A multi-agent state of charge planning and dynamic prediction control method and system

Abstract

The application provides a multi-agent state of charge planning and dynamic prediction control method and system, which comprises the following steps: acquiring load power, photovoltaic output and environmental temperature prediction, extracting peak load gap, photovoltaic surplus degree, discharge standby demand and charging margin demand, and determining the upper and lower boundaries of the energy storage SOC allowable corridor; according to the above, an SOC reference trajectory containing a falling section, a rising section and a terminal recovery section is generated; each section is identified and three modes of photovoltaic charging margin creation, discharge standby formation and terminal SOC recovery are set correspondingly; according to the modes, the photovoltaic light abandonment penalty, the SOC tracking and the terminal SOC penalty weight in the nonlinear model prediction control objective function are updated; and finally, the power instructions of the diesel generator, the energy storage and the photovoltaic are generated. Through adaptive SOC trajectory planning and dynamic weight adjustment, the method solves the problems of light abandonment and fuel waste.

Description

Technical Field

[0001] This application belongs to the field of vehicle-mounted power source management, and in particular relates to a multi-agent state of charge planning and dynamic predictive control method and system. Background Technology

[0002] Vehicle-mounted emergency microgrids typically integrate diesel generator sets, photovoltaic arrays, energy storage systems, public AC buses, and emergency loads, enabling them to provide independent power supply in scenarios such as natural disasters, large-scale power outages, field construction, and temporary backup. Compared to stationary microgrids, vehicle-mounted emergency microgrids are characterized by limited capacity, rapid changes in operating environment, frequent load switching, and high power supply security requirements.

[0003] In a diesel-PV-energy storage hybrid power supply structure, diesel generator sets provide controllable power support, PV arrays provide renewable energy, and energy storage systems perform functions such as peak shaving and valley filling, smoothing PV fluctuations, and improving reserve margins. The state of charge (SOC) of the energy storage system determines not only its discharge capacity during future high-load periods but also its charging capacity during high-output PV periods. Therefore, the design of the SOC reference trajectory directly affects PV absorption, fuel consumption, energy storage safety, and the initial state the following day. Existing energy management methods for vehicle-mounted or islanded microgrids mostly employ fixed SOC references, fixed-rule scheduling, or fixed-weight model predictive control. These methods guide energy storage charging and discharging through preset fixed SOC targets or fixed rules, and perform rolling optimization with fixed target weights.

[0004] The aforementioned technical solutions employ a fixed SOC reference, fixed rules, and fixed weights for energy management. This approach fails to adaptively plan the energy storage SOC reference trajectory and dynamically adjust control priorities based on forecasts of future photovoltaic output and load changes, as well as the differentiated needs of different operational phases. This lack of adaptive planning and dynamic adjustment capabilities makes it difficult to simultaneously meet the coordinated needs of photovoltaic power consumption, high-load period discharge reserves, and end-of-day SOC recovery under conditions of photovoltaic fluctuations and load uncertainty. This results in problems such as curtailment of solar power, increased fuel consumption, and decreased power supply security. Summary of the Invention

[0005] The purpose of this application is to overcome the deficiencies in the prior art and provide a multi-agent charge state planning and dynamic predictive control method and system.

[0006] This application provides a multi-agent charge state planning and dynamic predictive control method, including: Obtain load power forecast, photovoltaic output forecast, and ambient temperature forecast; Peak load gap, photovoltaic surplus, discharge reserve requirement, and charging margin requirement are extracted from the load power prediction, photovoltaic output prediction, and ambient temperature prediction. Based on the peak load gap, the photovoltaic surplus, the discharge reserve requirement, and the charging margin requirement, determine the upper and lower boundaries of the allowable state of charge corridor for the energy storage system; A state of charge reference trajectory is generated based on the upper boundary and the lower boundary. The state of charge reference trajectory is located between the upper boundary and the lower boundary and has a falling segment, a rising segment and a terminal recovery segment. Based on the state of charge reference trajectory, the falling segment, the rising segment, and the terminal recovery segment are identified, and the photovoltaic charging margin creation mode, the discharge reserve formation mode, and the terminal state of charge recovery mode are determined based on the falling segment, the rising segment, and the terminal recovery segment, respectively. Based on the photovoltaic charging margin creation mode, the discharge reserve formation mode, or the terminal state of charge recovery mode, update the weight coefficients in the objective function of the nonlinear model predictive control. The weight coefficients include the photovoltaic curtailment penalty weight, the state of charge tracking weight, and the terminal state of charge penalty weight. Based on the state of charge reference trajectory and the updated weighting coefficients, the output power command of the diesel generator set, the charging and discharging power command of the energy storage system, and the photovoltaic utilization power command are generated through nonlinear model predictive control.

[0007] Optionally, the method is executed by two or more functional agents in a multi-agent system, who cooperate and divide tasks. The functional agents exchange prediction features, charge state allowed corridors, candidate trajectories, feasibility review results, and evaluation results through a shared data space.

[0008] Optionally, determining the upper and lower boundaries of the energy storage system's state of charge allowable corridor includes: The lower boundary is raised based on the peak load gap and the discharge reserve requirement; The lower boundary is lowered based on the photovoltaic surplus and the charging margin requirement. Based on the aforementioned discharge backup requirements, the upper boundary is raised; The upper boundary is lowered based on the photovoltaic surplus level. Based on the floating directions of the upper and lower boundaries, the final positions of the upper and lower boundaries are determined, wherein the upper boundary does not exceed the maximum state of charge of the energy storage system, and the lower boundary is not lower than the minimum state of charge of the energy storage system.

[0009] Optionally, generating a state-of-charge reference trajectory based on the upper boundary and the lower boundary includes: Generate a backup priority trajectory, wherein the state of charge of the backup priority trajectory is higher than the first state of charge level; A photovoltaic (PV) grid connection priority trajectory is generated, wherein the state of charge of the PV grid connection priority trajectory is lower than the second state of charge level; A balanced operating trajectory is generated, wherein the state of charge of the balanced operating trajectory is between the state of charge of the backup priority trajectory and the state of charge of the photovoltaic consumption priority trajectory. Generate a terminal recovery trajectory, wherein the terminal charge state of the terminal recovery trajectory tends to the target charge state; A comprehensive coordinated trajectory is generated, and the state of charge of the comprehensive coordinated trajectory is determined based on the backup priority trajectory, the photovoltaic consumption priority trajectory, the balanced operation trajectory, and the terminal recovery trajectory; The backup priority trajectory, the photovoltaic consumption priority trajectory, the balanced operation trajectory, the terminal recovery trajectory, and the comprehensive coordination trajectory are projected between the upper boundary and the lower boundary, respectively. Each projected trajectory is smoothed to obtain multiple smoothed candidate trajectories, and one of these smoothed candidate trajectories is selected as the reference trajectory for the state of charge.

[0010] Optionally, generating a state-of-charge reference trajectory based on the upper boundary and the lower boundary includes: Generate multiple candidate trajectories; A feasibility review is performed on the multiple candidate trajectories, the feasibility review including: A first penalty is applied to candidate trajectories whose change in state of charge exceeds a first change threshold in adjacent time periods; A second penalty is applied to candidate trajectories whose second-order change in state of charge exceeds a second change threshold in adjacent time periods; A third penalty is applied to candidate trajectories whose initial state of charge deviates from the current actual state of charge by more than a first deviation threshold. A fourth penalty is applied to candidate trajectories that exceed the upper or lower boundary. A fifth penalty is applied to candidate trajectories whose pre-charge state is below the first reserve threshold during high load periods; The state of charge reference trajectory is determined from the plurality of candidate trajectories based on the first penalty, the second penalty, the third penalty, the fourth penalty, and the fifth penalty.

[0011] Optionally, determining the state-of-charge reference trajectory from the plurality of candidate trajectories based on the first penalty, the second penalty, the third penalty, the fourth penalty, and the fifth penalty includes: Calculate fuel consumption proxy indicators for candidate trajectories that have passed the feasibility review; Calculate photovoltaic grid integration proxy indicators for candidate trajectories that have passed the feasibility review; Calculate the terminal state of charge recovery index for candidate trajectories that have passed the feasibility review; Calculate the all-day coordination index for candidate trajectories that have passed the feasibility review; The constraints for calculating candidate trajectories that have passed the feasibility review are violated. The state of charge reference trajectory is selected from the candidate trajectories that have passed the feasibility review, based on the fuel consumption proxy index, the photovoltaic absorption proxy index, the terminal state of charge recovery index, the all-day coordination index, and the constraint violation index.

[0012] Optionally, it also includes: Obtain battery current prediction; The battery temperature prediction is determined based on the ambient temperature prediction and the battery current prediction. When the predicted battery temperature exceeds a first temperature threshold or the predicted battery current exceeds a first current threshold, an energy storage protection priority mode is determined. When the energy storage protection priority mode is determined, the weighting coefficient is updated. The weighting coefficient also includes battery current penalty weight and battery temperature penalty weight, and the state of charge tracking weight is reduced.

[0013] Optionally, it also includes: Acquire load power fluctuations, photovoltaic output fluctuations, and current state of charge; When the load power fluctuation is less than the first fluctuation threshold, the photovoltaic output fluctuation is less than the second threshold, and the current state of charge is within the comfort zone, an economic smoothing mode is determined. When the economic smoothing mode is determined, the load change penalty weight of the diesel generator set is increased, and the fuel economy weight is increased.

[0014] This application also provides a multi-agent charge state planning and dynamic prediction control system, including: The acquisition module is used to acquire load power prediction, photovoltaic output prediction, and ambient temperature prediction. The extraction module is used to extract peak load gap, photovoltaic surplus, discharge reserve requirement and charging margin requirement from the load power prediction, the photovoltaic output prediction and the ambient temperature prediction. The boundary module is used to determine the upper and lower boundaries of the allowable state of charge corridor of the energy storage system based on the peak load gap, the photovoltaic surplus, the discharge reserve requirement, and the charging margin requirement. The segmentation module is used to generate a state of charge reference trajectory based on the upper boundary and the lower boundary. The state of charge reference trajectory is located between the upper boundary and the lower boundary and has a falling segment, a rising segment and a terminal recovery segment. The mode module is used to identify the falling segment, the rising segment, and the terminal recovery segment based on the state of charge reference trajectory, and to determine the photovoltaic charging margin creation mode, the discharge reserve formation mode, and the terminal state of charge recovery mode based on the falling segment, the rising segment, and the terminal recovery segment, respectively. The penalty module is used to update the weight coefficients in the objective function of the nonlinear model predictive control according to the photovoltaic charging margin creation mode, the discharge reserve formation mode, or the terminal state of charge recovery mode. The weight coefficients include photovoltaic curtailment penalty weights, state of charge tracking weights, and terminal state of charge penalty weights. The instruction module is used to generate output power instructions for diesel generator sets, charging and discharging power instructions for energy storage systems, and photovoltaic utilization power instructions through nonlinear model predictive control based on the state of charge reference trajectory and the updated weighting coefficients.

[0015] Optionally, the boundary module includes: The first floating unit is used to raise the lower boundary according to the peak load gap and the discharge reserve requirement; The second floating unit is used to lower the lower boundary according to the photovoltaic surplus and the charging margin requirement; The third floating unit is used to raise the upper boundary according to the discharge backup requirements; The fourth floating unit is used to lower the upper boundary according to the photovoltaic surplus level; A boundary determination unit is used to determine the final position of the upper boundary and the final position of the lower boundary based on the floating direction of the upper boundary and the lower boundary, wherein the upper boundary does not exceed the maximum state of charge of the energy storage system and the lower boundary is not lower than the minimum state of charge of the energy storage system.

[0016] Optionally, the segmentation module includes: The first trajectory generation unit is used to generate a backup priority trajectory, wherein the state of charge of the backup priority trajectory is higher than the first state of charge level. The second trajectory generation unit is used to generate a photovoltaic consumption priority trajectory, wherein the state of charge of the photovoltaic consumption priority trajectory is lower than the second state of charge level; The third trajectory generation unit is used to generate a balanced operation trajectory, wherein the state of charge of the balanced operation trajectory is between the state of charge of the backup priority trajectory and the state of charge of the photovoltaic consumption priority trajectory. The fourth trajectory generation unit is used to generate a terminal recovery trajectory, wherein the terminal charge state of the terminal recovery trajectory tends to the target charge state; The fifth trajectory generation unit is used to generate a comprehensive coordinated trajectory, the state of charge of which is determined based on the backup priority trajectory, the photovoltaic consumption priority trajectory, the balanced operation trajectory, and the terminal recovery trajectory. The projection unit is used to project the backup priority trajectory, the photovoltaic absorption priority trajectory, the balanced operation trajectory, the terminal recovery trajectory, and the integrated coordination trajectory onto the area between the upper boundary and the lower boundary, respectively. A smoothing unit is used to smooth each projected trajectory to obtain multiple smoothed candidate trajectories, and to determine one of the multiple smoothed candidate trajectories as the reference trajectory for the state of charge.

[0017] The beneficial effects of this application are: This application provides a multi-agent state of charge (SOC) planning and dynamic predictive control method, comprising: acquiring load power prediction, photovoltaic (PV) output prediction, and ambient temperature prediction; extracting peak load gap, PV surplus, discharge reserve requirement, and charging margin requirement from the load power prediction, PV output prediction, and ambient temperature prediction; determining the upper and lower boundaries of the allowable SOC corridor of the energy storage system based on the peak load gap, PV surplus, discharge reserve requirement, and charging margin requirement; generating a SOC reference trajectory based on the upper and lower boundaries, the SOC reference trajectory being located between the upper and lower boundaries and having a falling segment, a rising segment, and a terminal recovery segment; and determining the upper and lower boundaries of the allowable SOC corridor of the energy storage system based on the SOC reference trajectory. The system identifies the descent segment, the ascent segment, and the terminal recovery segment, and determines the photovoltaic charging margin creation mode, the discharge reserve formation mode, and the terminal state of charge recovery mode based on the descent segment, the ascent segment, and the terminal state of charge recovery mode, respectively. Based on the photovoltaic charging margin creation mode, the discharge reserve formation mode, or the terminal state of charge recovery mode, the system updates the weight coefficients in the objective function of the nonlinear model predictive control. These weight coefficients include photovoltaic curtailment penalty weights, state of charge tracking weights, and terminal state of charge penalty weights. Based on the state of charge reference trajectory and the updated weight coefficients, the system generates output power commands for diesel generator sets, charging and discharging power commands for energy storage systems, and photovoltaic utilization power commands through nonlinear model predictive control. This application achieves adaptive planning of the SOC trajectory and dynamic adjustment of control priorities by dynamically constructing a SOC reference trajectory with descent, ascent, and terminal recovery segments based on predictive information, and identifying the operating mode accordingly to update control weights. This overcomes the problems of curtailment and fuel waste caused by the lack of this capability in existing technologies. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the multi-agent charge state planning and dynamic predictive control method in this application; Figure 2 This is a schematic diagram of the permitted corridor and multi-objective reference trajectory for energy storage SOC in this application; Figure 3 This is a schematic diagram of the operation mode recognition and NMPC weight dynamic adjustment in this application; Figure 4 This is a schematic diagram of the multi-agent charge state planning and dynamic prediction control system in this application. Detailed Implementation

[0019] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, the embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0020] This application provides a multi-agent state of charge planning and dynamic predictive control method, which is applied to the fields of energy management and predictive control technology for vehicle-mounted emergency power supplies, diesel, photovoltaic, and energy storage microgrids. It is used to solve the problem of difficulty in balancing photovoltaic absorption, operational economy and power supply security under conditions of photovoltaic fluctuations, load uncertainty and energy storage constraints.

[0021] The vehicle-mounted emergency microgrid includes a diesel generator set, a photovoltaic array, an energy storage system, a common AC bus, and loads.

[0022] This application constructs a two-layer energy management structure consisting of upper-layer multi-agent state-of-charge (SOC) reference trajectory planning and lower-layer dynamic weighted nonlinear model predictive control. Expert knowledge is encoded into SOC allowable corridors, stage objectives, and trajectory review rules. The upper layer outputs SOC reference trajectories and operating modes to the lower layer, which uses the SOC reference trajectory as a tracking target and dynamically adjusts the objective function weights according to the operating mode.

[0023] Please refer to Figure 1 As shown, the multi-agent charge state planning and dynamic predictive control method includes: S101, Obtain load power prediction, photovoltaic output prediction and ambient temperature prediction.

[0024] The controller acquires load power prediction, photovoltaic output prediction, and ambient temperature prediction for the future planning time domain, such as the next 24 hours, and can be updated according to the rolling prediction time domain during operation.

[0025] The forecast information can come from day-ahead forecasting models, meteorological data interfaces, historical load models, or manually set emergency load plans. Load power forecasting refers to the estimation of load power consumption over a future period, photovoltaic output forecasting refers to the estimation of the power that the photovoltaic array may generate over a future period, and ambient temperature forecasting refers to the estimation of ambient temperature over a future period.

[0026] In addition, the controller also collects operating status data such as current load power, diesel generator output power, energy storage system state of charge (SOC), battery temperature, actual photovoltaic output, and available photovoltaic power at a set sampling period, providing input for upper-level state of charge planning and lower-level rolling optimization.

[0027] S102. Extract the peak load gap, photovoltaic surplus, discharge reserve requirement and charging margin requirement from the load power prediction, photovoltaic output prediction and ambient temperature prediction.

[0028] The planning window is divided into several hourly segments or phase windows, and the operational demand characteristics are calculated within each window.

[0029] Peak load gap is used to characterize the risk of future loads exceeding the controllable supply capacity of diesel generator sets and available photovoltaic power. Peak load gap is determined based on load power forecasts, the maximum available power of diesel generator sets, and photovoltaic output forecasts.

[0030] The photovoltaic surplus is used to characterize the extent to which future photovoltaic output exceeds load demand. The photovoltaic surplus is determined based on the degree to which the photovoltaic output forecast exceeds the load power forecast.

[0031] Discharge reserve requirement is used to characterize the energy storage discharge capacity that should be reserved before the high load period. The discharge reserve requirement is determined based on the load power forecast for the future high load period, the power margin of the diesel generator set, and the photovoltaic output forecast.

[0032] Charging margin requirement is used to characterize the energy storage charging space that should be reserved before the high output window of photovoltaics. The charging margin requirement is determined based on the future surplus energy of photovoltaics, the charging power constraint of the energy storage system, and the upper limit of the safe state of charge.

[0033] To avoid planning deviations caused by different feature dimensions, peak load gap, photovoltaic surplus, discharge reserve demand and charging margin demand can be normalized, and related features can be fused.

[0034] For example, the discharge reserve requirement can be modified by the future high load demand and peak load gap, and the charging margin requirement can be modified by the future photovoltaic surplus and photovoltaic consumption window, thus obtaining a normalized feature set for state of charge planning.

[0035] S103. Based on the peak load gap, the photovoltaic surplus, the discharge reserve requirement, and the charging margin requirement, determine the upper and lower boundaries of the allowable state of charge corridor of the energy storage system.

[0036] Please refer to Figure 2 As shown, a charge state allowable corridor is constructed based on the normalized demand characteristics.

[0037] The charged state allows the corridor to include the charged state boundary and the charged state upper boundary.

[0038] The upper boundary of the state of charge increases with the increase of peak load gap or discharge reserve demand, and decreases with the increase of photovoltaic surplus or charging margin demand. The upper boundary of the state of charge increases with the increase of future discharge reserve demand, and decreases with the increase of future photovoltaic consumption demand.

[0039] The upper boundary of the state of charge is used to ensure sufficient discharge reserves before the high load period, while the upper boundary of the state of charge is used to ensure sufficient charging acceptance capacity before the high output period of photovoltaics.

[0040] The state of charge (SOC) allows the corridor to be constrained by the energy storage system's lower safety limit, upper safety limit, and engineering protection threshold. For example, the SOC of the energy storage system can be limited to between 20% and 80%, and the daily target SOC range can be set to between 40% and 50%; the target SOC can also be set as a single target value according to task requirements.

[0041] In other words, the lower boundary is raised based on the peak load gap and the discharge reserve requirement; the lower boundary is lowered based on the photovoltaic surplus and the charging margin requirement; the upper boundary is raised based on the discharge reserve requirement; the upper boundary is lowered based on the photovoltaic surplus; the final position of the upper boundary and the final position of the lower boundary are determined based on the floating direction of the upper and lower boundaries, and the upper boundary does not exceed the maximum state of charge of the energy storage system, and the lower boundary is not lower than the minimum state of charge of the energy storage system.

[0042] The charge state allows the corridor to be constructed based on operational requirements and pre-defined expert knowledge.

[0043] S104. Generate a state of charge reference trajectory based on the upper boundary and the lower boundary. The state of charge reference trajectory is located between the upper boundary and the lower boundary and has a falling segment, a rising segment and a terminal recovery segment.

[0044] Please refer to Figure 3As shown, after constructing the state-of-charge (SOC) allowance corridor, the planning layer further formulates reserve targets, charging margin targets, and all-day coordination targets. The reserve target causes the SOC to rise before the high-load period, the charging margin target causes the SOC to decrease before the high-output period of photovoltaics, and the all-day coordination target is used to limit the large oscillations of the SOC trajectory throughout the day and maintain a sustainable operating state at the end of the day.

[0045] In one optional implementation, the upper-level state-of-charge trajectory planning is accomplished through the collaborative efforts of multiple functional agents in a multi-agent system. This multi-agent system may include a feature extraction agent, a corridor construction agent, a candidate trajectory generation agent, a feasibility review agent, a proxy evaluation agent, and a decision-making agent. Each functional agent operates according to a preset processing sequence and exchanges predicted features, permitted state-of-charge corridors, candidate trajectories, review results, and proxy evaluation results through a shared data space. The decision-making agent selects candidate trajectories based on the shared data and outputs the state-of-charge reference trajectory and operating mode to the lower-level nonlinear model prediction and control module.

[0046] The candidate trajectory generation agent generates backup priority trajectory, photovoltaic consumption priority trajectory, balanced operation trajectory, terminal recovery trajectory, and comprehensive coordination trajectory.

[0047] The state of charge (SOC) of the backup priority trajectory is higher than the first SOC level, the SOC of the photovoltaic (PV) absorption priority trajectory is lower than the second SOC level, the SOC of the balanced operation trajectory is between the SOC of the backup priority trajectory and the SOC of the PV absorption priority trajectory, the terminal SOC of the terminal recovery trajectory tends towards the target SOC, and the SOC of the integrated coordination trajectory is determined based on the backup priority trajectory, the PV absorption priority trajectory, the balanced operation trajectory, and the terminal recovery trajectory.

[0048] The aforementioned backup priority trajectory, photovoltaic consumption priority trajectory, balanced operation trajectory, terminal recovery trajectory, and integrated coordination trajectory belong to candidate trajectory types; the falling segment, rising segment, and terminal recovery segment belong to the operation stages of the final state of charge reference trajectory, divided according to the changing trend and terminal time period, and belong to different levels. In one embodiment, each candidate trajectory is projected between the upper and lower boundaries and smoothed to obtain multiple smoothed candidate trajectories; in another embodiment, a corridor boundary violation penalty is calculated for unprojected candidate trajectories, and then projected and smoothed after the selected trajectory is selected. One of the two methods is implemented, and finally only the selected trajectory is used as the state of charge reference trajectory.

[0049] In the implementation of the punitive feasibility review, the feasibility review agent filters candidate trajectories. The filtering indicators include changes in adjacent anchor points, second-order changes, deviation between the initial SOC and the current SOC, degree of corridor boundary crossing, degree of insufficiency of backup before high-load periods, and terminal SOC deviation.

[0050] Specifically, the feasibility review includes applying a first penalty to candidate trajectories where the change in state of charge exceeds a first change threshold in adjacent time periods, applying a second penalty to candidate trajectories where the second-order change in state of charge exceeds a second change threshold in adjacent time periods, applying a third penalty to candidate trajectories where the deviation between the initial state of charge and the current actual state of charge exceeds a first deviation threshold, applying a fourth penalty to candidate trajectories that exceed the upper or lower boundary, and applying a fifth penalty to candidate trajectories where the state of charge is lower than a first reserve threshold before the high load time.

[0051] The first penalty is a penalty value applied to candidate trajectories where the change in state of charge (SOC) exceeds a first threshold value in adjacent time periods, used to suppress excessively large jumps in SOC within a short period. The second penalty is a penalty value applied to candidate trajectories where the second-order change in SOC exceeds a second threshold value in adjacent time periods, used to limit excessive curvature of the SOC trajectory and avoid severe oscillations. The third penalty is a penalty value applied to candidate trajectories where the deviation between the initial SOC and the current actual SOC exceeds a first deviation threshold value, used to ensure that the starting point of the planned trajectory remains consistent with the actual operating state. The fourth penalty is a penalty value applied to candidate trajectories exceeding the upper or lower boundary, used to force the trajectory to always remain within the allowed SOC corridor. The fifth penalty is a penalty value applied to candidate trajectories where the SOC is below a first reserve threshold before high load conditions, used to ensure that the energy storage system has sufficient discharge reserve capacity before high load conditions arrive.

[0052] The penalty value is used in the feasibility review to quantitatively assess the degree to which each candidate trajectory violates the constraints, thereby assisting the decision-making agent in selecting the charge state reference trajectory that best meets the operational requirements.

[0053] The agent-evaluation intelligent agent calculates a comprehensive evaluation value for candidate trajectories that have passed the feasibility review. The comprehensive evaluation value may include fuel consumption agent items, photovoltaic consumption agent items, terminal recovery agent items, all-day coordination agent items, and constraint violation penalty items.

[0054] Specifically, for candidate trajectories that have passed the feasibility review, fuel consumption proxy indicators, photovoltaic absorption proxy indicators, terminal state of charge recovery indicators, all-day coordination indicators, and constraint violation indicators are calculated. The decision-making agent selects the candidate trajectory with the optimal comprehensive evaluation value based on the fuel consumption proxy indicators, photovoltaic absorption proxy indicators, terminal state of charge recovery indicators, all-day coordination indicators, and constraint violation indicators, and interpolates its hourly anchor points as minute-level SOC reference trajectories.

[0055] S105. Based on the state of charge reference trajectory, identify the falling segment, the rising segment, and the terminal recovery segment, and determine the photovoltaic charging margin creation mode, the discharge reserve formation mode, and the terminal state of charge recovery mode based on the falling segment, the rising segment, and the terminal recovery segment, respectively.

[0056] The operation mode recognition and dynamic weight design module calculates mode trigger indicators based on the current demand characteristics, SOC reference trajectory, and real-time operation status. If the future photovoltaic surplus level is high enough to reach the first mode threshold and the current SOC needs to be reduced to create charging space, the declining segment is identified and the photovoltaic charging margin creation mode is selected. If a high-load period is approaching and the discharge reserve demand is high enough to reach the second mode threshold, the rising segment is identified and the discharge reserve formation mode is selected. If the day enters the terminal period and the SOC deviates from the target SOC range, the terminal recovery segment is identified and the terminal state of charge recovery mode is selected.

[0057] When the battery temperature, battery current, or SOC boundary risk is high, the energy storage protection priority mode is determined.

[0058] When the load power fluctuation is less than the first fluctuation threshold, the photovoltaic output fluctuation is less than the second fluctuation threshold, and the current state of charge is within the comfort zone, the economic smoothing mode is determined.

[0059] When multiple modes meet the triggering conditions simultaneously, the energy storage protection priority mode has the highest priority, and the other modes are determined according to the preset priority or comprehensive triggering indicators; if the triggering conditions of any mode are not met, the preset basic weight combination is adopted.

[0060] Obtain battery current prediction, determine battery temperature prediction based on ambient temperature prediction and battery current prediction, and determine energy storage protection priority mode when battery temperature prediction exceeds a first temperature threshold or battery current prediction exceeds a first current threshold.

[0061] S106. According to the photovoltaic charging margin creation mode, the discharge reserve formation mode, or the terminal state of charge recovery mode, update the weight coefficients in the objective function of the nonlinear model predictive control. The weight coefficients include photovoltaic curtailment penalty weights, state of charge tracking weights, and terminal state of charge penalty weights.

[0062] The weights of the objective function in the control layer differ under different operating modes.

[0063] When in the photovoltaic charging margin creation mode, increase the weight of photovoltaic curtailment penalty or reduce the tendency to maintain excessively high SOC, so that energy storage can prioritize the absorption of photovoltaic power.

[0064] When in discharge standby formation mode, increase the weight of SOC maintenance and standby related factors to ensure that the energy storage maintains usable energy before high load.

[0065] When in terminal state of charge recovery mode, increase the terminal SOC weight so that the end-of-day SOC returns to the target range.

[0066] When in protection priority mode, the weights of battery current, battery temperature, and boundary risk are increased, while the weight of state of charge tracking is decreased.

[0067] When in economic smoothing mode, increase the penalty weight for load changes in diesel generator sets and increase the weight for fuel economy.

[0068] The weighting coefficients are updated according to the operating mode through table lookup, rule mapping, or online calculation.

[0069] The dynamically updated weighting coefficients include at least two of the following: fuel economy weight, state of charge tracking weight, terminal state of charge weight, battery current weight, battery temperature weight, diesel generator set operation weight, and photovoltaic curtailment penalty weight.

[0070] S107. Based on the state of charge reference trajectory and the updated weighting coefficients, generate the diesel generator set output power command, the energy storage system charging and discharging power command, and the photovoltaic utilization power command through nonlinear model predictive control.

[0071] The nonlinear model predictive control module establishes a predictive model for the vehicle-mounted emergency microgrid.

[0072] The power balance relationship of the common bus can be expressed as the combined output power of the diesel generator set, the charging and discharging power of the energy storage, and the photovoltaic utilization power meeting the load power demand. The fuel consumption of the diesel generator set can be modeled using a quasi-steady-state model related to the output power. The state of charge (SOC) of the energy storage can be updated recursively based on the battery current or charging and discharging power. The battery temperature can be predicted using a first-order thermal model.

[0073] The nonlinear model predictive control objective function may include fuel consumption terms, SOC reference trajectory tracking terms, battery current penalty terms, battery temperature penalty terms, energy storage operating pressure penalty terms, diesel generator power change penalty terms, photovoltaic curtailment penalty terms, and terminal SOC recovery terms at the end of the prediction time domain. Among them, the SOC reference trajectory tracking term is used to measure the deviation between the predicted SOC and the SOC reference trajectory output by the upper layer, and the weights of each term are dynamically given by the operating mode determined by the upper layer.

[0074] Nonlinear model predictive control constraints include, but are not limited to: The constraints include: power balance of the common bus, minimum and maximum output power of the diesel generator set, charging and discharging power of the energy storage system, upper and lower limits of energy storage SOC, upper and lower limits of battery current, upper limit of battery temperature, and the constraint that the photovoltaic utilization power shall not exceed the available photovoltaic power.

[0075] The controller solves the predictive control problem in each control cycle, only executing the diesel generator power command, energy storage power command, and photovoltaic utilization power command for the current cycle, and then re-optimizes based on the new operating conditions in the next cycle.

[0076] Please refer to Figure 4 As shown, this application also provides a multi-agent charge state planning and dynamic prediction control system, including: The acquisition module 201 is used to acquire load power prediction, photovoltaic output prediction and ambient temperature prediction; Extraction module 202 is used to extract peak load gap, photovoltaic surplus, discharge reserve requirement and charging margin requirement from the load power prediction, the photovoltaic output prediction and the ambient temperature prediction. Boundary module 203 is used to determine the upper and lower boundaries of the allowable state of charge corridor of the energy storage system based on the peak load gap, the photovoltaic surplus, the discharge reserve requirement, and the charging margin requirement; The segmentation module 204 is used to generate a state of charge reference trajectory based on the upper boundary and the lower boundary. The state of charge reference trajectory is located between the upper boundary and the lower boundary and has a falling segment, a rising segment and a terminal recovery segment. The mode module 205 is used to identify the falling segment, the rising segment, and the terminal recovery segment according to the state of charge reference trajectory, and to determine the photovoltaic charging margin creation mode, the discharge reserve formation mode, and the terminal state of charge recovery mode according to the falling segment, the rising segment, and the terminal recovery segment, respectively. The penalty module 206 is used to update the weight coefficients in the objective function of the nonlinear model predictive control according to the photovoltaic charging margin creation mode, the discharge reserve formation mode, or the terminal state of charge recovery mode. The weight coefficients include photovoltaic curtailment penalty weights, state of charge tracking weights, and terminal state of charge penalty weights. The instruction module 207 is used to generate output power instructions for diesel generator sets, charging and discharging power instructions for energy storage systems, and photovoltaic utilization power instructions through nonlinear model predictive control based on the state of charge reference trajectory and the updated weight coefficients.

[0077] Optionally, the boundary module includes: The first floating unit is used to raise the lower boundary according to the peak load gap and the discharge reserve requirement; The second floating unit is used to lower the lower boundary according to the photovoltaic surplus and the charging margin requirement; The third floating unit is used to raise the upper boundary according to the discharge backup requirements; The fourth floating unit is used to lower the upper boundary according to the photovoltaic surplus level; A boundary determination unit is used to determine the final position of the upper boundary and the final position of the lower boundary based on the floating direction of the upper boundary and the lower boundary, wherein the upper boundary does not exceed the maximum state of charge of the energy storage system and the lower boundary is not lower than the minimum state of charge of the energy storage system.

[0078] Optionally, the segmentation module includes: The first trajectory generation unit is used to generate a backup priority trajectory, wherein the state of charge of the backup priority trajectory is higher than the first state of charge level. The second trajectory generation unit is used to generate a photovoltaic consumption priority trajectory, wherein the state of charge of the photovoltaic consumption priority trajectory is lower than the second state of charge level; The third trajectory generation unit is used to generate a balanced operation trajectory, wherein the state of charge of the balanced operation trajectory is between the state of charge of the backup priority trajectory and the state of charge of the photovoltaic consumption priority trajectory. The fourth trajectory generation unit is used to generate a terminal recovery trajectory, wherein the terminal charge state of the terminal recovery trajectory tends to the target charge state; The fifth trajectory generation unit is used to generate a comprehensive coordinated trajectory, the state of charge of which is determined based on the backup priority trajectory, the photovoltaic consumption priority trajectory, the balanced operation trajectory, and the terminal recovery trajectory. The projection unit is used to project the backup priority trajectory, the photovoltaic absorption priority trajectory, the balanced operation trajectory, the terminal recovery trajectory, and the integrated coordination trajectory onto the area between the upper boundary and the lower boundary, respectively. A smoothing unit is used to smooth each projected trajectory to obtain multiple smoothed candidate trajectories, and to determine one of the multiple smoothed candidate trajectories as the reference trajectory for the state of charge.

[0079] The above embodiments are provided to enable those skilled in the art to understand and apply this application. Those skilled in the art will readily make various modifications to the above embodiments and apply the general principles described herein to other embodiments without inventive effort. Therefore, this application is not limited to the above embodiments, and any improvements and modifications made to this application based on the disclosure thereof should be within the scope of protection of this application.

Claims

1. A multi-agent charge state planning and dynamic predictive control method, characterized in that, include: Obtain load power forecast, photovoltaic output forecast, and ambient temperature forecast; Peak load gap, photovoltaic surplus, discharge reserve requirement, and charging margin requirement are extracted from the load power prediction, photovoltaic output prediction, and ambient temperature prediction. Based on the peak load gap, the photovoltaic surplus, the discharge reserve requirement, and the charging margin requirement, determine the upper and lower boundaries of the allowable state of charge corridor for the energy storage system; A state of charge reference trajectory is generated based on the upper boundary and the lower boundary. The state of charge reference trajectory is located between the upper boundary and the lower boundary and has a falling segment, a rising segment and a terminal recovery segment. Based on the state of charge reference trajectory, the falling segment, the rising segment, and the terminal recovery segment are identified, and the photovoltaic charging margin creation mode, the discharge reserve formation mode, and the terminal state of charge recovery mode are determined based on the falling segment, the rising segment, and the terminal recovery segment, respectively. Based on the photovoltaic charging margin creation mode, the discharge reserve formation mode, or the terminal state of charge recovery mode, update the weight coefficients in the objective function of the nonlinear model predictive control. The weight coefficients include the photovoltaic curtailment penalty weight, the state of charge tracking weight, and the terminal state of charge penalty weight. Based on the state of charge reference trajectory and the updated weighting coefficients, the output power command of the diesel generator set, the charging and discharging power command of the energy storage system, and the photovoltaic utilization power command are generated through nonlinear model predictive control.

2. The multi-agent state-of-charge planning and dynamic predictive control method according to claim 1, characterized in that, The determination of the upper and lower boundaries of the allowable corridor for the state of charge of the energy storage system includes: The lower boundary is raised based on the peak load gap and the discharge reserve requirement; The lower boundary is lowered based on the photovoltaic surplus and the charging margin requirement. Based on the aforementioned discharge backup requirements, the upper boundary is raised; The upper boundary is lowered based on the photovoltaic surplus level. Based on the floating directions of the upper and lower boundaries, the final positions of the upper and lower boundaries are determined, wherein the upper boundary does not exceed the maximum state of charge of the energy storage system, and the lower boundary is not lower than the minimum state of charge of the energy storage system.

3. The multi-agent state-of-charge planning and dynamic predictive control method according to claim 1, characterized in that, Based on the upper boundary and the lower boundary, a charge state reference trajectory is generated, including: Generate a backup priority trajectory, wherein the state of charge of the backup priority trajectory is higher than the first state of charge level; A photovoltaic (PV) grid connection priority trajectory is generated, wherein the state of charge of the PV grid connection priority trajectory is lower than the second state of charge level; A balanced operating trajectory is generated, wherein the state of charge of the balanced operating trajectory is between the state of charge of the backup priority trajectory and the state of charge of the photovoltaic consumption priority trajectory. Generate a terminal recovery trajectory, wherein the terminal charge state of the terminal recovery trajectory tends to the target charge state; A comprehensive coordinated trajectory is generated, and the state of charge of the comprehensive coordinated trajectory is determined based on the backup priority trajectory, the photovoltaic consumption priority trajectory, the balanced operation trajectory, and the terminal recovery trajectory; The backup priority trajectory, the photovoltaic consumption priority trajectory, the balanced operation trajectory, the terminal recovery trajectory, and the comprehensive coordination trajectory are projected between the upper boundary and the lower boundary, respectively. Each projected trajectory is smoothed to obtain multiple smoothed candidate trajectories, and one of these smoothed candidate trajectories is selected as the reference trajectory for the state of charge.

4. The multi-agent state-of-charge planning and dynamic predictive control method according to claim 1, characterized in that, The step of generating a charge state reference trajectory based on the upper boundary and the lower boundary includes: Generate multiple candidate trajectories; A feasibility review is performed on the multiple candidate trajectories, the feasibility review including: A first penalty is applied to candidate trajectories whose change in state of charge exceeds a first change threshold in adjacent time periods; A second penalty is applied to candidate trajectories whose second-order change in state of charge exceeds a second change threshold in adjacent time periods; A third penalty is applied to candidate trajectories whose initial state of charge deviates from the current actual state of charge by more than a first deviation threshold. A fourth penalty is applied to candidate trajectories that exceed the upper or lower boundary. A fifth penalty is applied to candidate trajectories whose pre-charge state is below the first reserve threshold during high load periods; The state of charge reference trajectory is determined from the plurality of candidate trajectories based on the first penalty, the second penalty, the third penalty, the fourth penalty, and the fifth penalty.

5. The multi-agent state-of-charge planning and dynamic predictive control method according to claim 4, characterized in that, The step of determining the state-of-charge reference trajectory from the plurality of candidate trajectories based on the first penalty, the second penalty, the third penalty, the fourth penalty, and the fifth penalty includes: Calculate fuel consumption proxy indicators for candidate trajectories that have passed the feasibility review; Calculate photovoltaic grid integration proxy indicators for candidate trajectories that have passed the feasibility review; Calculate the terminal state of charge recovery index for candidate trajectories that have passed the feasibility review; Calculate the all-day coordination index for candidate trajectories that have passed the feasibility review; The constraints for calculating candidate trajectories that have passed the feasibility review are violated. The state of charge reference trajectory is selected from the candidate trajectories that have passed the feasibility review, based on the fuel consumption proxy index, the photovoltaic absorption proxy index, the terminal state of charge recovery index, the all-day coordination index, and the constraint violation index.

6. The multi-agent state-of-charge planning and dynamic predictive control method according to claim 1, characterized in that, Also includes: Obtain battery current prediction; The battery temperature prediction is determined based on the ambient temperature prediction and the battery current prediction. When the predicted battery temperature exceeds a first temperature threshold or the predicted battery current exceeds a first current threshold, an energy storage protection priority mode is determined. When the energy storage protection priority mode is determined, the weighting coefficient is updated. The weighting coefficient also includes battery current penalty weight and battery temperature penalty weight, and the state of charge tracking weight is reduced.

7. The multi-agent state-of-charge planning and dynamic predictive control method according to claim 1, characterized in that, Also includes: Acquire load power fluctuations, photovoltaic output fluctuations, and current state of charge; When the load power fluctuation is less than the first fluctuation threshold, the photovoltaic output fluctuation is less than the second threshold, and the current state of charge is within the comfort zone, an economic smoothing mode is determined. When the economic smoothing mode is determined, the load change penalty weight of the diesel generator set is increased, and the fuel economy weight is increased.

8. A multi-agent state-of-charge planning and dynamic predictive control system, characterized in that, include: The acquisition module is used to acquire load power prediction, photovoltaic output prediction, and ambient temperature prediction. The extraction module is used to extract peak load gap, photovoltaic surplus, discharge reserve requirement and charging margin requirement from the load power prediction, the photovoltaic output prediction and the ambient temperature prediction. The boundary module is used to determine the upper and lower boundaries of the allowable state of charge corridor of the energy storage system based on the peak load gap, the photovoltaic surplus, the discharge reserve requirement, and the charging margin requirement. The segmentation module is used to generate a state of charge reference trajectory based on the upper boundary and the lower boundary. The state of charge reference trajectory is located between the upper boundary and the lower boundary and has a falling segment, a rising segment and a terminal recovery segment. The mode module is used to identify the falling segment, the rising segment, and the terminal recovery segment based on the state of charge reference trajectory, and to determine the photovoltaic charging margin creation mode, the discharge reserve formation mode, and the terminal state of charge recovery mode based on the falling segment, the rising segment, and the terminal recovery segment, respectively. The penalty module is used to update the weight coefficients in the objective function of the nonlinear model predictive control according to the photovoltaic charging margin creation mode, the discharge reserve formation mode, or the terminal state of charge recovery mode. The weight coefficients include photovoltaic curtailment penalty weights, state of charge tracking weights, and terminal state of charge penalty weights. The instruction module is used to generate output power instructions for diesel generator sets, charging and discharging power instructions for energy storage systems, and photovoltaic utilization power instructions through nonlinear model predictive control based on the state of charge reference trajectory and the updated weighting coefficients.

9. A multi-agent state-of-charge planning and dynamic prediction control system according to claim 8, characterized in that, The boundary module includes: The first floating unit is used to raise the lower boundary according to the peak load gap and the discharge reserve requirement; The second floating unit is used to lower the lower boundary according to the photovoltaic surplus and the charging margin requirement; The third floating unit is used to raise the upper boundary according to the discharge backup requirements; The fourth floating unit is used to lower the upper boundary according to the photovoltaic surplus level; A boundary determination unit is used to determine the final position of the upper boundary and the final position of the lower boundary based on the floating direction of the upper boundary and the lower boundary, wherein the upper boundary does not exceed the maximum state of charge of the energy storage system and the lower boundary is not lower than the minimum state of charge of the energy storage system.

10. A multi-agent state-of-charge planning and dynamic prediction control system according to claim 8, characterized in that, The segmentation module includes: The first trajectory generation unit is used to generate a backup priority trajectory, wherein the state of charge of the backup priority trajectory is higher than the first state of charge level. The second trajectory generation unit is used to generate a photovoltaic consumption priority trajectory, wherein the state of charge of the photovoltaic consumption priority trajectory is lower than the second state of charge level; The third trajectory generation unit is used to generate a balanced operation trajectory, wherein the state of charge of the balanced operation trajectory is between the state of charge of the backup priority trajectory and the state of charge of the photovoltaic consumption priority trajectory. The fourth trajectory generation unit is used to generate a terminal recovery trajectory, wherein the terminal charge state of the terminal recovery trajectory tends to the target charge state; The fifth trajectory generation unit is used to generate a comprehensive coordinated trajectory, the state of charge of which is determined based on the backup priority trajectory, the photovoltaic consumption priority trajectory, the balanced operation trajectory, and the terminal recovery trajectory. The projection unit is used to project the backup priority trajectory, the photovoltaic absorption priority trajectory, the balanced operation trajectory, the terminal recovery trajectory, and the integrated coordination trajectory onto the area between the upper boundary and the lower boundary, respectively. A smoothing unit is used to smooth each projected trajectory to obtain multiple smoothed candidate trajectories, and to determine one of the multiple smoothed candidate trajectories as the reference trajectory for the state of charge.

Images