A battery and high-energy consumption load state evolution energy system scheduling method

Abstract

This application discloses an energy system scheduling method based on the state evolution of batteries and high-energy-consumption loads, belonging to the field of energy scheduling. The method includes: acquiring and correcting historical operating data within the park to obtain corrected data; obtaining ultra-short-term forecast data from the corrected data; establishing an optimization-level forecast model based on the ultra-short-term forecast data and obtaining the optimal forecast sequence; obtaining the forecast value for the next time step based on the first value of the optimal forecast sequence, and updating the decision variables based on the forecast value for the next time step to obtain updated decision variables; inputting the updated decision variables into the actual-level model and comparing them with the actual SOC of the energy system and the actual operating temperature of the aluminum electrolysis process to obtain comparison results; and updating the optimization-level forecast model based on the comparison results to obtain a new optimal forecast sequence. This method solves the problem in existing technologies of failing to control operational deviations during energy system scheduling.

Description

Technical Field

[0001] This application belongs to the field of energy dispatching technology, specifically, it relates to an energy system dispatching method for the state evolution of batteries and high-energy-consuming loads. Background Technology

[0002] With the increasing penetration rate of renewable energy, industrial parks face growing operational uncertainties. Existing scheduling methods based on Model Predictive Control (MPC) primarily utilize predictive information to update optimization decisions, but they fail to adequately consider the state deviations caused by the mismatch between scheduling commands and actual equipment responses, and the cumulative process of these deviations across multiple time scales. Such deviations can lead to battery state of charge (SOC) drift, thermal state shifts in electrolytic aluminum loads, and reduced executability of scheduling schemes.

[0003] Therefore, how to effectively control operational deviations based on the rational scheduling of the energy system is a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0004] To address the problem of unreasonable energy system scheduling in existing technologies, this application discloses an energy system scheduling method based on the state evolution of batteries and high-energy-consuming loads, specifically: An energy system scheduling method for battery and high-energy-consumption load state evolution, the scheduling method comprising: Obtain historical operational data within the park and correct it to obtain the corrected data; Based on the time interval of the intraday scheduling layer, ultra-short-term forecast data is obtained from the corrected data; An optimized prediction model is established based on the ultra-short-term prediction data, and the optimal prediction sequence is obtained. Based on the first value of the optimal prediction sequence, the predicted value for the next time step is obtained, and the predicted value for the next time step is used to update the decision variables to obtain the updated decision variables. The updated decision variables are input into the actual layer model and compared with the actual SOC of the energy system and the actual execution temperature of the electrolytic aluminum process to obtain the comparison results respectively. Based on the comparison results, the prediction model of the optimization layer is updated to obtain a new optimal prediction sequence.

[0005] Optionally, the step of acquiring and correcting historical operational data within the park to obtain corrected data includes: Obtain the energy system structure within the park to determine the energy types within the park; Historical operating data for all energy types are rolled over and corrected according to a specific time period to obtain corrected data.

[0006] Optionally, obtaining ultra-short-term forecast data from the corrected data based on the time interval of the intraday scheduling layer includes: Set the time interval of the intraday scheduling layer, wherein the time interval of the intraday scheduling layer is not longer than the specific time period length of the data correction process after correction. Based on the time interval of the intraday scheduling layer, the corresponding corrected data is obtained from the time axis of the corrected data to obtain ultra-short-term forecast data.

[0007] Optionally, the step of establishing an optimized prediction model based on the ultra-short-term prediction data and obtaining the optimal prediction sequence includes: Based on the aforementioned ultra-short-term forecast data, a day-ahead economic model for the energy system is obtained, with the following equation: , Among them, J DA Represents the day-ahead economic value of the energy system; C grid C represents the total cost of purchasing electricity from the external network; bat Indicates the total operating cost of energy storage; C PV / WT Indicates the total generation cost of distributed power units; C curtail This indicates the losses caused by wind and solar power curtailment; C Al This indicates the total compensation cost for peak shaving in electrolytic aluminum production; Obtain parameter constraints in the energy system and determine parameters based on these constraints; Based on the constraints of high-energy-consuming loads, grid economics, and battery energy storage regulation in the energy system, the comprehensive objective function for intraday rolling optimization is obtained as follows: , Where J represents the intraday rolling optimization comprehensive target; J op J represents the operating cost of the power grid; al This indicates the adjustment cost of high energy consumption; J smooth Indicates the constraint cost of battery energy storage regulation; J track This indicates the tracking items for the battery energy storage system planned one day in advance; J act J represents a penalty item for high-energy-consuming actions; ter Indicates the terminal dynamic penalty item; J slack Indicates a relaxation variable penalty term; Based on the day-ahead economic model and the intraday rolling optimization comprehensive objective function, the optimal forecast sequence for the day is obtained.

[0008] Optionally, the step of obtaining parameter constraints in the energy system and determining parameters based on these constraints includes: The parameter constraints include: The power balance constraint is given by the following equation: , in, This represents the photovoltaic power generation at time t; This represents the wind power output at time t; This represents the net power of the battery at time t; This represents the power exchange between the power grids at time t. This represents the electrolytic aluminum load power at time t; Indicates the power of other loads at time t; The power grid constraints are given by the following equations: in, Indicates the maximum power exchanged between the power grids; New energy constraints, the equation is: in, Indicates the maximum power output of photovoltaic power generation; Indicates the maximum power output of the wind turbine; Battery energy storage constraints include: The charge / discharge constraint equation is as follows: in, Indicates the maximum discharge power of the battery; Indicates the maximum charging power of the battery; The SOC state transition constraint equation is: Where SOC(t+1) represents the SOC state parameter at time t+1; SOC(t) represents the SOC state parameter at time t. Indicates the time difference; S bat Indicates battery capacity; η represents battery charge / discharge efficiency; P bat (t) represents the battery power at time t; The initial SOC constraint is given by the following equation: Among them, SOC int Represents the initial SOC; SOC end Indicates the end of SOC; The high-energy-consumption load constraint is represented by the following equation: in, This indicates the minimum load for electrolytic aluminum production; P represents the maximum load of electrolytic aluminum;Al (t) represents the electrolytic aluminum load at time t; Indicates the lowest temperature value; Indicates the highest temperature value; T Al (t) represents the temperature of the electrolytic aluminum at time t.

[0009] Optionally, the intraday rolling optimization comprehensive objective function includes: In the intraday rolling optimization comprehensive objective function, the power grid operating cost equation is: Where t represents the t-th scheduling time; N represents the total number of time periods within a day; c grid (t) represents the external grid purchase price of electricity at time t; P grid (t) represents the power exchange between the power grids at time t; Indicates the time step size for each time period; P represents the unit price of the penalty for curtailment of photovoltaic power. pv，curt (t) represents the amount of solar power curtailed at time t; P represents the unit price of wind power curtailment penalty; wt，curt (t) represents the wind power curtailment at time t; c bat P represents the unit operating cost of energy storage; ch (t) represents the energy storage charging power at time t; P dis (t) represents the energy storage discharge power at time t; In the intraday rolling optimization comprehensive objective function, the battery energy storage regulation constraint cost equation is: , Among them, P bat (·) represents the net power of the battery at a certain moment; In the intraday rolling optimization comprehensive objective function, the tracking term equation for the battery energy storage system's plan one day in advance is: , in, This indicates the reference power of energy storage in the day-ahead dispatch plan; In the intraday rolling optimization comprehensive objective function, the high-energy-consumption adjustment cost equation is: Among them, S al e represents the cost compensation coefficient for electrolytic aluminum. al，comp (t) represents the adjustment and compensation error of electrolytic aluminum; w track，al Indicates the tracking penalty weight for electrolytic aluminum; e al，tune (t) represents the electrolytic aluminum adjustment tracking error; In the intraday rolling optimization comprehensive objective function, the high-energy-consumption action penalty term and the terminal dynamic penalty term are respectively expressed as: , , Among them, z al (t) represents the penalty value for the electrolytic aluminum load action at time t; e Tal，terminal This indicates the deviation of the electrolytic aluminum temperature at the end of the predicted time domain; In the intraday rolling optimization comprehensive objective function, the equation for the relaxation variable penalty term is: , in, s represents the slack variable penalty coefficient corresponding to the energy constraint; Emin s represents the lower slack variable representing the energy constraint of the energy storage system; Emax Represents the upper relaxation variable of the energy constraint of the energy storage system; This represents the slack variable penalty coefficient corresponding to the temperature constraint; This represents the lower relaxation variable representing the temperature constraint in the aluminum electrolysis system. This represents the upper relaxation variable of the temperature constraint in the aluminum electrolysis system.

[0010] Optionally, the step of obtaining the predicted value for the next time step based on the first value of the optimal prediction sequence, and updating the decision variables using the predicted value for the next time step to obtain the updated decision variables, includes: Based on the time point corresponding to the optimal prediction sequence and the interval between the next time point and the corresponding time point, and based on the optimization layer prediction model, the predicted value for the next time point is obtained. The predicted value for the next moment is used to replace the predicted value for the next moment in the optimal prediction sequence to update the decision variables and obtain the updated decision variables.

[0011] Optionally, the step of inputting the updated decision variables into the actual layer model and comparing them with the actual SOC of the energy system and the actual execution temperature of the electrolytic aluminum process to obtain comparison results includes: The updated decision variables are input into the actual layer model, and the actual execution SOC of the energy system and the actual execution temperature of the electrolytic aluminum process are obtained respectively. The actual SOC of the energy system and the updated SOC of the decision variables are compared to obtain the SOC comparison result, which is: , in, SOC t+1|t This represents the difference between the actual execution SOC and the optimal decision SOC; a t e represents the perturbation coefficient at time t; L，t (·) represents the tracking error on the left side at time t; (eR，t ) — This represents the portion of the tracking error on the right side at time t that is less than 0. The actual operating temperature of the aluminum electrolysis process in the energy system is compared with the updated decision variable of the aluminum electrolysis process operating temperature to obtain the temperature comparison result, which is: , in, This represents the temperature deviation predicted at time t+1 from the predicted temperature at time t; cm is the equivalent thermal inertia; k real，t k represents the actual equivalent thermal coefficient at time t; plan Indicates the planned heat coefficient; T represents the actual temperature of the electrolytic cell at time t; T0 represents the rated temperature.

[0012] Optionally, updating the optimization layer prediction model based on the comparison results to obtain a new optimal prediction sequence includes: The temperature comparison result and the SOC comparison result in the comparison results are respectively compared with the corresponding thresholds to obtain the threshold comparison result; When the threshold comparison result is not less than 0, the actual SOC of the obtained energy system and the actual execution temperature of the electrolytic aluminum process are recalculated, and a new optimal prediction sequence is generated.

[0013] The beneficial effects of this application are as follows: (1) An execution-level state deviation model is established to quantitatively describe the inconsistency between the planned trajectory and the actual operating state. For battery energy storage systems, the model established in this application describes the SOC execution deviation caused by the uncertainty of new energy output and load prediction errors. For the electrolytic aluminum production process, the model further characterizes the temperature deviation caused by the dynamic characteristics of the industrial process and the uncertainty of the model. Through the above modeling, a state-level description method for the planning-execution inconsistency in the scheduling process of high-energy-consuming industrial parks is established.

[0014] (2) A cross-timescale state deviation propagation mechanism was established to reveal the accumulation and propagation law of battery SOC and electrolytic aluminum cell temperature execution deviation in the rolling scheduling process.

[0015] Further analysis of the relationship between execution deviation and scheduling correction provides a theoretical explanation of why the planned trajectory and the actual operating state gradually deviate under open-loop scheduling mode, and why the closed-loop state correction mechanism can effectively reduce the accumulation of execution deviation.

[0016] (3) A planning-execution-feedback-re-optimization framework integrating day-ahead scheduling, intraday MPC, and execution-level state feedback is proposed. This framework incorporates the actual operating states of the battery energy storage system and the electrolytic aluminum production process into the rolling optimization process, enabling adaptive correction of scheduling decisions. This method improves the consistency between the planned operating trajectory and the actual operating state while ensuring the operating constraints of the energy storage system and the industrial production process. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the accompanying drawings used in the embodiments of this application or the prior art will be briefly introduced below. Obviously, the following description is only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. The drawings are used to provide a further understanding of this disclosure and constitute a part of the specification. They are used together with the following detailed description to explain this disclosure, but do not constitute a limitation of this disclosure. In the drawings: Figure 1 A flowchart of an energy system scheduling method for battery and high-energy-consumption load state evolution provided in this application embodiment; Figure 2 A schematic diagram of an industrial park energy system configuration in an energy system scheduling method for battery and high-energy-consumption load state evolution provided in an embodiment of this application; Figure 3 A schematic diagram of intraday time period configuration in an energy system scheduling method for battery and high-energy-consumption load state evolution provided in an embodiment of this application; Figure 4 A schematic diagram of the actual execution layer modeling framework in an energy system scheduling method for battery and high-energy-consumption load state evolution provided in this application embodiment; Figure 5 A smoothed heat dissipation coefficient diagram for temperature updating in an energy system scheduling method for battery and high-energy-consumption load state evolution provided in an embodiment of this application; Figure 6 A flowchart of the intraday / real-time scheduling process in an energy system scheduling method for battery and high-energy-consumption load state evolution provided in this application embodiment; Figure 7 A time-of-use electricity price map is provided in an energy system scheduling method for battery and high-energy-consumption load state evolution according to an embodiment of this application; Figure 8 A diagram showing the advance scheduling results of electrolytic aluminum load in an energy system scheduling method for battery and high-energy-consumption load state evolution provided in an embodiment of this application; Figure 9A comparison chart of the predetermined and actual trajectories of SOC and electrolytic aluminum temperature under conventional MPC in an energy system scheduling method for battery and high-energy-consumption load state evolution provided in this application embodiment: Wherein, (a) represents a comparison diagram of the predetermined trajectory and the actual trajectory of SOC under conventional MPC; (b) shows a comparison between the predetermined trajectory and the actual trajectory of the electrolytic aluminum temperature under conventional MPC. Detailed Implementation

[0018] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application. Furthermore, in the embodiments of this application, "first," "second," etc., are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0019] This application proposes an Execution-State-Feedback Multi-Timescale Scheduling (ESF-MTS) framework for high-proportion new energy industrial parks. This framework coordinates day-ahead scheduling, intraday rolling optimization, and the actual execution process, and establishes execution-level battery SOC deviation models caused by new energy output and load forecast errors, as well as electrolytic aluminum temperature deviation models caused by thermal model uncertainties. By periodically feeding back the actual SOC state and electrolytic aluminum temperature state, the initial state and scheduling decisions in the subsequent rolling optimization process are updated. This framework can reduce the continuous accumulation of execution deviations and improve the consistency and executability between the scheduling plan and the actual operating state. Case analysis results under different uncertainty levels show that ESF-MTS can effectively limit the propagation of state deviations and keep the equipment state within the allowable operating range. Compared with the traditional MPC method, this method can reduce the maximum SOC deviation by 89.1%–92.4% and the maximum electrolytic aluminum temperature deviation by 51.6%. Furthermore, this framework does not increase battery energy throughput, and its computation time can meet the requirement of a 15-minute intraday rolling optimization cycle. The results validate the effectiveness of the proposed method in improving planning-execution consistency and scheduling executability.

[0020] like Figure 1 The diagram shown is a flowchart of an energy system scheduling method for battery and high-energy-consumption load state evolution provided in an embodiment of this application, specifically including: S110. Obtain historical operational data within the park and correct it to obtain the corrected data.

[0021] S120. Based on the time interval of the intraday scheduling layer, obtain ultra-short-term forecast data from the corrected data.

[0022] S130. Based on the ultra-short-term forecast data, establish an optimized layer forecast model and obtain the optimal forecast sequence.

[0023] S140. Based on the first value of the optimal prediction sequence, obtain the predicted value for the next moment, and update the decision variables using the predicted value for the next moment to obtain the updated decision variables.

[0024] S150. Input the updated decision variables into the actual layer model and compare them with the actual SOC of the energy system and the actual execution temperature of the electrolytic aluminum process to obtain the comparison results respectively.

[0025] S160. Based on the comparison results, the prediction model of the optimization layer is updated to obtain a new optimal prediction sequence.

[0026] The process of acquiring and correcting historical operational data within the park to obtain corrected data includes: Obtain the energy system structure within the park to determine the energy types within the park; Historical operating data for all energy types are rolled over and corrected according to a specific time period to obtain corrected data.

[0027] The ultra-short-term forecast data obtained from the corrected data based on the time interval of the intraday scheduling layer includes: Set the time interval of the intraday scheduling layer, wherein the time interval of the intraday scheduling layer is not longer than the specific time period length of the data correction process after correction. Based on the time interval of the intraday scheduling layer, the corresponding corrected data is obtained from the time axis of the corrected data to obtain ultra-short-term forecast data.

[0028] The step of establishing an optimized prediction model based on the ultra-short-term prediction data and obtaining the optimal prediction sequence includes: Based on the aforementioned ultra-short-term forecast data, a day-ahead economic model for the energy system is obtained, with the following equation: , Among them, J DA Represents the day-ahead economic value of the energy system; C grid C represents the total cost of purchasing electricity from the external network; bat Indicates the total operating cost of energy storage; C PV / WT Indicates the total generation cost of distributed power units; C curtail This indicates the losses caused by wind and solar power curtailment; C AlThis indicates the total compensation cost for peak shaving in electrolytic aluminum production; Obtain parameter constraints in the energy system and determine parameters based on these constraints; Based on the constraints of high-energy-consuming loads, grid economics, and battery energy storage regulation in the energy system, the comprehensive objective function for intraday rolling optimization is obtained as follows: , Where J represents the intraday rolling optimization comprehensive target; J op J represents the operating cost of the power grid; al This indicates the adjustment cost of high energy consumption; J smooth Indicates the constraint cost of battery energy storage regulation; J track This indicates the tracking items for the battery energy storage system planned one day in advance; J act J represents a penalty item for high-energy-consuming actions; ter Indicates the terminal dynamic penalty item; J slack Indicates a relaxation variable penalty term; Based on the day-ahead economic model and the intraday rolling optimization comprehensive objective function, the optimal forecast sequence for the day is obtained.

[0029] The process of acquiring parameter constraints in the energy system and determining parameters based on these constraints includes: The parameter constraints include: The power balance constraint is given by the following equation: , in, This represents the photovoltaic power generation at time t; This represents the wind power output at time t; This represents the net power of the battery at time t; This represents the power exchange between the power grids at time t. This represents the electrolytic aluminum load power at time t; Indicates the power of other loads at time t; The power grid constraints are given by the following equations: in, Indicates the maximum power exchanged between the power grids; New energy constraints, the equation is: in, Indicates the maximum power output of photovoltaic power generation; Indicates the maximum power output of the wind turbine; Battery energy storage constraints include: The charge / discharge constraint equation is as follows: in, Indicates the maximum discharge power of the battery; Indicates the maximum charging power of the battery; The SOC state transition constraint equation is: Where SOC(t+1) represents the SOC state parameter at time t+1; SOC(t) represents the SOC state parameter at time t. Indicates the time difference; S bat Indicates battery capacity; η represents battery charge / discharge efficiency; P bat (t) represents the battery power at time t; The initial SOC constraint is given by the following equation: Among them, SOC int Represents the initial SOC; SOC end Indicates the end of SOC; The high-energy-consumption load constraint is represented by the following equation: in, This indicates the minimum load for electrolytic aluminum production; P represents the maximum load of electrolytic aluminum; Al (t) represents the electrolytic aluminum load at time t; Indicates the lowest temperature value; Indicates the highest temperature value; T Al (t) represents the temperature of the electrolytic aluminum at time t.

[0030] The intraday rolling optimization comprehensive objective function includes: In the intraday rolling optimization comprehensive objective function, the power grid operating cost equation is: Where t represents the t-th scheduling time; N represents the total number of time periods within a day; c grid (t) represents the external grid purchase price of electricity at time t; P grid (t) represents the power exchange between the power grids at time t; Indicates the time step size for each time period; P represents the unit price of the penalty for curtailment of photovoltaic power. pv，curt (t) represents the amount of solar power curtailed at time t; P represents the unit price of wind power curtailment penalty; wt，curt (t) represents the wind power curtailment at time t; c bat P represents the unit operating cost of energy storage; ch (t) represents the energy storage charging power at time t; P dis (t) represents the energy storage discharge power at time t; In the intraday rolling optimization comprehensive objective function, the battery energy storage regulation constraint cost equation is: , Among them, P bat (·) represents the net power of the battery at a certain moment; In the intraday rolling optimization comprehensive objective function, the tracking term equation for the battery energy storage system's plan one day in advance is: , in, This indicates the reference power of energy storage in the day-ahead dispatch plan; In the intraday rolling optimization comprehensive objective function, the high-energy-consumption adjustment cost equation is: Among them, S al e represents the cost compensation coefficient for electrolytic aluminum. al，comp (t) represents the adjustment and compensation error of electrolytic aluminum; w track，al Indicates the tracking penalty weight for electrolytic aluminum; e al，tune (t) represents the electrolytic aluminum adjustment tracking error; In the intraday rolling optimization comprehensive objective function, the high-energy-consumption action penalty term and the terminal dynamic penalty term are respectively expressed as: , , Among them, z al (t) represents the penalty value for the electrolytic aluminum load action at time t; e Tal，terminal This indicates the deviation of the electrolytic aluminum temperature at the end of the predicted time domain; In the intraday rolling optimization comprehensive objective function, the equation for the relaxation variable penalty term is: , in, s represents the slack variable penalty coefficient corresponding to the energy constraint; Emin s represents the lower slack variable representing the energy constraint of the energy storage system; Emax Represents the upper relaxation variable of the energy constraint of the energy storage system; This represents the slack variable penalty coefficient corresponding to the temperature constraint; This represents the lower relaxation variable representing the temperature constraint in the aluminum electrolysis system. This represents the upper relaxation variable of the temperature constraint in the aluminum electrolysis system.

[0031] The step of obtaining the predicted value for the next time step based on the first value of the optimal prediction sequence, and updating the decision variables using the predicted value for the next time step to obtain the updated decision variables, includes: Based on the time point corresponding to the optimal prediction sequence and the interval between the next time point and the corresponding time point, and based on the optimization layer prediction model, the predicted value for the next time point is obtained. The predicted value for the next moment is used to replace the predicted value for the next moment in the optimal prediction sequence to update the decision variables and obtain the updated decision variables.

[0032] The updated decision variables are input into the actual layer model and compared with the actual SOC of the energy system and the actual operating temperature of the aluminum electrolysis process to obtain comparison results, including: The updated decision variables are input into the actual layer model, and the actual execution SOC of the energy system and the actual execution temperature of the electrolytic aluminum process are obtained respectively. The actual SOC of the energy system and the updated SOC of the decision variables are compared to obtain the SOC comparison result, which is: , in, SOC t+1|t This represents the difference between the actual execution SOC and the optimal decision SOC; a t e represents the perturbation coefficient at time t; L，t (·) represents the tracking error on the left side at time t; (e R，t ) — This represents the portion of the tracking error on the right side at time t that is less than 0. The actual operating temperature of the aluminum electrolysis process in the energy system is compared with the updated decision variable of the aluminum electrolysis process operating temperature to obtain the temperature comparison result, which is: , in, This represents the temperature deviation predicted at time t+1 from the predicted temperature at time t; cm is the equivalent thermal inertia; k real，t k represents the actual equivalent thermal coefficient at time t; plan Indicates the planned heat coefficient; T represents the actual temperature of the electrolytic cell at time t; T0 represents the rated temperature.

[0033] The step of updating the optimization layer prediction model based on the comparison results to obtain a new optimal prediction sequence includes: The temperature comparison result and the SOC comparison result in the comparison results are respectively compared with the corresponding thresholds to obtain the threshold comparison result; When the threshold comparison result is not less than 0, the actual SOC of the obtained energy system and the actual execution temperature of the electrolytic aluminum process are recalculated, and a new optimal prediction sequence is generated.

[0034] To provide a clearer understanding of the technical features of this application, the following practical examples will be used to explain and illustrate its content: like Figure 2 The diagram shown is a schematic of the energy system configuration in an industrial park within an energy system scheduling method for battery and high-energy-consumption load state evolution, as provided in an embodiment of this application. It defines the energy system configuration within the industrial park, including a 150MW photovoltaic power generation capacity, a 100MW wind power generation capacity, and a centralized battery energy storage system with a rated power of 60MW / 4h. The total load of the industrial park is approximately 110MW, of which electrolytic aluminum accounts for about 80%. To improve energy independence, support near-zero carbon operation, and minimize dependence on external coal-fired power generation, multiple substations are set up between the energy system and the loads to ensure that various loads within the park can be powered by the energy system. The 100MW photovoltaic power generation unit, 50MW photovoltaic power generation unit, and 100MW wind power generation unit are processed via 220kV substation 2, while the nearby 6*330MW thermal power plant is processed via 220kV substation 1. The electricity processed by 220kV substations 1 and 2 is then sent to 220kV substation 3. The electricity generated by the 60MW / 4h battery energy storage system is also processed by 220kV substation 3. Substation 3 is the core substation within the industrial park. After processing at substation 3, the electricity is sent to 66kV substations 1, 2, and 3. Substation 1 supplies power to load group 1, substation 2 to load group 2, and substation 3 to load group 3. It is important to note that a nearby thermal power plant (6×330MW) primarily serves other industrial parks, and the system under study cannot output electricity there. Therefore, when renewable energy production exceeds local demand, excess wind and solar power generation must be reduced.

[0035] This application constructs a multi-timescale coordination and scheduling framework, such as... Figure 3 The diagram illustrates the intraday time-period configuration in an energy system scheduling method for battery and high-energy-consumption load state evolution provided in this application embodiment. In an industrial park energy system, renewable power generation and load demand are significantly affected by uncertainties. Simultaneously, the system's operating state dynamically evolves over time. Under these circumstances, single-time-scale scheduling methods cannot simultaneously guarantee global optimality and real-time feasibility. Therefore, this paper establishes a multi-time-scale scheduling framework, which consists of three layers: day-ahead scheduling, intraday rolling optimization, and the execution layer.

[0036] The system employs an operational plan based on source load forecasting, providing a baseline scheduling scheme. The intraday layer utilizes an MPC strategy. It performs rolling corrections to the previous day's plan at a higher resolution of 15 minutes to address forecast errors and operational disturbances. Unlike traditional predictive control where feedback information is primarily used to update the optimization state and forecast information within the forecast range, this framework focuses on the inconsistencies between the planned trajectory and the execution state. Deviations caused by the execution of critical states, including battery SOC and electrolytic aluminum tank temperature, are fed back to the optimization layer for subsequent decision adjustments.

[0037] The execution layer operates at a 5-minute resolution, reflecting the actual operation of the system. Due to equipment dynamics and control delays, the achieved operational results may deviate from the predetermined commands. Within this multi-timescale framework, the scheduling results from the previous day provide a reference for intraday optimization, and the intraday layer generates control commands for the execution layer. However, existing multi-timescale scheduling methods typically focus on updating optimization decisions based on the latest information, while the inconsistency between the scheduling trajectory and the actual execution state is rarely explicitly characterized. Therefore, the cumulative evolution of state deviations, such as SOC drift and electrolytic aluminum temperature deviations, may not be effectively addressed. This can lead to inconsistencies between the scheduling plan and actual execution. It can further affect SOC tracking accuracy, temperature control of high-energy-intensive loads, and system operational stability. Therefore, a multi-timescale closed-loop scheduling method that combines state evolution and execution feedback is needed.

[0038] After obtaining the above basic information, a day-ahead economic model is constructed, which is obviously constrained and purposeful, with the goal of minimizing total operating costs through advance scheduling. Based on advance forecasts of renewable energy generation and load demand, factors such as electricity price, load characteristics, energy storage costs, and five reliability indicators are comprehensively considered. This is achieved by analyzing the charging and discharging power of the energy storage system, the output of controllable micro-sources, electricity purchased from seven external power grids, and the power consumption of electrolytic aluminum.

[0039] Optimization was performed to achieve the lowest operating cost for the eight-energy storage system. The objective function is expressed as follows: (1) The previous day's model provides an economic reference trajectory, while subsequent intraday and execution layers focus on improving the consistency between this reference trajectory and actual operations. The previous day's optimal scheduling model is subject to the following constraints: (1) Power balance constraint (2) (2) Power grid constraints (3) (3) New energy constraints (4) (4) Energy storage constraints These constraints mainly include charge / discharge power limits, energy capacity limits, SOC state transition constraints, SOC slope constraints, and 25-terminal SOC constraints.

[0040] 1) Charge and discharge constraints (5) 2) SOC state transition (6) 3) Initial SOC (7) The scheduling period should be as close as possible to the initial time period so that the energy storage system can meet the operating requirements of the next day.

[0041] (5) Electrolytic aluminum constraint (8) (9) This requires determining the intraday MPC scheduling model fed back from the layer to be executed, specifically: (1) Establishment of intraday scheduling objective function.

[0042] To balance the economic efficiency of microgrid operation, stable energy storage regulation, and the process constraints of flexible electrolytic aluminum load, the comprehensive objective function formula for intraday rolling optimization is as follows: (10) Wherein, J represents the intraday rolling optimization comprehensive target; J op J represents the operating cost of the power grid; al This indicates the adjustment cost of high energy consumption; J smooth Indicates the constraint cost of battery energy storage regulation; J track This indicates the tracking items for the battery energy storage system planned one day in advance; J act J represents a penalty item for high-energy-consuming actions; ter Indicates the terminal dynamic penalty item; J slack This represents a relaxation variable penalty term.

[0043] The operating cost of a microgrid is defined as follows: (11) It includes the cost of electricity purchase, the punitive costs of wind and solar power cuts, and the operating costs of the energy storage system. Considering that frequent and aggressive charging and discharging of the energy storage system may affect battery life and system stability, an additional smoothing term is introduced for the net power output of the energy storage system: (12) To maintain consistency between the intraday rolling plan and the previous day's plan, a tracking penalty clause is imposed on deviations of energy storage power from the previous day's reference: (13) In addition, the adjustment cost of electrolytic aluminum is defined to reflect the deviation cost generated during load adjustment: (14) To avoid frequent changes in the electrolytic aluminum load and to ensure that the process state meets the required conditions at the end of the prediction range, action penalty terms and terminal state penalty terms are introduced respectively: (15) (16) Finally, to improve the feasibility of the optimization problem under complex operating conditions, relaxation variables were introduced into the energy and temperature constraints, along with a large penalty factor: (17) Unlike traditional rolling optimization methods that use fixed or predicted initial states during the optimization process, the proposed framework uses measured states from the execution layer to periodically update the initial states of intraday optimization. Therefore, the optimization trajectory can continuously align with the actual operating state of the system. Definitions of other symbols are provided in Appendix A. Through the aforementioned objective function, intraday scheduling of the microgrid can be optimized in terms of economy, stability, and feasibility while satisfying system power balance, energy storage constraints, and electrolytic aluminum process constraints.

[0044] (2) Modeling of execution layer state deviation Within the closed-loop multi-timescale scheduling framework, the intraday scheduling layer generates rolling optimization results based on MPC. These results serve as reference commands for the actual execution layer. At the beginning of each intraday rolling interval, the scheduling layer uses the current system state and ultra-short-term forecast information to solve for the optimal control sequence within the forecast range. Decision variables include the charging and discharging power of the energy storage system, the power purchased by the grid, and the adjustment amount of the flexible electrolytic aluminum load. The actual execution layer only implements the pre-control steps. The remaining control actions are re-optimized in the next rolling interval using updated state information, such as... Figure 4 The diagram shown is a schematic of the actual execution layer modeling framework in an energy system scheduling method for battery and high-energy-consumption load state evolution provided in this application embodiment. The actual execution layer operates at a 5-minute resolution. It represents the implementation of the intraday schedule under actual operating conditions and captures the evolution of key states. Actual wind power generation, photovoltaic power generation, and load demand are generated by introducing time-related prediction errors into ultra-short-term forecasts. Specifically, an autoregressive first-order (AR(1)) model is used to characterize the time correlation of prediction bias: (18) Among them, e k k represents the relative prediction error at the time step. The time correlation coefficient represents 1. This is a zero-mean Gaussian innovation term. Wind power, photovoltaic power generation, and load demand are assigned different correlation coefficients and error variances based on their varying degrees of uncertainty. Assuming grid purchases initially follow intraday planning, the resulting power imbalance is calculated based on actual renewable energy and load conditions. Subsequently, the imbalance is compensated through the execution logic of the energy storage system and grid interaction.

[0045] This imbalance is initially absorbed by the energy storage system. Its actual charging and discharging power is limited by its rated power, and the State of Charge (SOC) is updated based on the actual operating power. To reflect the impact of prediction errors and execution deviations on the storage state under open-loop execution, the actual execution layer employs a wider SOC threshold range than the intraday scheduling layer. Correction is only activated when the SOC exceeds a specified boundary. Any remaining sub-power deviations are then addressed through grid compensation or wind and solar power reduction, thereby maintaining the system's power balance.

[0046] The actual execution layer primarily follows the wind and solar grid connection reduction plans issued by the intraday dispatch layer. Updateable forecast deviations are handled asymmetrically. When actual output exceeds the planned value, the excess is reduced. When actual output falls below the planned value, the deficit is first compensated by the energy storage system. Any remaining shortfall is covered by thermal power generation. For electrolytic aluminum load, the actual execution layer follows dispatch instructions determined by the intraday dispatch layer. According to the discrete-time model, the temperature state evolves with power variations. This reflects the impact of execution on the process state.

[0047] This also requires analysis of the evolution of state deviations, specifically: (1) Battery energy storage system During intraday scheduling, the planned SOC trajectory of the energy storage system is updated based on charge / discharge power commands issued by the planning layer. In contrast, at the actual execution layer, storage power is dynamically adjusted based on real-time source load conditions and wind and solar PV reduction rules. As a result, a discrepancy arises between actual and planned storage power, causing the actual SOC trajectory to deviate from the planned trajectory. This mismatch primarily stems from the propagation of source-load forecasting errors through storage power adjustments and SOC evolution. The analysis in this paper follows the fundamental mechanism identified in previous studies: forecasting errors propagate through storage power and accumulate in the SOC state. Unlike existing methods, the actual execution layer considered in this paper explicitly incorporates the control logic for wind and solar grid connections. When actual renewable energy production exceeds a predetermined value, the surplus is first reduced. When actual production falls below the planned value, the deficit is compensated through water storage discharge. Based on the execution mechanism, a recursive relation is derived to describe how the deviation caused by execution accumulates in the SOC evolution.

[0048] Load forecasting error and updatable forecasting error are defined as... (19) (20) The impact of renewable energy prediction errors on batteries is asymmetric. When When actual renewable power generation exceeds planned levels, the remaining power will be prioritized for reduction. At that time, actual renewable energy production is lower than planned, and the shortfall can be compensated for by increasing battery discharge. (Definition) (twenty one) Then, the deviation between the actual battery power and the predetermined battery power 2 can be expressed as: (twenty two) The SOC recursion of the planning layer is given by the following formula. (twenty three) The actual SOC recursion at the execution layer is given by the following formula. (twenty four) in, This represents the power-energy mapping function that considers charging and discharging efficiency. SOC deviation is defined as... (25) (26) When no switching occurs between charging and discharging states within a single time interval It can be linearized as follows: (27)

[0049] Therefore, the SOC bias in open-loop mode can be further written: (28) As can be seen from the expression above, the mismatch between the actual and planned states stems from load forecasting errors and renewable energy under-generation errors, which gradually accumulate over time through battery power deviations. Excess renewable generation does not generate symmetrical compensation in the battery's State of Charge (SOC) because it is handled through reduction. The formula indicates that the SOC deviation develops recursively and can accumulate over time due to deviations caused by execution, reflecting the impact of actual execution errors on the storage state.

[0050] (2) Electrolytic aluminum process Electrolyzer temperature was chosen as the feedback state because it directly reflects the thermal equilibrium of the electrolysis process and determines the feasible operating range of power regulation. Compared to electrical variables, temperature has a slower dynamic response and stronger memory characteristics, making it a key state variable for evaluating execution consistency. In this study, the purpose of the temperature model is not to reproduce the detailed electrochemical and thermal field distribution within the industrial cell; instead, it is used as an equivalent execution layer model to capture state deviations caused by model uncertainties and dynamic thermal responses. Although the actual execution layer of the aluminum electrolysis process can strictly follow the power or current schedule allocated within the day, the electrolyzer temperature may still deviate from the predicted intraday trajectory. This is mainly because the cell temperature is a slowly changing thermal equilibrium state. Its evolution is influenced not only by power input but also by a variety of factors, including wall thickness, top cover material / shell, endothermic effect of alumina feed, local electrothermal processes in the interelectrode region, and thermal inertia. Furthermore, intraday scheduling models typically employ simplified first-order linear models or lumped thermal capacitance approximations, while actual electrothermal coupling, heat dissipation coefficients, and equivalent thermal capacitance within the cell vary with power and operating conditions. Therefore, even when the execution layer operates exactly as planned, the actual temperature may still deviate from the intraday forecast. In the actual execution layer considered in this paper, the temperature evolution of the aluminum electrolytic cell is described by the following discrete update equation: (29) Among them, T k and T k+1 P represents the battery temperature at the current and next time steps, respectively. act This refers to the actual electrical power delivered; P0 and T0 are the rated operating points; k target The target heat dissipation coefficient is calculated based on the fixed bias term, power-related term, and bias correction factor; k real It is the smoothed thermal coefficient used for temperature updates, obtained by eliminating abrupt changes through exponential smoothing; a k It is the smoothing coefficient; cm This refers to the equivalent thermal inertia. Based on the thermal balance characteristics of industrial aluminum electrolytic cells and previous research on thermal models of aluminum electrolysis, thermal model parameters were selected. Specifically, the rated power and operating temperature were determined based on the equivalent thermal inertia and heat dissipation coefficient of typical large-scale aluminum electrolysis production conditions, representing the slow thermal response characteristic of the aluminum electrolysis production process. These parameters were used to construct an equivalent execution layer model for deviation analysis, rather than reproducing the detailed internal thermal field of a specific industrial unit. The basic idea of ​​this model is that the heat dissipation coefficient dynamically changes with power and operating conditions, thus reflecting the actual thermal balance characteristics of the electrolytic cell. Even if the actual execution strictly follows the intraday planned input, the temperature may still deviate from the intraday prediction because of the introduction of k... real As an equivalent variable parameter, k represents the uncertainty of heat dissipation characteristics under different operating conditions. real The outline is like Figure 5 The diagram shown is a smoothed heat dissipation coefficient diagram for temperature updating in an energy system scheduling method for battery and high-energy-consumption load state evolution provided in an embodiment of this application.

[0051] In open-loop mode, the planned intraday temperature always uses the previously planned value as its initial value, while the temperature update equation for the actual execution layer is given by the following formula. (30) The recursive equation for the planned temperature is given by the following formula. (31) Among them, T tact Indicates that at time t, T tplan Actual battery temperature, t, P tact P represents the planned (predicted) battery temperature over time. tplan This represents the planned power input. Here, P... tact and P tplan They are equal. Temperature error is defined as the difference between the actual temperature and the planned temperature: (32) (33) Equation (33) shows that, under open-loop operation, the upper bound of the temperature error in electrolytic aluminum consists of two parts. One part comes from the recursive propagation of the error from the previous time step. The other part is due to parameter mismatch caused by the inconsistency between the actual heat dissipation coefficient and the planned heat dissipation coefficient. Since the open-loop strategy does not use the actual temperature to correct the initial state of subsequent optimizations, the error continues to propagate during recursion. The magnitude of the temperature deviation is jointly affected by the propagation of the previous state deviation and the mismatch between the actual heat dissipation characteristics and the planned heat dissipation characteristics. In summary, the SOC error of the energy storage system gradually accumulates over time through charge and discharge deviations, while the temperature deviation is mainly affected by thermal inertia and heat dissipation changes, gradually evolving towards a bounded operating range. Therefore, the two systems need to be treated differently according to their dynamic characteristics. By introducing rolling optimization and feedback mechanisms, the accumulation or propagation of SOC and temperature deviations can be effectively suppressed, thereby improving the consistency between the actual operating state and intraday scheduling. It should be noted that the proposed feedback mechanism does not perform online parameter identification of the thermal model. Instead, it compensates for the cumulative effects of model-plant mismatch by periodically correcting the initial state of optimization using measured temperatures.

[0052] The key technical feature of this application is that it checks the execution status of the obtained execution result and makes corresponding adjustments, specifically: like Figure 6 The diagram shows a daily / real-time scheduling process flowchart in an energy system scheduling method for battery and high-energy-consumption load state evolution provided in this application embodiment. An execution-state coordination mechanism is incorporated into a closed-loop multi-timescale scheduling framework. Unlike traditional MPC feedback, which primarily updates the optimized state during the prediction process, the proposed mechanism focuses on coordinating the inconsistencies between the scheduling trajectory and the actual execution state. The daily scheduling layer performs rolling optimization at 15-minute intervals. The actual execution layer executes the first control step at a finer resolution of 5 minutes. At the end of each interval, the latest energy storage SOC and electrolytic aluminum temperature are fed back to the scheduling layer as the initial state for the next optimization cycle. Based on the above execution mechanism, the evolution of SOC and temperature deviations under closed-loop operation is analyzed to clarify how execution-state coordination limits the accumulation of state deviations.

[0053] (1) Battery energy storage system Under the closed-loop strategy, let t represent the rolling optimization time. The initial SOC of the next round of optimization is used as the feedback value from the actual execution layer, i.e., (34) in This represents the initial SOC of the plan at time t when rolling optimization begins. Therefore, at the beginning of each rolling interval... (35) In other words, the state deviation between the planning layer and the actual execution layer is reset to zero at each interval boundary. Within the current rolling interval, the one-step SOC prediction of the planning layer is given by the following formula. (36) The SOC after execution in the actual layer is given by the following formula. (37) Therefore, the SOC deviation at the end of the current rolling interval is (38) By combining it with the execution logic of the actual layer, we can further derive... (39) Under linearization conditions, the above expression can be written as (40) By comparing (28) and (40), it can be seen that under the closed-loop rolling correction strategy, the SOC deviation no longer accumulates continuously across the entire level as it does in the open-loop mode. Instead, it is determined only by the error within the current rolling interval. The fundamental reason is that the actual SOC is fed back to the planning level at the beginning of each rolling interval and used as the new initial state. This prevents the deviation accumulated in previous rolling intervals from being continuously propagated, while the remaining deviation is confined to the current rolling interval. Therefore, this strategy transforms the long-term accumulation of deviations caused by execution in open-loop operation into bounded local deviations within each rolling interval.

[0054] (2) Electrolytic aluminum process Under the closed-loop strategy, the temperature update equation for the actual execution layer of the aluminum electrolysis process is given: (41) In closed-loop scheduling, the initial temperature of the next day's plan is directly used as the actual temperature of the execution layer, that is, (42) Then calculate the planned temperature for the next step. (43) Temperature error is defined as the difference between the actual temperature and the planned temperature: (44) Substituting this into the planned temperature equation of the closed-loop system will produce (45) Take the absolute value of the equation. (44) Further give (46) (47) The error is (48) Equation (47) indicates that, under the closed-loop strategy, the temperature deviation caused by previous interval initialization errors is eliminated through state reset. The remaining deviation is mainly related to the mismatch between the actual heat dissipation characteristics and the predetermined heat dissipation characteristics. It should be noted that the proposed method does not perform online identification of thermodynamic parameters. Instead, it compensates for the effects of model-object mismatch by periodically correcting the optimized initial state using measured temperatures. Overall, the proposed execution-state coordination mechanism mitigates the accumulation of SOC deviations and improves the consistency between the electrolytic aluminum temperature trajectory and its predetermined reference. This enhances the executability and operational stability of multi-timescale scheduling.

[0055] The specific implementation effects of the above content will be explained below to illustrate the beneficial effects of this application. Specifically:

[0056] 1. Case Studies To evaluate the effectiveness of the proposed closed-loop system multi-timescale scheduling method, a simulation case of a typical high-energy-consuming industrial park was constructed. The studied system consists of industrial loads such as renewable energy generation, grid power supply, battery energy storage system, and electrolytic aluminum load. Renewable energy generation is prioritized to meet the park's needs, while the energy storage system is used for energy transfer and mitigating renewable energy fluctuations. The electrolytic aluminum load is modeled as a flexible industrial load with thermal state constraints, where the cell temperature is considered a critical process state to ensure safe operation.

[0057] The optimization problem was solved using the Gurobi algorithm. The thermodynamic parameters of the electrolytic aluminum model were selected based on typical industrial operating characteristics and equivalent thermal model requirements. To evaluate the economic benefits of energy dispatch in the industrial park, time-of-use pricing was introduced to represent the electricity transaction costs between the studied park and neighboring coal-fired power plants. Figure 7 The diagram shown illustrates a time-of-use electricity price chart in an energy system scheduling method for battery and high-energy-consumption load state evolution, provided in an embodiment of this application. The electricity price consists of three periods: peak, off-peak, and flat periods. A typical industrial electricity price profile was selected. Price changes provide an economic incentive for energy storage systems to perform energy transfer, i.e., charging during low-price periods and discharging during high-price periods. Simultaneously, tariff signals are coordinated with regulations on renewable and flexible electrolytic aluminum loads to reduce the overall operating costs of the industrial park.

[0058] To evaluate the robustness of the proposed method under different execution conditions, three uncertainty scenarios were considered, including low, medium, and high levels of uncertainty. Based on the uncertainty model, actual wind power, photovoltaic power generation, and load demand profiles were generated for each scenario. Different uncertainty intensities were obtained by adjusting the standard deviations of seven forecast error processes. The robustness of the proposed closed-loop scheduling strategy under different renewable power generation and load uncertainties was evaluated using these eight scenarios. Detailed parameters for the uncertainty scenarios are shown in Table 1.

[0059] Table 1 shows the parameters for uncertain scenarios.

[0060]

[0061] 2. Typical daily scheduling results under medium uncertainty scenarios

[0062] like Figure 8 The figure shown illustrates the one-day advance scheduling results of the electrolytic aluminum load in an energy system scheduling method for battery and high-energy-consumption load state evolution provided in this application embodiment. It displays the optimal scheduling result for the previous day under uncertain scenarios. The optimization model considers factors such as renewable energy generation forecasts, electricity prices, and load flexibility to determine a 24-hour scheduling strategy. The electrolytic aluminum load can be flexibly adjusted according to system operating conditions, while simultaneously meeting power, temperature, and energy constraints. The energy storage system provides power balance support through charging and discharging operations, while the electrolytic aluminum load utilizes its process flexibility and thermal inertia to achieve load transfer. The results show that the proposed day-ahead scheduling method can effectively coordinate renewable energy generation, energy storage, and flexible industrial loads, thereby improving the utilization rate of renewable energy.

[0063] In the rolling optimization process of traditional MPC (Multi-Process Control) intraday dispatching, the dispatching strategy is updated based on updated forecast information to coordinate renewable power generation, energy storage, and electrolytic aluminum loads. However, due to forecasting errors in renewable energy generation and load fluctuations, deviations occur between the planned operating trajectory and the actual operation of the units. In particular, during periods of rapid changes in renewable energy output, significant adjustments are made to the exchange of battery power and grid power to maintain system power balance.

[0064] The scheduling trajectory and actual trajectory of SOC and electrolytic aluminum temperature under conventional MPC are as follows: Figure 9The figure shows a comparison between the predetermined and actual trajectories of SOC and electrolytic aluminum temperature under conventional MPC in an energy system scheduling method for battery and high-energy-consumption load state evolution provided in this application embodiment. It can be observed that although the intraday scheduling process continuously updates the scheduling strategy through rolling optimization and applies terminal SOC constraints to ensure that the storage state reaches the expected target at the end of the scheduling period, the actual SOC gradually deviates from the scheduling trajectory during operation. The actual SOC exceeds the predefined safety boundary, indicating that the constraints applied at the scheduling layer are insufficient to effectively regulate state evolution during actual execution. This deviation is mainly due to the mismatch between the actual battery charging and discharging behavior and the planned state trajectory under renewable power generation prediction errors and power fluctuations. Therefore, terminal SOC constraints are mainly applicable to the predicted trajectory and cannot reflect the actual operating state in a timely manner.

[0065] Similarly, the actual electrolytic aluminum temperature deviates significantly from the predetermined trajectory and exceeds the predetermined safe temperature range at certain times. Due to the strong thermal inertia of the electrolytic aluminum process, temperature evolution is affected not only by predetermined power adjustments but also by heat dissipation characteristics, changes in operating conditions, and uncertainties in model parameters. Therefore, even when the electrolytic aluminum load follows intraday scheduling commands, accumulated temperature deviations can still occur due to prediction errors and external disturbances, leading to violations of temperature constraints. These results indicate that while traditional MPC based on updated prediction information can achieve effective power-level scheduling, significant inconsistencies exist between the scheduled state and the actual operating state during actual execution. As operation progresses, deviations in SOC and electrolytic aluminum temperature may further accumulate, causing the actual state to exceed its permissible operating range. Therefore, it is necessary to further consider changes in the execution state to improve the adaptability of multi-timescale scheduling strategies under uncertain operating conditions.

[0066] 3. Performance of the execution state feedback scheduling framework

[0067] Compared to traditional MPC, the intraday scheduling and actual execution results of the execution state feedback MPC framework incorporate actual operating states into the rolling optimization process and adjust subsequent scheduling decisions based on changes in SOC and electrolytic aluminum temperature, enabling battery power, renewable energy utilization, and grid power exchange to better adapt to actual operating conditions. It can be determined that under renewable generation fluctuations, the system maintains power balance by coordinating battery charging and discharging and grid power exchange. The actual execution results show good consistency with the intraday scheduling trajectory. Particularly during periods of rapid updates, the execution state feedback mechanism updates subsequent scheduling decisions based on actual operating conditions, thereby reducing power allocation deviations caused by prediction errors and improving the adaptability of the scheduling strategy. These results demonstrate that the proposed framework improves the consistency between predetermined power level operations and actual power level operations, laying the foundation for further reducing SOC and electrolytic aluminum temperature deviations.

[0068] For the SOC trajectory, compared to traditional MPC where the actual SOC gradually deviates from the scheduled trajectory and may violate predefined boundaries, the proposed framework keeps the actual SOC within the allowable operating range throughout the scheduling scope. Although deviations between the actual SOC and the previous day / intraday trajectory still exist due to updatable prediction errors, power fluctuations, and battery power adjustments, the rolling optimization process updates subsequent charging and discharging decisions based on the current SOC state. Therefore, it effectively prevents the continuous accumulation of SOC deviations, and the actual SOC gradually approaches the intraday scheduling trajectory. Furthermore, the actual SOC at the end of the scheduling level is close to the desired terminal value, indicating that terminal SOC constraints can better regulate the actual operating state when considering execution state information.

[0069] Using the proposed framework, the temperature trajectory of electrolytic aluminum was also significantly improved. Due to the strong thermal inertia of the electrolytic aluminum process, the actual temperature exhibits a certain lag and local deviation during load adjustments. However, by incorporating the actual temperature state into the rolling optimization process, subsequent scheduling decisions can be adjusted according to the current process conditions. Therefore, the actual temperature remains within the allowable operating range and gradually follows the intraday scheduling trajectory. Compared to traditional predictive control, where temperature deviations accumulate over time and may lead to state violations, the proposed framework effectively limits the propagation of temperature deviations.

[0070] The results demonstrate that the proposed execution state feedback scheduling framework improves the consistency between the scheduling state and the actual operating state by utilizing execution state information to update subsequent scheduling decisions. Under renewable uncertainties and dynamic industrial loads, this framework effectively reduces the accumulation of SOC and electrolytic aluminum temperature deviations, thereby enhancing the security and practical feasibility of multi-timescale scheduling strategies.

[0071] 4. Performance comparison under different levels of uncertainty

[0072] To evaluate the state regulation capability of the proposed ESF-MTS framework under different levels of renewable generation and load uncertainty, low, medium, and high uncertainty scenarios were considered, and it was compared with NS-MPC.

[0073] Under NS-MPC, SOC errors gradually accumulate within the scheduling range, and this accumulation becomes more pronounced with increasing uncertainty levels. The maximum SOC deviation reaches 0.2219, 0.3007, and 0.4596 in low, medium, and high uncertainty scenarios, respectively. Specifically, after approximately 10 hours, updatable prediction errors and load fluctuations continuously modify the actual battery charge / discharge power, while the scheduling model continues to propagate storage state based on the planned trajectory. Therefore, the difference between the actual SOC and the planned SOC increases over time. This result 6 indicates that rolling optimization based solely on updated prediction information cannot effectively prevent storage state errors from propagating across consecutive scheduling intervals.

[0074] In contrast, the SOC error obtained using the ESF-MTS framework remained near zero for most of the scheduling range, with only small short-term fluctuations during periods of significant changes in renewable energy and load. The maximum SOC deviation at the three uncertainty levels decreased to 0.0168, 0.0233, and 0.0499, respectively, representing reductions of approximately 92.4%, 92.3%, and 89.1% compared to NS-MPC.14 While the SOC error increased slightly with increasing uncertainty, the increase was much smaller than that observed under NS16 MPC. This suggests that execution state information can be used to update the initial state for subsequent rolling optimization intervals, thus limiting the continuous propagation of prediction errors to SOC deviations. Even under high determinism, the maximum SOC deviation remained below 0.05%, while the maximum deviation for NS-MPC reached 0.4596.

[0075] It is important to note that the temperature error primarily stems from the mismatch between the scheduling model and the actual thermal processes, including simplified representations of thermal inertia, heat dissipation parameters, and other unmodeled operational factors. Therefore, this error is not directly related to the low, medium, and high uncertainty scenarios defined for wind power, photovoltaic power generation, and load forecasting. Thus, only a representative temperature error comparison is presented.

[0076] Under the NS-MPC model, temperature errors during electrolytic aluminum power adjustment exhibit significant accumulation and hysteresis, with a maximum deviation of 0.6950℃. Around 6h, 16h, 18h, and 22h, changes in the planned aluminum load power lead to significant positive or negative temperature errors because the actual thermal state cannot immediately follow the predicted trajectory due to the strong thermal inertia of the process. Under the ESF-MTS framework, transient temperature deviations still occur at the power conversion moment, but the decay is faster thereafter. The maximum temperature deviation is reduced to 0.3364℃, a reduction of approximately 51.6% compared to NS-MPC. This result indicates that updating subsequent rolling optimization decisions using the actual temperature of electrolytic aluminum can reduce the accumulation of temperature deviations caused by model mismatch. However, due to the inherent thermal inertia of the aluminum process, execution state feedback cannot completely eliminate instantaneous tracking errors during load adjustment, which explains the remaining local deviations.

[0077] Table 2. Performance Comparison at Different Uncertainty Levels

[0078] The ESF-MTS framework achieved significantly lower SOC 16 bias across all three uncertainty levels without increasing battery throughput. Specifically, battery throughput decreased from 366.75, 361.96, and 361.04 MWh under NS-MPC to 363.88, 350.08, and 350.92 MWh under ESF-MTS. Therefore, the improvement in SOC performance was not achieved through more intensive battery charge and discharge cycles. Instead, it was primarily due to timely execution state updates and corresponding adjustments to subsequent scheduling decisions.

[0079] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to computer program instructions. The aforementioned computer program can be stored in a non-volatile storage medium, and when executed, it performs the steps of the above method embodiments. Alternatively, if the integrated unit of the present invention is implemented as a software functional module and sold or used as an independent product, it can also be stored in a non-volatile storage medium. Based on this understanding, the technical solution of the embodiments of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a non-volatile storage medium and includes several instructions to cause an electronic device (which may be a personal computer, server, network device, etc.) to execute all or part of the methods described in the various embodiments of the present invention.

[0080] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. An energy system scheduling method based on the state evolution of batteries and high-energy-consuming loads, characterized in that, The scheduling method includes: Obtain historical operational data within the park and correct it to obtain the corrected data; Based on the time interval of the intraday scheduling layer, ultra-short-term forecast data is obtained from the corrected data; An optimized prediction model is established based on the ultra-short-term prediction data, and the optimal prediction sequence is obtained. Based on the first value of the optimal prediction sequence, the predicted value for the next time step is obtained, and the predicted value for the next time step is used to update the decision variables to obtain the updated decision variables. The updated decision variables are input into the actual layer model and compared with the actual SOC of the energy system and the actual execution temperature of the electrolytic aluminum process to obtain the comparison results respectively. Based on the comparison results, the prediction model of the optimization layer is updated to obtain a new optimal prediction sequence.

2. The energy system scheduling method for battery and high-energy-consumption load state evolution according to claim 1, characterized in that, The process of acquiring and correcting historical operational data within the park to obtain corrected data includes: Obtain the energy system structure within the park to determine the energy types within the park; Historical operating data for all energy types are rolled over and corrected according to a specific time period to obtain corrected data.

3. The energy system scheduling method for battery and high-energy-consumption load state evolution according to claim 1, characterized in that, The ultra-short-term forecast data obtained from the corrected data based on the time interval of the intraday scheduling layer includes: Set the time interval of the intraday scheduling layer, wherein the time interval of the intraday scheduling layer is not longer than the specific time period length of the data correction process after correction. Based on the time interval of the intraday scheduling layer, the corresponding corrected data is obtained from the time axis of the corrected data to obtain ultra-short-term forecast data.

4. The energy system scheduling method for battery and high-energy-consumption load state evolution according to claim 1, characterized in that, The step of establishing an optimized prediction model based on the ultra-short-term prediction data and obtaining the optimal prediction sequence includes: Based on the aforementioned ultra-short-term forecast data, a day-ahead economic model for the energy system is obtained, with the following equation: ， Among them, J DA Represents the day-ahead economic value of the energy system; C grid C represents the total cost of purchasing electricity from the external network; bat Indicates the total operating cost of energy storage; C PV / WT Indicates the total generation cost of distributed power units; C curtail This indicates the losses caused by wind and solar power curtailment; C Al This indicates the total compensation cost for peak shaving in electrolytic aluminum production; Obtain parameter constraints in the energy system and determine parameters based on these constraints; Based on the constraints of high-energy-consuming loads, grid economics, and battery energy storage regulation in the energy system, the comprehensive objective function for intraday rolling optimization is obtained as follows: ， Where J represents the intraday rolling optimization comprehensive target; J op J represents the operating cost of the power grid; al This indicates the adjustment cost of high energy consumption; J smooth Indicates the constraint cost of battery energy storage regulation; J track This indicates the tracking items for the battery energy storage system planned one day in advance; J act J represents a penalty item for high-energy-consuming actions; ter Indicates the terminal dynamic penalty item; J slack Indicates a relaxation variable penalty term; Based on the day-ahead economic model and the intraday rolling optimization comprehensive objective function, the optimal forecast sequence for the day is obtained.

5. The energy system scheduling method for battery and high-energy-consumption load state evolution according to claim 4, characterized in that, The process of acquiring parameter constraints in the energy system and determining parameters based on these constraints includes: The parameter constraints include: The power balance constraint is given by the following equation: ， in, This represents the photovoltaic power generation at time t; This represents the wind power output at time t; This represents the net power of the battery at time t; This represents the power exchange between the power grids at time t. This represents the load power of the electrolytic aluminum at time t; This represents the power of other loads at time t; The power grid constraints are given by the following equations: ， in, Indicates the maximum power exchanged between the power grids; New energy constraints, the equation is: ， in, Indicates the maximum power output of photovoltaic power generation; Indicates the maximum power output of the wind turbine; Battery energy storage constraints include: The charge / discharge constraint equation is as follows: ， in, Indicates the maximum discharge power of the battery; Indicates the maximum charging power of the battery; The SOC state transition constraint equation is: ， Where SOC(t+1) represents the SOC state parameter at time t+1; SOC(t) represents the SOC state parameter at time t. Indicates the time difference; S bat Indicates battery capacity; η represents battery charge / discharge efficiency; P bat (t) represents the battery power at time t; The initial SOC constraint is given by the following equation: ， Among them, SOC int Represents the initial SOC; SOC end Indicates the end of SOC; The high-energy-consumption load constraint is represented by the following equation: ， in, This indicates the minimum load for electrolytic aluminum production; P represents the maximum load of electrolytic aluminum; Al (t) represents the electrolytic aluminum load at time t; Indicates the lowest temperature value; Indicates the highest temperature value; T Al (t) represents the temperature of the electrolytic aluminum at time t.

6. The energy system scheduling method for battery and high-energy-consumption load state evolution according to claim 4, characterized in that, The intraday rolling optimization comprehensive objective function includes: In the intraday rolling optimization comprehensive objective function, the power grid operating cost equation is: ， Where t represents the t-th scheduling time; N represents the total number of time periods within a day; c grid (t) represents the external grid purchase price of electricity at time t; P grid (t) represents the power exchange between the power grids at time t; Indicates the time step size for each time period; P represents the unit price of the penalty for curtailment of photovoltaic power. pv，curt (t) represents the amount of solar power curtailed at time t; P represents the unit price of wind power curtailment penalty; wt，curt (t) represents the wind power curtailment at time t; c bat P represents the unit operating cost of energy storage; ch (t) represents the energy storage charging power at time t; P dis (t) represents the energy storage discharge power at time t; In the intraday rolling optimization comprehensive objective function, the battery energy storage regulation constraint cost equation is: ， Among them, P bat (·) represents the net power of the battery at a certain moment; In the intraday rolling optimization comprehensive objective function, the tracking term equation for the battery energy storage system's plan one day in advance is: ， in, This indicates the reference power of energy storage in the day-ahead dispatch plan; In the intraday rolling optimization comprehensive objective function, the high-energy-consumption adjustment cost equation is: ， Among them, S al e represents the cost compensation coefficient for electrolytic aluminum. al，comp (t) represents the adjustment and compensation error of electrolytic aluminum; w track，al Indicates the tracking penalty weight for electrolytic aluminum; e al，tune (t) represents the electrolytic aluminum adjustment tracking error; In the intraday rolling optimization comprehensive objective function, the high-energy-consumption action penalty term and the terminal dynamic penalty term are respectively expressed as: ， ， Among them, z al (t) represents the penalty value for the electrolytic aluminum load action at time t; e Tal，terminal This indicates the deviation of the electrolytic aluminum temperature at the end of the predicted time domain; In the intraday rolling optimization comprehensive objective function, the equation for the relaxation variable penalty term is: ， in, s represents the slack variable penalty coefficient corresponding to the energy constraint; Emin s represents the lower slack variable representing the energy constraint of the energy storage system; Emax Represents the upper relaxation variable of the energy constraint of the energy storage system; This represents the slack variable penalty coefficient corresponding to the temperature constraint; This represents the lower relaxation variable representing the temperature constraint in the aluminum electrolysis system. This represents the upper relaxation variable of the temperature constraint in the aluminum electrolysis system.

7. The energy system scheduling method for battery and high-energy-consumption load state evolution according to claim 1, characterized in that, The step of obtaining the predicted value for the next time step based on the first value of the optimal prediction sequence, and updating the decision variables using the predicted value for the next time step to obtain the updated decision variables, includes: Based on the time point corresponding to the optimal prediction sequence and the interval between the next time point and the corresponding time point, and based on the optimization layer prediction model, the predicted value for the next time point is obtained. The predicted value for the next moment is used to replace the predicted value for the next moment in the optimal prediction sequence to update the decision variables and obtain the updated decision variables.

8. The energy system scheduling method for battery and high-energy-consumption load state evolution according to claim 1, characterized in that, The updated decision variables are input into the actual layer model and compared with the actual SOC of the energy system and the actual operating temperature of the aluminum electrolysis process to obtain comparison results, including: The updated decision variables are input into the actual layer model, and the actual execution SOC of the energy system and the actual execution temperature of the electrolytic aluminum process are obtained respectively. The actual SOC of the energy system and the updated SOC of the decision variables are compared to obtain the SOC comparison result, which is: ， in, SOC t+1|t This represents the difference between the actual execution SOC and the optimal decision SOC; a t e represents the perturbation coefficient at time t; L，t (·) represents the tracking error on the left side at time t; (e R，t ) — This represents the portion of the tracking error on the right side at time t that is less than 0. The actual operating temperature of the aluminum electrolysis process in the energy system is compared with the updated decision variable of the aluminum electrolysis process operating temperature to obtain the temperature comparison result, which is: ， in, This represents the temperature deviation predicted at time t+1 from the predicted temperature at time t; cm is the equivalent thermal inertia; k real，t k represents the actual equivalent thermal coefficient at time t; plan Indicates the planned heat coefficient; T represents the actual temperature of the electrolytic cell at time t; T0 represents the rated temperature.

9. The energy system scheduling method for battery and high-energy-consumption load state evolution according to claim 1, characterized in that, The step of updating the optimization layer prediction model based on the comparison results to obtain a new optimal prediction sequence includes: The temperature comparison result and the SOC comparison result in the comparison results are respectively compared with the corresponding thresholds to obtain the threshold comparison result; When the threshold comparison result is not less than 0, the actual SOC of the obtained energy system and the actual execution temperature of the electrolytic aluminum process are recalculated, and a new optimal prediction sequence is generated.

Images