and constructs a deterministic optimization problem within a rolling updated short-timescale prediction domain. This problem is based on the real-time acquired multi-source heterogeneous state dataset and the updated battery multi-physics digital twin model state, aiming to minimize the tracking error of the reference trajectory and the penalty for real-time power imbalance. The deterministic optimization problem is solved online and iteratively to generate the final energy storage system charge-discharge power command sequence with real-time feedback correction capability. The final energy storage system charge and discharge power command sequence is sent to the energy management system and converter of the energy storage system to control the energy storage system to perform charge and discharge operations.

[0008] Furthermore, the construction and real-time updating of the battery multiphysics digital twin model specifically includes: The electrochemical mechanism model adopts a single-particle model. By solving the solid-phase lithium-ion diffusion equation and the liquid-phase charge conservation equation, the terminal voltage, electrode potential and solid-liquid phase lithium-ion concentration distribution of the battery under a specific charge and discharge current are calculated. The three-dimensional thermal field distribution model is constructed using the finite element method, which meshes the battery cell, module structure, and heat dissipation channels. Its heat source term comes from the irreversible Joule heat and reversible entropy change heat calculated by the electrochemical mechanism model. The three-dimensional thermal field distribution model solves for the three-dimensional temperature distribution inside and on the surface of the battery. The multi-mechanism coupled battery aging model includes a solid electrolyte interface film growth model and a lithium plating effect model; the solid electrolyte interface film growth rate is determined by the negative potential of the electrode surface, local temperature, and lithium ion concentration; the occurrence and evolution of the lithium plating effect are jointly triggered by the negative electrode overpotential and local low temperature conditions; the output of the model is the battery capacity decay and internal resistance increase. The nonlinear state observer employs an unscented Kalman filter algorithm, using the battery multiphysics digital twin model as the state equation and the real-time collected battery cell terminal voltage and surface temperature as the observation equation. Through iterative prediction and update steps, it achieves closed-loop estimation of state variables that cannot be directly measured within the model, such as solid-phase lithium ion concentration, internal temperature gradient, and health state parameters.

[0009] Furthermore, driving a probabilistic joint prediction model specifically includes: Construct a sequence-to-sequence neural network model based on an attention mechanism; Historical user-side load power, renewable energy generation power, time-of-use electricity price series, and related exogenous variables, such as time codes and weather forecast data, are used as inputs to the neural network model. The output layer of the neural network model is designed to directly output multiple quantile values, which together constitute a discretized approximation of the joint probability density function of future load, power generation, and electricity price, thereby quantifying the uncertainty of the prediction.

[0010] Furthermore, the solution of the intermediate stochastic scheduling optimization layer specifically includes: Using a scenario generation method, a set of representative typical scenarios are generated based on the probability distribution prediction data. Each scenario corresponds to a specific load, power generation, and electricity price realization path and its probability of occurrence. The original stochastic optimization problem is transformed into a large-scale deterministic multi-scenario optimization problem, whose objective function is the weighted average of the comprehensive economic benefits under all scenarios. The battery degradation cost is calculated as follows: for any alternative scheduling strategy, it is used as input to drive the battery multiphysics digital twin model to perform full-cycle simulation, obtain the change in battery health status at the end of the simulation, and then monetize the change in health status into a cost value through a preset battery asset depreciation function. The deterministic multi-scenario optimization problem is solved using a column and constraint generation algorithm or other large-scale optimization decomposition algorithms.

[0011] Furthermore, the solution of the short-term model prediction control layer specifically includes: Define a prediction time domain and a control time domain, wherein the length of the prediction time domain is fifteen minutes to two hours, and the length of the control time domain is five minutes; At the beginning of each control step, the latest system state measurement is acquired and used to update the initial state of the battery multiphysics digital twin model; Within the prediction time domain, dynamic models of energy storage systems, loads, and renewable energy are established based on high-resolution deterministic predictions. Solving the deterministic optimization problem yields the optimal control sequence in the prediction time domain. Only the first control instruction in the optimal control sequence, i.e. the charging and discharging power instruction corresponding to the current control time domain, is issued and executed as the final instruction. In the next control step, the above process is repeated to form a continuous rolling optimization and feedback correction closed loop.

[0012] According to another aspect of the present invention, an energy resource dispatching system for industrial and commercial energy storage systems is provided, comprising: A multi-source data acquisition and fusion unit is configured to acquire energy storage system operation data, grid interaction data, renewable energy power generation data and user-side load data in real time through industrial bus interface, wireless sensor network and application programming interface, and perform timestamp alignment, cleaning and structured storage on the acquired data to form a unified multi-source heterogeneous state dataset. A battery multiphysics digital twin module integrates an electrochemical mechanism model, a three-dimensional thermal field distribution model, and a multi-mechanism coupled battery aging model, and includes a nonlinear state observer. The battery multiphysics digital twin module is connected to the multi-source data acquisition and fusion unit to receive real-time operating data to drive the nonlinear state observer, perform online calibration and estimation of the model's internal state, and provide high-precision battery state information and degradation cost simulation functions. A probabilistic joint prediction module is configured to run an attention-based sequence-to-sequence neural network model based on a historical multi-source heterogeneous state dataset to generate probability distribution prediction data of user-side load, renewable energy power generation, and time-of-use electricity price for a future scheduling cycle. A hierarchical collaborative scheduling optimization engine is configured to execute a hierarchical collaborative energy resource scheduling optimization strategy. The engine internally includes a long-term degradation trajectory planner, a medium-term stochastic scheduling optimizer, and a short-term model prediction controller. The long-term degradation trajectory planner is used to set an upper limit for the battery health state decay rate over a long period. The medium-term stochastic scheduling optimizer is connected to the probabilistic joint prediction module and the battery multiphysics digital twin module to solve stochastic optimization problems and generate a reference charge / discharge power trajectory. The short-term model prediction controller is connected to the multi-source data acquisition and fusion unit and the battery multiphysics digital twin module to perform rolling optimization and generate the final energy storage system charge / discharge power command sequence. A scheduling instruction execution unit, connected to the hierarchical collaborative scheduling optimization engine, is used to convert the final energy storage system charging and discharging power instruction sequence into a message conforming to a specific communication protocol, and send it to the energy management system and converter of the energy storage system through a physical communication interface.

[0013] Furthermore, the battery multiphysics digital twin module is deployed on an edge computing device, which includes a graphics processing unit for parallel computing and a multi-core central processing unit for numerical computing to meet the high real-time solution requirements of the model.

[0014] Furthermore, the three internal components of the hierarchical collaborative scheduling optimization engine—the long-term degradation trajectory planner, the mid-term stochastic scheduling optimizer, and the short-term model prediction controller—have progressively decreasing decision cycles, with decision information being passed down level by level. The decision cycle of the long-term degradation trajectory planner is weekly or monthly, and its output decay rate upper limit is passed to the mid-term stochastic scheduling optimizer as a constraint. The decision cycle of the mid-term stochastic scheduling optimizer is daily, and its output baseline power trajectory is passed to the short-term model prediction controller as a reference. The decision cycle of the short-term model prediction controller is minute-level, forming a separation of time scales and coupling of information at the decision-making level.

[0015] In summary, this application includes at least one of the following beneficial technical effects: (1): By constructing a digital twin model that couples electrochemical, thermal and multi-physical degradation mechanisms, this invention can accurately quantify the profound impact of different scheduling strategies on the internal physicochemical state of the battery, endogenously integrate the health degradation cost of the battery into the economic scheduling model, realize the unified optimization of short-term economic benefits and the full life cycle value of the energy storage system, and fundamentally avoid the decision trap of "optimal economy, worst lifespan" caused by the use of simplified degradation models in traditional scheduling methods.

[0016] (2): This invention uses a probabilistic joint prediction model to replace the traditional deterministic point prediction, which can accurately capture and quantify the uncertainty of future load, new energy output and market electricity price, and seamlessly integrate them into the medium-term stochastic scheduling optimization framework, so that the generated scheduling strategy has inherent robustness, can effectively resist the impact of prediction error, and significantly improve the actual economic benefits and operational reliability of energy storage system in real and variable operating environments.

[0017] (3): The long-term, medium-term and short-term three-layer collaborative scheduling architecture designed in this invention decomposes the macro asset health management goal, namely the long-term degradation trajectory, into medium-term day-ahead economic scheduling planning and short-term minute-level real-time feedback control. This achieves logical consistency of decision-making goals and smooth transmission of information flow under different time scales, overcomes the inherent short-sightedness and decision fragmentation problems of existing single-scale optimization methods, and forms a globally collaborative and dynamically adaptive closed-loop optimization control system. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the overall technical architecture of an energy resource scheduling method and system for industrial and commercial energy storage systems proposed in this invention; Figure 2 This is a schematic diagram illustrating the data-driven and decision-making interaction relationship between the probabilistic joint prediction model and the mid-term stochastic scheduling optimization layer in this invention; Figure 3This is a schematic diagram of the multi-level interaction relationship and data flow of multi-source heterogeneous state data acquisition, fusion and scheduling instruction execution in this invention. Detailed Implementation

[0019] This invention provides an energy resource scheduling method and system for industrial and commercial energy storage systems, aiming to solve core problems in existing technologies such as the disconnect between economic optimization and battery health management, overly simplistic degradation models, and insufficient adaptability to load and electricity price uncertainties. The complete execution flow of this application will be described in detail below.

[0020] The energy resource scheduling method for industrial and commercial energy storage systems begins with step S1: real-time acquisition and fusion to form a multi-source heterogeneous state dataset. This dataset consists of four types of data: energy storage system operation data, grid interaction data, renewable energy generation data, and user-side load data.

[0021] The energy storage system's operating data comes from a high-precision sensor array deployed on each battery cell, including internal temperature data measured by voltage sensors, current sensors, surface temperature sensors, and embedded thermocouples; at the same time, the battery management system periodically reports its estimated state of charge and state of health values.

[0022] Grid interaction data is obtained through smart meters at the grid connection point and the electricity market information interface, including active power, reactive power, cumulative energy metering values, real-time time-of-use pricing signals, and grid frequency fluctuation data. Renewable energy generation data is provided by the photovoltaic inverter monitoring unit, which records the DC-side output power of the photovoltaic array and simultaneously collects irradiance, ambient temperature, and wind speed data provided by the environmental monitoring station.

[0023] User-side load data is collected through multi-functional power meters in the plant's main distribution cabinet, forming a total power load curve with granularity at the second or minute level. All of the above data flows through industrial Ethernet or wireless sensor networks to the central data processing node, where timestamp alignment, outlier removal, missing value imputation, and format standardization are performed. Finally, a unified, time-consistent, and dimensionally complete multi-source heterogeneous state dataset is constructed, serving as the input basis for all subsequent decision-making modules.

[0024] Next, step S2 is executed: a multi-physics digital twin model of the battery mapped to the physical energy storage system is constructed and updated in real time. This model is not a static simulation tool, but a dynamically evolving closed-loop system, whose core consists of three coupled sub-models. The first sub-model is an electrochemical mechanism model, which adopts a single-particle model architecture. By solving the partial differential equation of lithium-ion diffusion in the solid phase and the charge conservation equation in the liquid phase, it calculates the spatial-temporal evolution trajectory of the battery terminal voltage, positive and negative electrode potential distribution, and lithium-ion concentration in the solid and liquid phases under a given charge and discharge current excitation.

[0025] The second sub-model is a three-dimensional thermal field distribution model. Based on the finite element method, the geometry of the battery cell is meshed, and the module shell, thermal pads, and air-cooled / liquid-cooled channels are included in the modeling scope. Its heat source term is composed of the irreversible Joule heat power density and the reversible entropy change heat power density output by the electrochemical mechanism model. This model solves the transient temperature field in the three-dimensional space inside the battery and outputs the temperature value and its gradient at any location.

[0026] The third sub-model is a multi-mechanism coupled battery aging model, including a solid electrolyte interface film growth sub-model and a lithium plating effect sub-model. The growth rate of the solid electrolyte interface film is determined by the local potential, temperature, and lithium-ion concentration of the negative electrode surface, following an Arrhenius-type kinetic equation. The lithium plating effect is triggered by the dual conditions of the negative electrode overpotential being lower than the lithium plating potential threshold and the local temperature being lower than the critical value. The amount of lithium metal deposited is calculated by integrating Faraday's law. The output of this aging model is the percentage of battery capacity decay and the increase in internal resistance, which directly reflects the degree of deterioration in the battery's health status.

[0027] To ensure synchronization between the digital twin model and the physical entity, a nonlinear state observer is introduced, implemented using an unscented Kalman filter algorithm. This observer uses the digital twin model as the state transition equation and real-time acquired cell terminal voltages and surface temperatures as observation vectors. It generates a Sigma point set through unscented transformation and sequentially executes four sub-steps: state prediction, observation prediction, covariance calculation, and state update. This iteratively estimates the unmeasurable state variables within the model, including the average solid-phase lithium concentration, maximum internal temperature difference, interface film thickness, and health parameters. After each observation update, the model's initial conditions are reset, thus achieving online calibration and closed-loop tracking of the digital twin.

[0028] Then proceed to step S3: Based on the historical multi-source heterogeneous state dataset, drive a probabilistic joint prediction model to generate probability distribution prediction data for user-side load, renewable energy power generation, and time-of-use electricity price within a future scheduling cycle.

[0029] The prediction model employs a sequence-to-sequence neural network architecture based on an attention mechanism. The input sequence consists of load power, photovoltaic output, and electricity price values ​​over the past 24 hours, with additional exogenous variable encoding, including hour number, weekday type, holiday markers, and cloud cover index and temperature forecast from the 24-hour weather forecast.

[0030] The encoder consists of multiple layers of gated recurrent units stacked together, extracting contextual features of the input sequence step by step. The decoder also employs gated recurrent units and introduces an attention mechanism to dynamically weight the hidden states of the encoder at each time step, focusing on the past information that has the greatest influence on the current prediction time. The key innovation lies in the output layer design: instead of outputting a single predicted value, it directly outputs predicted values ​​corresponding to multiple preset quantiles (such as 10%, 30%, 50%, 70%, and 90%), with each quantile corresponding to a future trajectory.

[0031] Therefore, the joint prediction result of the three variables (load, generation, and electricity price) manifests as a set of discrete quantile trajectories, which together approximate its joint probability density function, thus explicitly quantifying the prediction uncertainty. The model is incrementally trained using the latest historical data at midnight each day to ensure its adaptability to seasonal changes and shifts in user behavior.

[0032] The core scheduling logic is embodied in step S4: executing a hierarchical and collaborative energy resource scheduling optimization strategy. This strategy comprises three levels, from top to bottom: a long-term degradation trajectory planning layer, a medium-term stochastic scheduling optimization layer, and a short-term model prediction and control layer. These three levels form a collaborative architecture characterized by time scale separation, hierarchical decomposition of objectives, and hierarchical propagation of constraints.

[0033] The long-term degradation trajectory planning layer runs first, with a decision-making cycle of one month. This layer is based on the full life-cycle asset value model of the energy storage system, which comprehensively considers the impact function of initial investment cost, operation and maintenance costs, residual value recovery rate, and battery health status on residual value.

[0034] By solving a long-term optimization problem, the maximum allowable rate of battery health degradation over the next month is determined, for example, by setting it to no more than 0.5% capacity degradation per month. This upper limit of the degradation rate is transformed into a hard constraint and passed to the intermediate stochastic scheduling optimization layer as a boundary condition that all its feasible scheduling strategies must satisfy.

[0035] The medium-term stochastic scheduling optimization layer runs before the start of each day, with a decision cycle of one day. This layer receives the upper limit of the health state decay rate set by the long-term layer and incorporates the probability distribution prediction data generated in step S3. Its goal is to construct and solve a stochastic optimization problem to maximize the expected value of the comprehensive economic benefits over the next 24 hours. The comprehensive economic benefits are defined as the algebraic sum of grid interaction benefits and battery degradation costs.

[0036] The grid interaction revenue is calculated based on time-of-use pricing and the net purchase and sale of electricity by the energy storage system; the battery degradation cost is obtained by proactively calling the battery multiphysics digital twin model in step S2. Specifically, for any candidate charge-discharge power sequence, it is used as the current excitation input to the digital twin model, driving the model to complete a full-cycle simulation of the next 24 hours, recording the change in health status ΔSOH at the end of the simulation; and then, a preset asset depreciation function is used... Specifically:

[0037] Where α and β are preset cost coefficients, ΔSOH is monetized into specific cost values.

[0038] To handle randomness, a scenario generation method is employed, extracting one hundred typical scenarios from the probability distribution prediction data. Each scenario contains a complete path for load, power generation, and electricity price realization, along with their corresponding probability weights. The original stochastic optimization problem is thus transformed into a large-scale deterministic multi-scenario optimization problem, with the objective function being the weighted average of the comprehensive economic benefits across all scenarios.

[0039] This problem is solved using a column and constraint generation algorithm. Through iteration of the main problem and subproblems, worst-case scenarios are gradually added until convergence. The final output is a reference trajectory of the baseline charge and discharge power for the next 24 hours. This trajectory achieves Pareto optimality in terms of economy and lifetime while satisfying long-term health constraints.

[0040] The short-term model predictive control layer executes on a rolling basis at a frequency of minutes throughout the operating day. Its prediction time domain is set to one hour, and the control time domain is five minutes. At the beginning of each control step, the system acquires the latest multi-source heterogeneous state dataset and uses it to update the initial state of the battery multiphysics digital twin model, ensuring that the model reflects the current real physical state.

[0041] Based on this, a dynamic model of the energy storage system (a simplified proxy model based on a digital twin) and a high-resolution point prediction model of load and photovoltaic (taking the median trajectory of probabilistic prediction) are established in the prediction time domain. A deterministic optimization problem is constructed, with the objective function being the sum of two terms: the first term is the integral of the squared deviation between the actual charging and discharging power in the prediction time domain and the baseline trajectory issued by the intermediate layer; the second term is a quadratic penalty for real-time power imbalance (i.e., the absolute value of the sum of net load and photovoltaic power and energy storage regulation power).

[0042] This optimization problem is constrained by the physical constraints of the energy storage system (power limits, upper and lower limits of state of charge) and grid security constraints. An interior-point method is used to solve this quadratic programming problem online, obtaining the optimal control sequence in the prediction time domain. Only the first control command in this sequence, i.e., the charging and discharging power setpoint for the next five minutes, is issued and executed as the final command. The above process is repeated for each subsequent control step, forming a continuous rolling optimization and feedback correction closed loop, effectively suppressing the influence of prediction errors and external disturbances.

[0043] Finally, step S5 is executed: the final energy storage system charge / discharge power command sequence is sent to the energy management system and converter of the energy storage system. This command is first received by the dispatch command execution unit, which is responsible for protocol conversion, encapsulating the internal power command into a communication message conforming to the IEC 61850 or Modbus TCP standard, and sending it to the field energy management system via fiber optic or industrial Ethernet.

[0044] After parsing the commands, the energy management system generates the corresponding active power setpoint and transmits it to the energy storage converter via the CAN bus or RS-485 interface. The converter then adjusts the pulse width modulation signal of its IGBT switches accordingly to precisely control the magnitude and direction of the current flowing into or out of the battery cluster, thereby performing charging or discharging operations. The entire command issuance and execution process has a millisecond-level delay, ensuring real-time control and accuracy.

[0045] Additionally, the energy resource dispatching system for industrial and commercial energy storage systems disclosed in this application specifically includes: A multi-source data acquisition and fusion unit is configured to acquire energy storage system operation data, grid interaction data, renewable energy power generation data and user-side load data in real time through industrial bus interface, wireless sensor network and application programming interface, and perform timestamp alignment, cleaning and structured storage on the acquired data to form a unified multi-source heterogeneous state dataset. A battery multiphysics digital twin module integrates an electrochemical mechanism model, a three-dimensional thermal field distribution model, and a multi-mechanism coupled battery aging model, and includes a nonlinear state observer. The battery multiphysics digital twin module is connected to a multi-source data acquisition and fusion unit to receive real-time running data to drive the nonlinear state observer, perform online calibration and estimation of the model's internal state, and provide high-precision battery state information and degradation cost simulation functions. A probabilistic joint prediction module is configured to run an attention-based sequence-to-sequence neural network model based on a historical multi-source heterogeneous state dataset to generate probability distribution prediction data of user-side load, renewable energy power generation, and time-of-use electricity price for a future scheduling cycle. A hierarchical collaborative scheduling optimization engine is configured to execute a hierarchical collaborative energy resource scheduling optimization strategy. The engine internally includes a long-term degradation trajectory planner, a medium-term stochastic scheduling optimizer, and a short-term model prediction controller. The long-term degradation trajectory planner is used to set the upper limit of the battery health state decay rate over a long period. The medium-term stochastic scheduling optimizer is connected to a probabilistic joint prediction module and a battery multi-physics digital twin module to solve stochastic optimization problems and generate a baseline charge / discharge power reference trajectory. The short-term model prediction controller is connected to a multi-source data acquisition and fusion unit and a battery multi-physics digital twin module to perform rolling optimization and generate the final energy storage system charge / discharge power command sequence. A scheduling instruction execution unit, connected to a hierarchical collaborative scheduling optimization engine, is used to convert the final energy storage system charging and discharging power instruction sequence into messages conforming to a specific communication protocol, and send them to the energy management system and converter of the energy storage system through a physical communication interface.

[0046] Furthermore, the battery multiphysics digital twin module is deployed on an edge computing device, which includes a graphics processing unit for parallel computing and a multi-core central processing unit for numerical computation to meet the high real-time solution requirements of the model.

[0047] Furthermore, the three internal components of the hierarchical collaborative scheduling optimization engine—the long-term degradation trajectory planner, the mid-term stochastic scheduling optimizer, and the short-term model prediction controller—have progressively decreasing decision cycles and pass-through decision information at each level. The decision cycle of the long-term degradation trajectory planner is weekly or monthly, and its output decay rate upper limit is passed to the mid-term stochastic scheduling optimizer as a constraint. The decision cycle of the mid-term stochastic scheduling optimizer is daily, and its output baseline power trajectory is passed to the short-term model prediction controller as a reference. The decision cycle of the short-term model prediction controller is minute-level, forming a separation of time scales and coupling of information at the decision level.

[0048] In summary, this embodiment achieves globally optimal scheduling of economic efficiency and lifespan for industrial and commercial energy storage systems under complex and uncertain environments by deeply integrating high-fidelity digital twins, probabilistic prediction, and three-layer collaborative optimization.

[0049] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention. Therefore, the embodiments should be regarded as exemplary and non-limiting in all respects.

[0050] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A method for energy resource scheduling of commercial and industrial energy storage systems, characterized in that, The application comprises: real-time acquisition and fusion of multi-source heterogeneous state data sets; construction and real-time updating of a battery multi-physical field digital twin model, which is coupled with an electrochemical mechanism model, a three-dimensional thermal field distribution model, and a multi-mechanism coupled battery aging model, and uses real-time data for online calibration and state estimation through a nonlinear state observer; based on the multi-source heterogeneous state data set, a probabilistic joint prediction model is driven to generate probabilistic distribution prediction data of load, renewable energy power generation, and time-of-use electricity price in the future scheduling period; a hierarchical collaborative energy resource scheduling optimization strategy is executed, including: a long-term degradation trajectory planning layer that sets an upper limit for the battery health state decay rate; a medium-term stochastic scheduling optimization layer that, based on the decay rate upper limit and the probabilistic distribution prediction data, constructs a stochastic optimization problem with the goal of maximizing the expected value of comprehensive economic benefits to generate a benchmark charge-discharge power reference trajectory, where the comprehensive economic benefits are the difference between grid interaction benefits and battery degradation costs, and the battery degradation costs are obtained through forward simulation of the digital twin model; a short-term model predictive control layer that, with the benchmark trajectory as the target, generates a final charge-discharge power instruction sequence based on real-time data and an updated digital twin model; the final charge-discharge power instruction sequence is sent to the energy storage system control to perform charge-discharge operations.

2. The method of claim 1, wherein, The construction and real-time updating of the battery multi-physical field digital twin model specifically includes: the electrochemical mechanism model uses a single particle model to calculate the terminal voltage, electrode potential, and solid-liquid lithium ion concentration distribution of the battery under a specific charge-discharge current by solving the solid-phase lithium ion diffusion equation and the liquid-phase charge conservation equation; the three-dimensional thermal field distribution model uses the finite element method to construct a grid processing of the battery monomer, module structure, and heat dissipation channel, with the heat source term coming from the irreversible Joule heat and reversible entropy change heat calculated by the electrochemical mechanism model; the three-dimensional thermal field distribution model solves the three-dimensional temperature distribution inside and on the surface of the battery; the multi-mechanism coupled battery aging model includes a solid-state electrolyte interface film growth model and a lithium precipitation effect model; the growth rate of the solid-state electrolyte interface film is determined by the negative potential on the electrode surface, the local temperature, and the lithium ion concentration; the occurrence and evolution of the lithium precipitation effect are triggered by the negative electrode overpotential and the local low temperature condition; the output of the model is the battery capacity attenuation and internal resistance increase; the nonlinear state observer uses the unscented Kalman filter algorithm, takes the battery multi-physical field digital twin model as the state equation, and takes the real-time acquired battery monomer terminal voltage and surface temperature as the observation equation, to realize the closed-loop estimation of the internal state variables of the model that cannot be directly measured through iterative prediction and update steps.

3. The method of claim 1, wherein, The driving of the probabilistic joint prediction model specifically includes: constructing a sequence-to-sequence neural network model based on an attention mechanism; using historical user-side load power, renewable energy power generation, time-of-use electricity price sequences, and related exogenous variables such as time encoding and weather forecast data as inputs of the neural network model; The output layer of the neural network model is designed to directly output a plurality of quantile values which collectively constitute a discretized approximation of a joint probability density function of future load, generation and price.

4. The method of claim 1, wherein, The solution of the mid-term stochastic scheduling optimization layer specifically includes: A scenario generation method is adopted to generate a set of representative typical scenarios according to the probability distribution prediction data, each scenario corresponding to a specific load, generation and price realization path and its occurrence probability; The original stochastic optimization problem is converted into a large-scale deterministic multi-scenario optimization problem, and the objective function is the weighted average of the comprehensive economic benefits in all scenarios; The battery degradation cost is calculated as follows: for any candidate scheduling strategy, it is used as input to drive the battery multi-physical field digital twin model for full-cycle simulation to obtain the change in battery health state at the end of the simulation, and then the battery asset depreciation function is used to monetize the health state change into a cost value; The column and constraint generation algorithm is used to solve the deterministic multi-scenario optimization problem.

5. The method of claim 4, wherein, The solution of the short-term model predictive control layer specifically includes: A prediction horizon and a control horizon are set, the length of the prediction horizon is fifteen minutes to two hours, and the length of the control horizon is five minutes; At the beginning of each control step, the latest system state measurement value is obtained, and the initial state of the battery multi-physical field digital twin model is updated; Within the prediction horizon, based on high-resolution deterministic prediction, dynamic models of the energy storage system, load and renewable energy are established; The deterministic optimization problem is solved to obtain the optimal control sequence within the prediction horizon, and only the first control instruction in the optimal control sequence is issued as the final instruction for execution; In the next control step, the above process is repeated to form a continuous rolling optimization and feedback correction closed loop.

6. The method of claim 1, wherein, The nonlinear state observer generates a Sigma point set through unscented transformation, and sequentially performs state prediction, observation prediction, covariance calculation and state update four sub-steps to estimate the solid-state average lithium concentration, internal maximum temperature difference, interfacial film thickness and health state parameters.

7. The method of claim 1, wherein, In the multi-mechanism coupled battery aging model, the growth rate of the solid-state electrolyte interfacial film follows the Arrhenius-type kinetic equation, and the precipitated lithium metal amount of the lithium precipitation effect is calculated by Faraday's law integration.

8. A commercial and industrial energy storage system energy resource scheduling system, characterized by, It includes: A multi-source data acquisition and fusion unit configured to acquire energy storage system operation data, grid interaction data, renewable energy generation data and user-side load data in real time through industrial bus interfaces, wireless sensor networks and application program interfaces, and to perform timestamp alignment, cleaning and structured storage on the collected data to form a unified multi-source heterogeneous state data set; A battery multi-physical field digital twin module, which integrates an electrochemical mechanism model, a three-dimensional thermal field distribution model, and a multi-mechanism coupled battery aging model internally, and contains a nonlinear state observer; the battery multi-physical field digital twin module is connected with the multi-source data acquisition and fusion unit, for receiving real-time operation data to drive the nonlinear state observer, performing online calibration and estimation on the internal state of the model, and providing high-precision battery state information and degradation cost simulation functions externally; A probabilistic joint prediction module configured to run an attention mechanism-based sequence-to-sequence neural network model based on a historical multi-source heterogeneous state data set to generate probabilistic distribution prediction data of user-side load, renewable energy power generation, and time-of-use electricity price in a future scheduling period; A hierarchical collaborative scheduling optimization engine configured to execute a hierarchical collaborative energy resource scheduling optimization strategy; the hierarchical collaborative scheduling optimization engine internally contains a long-term degradation trajectory planner, a medium-term stochastic scheduling optimizer, and a short-term model predictive controller; the long-term degradation trajectory planner is used to set the upper limit of the long-cycle battery health state decay rate; the medium-term stochastic scheduling optimizer is connected with the probabilistic joint prediction module and the battery multi-physical field digital twin module, for solving a stochastic optimization problem to generate a benchmark charge-discharge power reference trajectory; the short-term model predictive controller is connected with the multi-source data acquisition and fusion unit and the battery multi-physical field digital twin module, for performing rolling optimization to generate a final energy storage system charge-discharge power instruction sequence; A scheduling instruction execution unit connected with the hierarchical collaborative scheduling optimization engine, for converting the final energy storage system charge-discharge power instruction sequence into a message conforming to a specific communication protocol, and sending it to the energy management system and the converter of the energy storage system through a physical communication interface.

9. The industrial and commercial energy storage system energy resource scheduling system of claim 8, wherein, The battery multi-physical field digital twin module is deployed on an edge computing device, which contains a graphics processing unit for parallel computing and a multi-core central processing unit for numerical computing; the decision-making periods of the three internal components of the hierarchical collaborative scheduling optimization engine decrease step by step, the decision-making period of the long-term degradation trajectory planner is weekly or monthly, the decision-making period of the medium-term stochastic scheduling optimizer is daily, and the decision-making period of the short-term model predictive controller is minute-level.

Images