Industrial and commercial energy storage system multi-target model prediction control method and system

Abstract

The invention designs a multi-target model prediction control method and system for an industrial and commercial energy storage system, and belongs to the technical field of energy storage systems. Establishing a battery life loss model based on the change quantity of the state of charge and the average state of charge, describing the differentiated influence of different working intervals on the service life of the battery through a piecewise linear function, and constructing a multi-objective optimization function comprising the cost of electric charge, the cost of battery loss and a power fluctuation penalty term under a model prediction control framework; the economy and the battery life are balanced; performing linearization processing on a quadratic term and a piecewise function in the loss model, and solving an optimal charging and discharging power sequence in a prediction time domain; rolling optimization and feedback correction are executed, and the optimization problem is solved again according to the actually measured state of charge and the updated prediction information. According to the method, organic combination of energy storage system economic dispatching and battery life protection is realized, the real-time solution bottleneck of complex multi-objective optimization is broken through, and the economical efficiency, reliability and adaptive capacity of the system are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of energy storage systems, in particular to a multi-objective model predictive control method and system for industrial and commercial energy storage systems. BACKGROUND

[0002] With the transformation of global energy structure and the promotion of carbon neutralization target, the role of energy storage systems in power systems is increasingly important. Industrial and commercial energy storage systems can effectively reduce enterprise electricity costs and improve power grid operation efficiency through peak-valley arbitrage, demand management and demand response, and become an important part of distributed energy development. The core of the energy storage system is the battery energy storage unit, and the battery is the most expensive and limited life component of the energy storage system, and its operation strategy directly affects the economy and reliability of the system. How to maximize economic benefits while extending the service life of the battery is a key technical challenge faced by industrial and commercial energy storage systems. Model predictive control, as an advanced control method, optimizes control strategies in the future time domain by establishing a system prediction model, and has been widely used in industrial process control. Applying model predictive control to energy storage system scheduling can make full use of the prediction information of load and electricity price to achieve forward-looking optimization control of the energy storage system. However, the existing energy storage system model predictive control method has many technical problems, which restricts the improvement of the overall performance of the energy storage system.

[0003] Firstly, the existing technology has obvious deficiencies in battery life loss modeling. Traditional energy storage scheduling methods usually use rough life models, only considering the number of charge and discharge times or equivalent cycle times, and fail to accurately reflect the differentiated loss characteristics of batteries under different charge and discharge depths and different state of charge intervals. The actual aging process of the battery is complexly affected by multiple factors. In the deep discharge area and high state of charge area, the loss rate of the battery is significantly higher than that in the normal working interval. The existing method cannot accurately quantify this differentiated impact, resulting in low accuracy of life prediction and inability to provide reliable loss evaluation basis for energy storage system scheduling. The limitations of this modeling method make it impossible to accurately quantify the trade-off between economic benefits and battery protection in the control strategy, either overemphasizing economic benefits and accelerating battery aging, or being too conservative and sacrificing economic efficiency. Secondly, the existing technology has a single limitation in the design of optimization objectives. Traditional energy storage control methods usually only focus on a single objective, either pursuing the minimization of electricity costs and ignoring the damage to battery life caused by frequent charging and discharging, or only considering battery protection and failing to fully realize the economic value of the energy storage system. In actual applications, energy storage systems need to consider multiple objectives such as economic benefits, battery life, and operational stability. Single-objective optimization methods cannot meet the actual needs of industrial and commercial energy storage systems and are difficult to achieve global optimization of system operation. Even if some methods try to build a multi-objective optimization framework, they often lack in-depth analysis of the trade-off relationship between objectives, and the setting of weight coefficients lacks theoretical basis and engineering practice guidance, resulting in optimization results that are difficult to meet the requirements of actual applications. Thirdly, the existing technology faces real-time bottlenecks in solving efficiency. The multi-objective optimization problem of energy storage systems usually contains nonlinear cost functions, quadratic constraints, and piecewise functions, which have complex mathematical forms. Directly solving such mixed integer nonlinear programming problems has high computational complexity and long solving time, making it difficult to meet the requirements of real-time control. Traditional heuristic algorithms, although fast in calculation, cannot guarantee the quality of the solution and are prone to local optimization, making it difficult to obtain global optimal solution. Commercial nonlinear optimization solvers, although high in solution quality, have an exponential increase in calculation time with the problem size. For typical twenty-four-hour prediction time domain problems, the solving time often takes several minutes or even several tens of minutes, which cannot meet the time requirements of real-time control. This bottleneck in solving efficiency seriously restricts the engineering application of complex optimization algorithms in actual energy storage systems. Fourthly, the existing technology lacks effective mechanisms to deal with uncertainties. Energy storage systems face various uncertain factors such as load prediction errors, price fluctuations, and battery parameter changes during operation. Traditional open-loop optimization control methods maintain the control strategy unchanged throughout the prediction time domain once the optimization is completed, which cannot be dynamically adjusted according to the actual operating state, resulting in a decline in control accuracy over time. Battery characteristics gradually change with the increase in cycle number and environmental conditions, and parameters such as charging and discharging efficiency and internal resistance gradually deviate from the initial values.The existing method usually adopts fixed model parameters without considering the time-varying characteristics of the parameters. When the model parameters deviate from the actual system characteristics, it cannot be found and corrected in time, resulting in gradual deterioration of control performance and poor system robustness. Fifth, the existing technology has engineering difficulty in system integration. Even if part of the advanced control algorithm has superiority in the theoretical level, due to the lack of perfect system architecture design and engineering implementation scheme, it is difficult to apply in practical energy storage projects. The implementation of the control algorithm needs to be deeply integrated with the hardware devices such as battery management system and energy storage converter, and a reliable communication mechanism and data interaction interface need to be established. The existing method often only focuses on the algorithm itself and ignores the complexity of system integration, resulting in poor compatibility of the algorithm with the hardware platform, low overall integration of the system, difficult maintenance, and restriction of the large-scale popularization and application of advanced control technology in industrial and commercial energy storage field.

[0004] Therefore, it is urgent to design a new industrial and commercial energy storage system control method and system, establish an accurate battery life loss model, accurately reflect the differentiated loss characteristics of different working intervals, realize the dynamic balance of economy and battery life under the multi-objective optimization framework, and comprehensively improve the economy, reliability and practicability of the industrial and commercial energy storage system. SUMMARY

[0005] To solve the problems in the background art, the present application provides an industrial and commercial energy storage system multi-objective model predictive control method, comprising the following steps: S1, establishing a battery life loss model: based on the change amount of state of charge SOC and the average state of charge in the battery charging and discharging process, an equivalent loss cost model of single charging and discharging is established, and the differentiated influence of different working intervals on the battery life is described by a piecewise linear function; S2, constructing a multi-objective optimization function: under the model predictive control MPC framework, a weighted optimization objective function including electricity cost, battery loss cost and power fluctuation penalty term is established, and the economy and battery life are balanced by weight coefficients; S3, solving the mixed integer linear programming problem: the quadratic term and the piecewise function in the loss model are linearized, the multi-objective optimization problem is converted into a mixed integer linear programming MILP problem, and the optimal charging and discharging power sequence is solved in the prediction time domain; S4, executing rolling optimization and feedback correction: only the optimal charging and discharging power at the current time is executed, the optimization problem is re-solved according to the measured SOC and updated prediction information at the next time, and the model parameter is corrected online when the SOC prediction error exceeds the threshold.

[0006] Further, the S1 comprises the following steps: S11, at each sampling time The change amount of the battery state of charge is calculated, and the formula is: ; in for Change in SOC at time (%) for Battery state of charge (%) at any given time; for Battery state of charge (%) at any given time; S12. Calculate the average operating time during the charging and discharging process using the following formula: ; in for The average state of charge (%) at any given moment during the charging and discharging process. S13. Determine the weighting factor based on the interval of the average state of charge. The piecewise function used is: when hour, ; when hour, ; when hour, , in The weighting factor is based on the average state of charge and is dimensionless. S14. Calculate the equivalent loss cost of a single charge-discharge cycle using the following formula: ; in for Equivalent loss cost per charge and discharge cycle (in yuan); Linear loss coefficient (yuan / %) for Absolute value of SOC change at time (%) Nonlinear loss coefficient (yuan / %²); for The square of the change in SOC at time t (%²); and The data was obtained by regression analysis of cycle life test data of the same type of battery at different charge-discharge depths.

[0007] Furthermore, step S2 includes the following steps: S21. Construct the electricity cost item for each time point in the prediction time domain, using the following formula: ; in To predict within the time domain Electricity cost at any given time (in yuan); for Electricity price at any given time (RMB / kWh); is the grid purchase power (kW) at time t; is the grid purchase power (kW) at time t; is the sampling time interval (h); is the time index in the prediction horizon; S22, construct the battery loss cost term at each time in the prediction horizon, calculate according to the method of claim 2 wherein is the battery loss cost (yuan) at time t; is the battery loss cost (yuan) at time t; S23, construct the power fluctuation penalty term at each time in the prediction horizon, the formula is:

[0008] wherein is the power fluctuation penalty term (yuan) at time t; is the power fluctuation penalty term (yuan) at time t; is the power fluctuation penalty coefficient (yuan / kW); is the battery charge-discharge power (kW) at time t; is the battery charge-discharge power (kW) at time t; is the battery charge-discharge power (kW) at time t; is the battery charge-discharge power (kW) at time t; is the absolute value of the battery power difference between adjacent times (kW); S24, establish a weighted multi-objective optimization function, the formula is: ; wherein is the multi-objective optimization function (yuan); is the summation operator; is the MPC prediction horizon length, i.e. the number of time steps predicted; is the current time index; is the electricity cost weight coefficient, dimensionless; is the loss cost weight coefficient, dimensionless; is the power fluctuation penalty weight coefficient, dimensionless; indicates the minimization optimization objective.

[0009] Further, the S3 comprises the following steps: S31, linearize the quadratic term: for the term in the loss model, introduce an auxiliary variable to represent the approximate value of , divide the SOC change range into subintervals, each subinterval length is , introduce binary variables , , through the constraint condition ensure that only one subinterval is activated, use the piecewise linear function Approximating the quadratic term, where for Approximate change in SOC at time (%). This represents the total number of sub-intervals. The maximum percentage of change in SOC. for Time of the first Binary selection variables for each sub-interval; For the first Representative point values ​​(%) of each sub-interval; S32. Linearize the piecewise function: This involves adjusting the weighting factor for the SOC working interval. Introducing three binary variables , , Through constraints Ensure mutual exclusion by selecting an interval, according to The interval is constrained by linear inequalities to activate the corresponding binary variables, and the loss coefficient is expressed as follows: ;in for Time-deep discharge region binary selection variables; for Always Healthy Workspace binary selection variables; for High SOC aging zone binary selection variables; for The weighting factor after linearization at time points is dimensionless. S33. Establish the standard form of MILP: Define continuous decision variables including battery charging and discharging power. and state of charge Discrete decision variables are defined, including the charging / discharging mode selection variable. , And linearize auxiliary variables, establish constraints, including the SOC dynamic equation: Power balance equation: Charge-discharge mutual exclusion constraint: Power boundary constraints: ; and SOC boundary constraints ; in for State of charge (%) at time t; Charging efficiency is dimensionless. for Charging power (kW) at any given time; Discharge efficiency, dimensionless; The rated capacity of the battery is (kWh). for Load power (kW) at any given time; for The variable for selecting the charging mode at any given time; for The variable for selecting the discharge mode at any given time; Maximum charging and discharging power (kW); The lower limit of SOC (%); SOC upper limit (%); S34. Call the MILP solver: Input the linearized objective function and constraints into the MILP solver, and perform the prediction in the time domain. The optimal charge / discharge power sequence is obtained by internal solution. , .

[0010] Furthermore, step S4 includes the following steps: S41. Extract the first-step control quantity from the optimal power sequence. The power command is then sent to the energy storage converter to perform the actual charging and discharging operation, whereby... For the current moment The optimal charge / discharge power (kW); S42, in At any given time, the measured SOC value is obtained from the battery management system via the data acquisition module. At the same time, obtain the updated future Load forecast information at each time point Electricity price forecast information ,in for The measured state of charge (%) fed back by the battery management system at all times; The updated load forecast sequence (kW); The updated electricity price forecast series (yuan / kWh); S43. Calculate the SOC prediction error using the following formula: ; in for SOC prediction error at time (%); In order to be in Predicted during time optimization SOC value at time (%) S44, Judgment Is it greater than the preset error threshold? ,like This triggers online parameter correction, re-identifying the battery charge / discharge efficiency. and and update the loss coefficient by recent measured data and wherein is the SOC prediction error threshold (%); S45, with measured as the new initial state, combined with the updated prediction information and and the corrected model parameters, repeat S2 to S3 to obtain a new optimal control sequence, realizing closed-loop rolling optimization control.

[0011] The application also provides a multi-objective model predictive control system for industrial and commercial energy storage systems, comprising: an edge computing gateway, including a loss calculation unit, a prediction module, a MILP modeling engine, a solver module, a rolling scheduler, and a parameter correction unit; The loss calculation unit receives the SOC data at the current time and the previous time, calculates the SOC change, the average state of charge, the weighting factor, and the equivalent loss cost, and outputs the loss cost to the MILP modeling engine; the prediction module obtains the future time domain electricity price and load information, including an electricity price prediction submodule and a load prediction submodule, and outputs the predicted electricity price sequence and load sequence to the MILP modeling engine; the MILP modeling engine receives the loss cost output by the loss calculation unit, the electricity price sequence and load sequence output by the prediction module, constructs the electricity cost term, the power fluctuation penalty term, and the multi-objective optimization function, linearizes the quadratic term and the piecewise function, establishes an optimization problem in the standard form of MILP, and outputs the optimization problem to the solver module; the solver module receives the optimization problem output by the MILP modeling engine, solves to obtain the optimal charging and discharging power sequence in the prediction time domain, and outputs the optimal power sequence to the rolling scheduler; the rolling scheduler receives the optimal power sequence output by the solver module, extracts the first step control amount, and sends the charging and discharging power instructions to the energy storage converter through the communication interface; the parameter correction unit receives the measured SOC feedback by the battery management system, calculates the SOC prediction error, judges whether it exceeds the threshold, and if it exceeds the threshold, re-identifies the charging and discharging efficiency and updates the loss coefficient, and outputs the corrected parameters to the MILP modeling engine; a battery management system BMS connected with the edge computing gateway through a first communication bus, collecting single cell voltage and current, calculating and providing real-time SOC data to the loss calculation unit and the parameter correction unit of the edge computing gateway; an energy storage converter PCS connected with the edge computing gateway through a second communication bus, receiving the charging and discharging power instructions sent by the rolling scheduler, controlling the battery energy storage unit to perform charging and discharging operations, and feeding back the actual operating power to the edge computing gateway. The battery energy storage unit is connected with the BMS and the PCS respectively through a battery pack bus.

[0012] Further, the loss calculation unit comprises an SOC variation calculation module, an average SOC calculation module, a weighting factor determination module, and a loss cost calculation module. The SOC variation calculation module receives the current SOC and the previous SOC, calculates the SOC variation, and outputs the SOC variation to the average SOC calculation module and the loss cost calculation module; the average SOC calculation module receives the current SOC and the previous SOC, calculates the average state of charge, and outputs the average state of charge to the weighting factor determination module; the weighting factor determination module determines the weighting factor according to the interval in which the average state of charge is located, and outputs the weighting factor to the loss cost calculation module; the loss cost calculation module receives the SOC variation and the weighting factor, and calculates the equivalent loss cost according to the pre-stored loss coefficient.

[0013] Further, the MILP modeling engine comprises a cost construction module, a linearization processing module, and a constraint construction module. The cost construction module receives the loss cost output by the loss calculation unit, the electricity price sequence and the load sequence output by the prediction module, constructs the electricity cost item and the power fluctuation penalty item, combines the loss cost item to establish a weighted multi-objective optimization function, and outputs the optimization function to the linearization processing module; the linearization processing module introduces auxiliary variables and binary variables for piecewise linear approximation of quadratic terms in the optimization function, linearizes the piecewise function by introducing binary variables, and outputs the linearized objective function to the constraint construction module; the constraint construction module establishes the SOC dynamic equation, the power balance equation, the charge-discharge mutual exclusion constraint, the power boundary constraint, the SOC boundary constraint, and the linearization auxiliary constraint, to form a complete MILP optimization problem.

[0014] Further, the parameter correction unit comprises an error calculation module, a threshold judgment module, and a parameter identification module. The error calculation module receives the measured SOC fed back by the BMS and the predicted SOC output by the solver module, calculates the SOC prediction error, and outputs the prediction error to the threshold judgment module; the threshold judgment module compares the prediction error with a preset threshold, and triggers the parameter identification module when the prediction error exceeds the threshold; the parameter identification module receives the measured SOC, the charge-discharge power, and the loss data at a plurality of recent time points, re-identifies the charge-discharge efficiency and the loss coefficient, and outputs the corrected parameters to the MILP modeling engine.

[0015] Further, the first communication bus is a CAN bus or an RS485 bus, the second communication bus is a ModbusTCP bus or a CAN bus, and the battery pack bus is a CAN bus; a master-slave communication architecture is adopted between the edge computing power gateway, the BMS and the PCS, the edge computing power gateway periodically requests SOC data from the BMS and issues power instructions to the PCS as a master station, and the BMS and the PCS respond to the request of the master station as slave stations.

[0016] The present application has the following beneficial effects: First, the present application introduces a classic fatigue analysis method into the energy storage scheduling field and linearizes it to adapt to the model predictive control framework, establishes a battery life loss model, and accurately describes the differentiated influence of different working intervals on battery life through piecewise linear functions based on the change in state of charge and average state of charge during battery charging and discharging, solving the technical problems of traditional life modeling methods that are rough and cannot reflect differentiated losses under different charging and discharging depths; improving the accuracy of battery life prediction, enabling the energy storage system to accurately quantify the trade-off between economic benefits and battery protection, effectively extending the battery life and reducing the total life cycle cost.

[0017] Second, the multi-objective optimization function constructed by the present application combines electricity cost, battery loss cost and power fluctuation penalty term under the model predictive control framework, and realizes dynamic balance optimization of economy and battery life through weight coefficients, breaking through the single target limitation of traditional energy storage control that only focuses on economic benefits or only considers battery protection, and achieving global optimization of the energy storage system operation strategy. By comprehensively considering the minimization of electricity cost, battery loss control and power smoothing adjustment, the energy storage system can maximize economic benefits while avoiding damage to the battery caused by frequent charging and discharging, improving the stability and reliability of system operation.

[0018] Third, the present application innovatively linearizes the quadratic term and piecewise function in the loss model, successfully transforming the complex multi-objective optimization problem into a mixed integer linear programming problem, achieving efficient and accurate real-time solution. By introducing auxiliary variables and binary selection variables for piecewise linear approximation, the mathematical properties of the original problem are maintained, and the calculation efficiency is significantly improved, solving the real-time bottleneck of complex multi-objective optimization, enabling the energy storage system to complete the optimal charging and discharging power sequence solution within the prediction time domain in a short time, ensuring the feasibility and practicality of the control algorithm in engineering practice.

[0019] Fourthly, the rolling optimization and feedback correction mechanism established by the application realizes the closed-loop control scheme of model predictive control, effectively deals with the influence of load prediction error and model uncertainty by only executing the optimal control quantity at the current time and re-optimizing according to the measured data at the next time. The online parameter correction function triggered automatically when the state of charge prediction error exceeds the threshold value enables the system to have strong adaptive ability to track the slow changes of battery characteristics and the fluctuations of environmental conditions, improves the control precision and system robustness, and ensures that the energy storage system maintains stable control performance and optimization effect in long-term operation.

[0020] Fifthly, the modular system architecture designed by the application realizes the organic combination of complex control algorithms and hardware platforms through the coordinated control of edge computing gateway, battery management system, energy storage converter and battery energy storage unit. The loss calculation unit, prediction module, mixed integer linear programming modeling engine, solver module, rolling scheduler and parameter correction unit integrated in the edge computing gateway realize efficient algorithm implementation and flexible function configuration through modular design. The system communication design using master-slave communication architecture and standardized industrial communication protocol ensures real-time data exchange and coordinated control between components. This system architecture innovation effectively improves the overall integration and maintainability of the energy storage system, and provides important technical support and engineering solutions for the large-scale popularization and application of industrial and commercial energy storage. BRIEF DESCRIPTION OF DRAWINGS

[0021] Fig. 1 is a multi-objective model predictive control system architecture diagram of an industrial and commercial energy storage system of the application; Fig. 2 is a multi-objective model predictive control method flow chart of an industrial and commercial energy storage system of the application. DETAILED DESCRIPTION

[0022] The technical solutions in the application will be described clearly and completely below with reference to the drawings in the application, and the forms of each structure described in the following embodiments are only examples, and the application is not limited to each structure described in the following embodiments. All other embodiments obtained by those of ordinary skill in the art without making creative efforts fall within the scope of protection of the application.

[0023] Referring to Figs. 1-2 The multi-objective model predictive control method of the industrial and commercial energy storage system of the application includes four main steps, which are establishing a battery life loss model, constructing a multi-objective optimization function, solving a mixed integer linear programming problem and executing rolling optimization and feedback correction. Specifically as follows: Step S1: Establish a battery life loss model. Based on the change in State of Charge (SOC) and average SOC during battery charging and discharging, Step S1 establishes an equivalent loss cost model for a single charge-discharge cycle. A piecewise linear function describes the differentiated impact of different operating intervals on battery life. This step is the first to introduce the classic rainflow counting method from fatigue analysis into energy storage scheduling, and performs linearization to adapt it to the model's predictive control framework. This solves the technical problem that traditional life modeling methods are coarse and cannot reflect differentiated losses at different charge-discharge depths.

[0024] Step S11: Calculate the change in battery state of charge; calculate the change in battery state of charge at each sampling time to establish fundamental parameters that accurately reflect the battery's charge and discharge intensity. The formula is: ;in, for Change in SOC at time (unit: %) for Battery state of charge at any given time (in %). for The battery's state of charge (SOC) at time 1.5 (%). This formula accurately captures the battery's depth of charge and discharge in each sampling period by calculating the difference in SOC between adjacent time points, providing a key input parameter for subsequent loss cost calculations.

[0025] Step S12: Calculate the average state of charge (SOC) during the charge and discharge process. The average SOC reflects the battery's operating range during charge and discharge and is a crucial factor affecting battery life, as the aging mechanism of batteries differs significantly across different SOC ranges. The formula is: ; in, for The formula represents the average state of charge (SOC) during the charging and discharging process at each time point (in %). It calculates the arithmetic mean of the SOC between two adjacent time points, accurately characterizing the battery's operating state within that time period. Based on research into the aging mechanism of lithium-ion batteries, the aging rate in the deep discharge region and high SOC region is significantly higher than in the healthy operating region. Therefore, it is necessary to determine the operating range by averaging the SOC and applying different weighting factors.

[0026] Step S13: Determine the weighting factor based on the interval of the average state of charge; based on the research results of battery aging mechanism, establish a piecewise linear weighting function to accurately reflect the differentiated impact of different working intervals on battery life.

[0027] when hour, ; when hour, ; when Time, .

[0028] wherein, is the weighting factor based on average state of charge, dimensionless. The design of this piecewise function is based on a large amount of battery cycle life test data, the weighting factor in the deep discharge region (0%-20%) is set to 1.5, which reflects the acceleration effect of deep discharge on battery capacity attenuation; the weighting factor in the healthy working region (20%-80%) is set to 1.0, which is the benchmark loss level; the weighting factor in the high SOC aging region (80%-100%) is set to 1.3, which reflects the additional loss caused by the intensification of electrolyte decomposition under high voltage. This segmented weighting strategy guides the energy storage system to preferentially operate in the healthy working region, effectively prolonging the battery service life.

[0029] Step S14: Calculate the equivalent loss cost of single charge-discharge; a loss cost model containing linear and quadratic terms is established to accurately quantify the economic impact of single charge-discharge operation on battery life. The formula is: ; wherein, is the equivalent loss cost of single charge-discharge at time t (unit: yuan); is the linear loss coefficient (unit: yuan / %); is the absolute value of SOC change at time t (unit: %); is the nonlinear loss coefficient (unit: yuan / % square); is the square of SOC change at time t (unit: % square). The linear term reflects the basic loss proportional to the charge-discharge depth, and the nonlinear term reflects the additional loss amplification effect of large current charge-discharge and extreme SOC working interval. The loss coefficients and are obtained by regression of cycle life test data of the same type of battery under different charge-discharge depths, ensuring the accuracy and practicality of the model parameters.

[0030] Step S2: Construct a multi-objective optimization function; step S2 establishes a weighted optimization objective function containing electricity cost, battery loss cost and power fluctuation penalty term under the model predictive control framework, and balances the economy and battery life through weight coefficients. This step first realizes the quantitative trade-off between electricity arbitrage income and battery life loss, breaking through the technical limitations of traditional energy storage scheduling methods that only focus on short-term income and ignore battery life.

[0031] ​​​Step S21: Construct the electricity cost term for each time point within the prediction time domain; the electricity cost term accurately calculates the grid purchase cost of the energy storage system at each time point within the prediction time domain, providing a basic objective function for economic optimization. The formula is: ; in, To predict within the time domain Electricity cost at any given time (unit: yuan); for Electricity price at any given time (unit: yuan per kilowatt-hour); for Power purchased from the power grid at any time (unit: kilowatt); Sampling time interval (unit: hours); This is used for predicting time indexes within the time domain. The formula calculates the electricity cost at each time point by multiplying the electricity price by the purchased power. A positive value indicates the cost of purchasing electricity from the grid, while a negative value indicates the revenue gained from selling electricity to the grid. This provides an accurate cost quantification basis for the economic dispatch of energy storage systems.

[0032] Step S22: Construct the battery loss cost item for each moment in the prediction time domain; the battery loss cost item is calculated based on the loss model established in step S1, and the battery life loss is transformed into a quantifiable economic cost, so that the optimization algorithm can make a reasonable trade-off between electricity cost savings and battery protection.

[0033] Battery loss cost Calculate according to the formula in step S1, where for Battery degradation cost at any given moment (unit: yuan). The introduction of this cost item means that when making charging and discharging decisions, energy storage systems must not only consider immediate electricity revenue, but also weigh the long-term impact on battery life. This avoids the problem of excessive battery degradation caused by frequent shallow charge and discharge cycles in traditional methods, and maximizes the economic benefits throughout the entire life cycle.

[0034] Step S23: Construct the power fluctuation penalty term for each time step in the prediction time domain. The power fluctuation penalty term, by penalizing power changes between adjacent time points, suppresses frequent switching of charge and discharge states in the energy storage system, reduces the impact on the battery and inverter, and improves system operational stability. The formula is: ; in, for Power fluctuation penalty at any given time (unit: yuan); Power fluctuation penalty factor (unit: yuan per kilowatt); for Battery charging and discharging power at any time (unit: kilowatt); for Battery power at time t (unit: kW); The absolute value of the power difference between adjacent time points (unit: kW). This penalty term guides the energy storage system to adopt a smooth power change trajectory by linearly penalizing the power change amplitude, avoiding the stress impact of sharp power fluctuations on the energy storage equipment, prolonging the service life of the equipment and improving the operating efficiency.

[0035] Step S24: Establish a weighted multi-objective optimization function; the optimization function combines the electricity cost, battery loss cost and power fluctuation penalty term by weighting, forming a unified optimization target, achieving a comprehensive balance of economy, battery protection and operation stability. The formula is: Wherein, The multi-objective optimization function (unit: yuan); The summation operator; The model predictive control prediction time domain length, i.e. the number of time steps predicted; The current time index; The electricity cost weight coefficient, dimensionless; The loss cost weight coefficient, dimensionless; The power fluctuation penalty weight coefficient, dimensionless; Indicates the minimization of the optimization target. The design of the weight coefficient allows the operator to flexibly adjust the importance of each cost according to different operation strategies and market conditions, Usually set to 1.0 as a benchmark, The value range of is 0.5 to 2.0, which can be adjusted according to the battery cost and service life requirements, Usually set to 0.1 to smooth the power change.

[0036] Step S3: Solve the mixed integer linear programming problem; step S3 linearizes the quadratic term and the segmented function in the loss model, converts the multi-objective optimization problem into a mixed integer linear programming problem, and solves the optimal charge and discharge power sequence within the prediction time domain. The core innovation of this step is the high-precision linearization technology of the quadratic term and the segmented function, which enables the complex nonlinear optimization problem to be solved within seconds, meeting the strict time requirements of real-time control.

[0037] Step S31: Linearize the quadratic term; for the quadratic term in the loss model, Introduce auxiliary variables for piecewise linear approximation, effectively solving the nonlinear optimization complexity problem caused by the quadratic term.

[0038] This linearization method divides the SOC change range into subintervals, each with a length of , and introduces binary variables ​,in Through constraints To ensure that only one subinterval is activated, use a piecewise linear function. Approximating the quadratic term, where for Approximate change in SOC at time (unit: %). This represents the total number of sub-intervals. The maximum value of the change in SOC (unit: %). for Time of the first Binary selection variables for each sub-interval; For the first The representative point values ​​of each subinterval (unit: %). This method achieves high-precision approximation of quadratic terms with an approximation error of less than 2% by adding a small number of binary variables and linear constraints, while maintaining the computational efficiency of linear programming.

[0039] Step S32: Linearize the piecewise function; weight the SOC working interval factors. By leveraging the piecewise characteristics of binary variables for linearization modeling, a precise expression of piecewise functions within the framework of mixed-integer linear programming is achieved.

[0040] This method introduces three binary variables , , Through constraints Ensure mutual exclusion by selecting an interval, according to The corresponding binary variables within the given interval are activated by linear inequality constraints, and the loss coefficient is expressed as: ; in, for Binary selection variables for the time-deep discharge region; for Binary selection variables for the constant health workspace; for Binary selection variables for the high SOC aging region at any given time; for The weighting factor after linearization is dimensionless. This linearization method, by introducing a small number of binary variables, transforms complex conditional logic into standard linear constraints, ensuring the accuracy of the piecewise function in the optimization process.

[0041] Step S33: Establish the standard form of mixed-integer linear programming; this step unifies the physical constraints, control logic and linearization conditions of the energy storage system into a standard mixed-integer linear programming problem, laying the foundation for efficient solution.

[0042] Continuous decision variables include battery charging and discharging power. and state of charge Discrete decision variables include charging / discharging mode selection variables. , And linearization auxiliary variables. The main constraints include the SOC dynamic equations: ; This equation describes the dynamic change of the battery's state of charge with charge and discharge power; power balance equation: ; Ensure power supply and demand balance at every moment; charge and discharge mutual exclusion constraints: ; Non-physical states that prevent simultaneous charging and discharging; power boundary constraints: ; And SOC boundary constraints: ; in, for State of charge at any given time (unit: %); Charging efficiency is dimensionless. for Charging power at any given time (unit: kilowatts); Discharge efficiency, dimensionless; for Discharge power at any given moment (unit: kilowatt); Rated capacity of the battery (unit: kilowatt-hours); for Load power at any given time (unit: kilowatts); for The variable for selecting the charging mode at any given time; for The variable for selecting the discharge mode at any given time; Maximum charging and discharging power (unit: kilowatt); SOC lower limit (unit: %) The SOC (State of Charge) is the upper limit (in %). These constraints comprehensively consider the electrochemical characteristics of the battery, the physical limitations of power electronic devices, and the operational safety requirements of the energy storage system, ensuring that the optimization results are technically feasible.

[0043] Step S34: Invoke the mixed-integer linear programming solver; input the linearized objective function and constraints into the commercial optimization solver, and solve for the optimal charge-discharge power sequence in the prediction time domain. , ,in for The optimal charging and discharging power at any given time (unit: kilowatts). Modern commercial solvers such as Gurobi and CPLEX have efficient algorithms for solving mixed-integer linear programming problems. For typical 24-hour prediction time-domain problems, the solution time is usually within 1 second, meeting the time requirements of real-time control. The solution process employs the branch and bound method to handle integer variables, the simplex method or interior-point method to handle linear relaxation problems, and accelerates convergence through pre-solution techniques and heuristic methods.

[0044] Step S4: Perform rolling optimization and feedback correction; Step S4 only executes the optimal charging and discharging power at the current moment. At the next moment, the optimization problem is resolved based on the measured SOC and the updated prediction information. When the SOC prediction error exceeds the threshold, online correction of the model parameters is triggered. This step realizes the rolling time-domain optimization mechanism of model predictive control, which improves control accuracy and robustness through continuous feedback correction, effectively addressing the impact of load prediction errors and model uncertainties.

[0045] Step S41: Extract the first control quantity from the optimal power sequence and execute it; extract the first control quantity from the optimal power sequence obtained in step S34. The power command is sent to the energy storage converter via the communication interface to perform the actual charging and discharging operation. For the current moment The optimal charging and discharging power (unit: kilowatt) is determined. Power commands are issued via industrial communication protocols such as Modbus TCP or CAN bus to ensure real-time and reliable command transmission. After receiving the power command, the energy storage converter executes the corresponding charging and discharging operations based on the current battery state and grid conditions, and simultaneously feeds back the actual executed power value to the control system for closed-loop control.

[0046] Step S42: Acquire real-time data and update prediction information; At any given time, the measured SOC value is obtained from the battery management system via the data acquisition module. At the same time, obtain the updated future Load forecast information at each time point Electricity price forecast information ,in for Measured state of charge (in %) fed back by the battery management system at all times; The updated load forecast sequence (unit: kilowatts); This is the updated electricity price forecast sequence (unit: yuan per kilowatt-hour). Measured SOC data is obtained through coulombic metering and open-circuit voltage method using a battery management system, ensuring high measurement accuracy. Load forecast information is updated using machine learning methods based on historical load data and weather forecasts. Electricity price forecast information is updated according to electricity market rules and policy documents, ensuring the timeliness and accuracy of the forecast information.

[0047] Step S43: Calculate the SOC prediction error and determine the threshold; the formula for calculating the SOC prediction error is: ; in, for SOC prediction error at time (unit: %); In order to be in Predicted during time optimization SOC value at any given time (unit: %). When the prediction error exceeds a preset threshold, it indicates a deviation between the current model parameters and the actual system characteristics, requiring the triggering of an online parameter correction mechanism. This threshold is typically set to 5%, which allows for timely detection of model deviations while avoiding frequent parameter adjustments due to measurement noise.

[0048] Step S44: Perform online parameter correction; when At this time, the online parameter correction mechanism is triggered to re-identify the battery charging and discharging efficiency. and And update the loss coefficient based on the latest measured data. and ,in The SOC prediction error threshold (unit: %).

[0049] The parameter identification process employs recursive least squares method, using recently measured SOC, charge / discharge power, and environmental condition data to re-estimate battery efficiency parameters. The efficiency parameter correction formula is based on the energy balance equation, adjusting parameters by comparing the theoretical energy change with the actual SOC change. The loss coefficient is updated based on health status data from the battery management system, combined with actual cycle count and capacity decay, using an online learning algorithm for gradual correction. and The numerical value. This adaptive parameter correction mechanism ensures that the control algorithm can track slow changes in battery characteristics, such as temperature effects and aging effects, while maintaining control accuracy and optimization performance.

[0050] Step S45: Execute rolling optimization control; based on actual measurements As a new initial state, combined with the updated prediction information and And the revised model parameters, repeat steps S2 to S3, solve the new optimal control sequence, realize the closed loop rolling optimization control.

[0051] The core of rolling optimization is to update the control strategy with the latest information constantly, and each control period is based on the most accurate system state and the latest prediction information to make decisions. This mechanism makes the control system have good anti-interference ability and adaptive performance, and can effectively deal with uncertain factors such as load fluctuation, price change and prediction error. At the same time, through the online correction of model parameters, the system can gradually learn and adapt to the changes of battery characteristics, and maintain long-term stable control performance.

[0052] The application also designs a multi-objective model predictive control system for industrial and commercial energy storage systems to realize the above control method, which includes an edge computing gateway, a battery management system, a storage converter and a battery storage unit as four main components, and each part realizes data exchange and coordinated control through a standardized communication interface. Specifically as follows: The edge computing gateway is the core of the whole control system, which contains six functional modules of loss calculation unit, prediction module, mixed integer linear programming modeling engine, solver module, rolling scheduler and parameter correction unit, and realizes efficient implementation and flexible configuration of complex control algorithms through modular design.

[0053] Loss calculation unit: receive the SOC data at the current time and the previous time, calculate the SOC change, average state of charge, weighting factor and equivalent loss cost according to the algorithm of step S1, and output the loss cost to the mixed integer linear programming modeling engine. The unit contains four sub-modules of SOC change calculation module, average SOC calculation module, weighting factor determination module and loss cost calculation module, and each module transmits data at high speed through the internal data bus.

[0054] Prediction module: obtain the future time domain of price and load information, including price prediction sub-module and load prediction sub-module, output the predicted price sequence and load sequence to the mixed integer linear programming modeling engine. The price prediction sub-module predicts the future 24-hour price trend based on historical price data, policy documents and market information, using time series analysis method; the load prediction sub-module predicts the future load demand based on historical load data, weather forecast and power consumption mode, using machine learning algorithm. Both sub-modules have online learning ability, which can automatically adjust the model parameters according to the prediction error to improve the prediction accuracy.

[0055] The mixed integer linear programming modeling engine receives the loss cost output by the loss calculation unit, the electricity price sequence and the load sequence output by the prediction module, constructs the electricity cost term, the power fluctuation penalty term and the multi-objective optimization function, linearizes the quadratic term and the segmented function, establishes the optimization problem in the standard form of the mixed integer linear programming, and outputs the optimization problem to the solver module. The engine internally includes a cost construction module, a linearization processing module and a constraint construction module. The cost construction module implements the multi-objective function construction of step S2, the linearization processing module executes the linearization algorithm of steps S31 and S32, and the constraint construction module establishes various constraint conditions of step S33.

[0056] The solver module receives the optimization problem output by the mixed integer linear programming modeling engine, solves the optimal charging and discharging power sequence in the prediction time domain by using a commercial optimization solver, and outputs the optimal power sequence to the rolling scheduler. The module integrates high-performance optimization solvers such as Gurobi or CPLEX, and through special hardware acceleration and algorithm optimization, ensures that the optimization solution of the 24-hour prediction time domain is completed within 1 second. The solver module also has a solving state monitoring and abnormal handling function, which can automatically degrade to a heuristic algorithm or a safety mode control strategy when the solving fails.

[0057] The rolling scheduler receives the optimal power sequence output by the solver module, extracts the first control quantity, and sends the charging and discharging power instructions to the energy storage converter through the communication interface. The module realizes the rolling time domain mechanism of model predictive control, and cyclically executes the optimization and control process according to the preset control period (usually 15 minutes). The rolling scheduler has multiple communication interfaces, supports ModbusTCP, CAN bus, Ethernet and other industrial standard protocols, and ensures compatibility with energy storage devices of different manufacturers.

[0058] The parameter correction unit receives the measured SOC feedback by the battery management system, calculates the SOC prediction error, judges whether the threshold is exceeded, and if the threshold is exceeded, re-identifies the charging and discharging efficiency and updates the loss coefficient, and outputs the corrected parameters to the mixed integer linear programming modeling engine. The unit includes an error calculation module, a threshold judgment module and a parameter identification module. The error calculation module implements the error calculation function of step S43, the threshold judgment module performs the threshold comparison of step S44, and the parameter identification module executes the online parameter update algorithm. The parameter identification adopts the recursive least squares method and Kalman filtering technology, which has strong numerical stability and convergence.

[0059] The battery management system is connected with the edge computing gateway through a first communication bus, collects single cell voltage and current, calculates and provides real-time SOC data to the loss calculation unit and parameter correction unit of the edge computing gateway. The battery management system adopts a distributed architecture, with the main controller responsible for overall coordination and communication, and the slave controller responsible for specific battery monitoring and protection functions. It includes high-precision collection of single cell voltage, current and temperature, with sampling accuracy reaching millivolt and milliampere levels, and sampling frequency not less than 1 Hz to meet real-time control requirements. The SOC estimation adopts a fusion algorithm combining Coulomb counting method and open circuit voltage method, and realizes state estimation through an extended Kalman filter, with SOC estimation accuracy better than 2%. The battery management system also has perfect safety protection functions, including overvoltage, undervoltage, overcurrent, overtemperature, insulation monitoring and other protection mechanisms, which can timely cut off the charging and discharging circuit and report to the upper system when detecting abnormal state.

[0060] The communication protocol adopts a master-slave mode, with the edge computing gateway as the master station periodically requesting SOC and other state data, and the battery management system as the slave station responding to data requests and actively reporting abnormal events. The communication protocol supports CAN bus and RS485 bus, with a communication frequency of 1 Hz to ensure that the control algorithm can timely obtain the latest battery state information. The data transmission adopts checksum and redundancy mechanism to ensure the reliability and integrity of data transmission.

[0061] The energy storage converter is connected with the edge computing gateway through a second communication bus, receives the charge and discharge power instructions issued by the rolling scheduler, controls the battery energy storage unit to perform charge and discharge operation, and feeds back the actual operating power to the edge computing gateway. The energy storage converter adopts a bidirectional DC-AC converter topology, has four-quadrant operation capability, and can realize bidirectional energy flow between the battery and the grid. The rated power is configured according to the capacity of the energy storage system, usually 0.5 to 1 times the capacity of the battery; the conversion efficiency is better than 95% in the full power range, and the peak efficiency can reach more than 98%; the power instruction response time is less than 100 milliseconds, meeting the requirement of fast power regulation; the power control accuracy is better than 1% of the rated power, ensuring accurate power tracking capability. Voltage frequency protection, anti-islanding protection, power quality control and other functional modules are set. Voltage frequency protection ensures that the grid voltage and frequency are within the allowed range before grid-connected operation; anti-islanding protection quickly detects and disconnects the grid-connected switch when the grid fails, avoiding power transmission to the fault grid; power quality control improves the power quality index of the access point through active filtering and reactive power compensation functions. The energy storage converter also supports multiple operating modes, including power control mode, voltage control mode and frequency control mode, which can adapt to different application scenario requirements.

[0062] Battery energy storage unit; the battery energy storage unit is connected with the battery management system and the energy storage converter through the battery pack bus respectively, and constitutes the energy storage core of the energy storage system. The battery energy storage unit adopts lithium iron phosphate battery technology, has the advantages of high safety, long cycle life, strong temperature adaptability and the like, and is suitable for long-term reliable operation requirements of industrial and commercial energy storage applications. Each battery module contains a plurality of battery clusters, and each battery cluster is composed of a plurality of battery monomers in series and parallel connection. The modular design facilitates system capacity expansion and maintenance, and the failure of a single module does not affect the normal operation of other modules. The battery monomer selects a standardized lithium iron phosphate battery cell, the monomer capacity is usually 100 Ah to 300 Ah, and the rated voltage is 3.2 V, and the required system voltage and capacity level are achieved through series and parallel connection. The thermal management system controls the battery temperature in the most suitable working range by using active air cooling or liquid cooling. The thermal management system automatically adjusts the cooling power according to the battery temperature and the environmental temperature, so that the battery can maintain good performance and service life under different seasons and working conditions. The battery energy storage unit is also provided with a fire safety system, including smoke detection, temperature monitoring, gas detection and automatic fire extinguishing functions, to ensure that safety hazards can be found and disposed of in time in abnormal conditions.

[0063] System communication architecture; the communication between the components of the system adopts a layered distributed architecture, the first communication bus is a CAN bus or an RS485 bus, which is used for the communication between the edge computing gateway and the battery management system; the second communication bus is a ModbusTCP bus or a CAN bus, which is used for the communication between the edge computing gateway and the energy storage converter; and the battery pack bus is a CAN bus, which is used for the communication between the battery management system and the internal modules of the battery energy storage unit. The edge computing gateway, the battery management system and the energy storage converter adopt a master-slave communication architecture, the edge computing gateway periodically requests SOC data from the battery management system and issues power instructions to the energy storage converter as a master station, and the battery management system and the energy storage converter respond to the request of the master station as slave stations. This communication architecture ensures the timing and consistency of data transmission, avoiding the problems of multi-master conflict and data confusion. The communication protocol design fully considers the requirements of real-time, reliability and scalability. In terms of real-time, the transmission delay of key control data is controlled within 50 milliseconds; in terms of reliability, the cyclic redundancy check and timeout retransmission mechanism are adopted to ensure the correctness of data transmission; and in terms of scalability, the communication interface supports hot plug and automatic identification function, facilitating system upgrade and equipment replacement.

[0064] The application realizes the quantitative trade-off between economic scheduling and battery life protection of the energy storage system by establishing a battery life loss model based on the rain flow counting method linearization; through the efficient solving technology of mixed integer linear programming, the real-time bottleneck of complex multi-objective optimization is broken through; through the modular system architecture design, the organic combination of control algorithm and hardware platform is realized, which significantly prolongs the service life of the battery while ensuring the economic benefit of the energy storage system.

[0065] The above merely provides the preferred embodiments of the present application, and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall fall within the protection scope of the present application.

Claims

1. A multi-objective model predictive control method for industrial and commercial energy storage systems, characterized in that, The method comprises the following steps: S1, establishing a battery life loss model: based on the change amount of state of charge SOC and the average state of charge in the battery charging and discharging process, an equivalent loss cost model of single charging and discharging is established, and the differential influence of different working intervals on the battery life is described by a piecewise linear function; S2, constructing a multi-objective optimization function: under the model predictive control MPC framework, a weighted optimization objective function including electricity cost, battery loss cost and power fluctuation penalty term is established, and the economic efficiency and battery life are balanced through the weight coefficient; S3, solving a mixed integer linear programming problem: the quadratic term and the piecewise function in the loss model are linearized, the multi-objective optimization problem is converted into a mixed integer linear programming MILP problem, and the optimal charging and discharging power sequence in the prediction time domain is solved; S4, performing rolling optimization and feedback correction: only the optimal charging and discharging power at the current time is executed, and the optimal problem is re-solved according to the measured SOC and updated prediction information at the next time, and the model parameter is online corrected when the SOC prediction error exceeds the threshold.

2. The method of claim 1, wherein, The S1 comprises the following steps: S11, at each sampling time The change amount of the battery state of charge is calculated, and the formula is: ; wherein is the SOC change (%) at the time point; is the battery state of charge (%) at the time point; is the battery state of charge (%) at the time point; S12, calculate the average work of the charging and discharging process, the formula is: ; wherein is the average state of charge (%) at the moment corresponding to the charging and discharging process; S13. Determine the weighting factor based on the interval in which the average state of charge lies The piecewise function used is: When Time, ; When Time, ; When time, , wherein is a weighting factor based on average state of charge, dimensionless; S14, calculating the equivalent loss cost of single charging and discharging, and the formula is: ； Wherein is Equivalent loss cost of single charge and discharge at the moment (yuan); is the linear loss coefficient (yuan / %); is Absolute value of SOC change amount at the moment (%); is the nonlinear loss coefficient (yuan / %2); is Square of SOC change amount at the moment (%2); the and It is obtained by regression of cycle life test data of different charge and discharge depths of the same type of battery.

3. The method of claim 1, wherein, The S2 comprises the following steps: S21, constructing the electricity cost item at each time in the prediction time domain, the formula is: ; wherein is the electricity cost (Yuan) at time instant is the electricity price (Yuan / kWh) at time instant is the electricity purchase power from the grid (kW) at time instant is the sampling time interval (h) is the time instant index within the prediction horizon​​​ S22, constructing a battery loss cost item at each time in the prediction time domain, calculating according to the method of claim 2 wherein is battery loss cost (yuan) at the time S23. Construct the power fluctuation penalty term for each time step in the prediction time domain, with the following formula: in for Power fluctuation penalty term (in yuan) at any given time; Power fluctuation penalty factor (yuan / kW); for Battery charging and discharging power at any time (kW); for Battery charging and discharging power at any time (kW); It represents the absolute value (kW) of the difference in battery power between adjacent time points. S24, a weighted multi-objective optimization function is established, and the formula is: ; wherein is a multi-objective optimization function (meta); is a summation operator; is the MPC prediction horizon length, i.e., the number of predicted time steps; is the current time index; is the electricity cost weight coefficient, dimensionless; is the loss cost weight coefficient, dimensionless; is the power fluctuation penalty weight coefficient, dimensionless; denotes the minimization optimization objective.

4. The method of claim 1, wherein, The S3 comprises the following steps: S31, linearization of quadratic term: for the quadratic term in the loss model , an auxiliary variable is introduced to represent the approximation of , the SOC change range is divided into sub-intervals, each with a length of , a binary variable is introduced, and through the constraint condition it is ensured that only one sub-interval is activated, using a piecewise linear function to approximate the quadratic term, where is the SOC change approximation at time ; is the total number of sub-intervals; is the maximum value of the SOC change (%); is the binary selection variable of the th sub-interval at time ; is the representative point value of the th sub-interval (%); S32. Linearize the piecewise function: This involves adjusting the weighting factor for the SOC working interval. Introducing three binary variables , , Through constraints Ensure mutual exclusion by selecting an interval, according to The interval is constrained by linear inequalities to activate the corresponding binary variables, and the loss coefficient is expressed as follows: ;in for Time-deep discharge region binary selection variables; for Always Healthy Workspace binary selection variables; for High SOC aging zone binary selection variables; for The weighting factor after linearization at time points is dimensionless. S33, establishing MILP standard form: defining continuous decision variables including battery charging and discharging power and state of charge , defining discrete decision variables including charging and discharging mode selection variables 、 and linearization auxiliary variables, establishing constraint conditions including SOC dynamic equation: ; power balance equation: ; charging and discharging mutual exclusion constraint: power boundary constraint: ; and SOC boundary constraint ; wherein is state of charge (%) at time t; is the charging efficiency, dimensionless; is charging power (kW) at time t; is the discharging efficiency, dimensionless; is the battery rated capacity (kWh); is load power (kW) at time t; is charging mode selection variable at time t; is discharging mode selection variable at time t; is the maximum charge and discharge power (kW); is the SOC lower limit (%); is the SOC upper limit (%); S34, calling MILP solver: input the linearized objective function and constraints into the MILP solver to obtain the optimal charging and discharging power sequence in the prediction time domain .​​ 5. The method of claim 1, wherein, The S4 comprises the following steps: S41, extracting a first step control quantity from the optimal power sequence The power instruction is sent to the energy storage converter to perform actual charging and discharging operation, wherein is the optimal charging and discharging power (kW) at the current moment ​ S42、in the SOC value measured at the moment is obtained from the battery management system through the data acquisition module at the same time, the updated future load prediction information and electricity price prediction information , wherein is the measured state of charge (%) fed back by the battery management system at the moment; is the updated load prediction sequence (kW); is the updated electricity price prediction sequence (yuan / kWh); S43, calculate the SOC prediction error, the formula is: ; wherein is the SOC prediction error (%) at the time instant; is the optimized SOC value (%) at the time instant predicted at the time instant; the SOC prediction error (%) at the time instant; S44, judging whether greater than a preset error threshold , if , trigger online parameter correction, re-identify battery charging and discharging efficiency and , and update the loss coefficient through the latest measured data and , wherein SOC prediction error threshold (%) S45, with the measured As a new initial state, combining the updated prediction information and and the corrected model parameters, repeat S2 to S3 to obtain a new optimal control sequence, and realize closed-loop rolling optimization control.

6. A multi-objective model predictive control system for industrial and commercial energy storage systems for implementing the method of any one of claims 1 to 5, characterized in that, It comprises: An edge computing gateway includes a loss calculation unit, a prediction module, a MILP modeling engine, a solver module, a rolling scheduler and a parameter correction unit; The loss calculation unit receives the SOC data at the current time and the previous time, calculates the SOC change amount, the average state of charge, the weighting factor and the equivalent loss cost, and outputs the loss cost to the MILP modeling engine; the prediction module obtains the electricity price and load information in the future time domain, including an electricity price prediction sub-module and a load prediction sub-module, and outputs the predicted electricity price sequence and load sequence to the MILP modeling engine; the MILP modeling engine receives the loss cost output by the loss calculation unit, the electricity price sequence and the load sequence output by the prediction module, constructs the electricity cost term, the power fluctuation penalty term and the multi-objective optimization function, linearizes the quadratic term and the piecewise function, establishes the optimization problem in the standard form of MILP, and outputs the optimization problem to the solver module; the solver module receives the optimization problem output by the MILP modeling engine, solves the optimal charging and discharging power sequence in the prediction time domain, and outputs the optimal power sequence to the rolling scheduler; the rolling scheduler receives the optimal power sequence output by the solver module, extracts the first control amount, and sends the charging and discharging power instruction to the energy storage converter through the communication interface; the parameter correction unit receives the measured SOC fed back by the battery management system, calculates the SOC prediction error, judges whether it exceeds the threshold, and if it exceeds the threshold, re-identifies the charging and discharging efficiency and updates the loss coefficient, and outputs the corrected parameters to the MILP modeling engine; A battery management system BMS is connected with the edge computing gateway through a first communication bus, collects single cell voltage and current, calculates and provides real-time SOC data to a loss calculation unit and a parameter correction unit of the edge computing gateway; A power conversion system PCS is connected with the edge computing gateway through a second communication bus, receives a charge-discharge power instruction issued by the rolling scheduler, controls the battery energy storage unit to perform charge-discharge operation, and feeds back actual operation power to the edge computing gateway; The battery energy storage unit is connected with the BMS and the PCS through a battery pack bus respectively.

7. The system of claim 6, wherein, The loss calculation unit includes an SOC change amount calculation module, an average SOC calculation module, a weighting factor determination module, and a loss cost calculation module; The SOC change amount calculation module receives the current time SOC and the previous time SOC, calculates the SOC change amount, and outputs the SOC change amount to the average SOC calculation module and the loss cost calculation module; the average SOC calculation module receives the current time SOC and the previous time SOC, calculates the average state of charge, and outputs the average state of charge to the weighting factor determination module; the weighting factor determination module determines the weighting factor according to the interval of the average state of charge, and outputs the weighting factor to the loss cost calculation module; the loss cost calculation module receives the SOC change amount and the weighting factor, and calculates the equivalent loss cost according to the pre-stored loss coefficient.

8. The system of claim 6, wherein, The MILP modeling engine includes a cost construction module, a linearization processing module, and a constraint construction module; The cost construction module receives the loss cost output by the loss calculation unit, the electricity price sequence and the load sequence output by the prediction module, constructs the electricity cost item and the power fluctuation penalty item, combines the loss cost item to establish a weighted multi-objective optimization function, and outputs the optimization function to the linearization processing module; the linearization processing module introduces auxiliary variables and binary variables for piecewise linear approximation of quadratic terms in the optimization function, linearizes the segmented function by introducing binary variables, and outputs the linearized objective function to the constraint construction module; the constraint construction module establishes SOC dynamic equation, power balance equation, charge-discharge mutual exclusion constraint, power boundary constraint, SOC boundary constraint and linearization auxiliary constraint to form a complete MILP optimization problem.

9. The system of claim 6, wherein, The parameter correction unit includes an error calculation module, a threshold judgment module, and a parameter identification module; The error calculation module receives the measured SOC fed back by the BMS and the predicted SOC output by the solver module, calculates the SOC prediction error, and outputs the prediction error to the threshold judgment module; the threshold judgment module compares the prediction error with the preset threshold, and triggers the parameter identification module when the prediction error exceeds the threshold; The parameter identification module receives the measured SOC, charge-discharge power and loss data at the recent multiple time points, re-identifies the charge-discharge efficiency and the loss coefficient, and outputs the corrected parameters to the MILP modeling engine.

10. The system of claim 6, wherein, The first communication bus is a CAN bus or an RS485 bus, the second communication bus is a ModbusTCP bus or a CAN bus, and the battery pack bus is a CAN bus; a master-slave communication architecture is adopted between the edge computing power gateway, the BMS and the PCS, the edge computing power gateway acts as a master station to periodically request SOC data from the BMS and issue power instructions to the PCS, and the BMS and the PCS act as slave stations to respond to the request of the master station.

Images