A scheduling method for integrated grid energy storage based on MPC of light-storage-chip coordination planning and operation

Abstract

A kind of scheduling method of MPC-based photovoltaic-storage-diesel collaborative planning-operation integrated grid energy storage, including steps: constructing the energy storage capacity optimization model of minimum annual operation frequency modulation cost in upper layer, establishing short-term photovoltaic load shedding frequency modulation optimization model in lower layer with the economy and frequency stability of system actual operation as target, global optimization is realized by using improved golden search optimization algorithm GSO combined with fuzzy membership decision;Traditional photovoltaic-storage-diesel combined frequency modulation is replaced by photovoltaic-storage-diesel collaborative planning-operation integrated regulation based on model predictive control (MPC), super short-term prediction of photovoltaic output, load and frequency modulation signal is fused, advanced rolling optimization and power distribution are realized;Synchronous completion of energy storage SOC adaptive recovery, and support on-grid and off-grid adaptive operation.The present application realizes deep cooperation of planning and operation, effectively improves frequency stability and reduces operation cost, solves problems such as low inertia, frequency modulation response lag and energy storage SOC imbalance in high proportion photovoltaic microgrid.

Description

Technical Field

[0001] This invention relates to the field of target detection technology, and specifically to a scheduling method for grid energy storage based on MPC-integrated planning and operation of photovoltaic-storage-diesel co-planning. Background Technology

[0002] High grid-connected photovoltaic (PV) systems reduce the system's equivalent inertia and weaken its frequency support capability. In traditional maximum power point tracking (MPPT) mode, PV units always output maximum power without any active power reserve. When grid frequency fluctuates, PV cannot actively participate in frequency regulation, placing the regulation burden entirely on conventional generating units (such as diesel generators) and energy storage systems. This operating method is simple but lacks flexibility, especially when PV penetration is high, as conventional generating units struggle to respond quickly to frequent power fluctuations.

[0003] Existing technologies largely focus on energy storage participation in frequency regulation. For example, they utilize energy storage systems to provide virtual inertia and primary frequency regulation support, or employ joint frequency regulation of energy storage and diesel engines. However, these studies typically assume that the energy storage capacity is predetermined, failing to integrate energy storage capacity planning with photovoltaic load shedding operations. This leads to problems such as unreasonable capacity allocation (too large a capacity results in wasted investment, while too small a capacity leads to insufficient frequency regulation), high frequency regulation costs, and insufficient frequency stability. Furthermore, single-timescale scheduling (such as day-ahead scheduling or real-time control only) struggles to balance long-term economic efficiency with short-term dynamic response: long-term models cannot capture second-level frequency fluctuations, while short-term models lack a holistic consideration of equipment lifespan and investment costs.

[0004] Traditional photovoltaic-storage-diesel frequency regulation is a passive response control: the system monitors frequency deviation in real time, and when the deviation exceeds the dead zone, it issues commands according to a fixed ratio (such as droop control) or simple logic (such as energy storage acting first, diesel engine acting later). This approach lacks advance prediction of photovoltaic output, load fluctuations, and frequency regulation signals, which can easily cause drastic fluctuations in energy storage SOC and lag in recovery—for example, during continuous low-frequency events, the energy storage continues to discharge, and once the SOC drops to the lower limit, it cannot continue to support the system, leading to frequency regulation failure.

[0005] Furthermore, a fragmented "plan first, then operate" model is commonly adopted: energy storage capacity is determined based on typical daily data in the annual planning phase, and operation is carried out independently, with no closed-loop feedback between medium- and long-term capacity configuration and real-time operation. This "insufficient global optimization capability" leads to the following drawbacks: ① Planned capacity cannot adapt to seasonal and random changes in actual operation, resulting in resource waste or insufficient capacity; ② Operational data cannot correct planning parameters, leaving the system's economy and reliability in a suboptimal state for a long time; ③ It is impossible to achieve dynamic coordination between energy storage lifespan and frequency regulation needs, accelerating equipment aging. Summary of the Invention

[0006] To address the aforementioned technical issues, this technical solution provides a grid energy storage scheduling method based on MPC-integrated planning and operation of photovoltaic, energy storage, and diesel energy storage. This method achieves optimal energy storage configuration, reasonable participation of photovoltaic load reduction, and a win-win situation of frequency stability and economical operation, effectively solving the aforementioned technical problems.

[0007] This invention is achieved through the following technical solution:

[0008] A scheduling method for grid energy storage based on MPC-integrated planning and operation of photovoltaic-storage-diesel collaborative grid energy storage includes the following steps:

[0009] Step 1: Construct an energy storage capacity optimization model that minimizes the annual frequency regulation cost;

[0010] Step 2: Construct a short-term photovoltaic load shedding frequency regulation optimization model with the goal of improving the economy and frequency stability of the system during actual operation.

[0011] Step 3: Use the improved Golden Search optimization algorithm (GSO) combined with fuzzy membership decision to achieve global optimization;

[0012] Step 4: Based on MPC, the photovoltaic-storage-diesel coordinated planning and operation integrated control is carried out. A multi-time-scale MPC control framework is built, which integrates ultra-short-term photovoltaic output forecasting, load forecasting, and grid frequency regulation signal forecasting to construct a system state prediction model. The model makes rolling predictions on frequency regulation demand and energy output changes in the next 1-24 hours, and realizes closed-loop coordination between planning and operation.

[0013] Step 5: Adaptive operation between grid and off-grid, real-time monitoring of grid status, participation in system frequency regulation in grid-connected mode; islanded mode maintains stable voltage and frequency to ensure power supply to critical loads.

[0014] Furthermore, the optimization model described in step one takes minimizing the overall system frequency regulation cost for the entire year as its sole optimization objective. It comprehensively considers the entire lifecycle cost of energy storage, the operating cost of new energy sources, the frequency regulation cost of conventional units, and the costs of benefits and penalties. This achieves optimal planning of energy storage power and capacity, providing a basic capacity configuration basis for the lower-level short-term (24-hour) photovoltaic load shedding frequency regulation optimization model, which aims at the economic efficiency and frequency stability of the system during actual operation. Its objective function comprehensively covers all dimensions of cost and benefit terms, and the specific expression and meaning of each sub-item are as follows:

[0015] ;

[0016] C stin Annualized investment cost for energy storage: The initial investment in energy storage is converted into an annualized cost using a capital recovery factor, reflecting the long-term investment in energy storage power and capacity configuration. The expression is as follows:

[0017] ;

[0018] in, This is the capital recovery coefficient.

[0019] ;

[0020] r is the benchmark discount rate, N set Design lifespan for energy storage; c p P represents the investment cost per unit power of energy storage. ess c is the rated power of the energy storage. c E represents the investment cost per unit capacity of energy storage. ess This refers to the rated capacity of the energy storage.

[0021] C stop Annualized operation and maintenance cost of energy storage: This includes fixed expenses such as daily inspection, monitoring, and manual operation and maintenance of energy storage equipment. It is calculated according to a fixed ratio of energy storage power and capacity and is an annual fixed cost item.

[0022] C pv Cost of photovoltaic curtailment: The cost of power generation loss caused by photovoltaic curtailment to meet system frequency regulation requirements is directly related to the photovoltaic curtailment power, curtailment duration, and unit electricity price.

[0023] C diesel The cost of diesel engine frequency regulation: fuel consumption, unit wear and maintenance costs during the frequency regulation process of diesel engine are positively correlated with diesel engine output, gradient, and running time.

[0024] C sell Revenue from electricity supply: The revenue from selling electricity generated by the photovoltaic-storage-diesel system to the grid is a cost deduction item.

[0025] C pen Subsidized revenue: Policy subsidies and ancillary service revenue obtained by the system from participating in power grid frequency regulation services further offset the overall costs.

[0026] C pun Frequency regulation penalty costs: When the system's frequency regulation response is not timely, the frequency deviation exceeds the standard, or the output fails to meet the standard, the penalty fees charged by the grid side are an additional cost item.

[0027] The constraints include:

[0028] Solar load shedding range constraints: , where P pv_cur P represents the actual load shedding power of the photovoltaic system. pv_max The maximum output power of photovoltaic power should be set at the current level, and excessive reduction of photovoltaic power output should be prohibited to avoid resource waste.

[0029] Energy storage SOC and power limit constraints: , To ensure that the energy storage's state of charge is within a safe range, the charging and discharging power does not exceed the rated value, and to avoid overcharging and over-discharging.

[0030] Diesel engine power and gradeability constraints: , This limits the minimum stable output, maximum output, and gradient of the diesel engine per unit time, thus protecting the operational stability of the unit.

[0031] Solution method: Typical daily data for each of the four seasons is used instead of hourly data for the whole year to simplify the calculation while ensuring seasonal differences. An improved Golden Search Optimization Algorithm (GSO) is used iteratively to solve the problem and output the optimal rated power P of the energy storage. ess With rated capacity E ess Complete the long-term capacity planning of the upper layer.

[0032] Furthermore, the frequency regulation optimization model described in step two has two objectives: minimizing frequency regulation cost and minimizing frequency deviation. It considers energy storage charging and discharging losses as well as lifespan degradation. Monte Carlo sampling is used to simulate photovoltaic prediction error scenarios on a 5-minute timescale. Constraints are the same as the upper layer, but adaptive recovery of energy storage SOC is added. With the lower layer's 5-minute short timescale as the core, the model has two optimization objectives: minimizing system frequency regulation cost and minimizing grid frequency deviation. It balances energy storage operating losses and lifespan degradation, accurately matches real-time frequency regulation requirements, and optimizes the allocation of reserve capacity for each unit. Its objective function is:

[0033] FM cost minimization objective C min Focusing on real-time operational loss costs, the expression is:

[0034] ;

[0035] Among them, C loss Real-time cost of energy storage charging and discharging power loss and line transmission loss; C stlife The cost of lifespan loss due to deep charging and discharging of energy storage and frequent switching of operating conditions is calculated and directly linked to the depth of charging and discharging and the number of cycles; the meanings of the remaining sub-items are consistent with the upper-level model, adapting to real-time calculations on short time scales. pv To reduce the load on photovoltaic systems, C diesel For the cost of diesel engine frequency regulation, C pun This is to reduce the cost of frequency modulation penalties.

[0036] The objective J for minimizing frequency deviation is to ensure the stability of the power grid frequency, expressed as follows: Δf is the deviation between the actual grid frequency and the rated frequency (50Hz). The goal is to control the frequency deviation within the allowable range and improve the power supply quality.

[0037] The scenario simulation and constraints include:

[0038] Photovoltaic prediction error scenario simulation: The Monte Carlo sampling method is used to generate a large number of photovoltaic power output prediction error scenarios to simulate the randomness and fluctuation of photovoltaic power output in actual operation, cover the error range under different illumination conditions, and improve the adaptability of the model.

[0039] Basic constraints: The photovoltaic load reduction, energy storage power, diesel engine power and ramping constraints of the upper-level model are used to ensure the consistency of constraints between the upper and lower-level models.

[0040] New constraint: Energy storage SOC adaptive recovery constraint, which sets a target range for dynamic SOC recovery, avoids excessive deviation of SOC from the optimal range due to frequent frequency adjustments on a short time scale, and provides a basic constraint basis for MPC regulation.

[0041] Time scale: It adopts a short time granularity of 5 minutes to match the rapid changes in the real-time frequency regulation demand of the power grid, so as to realize the fine allocation of frequency regulation reserve capacity.

[0042] Furthermore, the improved Gold Search Optimization Algorithm (GSO) described in step three introduces a variable step size operator and an adaptive mutation strategy to balance global search and local exploitation, and avoid premature convergence.

[0043] The variable step size operator is:

[0044] ;

[0045] Where T is the variable step size operator, exp is the natural exponential function, q is the current iteration number, and qmax=100 is the maximum iteration number;

[0046] The step size update formula is:

[0047] ;

[0048] Where T is the variable step size operator, Sstep'(q) represents the step size of the q-th iteration, and Sstep'(q+1) is the updated step size; TSstep'(q) represents the product of the variable step size operator and the current step size, used to decay the step size to balance global search and local exploitation; cos and sin represent cosine and sine respectively; c1 and c2 are learning factors, with c1=c2=1.5; r1 and r2 are uniformly distributed random numbers in [0,1]; Xpbest is the historical best position of the i-th particle; Xgbest is the global best position; and Xi is the current position of the i-th particle in the q-th iteration.

[0049] The particle position update formula is:

[0050] ;

[0051] In the above formula, Xi(q) is the current position of the i-th particle; Xi(q+1) is the updated position;

[0052] The adaptive mutation probability is:

[0053] ;

[0054] pm is the adaptive mutation probability, which is used to control the probability of particle mutation; qmax is the maximum number of iterations; that is, the mutation probability is low in the early iteration stage and increases in the late iteration stage; for each particle, a random number randϵ[0,1] is generated, if rand<pm, the position of the particle is re-randomly initialized within the solution space to maintain population diversity.

[0055] Further, in step three, the fuzzy membership decision is constructed to perform lower-layer bi-objective optimization to generate a Pareto optimal solution set, and the global optimal compromise solution is selected by the fuzzy membership function; the membership degree of the j-th objective of the i-th solution is:

[0056] For the Pareto solution set, calculate the membership degree of each solution to each objective:

[0057] ;

[0058] Where Fi,j is the function value of the i-th solution under the j-th objective, Ij and Mj are the minimum and maximum values of this objective respectively; λ ij is the membership degree of the i-th solution to the j-th objective, and a value closer to 1 indicates that the solution is better on this objective.

[0059] Comprehensive membership degree:

[0060] ;

[0061] Where, n p is the number of solutions in the Pareto solution set; λ i is the comprehensive membership degree value of the i-th solution; select the solution with the maximum λ i as the global optimal solution, and determine the photovoltaic load shedding rate d and the output Pgs of the diesel engine, so as to provide an optimization reference for the subsequent real-time MPC regulation.

[0062] The lower-layer bi-objective optimization model will generate a Pareto optimal solution set, the fuzzy membership function is used for quantitative evaluation of the solution set, the weights of frequency regulation cost and frequency deviation are comprehensively considered, the global optimal compromise solution is screened out from the Pareto solution set, the optimal photovoltaic load shedding rate and real-time output scheme of the diesel engine are finally determined, which provides an optimization reference for subsequent real-time MPC regulation.

[0063] Furthermore, the integrated control described in step four includes a segmented nonlinear adaptive recovery mechanism for the energy storage SOC. This mechanism uses a charging coefficient Kc and a discharging coefficient Kd based on an S-shaped curve to dynamically adjust the recovery intensity according to the current SOC's location in the danger zone, warning zone, or normal zone.

[0064] The multi-timescale MPC control framework includes a prediction module, a rolling optimization module, and a feedback correction module;

[0065] The prediction module integrates three types of ultra-short-term prediction data:

[0066] Photovoltaic output ultra-short-term forecast: Based on historical irradiance and meteorological data, a time series forecast model is used to generate photovoltaic output forecasts for the next 1 to 24 hours in a rolling manner, with a resolution of 5 minutes;

[0067] Load forecasting: Based on historical load patterns and temperature and holiday information, generate load forecast curves for the next 1 to 24 hours;

[0068] Power grid frequency regulation signal prediction: Receive historical sequences of AGC commands from the power grid dispatch center and combine them with load fluctuation trends to predict frequency regulation demand for the next 1 to 24 hours: power increase / decrease commands;

[0069] The rolling optimization module constructs a system state prediction model and adopts a model predictive control algorithm. It takes the frequency regulation demand and energy output changes in the next 1 to 24 hours as inputs, and aims to minimize the system operating cost and frequency deviation. It solves the optimal control sequence in real time and executes only the first control command at the current moment.

[0070] The feedback correction module compares the actual sampled values ​​with the predicted values, corrects the predicted model state at the next moment, and forms a closed-loop control. The actual sampled values ​​include: energy storage SOC, frequency deviation, actual diesel engine output, and actual photovoltaic load reduction.

[0071] Furthermore, the system state prediction model described in step four adopts a linear discrete state-space model as the prediction model, and its structure includes:

[0072] ;

[0073] Where x(k), u(k), w(k), and v(k) are the state variable, control variable, process noise, and measurement noise, respectively, and x(k+1) represents the state variable vector of the system at the next moment, which is one of the core outputs of the MPC prediction model; A, B, and C are the system matrices, which are identified based on the actual parameters of the microgrid (inertia constant, damping coefficient, energy storage charging and discharging efficiency, etc.).

[0074] Furthermore, the MPC described in step four executes a two-layer rolling optimization process, which includes the outer layer SOC performing trajectory optimization once every 1 hour and the inner layer power allocation performing rolling optimization once every 5 minutes.

[0075] The trajectory optimization includes the following steps:

[0076] (1) Read the actual SOCS(k) and diesel engine output Pgs(k) at the current time k;

[0077] (2) Obtain photovoltaic output, load, and frequency modulation signal prediction data for the next 6 hours from the prediction module;

[0078] (3) With the objective of minimizing the system operating cost over the next 6 hours, solve the quadratic programming problem to obtain the optimal SOC reference trajectory for the next 6 hours. } and the corresponding charge and discharge power sequence;

[0079] (4) The first control command u(k) is sent to each execution unit. The first control command u(k) includes the energy storage charging and discharging power, photovoltaic load reduction rate, and diesel engine adjustment amount.

[0080] (5) Discard the remaining control sequence and wait for the next sampling period to re-optimize;

[0081] The aforementioned scrolling optimization includes the following steps:

[0082] (1) Based on the current measured SOC value, ultra-short-term prediction (future 15 minutes, resolution 5 minutes) and actual frequency deviation;

[0083] (2) Allocate energy storage charging and discharging power, photovoltaic load reduction and diesel engine adjustment in real time according to the scenario-specific control strategy;

[0084] (3) Generate a power instruction sequence within 5 minutes, and execute only the first instruction;

[0085] (4) Collect feedback quantities (frequency, SOC, power) and correct the inner prediction model.

[0086] Furthermore, the MPC scenario-specific frequency over-limit control strategy described in step four:

[0087] Frequency overshoot limit: MPC prioritizes energy storage charging, reduces diesel engine output, and, if necessary, reduces photovoltaic load to proactively suppress frequency rise; frequency overshoot limit regulation ( The actual frequency is higher than the rated frequency.

[0088] When the grid frequency is too high, the MPC allocates control commands according to the priority order of "energy storage first - diesel engines second - photovoltaic as a backup" to proactively suppress the frequency rise:

[0089] First priority: Issue energy storage charging instructions to utilize the excess active power of the energy storage system, quickly suppress the rising frequency trend, and give full play to the advantage of fast response speed of energy storage.

[0090] Second priority: Control the diesel engine to reduce its output, reduce the active power supply of the system, avoid frequent start-stop of the diesel engine, and ensure stable operation of the unit;

[0091] Third priority: If the first two steps of regulation still cannot suppress the frequency exceeding the standard, start the photovoltaic system to reduce the photovoltaic grid-connected power, reduce the excess active power of the system from the source, until the frequency returns to the rated range.

[0092] Frequency low limit violation: MPC prioritizes photovoltaic power generation and energy storage discharge support, with the insufficient portion quickly supplemented by diesel engines; frequency low limit violation regulation ( The actual frequency is lower than the rated frequency.

[0093] When the grid frequency is low, the MPC issues commands according to the priority of "photovoltaics first, energy storage second, diesel engine supplement" to quickly make up for the active power deficit and suppress frequency drop:

[0094] First priority: instruct the photovoltaic power generation to be increased to full capacity within the maximum output range of the photovoltaic system, making full use of clean energy and reducing the operating time of diesel engines;

[0095] Second priority: Control the discharge of energy storage, quickly release active power to support the grid, make up for short-term active power deficit, and control the response time to the millisecond level;

[0096] Third priority: If energy storage and photovoltaic output cannot meet frequency regulation requirements, immediately instruct the diesel engine to quickly increase its output, utilize the stable output characteristics of the diesel engine to make up for the remaining active power gap, and ensure rapid frequency recovery.

[0097] By relying on MPC rolling optimization to dynamically set the energy storage SOC segment adaptive recovery coefficient, the SOC is maintained in the optimal range, avoiding SOC imbalance caused by passive frequency regulation and ensuring continuous frequency regulation capability.

[0098] Furthermore, the MPC mentioned in step four is dynamically set by the segmented adaptive recovery of the energy storage SOC:

[0099] Set the energy storage SOC segmented adaptive recovery coefficient, including the charging coefficient Kc and the discharging coefficient Kd; this recovery coefficient adopts a piecewise nonlinear function, with a smooth transition through an S-curve to avoid sudden changes in recovery power; the parameters are set based on engineering experience of lithium iron phosphate battery BMS (Battery Management System): reference recovery coefficient K E =0.8 dimensionless, representing the ratio of maximum recovery power to rated power), steepness coefficient n=10 (controlling the slope of the S-curve).

[0100] The SOC threshold is defined as follows: (Lower limit danger value) (Upper limit danger value) (Lower warning value) (Lower value) (Higher warning level) (One-order threshold too high);

[0101] The expression for the charging coefficient Kc is:

[0102] ;

[0103] The expression for the discharge coefficient Kd is:

[0104] .

[0105] Furthermore, the energy storage SOC segmented adaptive recovery MPC dynamic optimization is as follows:

[0106] To address the issues of energy storage SOC imbalance and insufficient continuous frequency regulation capability caused by traditional passive frequency regulation, this paper utilizes MPC rolling optimization to dynamically set segmented adaptive recovery coefficients for energy storage SOC.

[0107] 1. Divide the SOC into optimal operating range, warning range, and danger range, and adaptively adjust the SOC recovery speed and target value based on the current SOC range and future frequency regulation demand prediction;

[0108] 2. When the SOC is in the optimal range, maintain stable charging and discharging, and only make small adjustments according to the frequency regulation requirements; when the SOC is close to the lower / upper warning value, the MPC will issue a charging / discharging recovery command first, and use the frequency regulation interval to bring the SOC back to the optimal range.

[0109] 3. By segmented adaptive control, overcharging and over-discharging of the SOC are prevented, ensuring that the energy storage always has sufficient frequency regulation reserve capacity and guaranteeing the system's continuous frequency regulation capability.

[0110] Furthermore, the specific process for dynamic adjustment described in step four includes:

[0111] S1: SOC interval division:

[0112] Lower danger zone: Warning; Lower limit zone: Normal area Warning upper limit zone: Dangerous upper limit zone: ;

[0113] S2: Dynamic adjustment of the recovery coefficient:

[0114] In each rolling optimization cycle, the MPC controller calculates the recovery coefficients Kc and Kd (discharge coefficients) in real time based on the current SOC range and future frequency regulation requirements.

[0115] When the SOC is in the danger lower limit area, the MPC forcibly issues a charging command, Kc=1, Kd=0, and stops discharging;

[0116] When the SOC is in the lower warning limit zone, MPC prioritizes charging recovery, Kc gradually increases from 0.2 to 0.8 according to the S-shaped curve, and Kd is limited to below 0.2;

[0117] When the SOC is in the normal range, Kc and Kd are adjusted in a balanced manner according to the frequency modulation requirements to maintain SOC stability.

[0118] When the SOC is in the warning upper limit area, MPC prioritizes discharge recovery, Kd increases, and Kc decreases;

[0119] When the SOC is in the dangerous upper limit area, the MPC forcibly issues a discharge command, Kd=1, Kc=0, and stops charging;

[0120] S3: Generation of recovery commands:

[0121] MPC multiplies the recovery factor by the frequency modulation power requirement to obtain the final charge and discharge power command, thereby actively pulling the SOC back to the normal range while meeting the frequency modulation requirements.

[0122] Furthermore, the iterative mechanism of the planning and operation closed loop described in step four is as follows: summarize the operation data every 24 hours, revise the cost coefficient in step one based on the accumulated operation data every 90 days and re-run the optimization model in step one, and use the newly obtained energy storage rated power and capacity as the updated boundary constraints for MPC regulation in steps two and four.

[0123] The MPC is used to achieve a two-way closed loop between upper-layer capacity planning and lower-layer real-time frequency modulation commands. The specific internal operation is as follows:

[0124] (1) Parameter transfer from upper layer to lower layer, that is, parameter transfer from offline to online:

[0125] ① The planning parameters obtained from the upper-level optimization, such as the rated energy storage power Pess, the rated capacity Eess, the optimal photovoltaic load reduction rate range [0, dmax], and the economic output range of the diesel engine, are stored in the parameter table of the central controller.

[0126] ② When the lower-level MPC starts, it reads these boundary constraints from the parameter table as hard constraints for rolling optimization;

[0127] (2) Real-time feedback from the lower to the upper layers, i.e., correction from online to offline:

[0128] During real-time operation of MPC, the central controller continuously records the following operational data, with a storage period of 1 day:

[0129] ① Energy storage charge / discharge depth, cycle count, number of times SOC exceeds limit, and duration;

[0130] ② Actual fuel consumption, number of start-stop cycles, and number of hill climbs of the diesel engine;

[0131] ③ The actual number of times and amount of photovoltaic load shedding;

[0132] ④ The number of times the frequency deviation exceeded the limit and the extent of the deviation;

[0133] ⑤ The actual cost of frequency modulation penalties and the subsidies received.

[0134] Furthermore, the specific operational steps of the closed-loop iteration described in step four are as follows:

[0135] (1) Data collection and cleaning: Every 24 hours, the controller will summarize the operating data, remove outliers, and calculate the daily average indicators;

[0136] (2) Cost parameter adjustment: Every quarter, the cost coefficients in the upper objective function are adjusted using accumulated operating data.

[0137] If the actual energy storage life loss is greater than the design value (e.g., the actual number of cycles exceeds the expectation), then the energy storage life loss cost coefficient should be increased.

[0138] If the actual frequency modulation penalty cost is higher than the preset value, then increase the penalty cost coefficient c. pun ;

[0139] If the actual diesel engine fuel consumption is higher than the theoretical value, then the k, m, and n coefficients in the diesel engine cost function are adjusted.

[0140] (3) Rerun the upper-level optimization: Substitute the corrected cost coefficient into the upper-level energy storage capacity optimization model, run the improved GSO algorithm again, and obtain the new optimal energy storage power and capacity;

[0141] (4) Parameter update and distribution: Update the new energy storage capacity configuration parameters to the parameter table and distribute them to the lower-level MPC controller; if the difference between the old and new parameters exceeds the threshold (e.g., capacity change > 5%), issue a prompt and suggest that the system expand or reduce the energy storage capacity.

[0142] (5) Rolling execution: The above correction and optimization process is automatically executed once a quarter, forming a long-term closed loop of "planning → operation → feedback → correction → replanning", so that the energy storage capacity configuration always adapts to the changes in actual operating conditions.

[0143] Furthermore, step five, the adaptive operation between the network and off-grid, includes:

[0144] Grid-connected mode operation:

[0145] Real-time monitoring of grid connection status, frequency and voltage signals; when the system is stably connected to the grid, it participates in the primary / secondary frequency regulation of the grid throughout the entire process.

[0146] 1. Receive frequency regulation commands from the power grid dispatch center, coordinate with the MPC control framework, quickly respond to power grid frequency fluctuations, and allocate photovoltaic, energy storage, and diesel power output according to predetermined strategies;

[0147] 2. Simultaneously calculate the benefits and costs of frequency regulation to maximize the economic benefits of the system while meeting the power grid frequency regulation assessment requirements and avoiding penalty costs.

[0148] Island mode operation:

[0149] When off-grid signals such as grid faults or power outages are detected, the system switches to islanded operation mode within milliseconds, with the core objective shifting to maintaining voltage and frequency stability in the islanded system.

[0150] 1. Suspend grid frequency regulation related commands, MPC quickly switches control logic, and prioritizes ensuring continuous power supply to critical loads;

[0151] 2. Real-time balancing of photovoltaic output, load demand, energy storage, and diesel engine output within the island; short-term power fluctuations are suppressed through rapid charging and discharging of energy storage, while the diesel engine provides stable base power;

[0152] 3. Strictly control the frequency and voltage deviation of the islanded system within the allowable range to avoid load power outages and equipment damage. After the power grid returns to normal, seamlessly switch back to grid-connected mode to achieve a smooth switching between grid connection and off-grid operation.

[0153] Beneficial effects

[0154] The present invention proposes a grid energy storage scheduling method based on MPC (Multi-Planet Control) for coordinated planning and operation of photovoltaic, energy storage, and diesel power grids. Compared with existing technologies, this method has the following advantages:

[0155] (1) The MPC advance prediction and rolling optimization in this invention significantly improves the response speed and accuracy of frequency regulation. It uses ultra-short-term prediction (photovoltaic output, load demand, and grid frequency regulation signal) to construct a linear discrete state-space prediction model and adopts dual-time-scale rolling optimization: the outer layer predicts the optimal SOC trajectory for the next 6 hours with a period of 1 hour, and the inner layer rolls power allocation with a period of 5 minutes. The MPC controller generates the control sequence in advance based on the prediction results, so that the frequency regulation command acts in advance, rather than the "lagging response" of traditional droop control, thus solving the problem of "lagging frequency regulation response".

[0156] (2) The planning-operation closed-loop coordination in this invention achieves optimal configuration of energy storage capacity and improved economic efficiency. It breaks the traditional "plan first, then operate" fragmented model and forms a closed loop by integrating the upper-level annual scale energy storage capacity optimization model with the lower-level MPC real-time operation through data feedback. Every 24 hours, the operation data (number of times the energy storage SOC exceeds the limit, frequency regulation penalty cost, diesel engine fuel consumption, actual number of cycles, etc.) is summarized. Every 90 days, the cost coefficient, energy storage life loss coefficient, and diesel engine cost coefficient in the upper-level objective function are corrected, and the improved GSO algorithm is run again to optimize the energy storage capacity.

[0157] (3) The SOC segmented nonlinear adaptive recovery in this invention uses a segmented S-shaped function of the designed charging and discharging coefficients. A threshold is set based on the engineering parameters of the lithium iron phosphate battery BMS, and the MPC dynamically adjusts the recovery coefficient according to the current SOC range in each rolling optimization cycle. When the SOC deviates from the normal range, the MPC prioritizes charging or discharging recovery commands to actively pull the SOC back to the optimal range, significantly enhancing the continuous frequency regulation capability; effectively solving the problem of "easy imbalance of energy storage SOC".

[0158] (4) This invention features grid-connected and off-grid adaptive operation to ensure power supply reliability in islanded mode; it also configures solid-state relays as grid-connected / off-grid switching devices with a switching time in milliseconds. The central controller monitors the grid status in real time. When connected to the grid, the MPC follows the AGC command to participate in primary / secondary frequency regulation; when islanded, the MPC automatically switches to VF (voltage-frequency) control mode, and the energy storage, as the main power source, adopts droop control to quickly respond to load fluctuations. The photovoltaic system maintains MPPT output, and the diesel engine serves as a backup. This achieves the goals of improving frequency stability, reducing operating costs, and enhancing the continuous frequency regulation capability of energy storage. Attached Figure Description

[0159] Figure 1 This is a schematic diagram of the overall process of the present invention.

[0160] Figure 2 This is the integrated control diagram of photovoltaic, energy storage and diesel co-planning and operation based on MPC in this invention.

[0161] Figure 3 This is a structural diagram of the multi-timescale bilayer optimization model in this invention. Detailed Implementation

[0162] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments. The described embodiments are merely some embodiments of the present invention, and not all embodiments. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the design concept of the present invention should fall within the protection scope of the present invention.

[0163] Example 1:

[0164] like Figure 1 As shown, a scheduling method for grid energy storage based on MPC-integrated planning and operation of photovoltaic-storage-diesel collaborative grid energy storage includes the following steps:

[0165] Step 1: Construct an energy storage capacity optimization model that minimizes the annual frequency regulation cost. This optimization model takes minimizing the overall annual system frequency regulation cost as its sole objective, comprehensively considering the entire lifecycle cost of energy storage, the operating cost of new energy sources, the frequency regulation cost of conventional units, and the costs of benefits and penalties. This achieves optimal planning of energy storage power and capacity, providing a basic capacity configuration basis for the lower-level short-term (24-hour) photovoltaic load shedding frequency regulation optimization model, which aims at the economic efficiency and frequency stability of the system during actual operation. Its objective function comprehensively covers all dimensions of cost and benefit items; the specific expression and meaning of each sub-item are as follows:

[0166] ;

[0167] C stinwei Annualized investment cost for energy storage: The initial investment in energy storage is converted into an annualized cost using a capital recovery factor, reflecting the long-term investment in energy storage power and capacity configuration. The expression is as follows:

[0168] ;

[0169] in, This is the capital recovery coefficient.

[0170] ;

[0171] r is the benchmark discount rate, N set Design lifespan for energy storage; c p P represents the investment cost per unit power of energy storage. ess c is the rated power of the energy storage. c E represents the investment cost per unit capacity of energy storage. ess This refers to the rated capacity of the energy storage.

[0172] C stop Annualized operation and maintenance cost of energy storage: This includes fixed expenses such as daily inspection, monitoring, and manual operation and maintenance of energy storage equipment. It is calculated according to a fixed ratio of energy storage power and capacity and is an annual fixed cost item.

[0173] C pv Cost of photovoltaic curtailment: The cost of power generation loss caused by photovoltaic curtailment to meet system frequency regulation requirements is directly related to the photovoltaic curtailment power, curtailment duration, and unit electricity price.

[0174] C dieselThe cost of diesel engine frequency regulation: fuel consumption, unit wear and maintenance costs during the frequency regulation process of diesel engine are positively correlated with diesel engine output, gradient, and running time.

[0175] C sell Revenue from electricity supply: The revenue from selling electricity generated by the photovoltaic-storage-diesel system to the grid is a cost deduction item.

[0176] C pen Subsidized revenue: Policy subsidies and ancillary service revenue obtained by the system from participating in power grid frequency regulation services further offset the overall costs.

[0177] C pun Frequency regulation penalty costs: When the system's frequency regulation response is not timely, the frequency deviation exceeds the standard, or the output fails to meet the standard, the penalty fees charged by the grid side are an additional cost item.

[0178] The constraints include:

[0179] Solar load shedding range constraints: , where P pv_cur P represents the actual load shedding power of the photovoltaic system. pv_max The maximum output power of photovoltaic power should be set at the current level, and excessive reduction of photovoltaic power output should be prohibited to avoid resource waste.

[0180] Energy storage SOC and power limit constraints: , To ensure that the energy storage's state of charge is within a safe range, the charging and discharging power does not exceed the rated value, and to avoid overcharging and over-discharging.

[0181] Diesel engine power and gradeability constraints: , This limits the minimum stable output, maximum output, and gradient of the diesel engine per unit time, thus protecting the operational stability of the unit.

[0182] Solution method: Typical daily data for each of the four seasons is used instead of hourly data for the whole year to simplify the calculation while ensuring seasonal differences. An improved Golden Search Optimization Algorithm (GSO) is used iteratively to solve the problem and output the optimal rated power P of the energy storage. ess With rated capacity E ess Complete the long-term capacity planning of the upper layer.

[0183] Step 2: Construct a short-term photovoltaic load shedding frequency regulation optimization model with the goal of improving the economy and frequency stability of the system during actual operation.

[0184] The frequency regulation optimization model described above has two objectives: minimizing frequency regulation cost and minimizing frequency deviation. It considers energy storage charging and discharging losses as well as lifespan degradation. Monte Carlo sampling is used to simulate photovoltaic prediction error scenarios on a 5-minute timescale. Constraints are the same as the upper layer, but adaptive recovery of energy storage SOC is added. The lower-level 5-minute short timescale is the core, with the dual optimization objectives of minimizing system frequency regulation cost and minimizing grid frequency deviation. It also considers energy storage operating losses and lifespan degradation, accurately matching real-time frequency regulation requirements and optimizing the allocation of reserve capacity for each unit. Its objective function is:

[0185] FM cost minimization objective C min Focusing on real-time operational loss costs, the expression is:

[0186] ;

[0187] Among them, C loss Real-time cost of energy storage charging and discharging power loss and line transmission loss; C stlife The cost of lifespan loss due to deep charging and discharging of energy storage and frequent switching of operating conditions is calculated and directly linked to the depth of charging and discharging and the number of cycles; the meanings of the remaining sub-items are consistent with the upper-level model, adapting to real-time calculations on short time scales. pv To reduce the load on photovoltaic systems, C diesel For the cost of diesel engine frequency regulation, C pun This is to reduce the cost of frequency modulation penalties.

[0188] The objective J for minimizing frequency deviation is to ensure the stability of the power grid frequency, expressed as follows: Δf is the deviation between the actual grid frequency and the rated frequency (50Hz). The goal is to control the frequency deviation within the allowable range and improve the power supply quality.

[0189] The scenario simulation and constraints include:

[0190] Photovoltaic prediction error scenario simulation: The Monte Carlo sampling method is used to generate a large number of photovoltaic power output prediction error scenarios to simulate the randomness and fluctuation of photovoltaic power output in actual operation, cover the error range under different illumination conditions, and improve the adaptability of the model.

[0191] Basic constraints: The photovoltaic load reduction, energy storage power, diesel engine power and ramping constraints of the upper-level model are used to ensure the consistency of constraints between the upper and lower-level models.

[0192] New constraint: Energy storage SOC adaptive recovery constraint, which sets a target range for dynamic SOC recovery, avoids excessive deviation of SOC from the optimal range due to frequent frequency adjustments on a short time scale, and provides a basic constraint basis for MPC regulation.

[0193] Time scale: It adopts a short time granularity of 5 minutes to match the rapid changes in the real-time frequency regulation demand of the power grid, so as to realize the fine allocation of frequency regulation reserve capacity.

[0194] Step 3: The improved Golden Search Optimization Algorithm (GSO) combined with fuzzy membership decision-making is used to achieve global optimization. The improved GSO algorithm introduces a variable step size operator and an adaptive mutation strategy to balance global search and local development, and avoid premature convergence.

[0195] The variable step size operator is:

[0196] ;

[0197] The variable step size operator is:

[0198] ;

[0199] Where T is the variable step size operator, exp is the natural exponential function, q is the current iteration number, and qmax=100 is the maximum iteration number;

[0200] The step size update formula is:

[0201] ;

[0202] Where T is the variable step size operator, Sstep'(q) represents the step size of the q-th iteration, and Sstep'(q+1) is the updated step size; TSstep'(q) represents the product of the variable step size operator and the current step size, used to decay the step size to balance global search and local exploitation; cos and sin represent cosine and sine respectively; c1 and c2 are learning factors, with c1=c2=1.5; r1 and r2 are uniformly distributed random numbers in [0,1]; Xpbest is the historical best position of the i-th particle; Xgbest is the global best position; and Xi is the current position of the i-th particle in the q-th iteration.

[0203] The particle position update formula is:

[0204] ;

[0205] In the above formula, Xi(q) is the current position of the i-th particle; Xi(q+1) is the updated position;

[0206] The adaptive mutation probability is:

[0207] ;

[0208] pm is the adaptive mutation probability, which is used to control the probability of particle mutation; qmax is the maximum number of iterations; that is, the mutation probability is low in the early iteration stage, and increases in the late iteration stage; for each particle, a random number randϵ[0,1] is generated, if rand<pm, the position of this particle is randomly reinitialized in the solution space to maintain population diversity.

[0209] The fuzzy membership degree decision described constructs the lower-level bi-objective optimization to generate a Pareto optimal solution set, and the global optimal compromise solution is selected by the fuzzy membership function; the membership degree of the j-th objective of the i-th solution is:

[0210] For the Pareto solution set, calculate the membership degree of each solution for each objective:

[0211] ;

[0212] Wherein, F i,j is the function value of the i-th solution under the j-th objective, Ij and Mj are the minimum value and maximum value of this objective respectively; λ ij is the membership degree of the i-th solution to the j-th objective, a value closer to 1 indicates that the solution is more optimal for this objective.

[0213] Comprehensive membership degree:

[0214] ;

[0215] Wherein, n p is the number of solutions in the Pareto solution set; select the solution with the maximum λ i as the global optimal solution, determine the photovoltaic load shedding rate d and the output Pgs of the diesel engine, so as to provide an optimization reference for subsequent MPC real-time regulation.

[0216] The lower-level bi-objective optimization model generates a Pareto optimal solution set, the fuzzy membership function is used to quantitatively evaluate the solution set, the weights of frequency regulation cost and frequency deviation are comprehensively considered, the global optimal compromise solution is screened from the Pareto solution set, the optimal photovoltaic load shedding rate and real-time output scheme of the diesel engine are finally determined, so as to provide an optimization reference for subsequent MPC real-time regulation.

[0217] Step 4: MPC-based integrated planning-operation coordinated regulation of photovoltaic-energy storage-diesel system, a multi-time scale MPC regulation framework is established, which integrates ultra-short-term photovoltaic output prediction, load prediction and power grid frequency regulation signal prediction, constructs a system state prediction model, and performs rolling prediction on frequency regulation demand and energy output change in the next 1 to 24 hours, so as to realize closed-loop coordination between planning and operation.

[0218] Integrated regulation includes a piecewise nonlinear adaptive recovery mechanism for energy storage SOC. This mechanism uses a charging coefficient Kc and a discharging coefficient Kd based on an S-shaped curve to dynamically adjust the recovery intensity according to the current SOC's location within the danger zone, warning zone, or normal zone. The specific steps of integrated regulation are as follows: Figure 2 As shown.

[0219] like Figure 3 As shown, the multi-timescale MPC control framework includes a prediction module, a rolling optimization module, and a feedback correction module.

[0220] The prediction module integrates three types of ultra-short-term prediction data:

[0221] Photovoltaic power output ultra-short-term forecast: Based on historical irradiance and meteorological data, a time series forecast model is used to generate photovoltaic power output forecasts for the next 1 to 24 hours in a rolling manner, with a resolution of 5 minutes.

[0222] Load forecasting: Based on historical load patterns, temperature, and holiday information, load forecast curves for the next 1 to 24 hours are generated.

[0223] Power grid frequency regulation signal prediction: Based on the historical sequence of AGC commands received from the power grid dispatch center and combined with load fluctuation trends, predict the frequency regulation demand for the next 1 to 24 hours: power increase / decrease commands.

[0224] The rolling optimization module constructs a system state prediction model and adopts a model predictive control algorithm. It takes the frequency regulation demand and energy output changes in the next 1 to 24 hours as inputs, and aims to minimize the system operating cost and frequency deviation. It solves the optimal control sequence in real time and executes only the first control command at the current moment.

[0225] The feedback correction module compares the actual sampled values ​​with the predicted values, corrects the predicted model state at the next moment, and forms a closed-loop control. The actual sampled values ​​include: energy storage SOC, frequency deviation, actual diesel engine output, and actual photovoltaic load reduction.

[0226] The system state prediction model uses a linear discrete state-space model as the prediction model, and its structure includes:

[0227] ;

[0228] Where x(k), u(k), w(k), and v(k) are the state variable, control variable, process noise, and measurement noise, respectively; x(k+1) represents the system's state variable vector at the next time step, which is one of the core outputs of the MPC prediction model; A, B, and C are the system matrices, identified based on the actual parameters of the microgrid (inertia constant, damping coefficient, energy storage charging and discharging efficiency, etc.). x(k+1) represents the system's state variable at the next time step.

[0229] MPC executes a two-layer rolling optimization process, which includes an outer layer SOC that performs trajectory optimization every hour and an inner layer power allocation that performs rolling optimization every 5 minutes.

[0230] The trajectory optimization includes the following steps:

[0231] (1) Read the actual SOCS(k) and diesel engine output Pgs(k) at the current time k;

[0232] (2) Obtain photovoltaic output, load, and frequency modulation signal prediction data for the next 6 hours from the prediction module;

[0233] (3) With the objective of minimizing the system operating cost over the next 6 hours, solve the quadratic programming problem to obtain the optimal SOC reference trajectory for the next 6 hours. } and the corresponding charge and discharge power sequence;

[0234] (4) The first control command u(k) is sent to each execution unit. The first control command u(k) includes the energy storage charging and discharging power, photovoltaic load reduction rate, and diesel engine adjustment amount.

[0235] (5) Discard the remaining control sequence and wait for the next sampling period to re-optimize.

[0236] The aforementioned scrolling optimization includes the following steps:

[0237] (1) Based on the current measured SOC value, ultra-short-term prediction (future 15 minutes, resolution 5 minutes) and actual frequency deviation;

[0238] (2) Allocate energy storage charging and discharging power, photovoltaic load reduction and diesel engine adjustment in real time according to the scenario-specific control strategy;

[0239] (3) Generate a power instruction sequence within 5 minutes, and execute only the first instruction;

[0240] (4) Collect feedback quantities (frequency, SOC, power) and correct the inner prediction model.

[0241] MPC frequency over-limit control strategy based on different scenarios:

[0242] Frequency overshoot limit: MPC prioritizes energy storage charging, reduces diesel engine output, and, if necessary, reduces photovoltaic load to proactively suppress frequency rise; frequency overshoot limit regulation ( The actual frequency is higher than the rated frequency.

[0243] When the grid frequency is too high, the MPC allocates control commands according to the priority order of "energy storage first - diesel engines second - photovoltaic as a backup" to proactively suppress the frequency rise:

[0244] First priority: Issue energy storage charging instructions to utilize the excess active power of the energy storage system, quickly suppress the rising frequency trend, and give full play to the advantage of fast response speed of energy storage.

[0245] Second priority: Control the diesel engine to reduce its output, reduce the active power supply of the system, avoid frequent start-stop of the diesel engine, and ensure stable operation of the unit;

[0246] Third priority: If the first two steps of regulation still cannot suppress the frequency exceeding the standard, start the photovoltaic system to reduce the photovoltaic grid-connected power, reduce the excess active power of the system from the source, until the frequency returns to the rated range.

[0247] Frequency low limit violation: MPC prioritizes photovoltaic power generation and energy storage discharge support, with the insufficient portion quickly supplemented by diesel engines; frequency low limit violation regulation ( The actual frequency is lower than the rated frequency.

[0248] When the grid frequency is low, the MPC issues commands according to the priority of "photovoltaics first, energy storage second, diesel engine supplement" to quickly make up for the active power deficit and suppress frequency drop:

[0249] First priority: instruct the photovoltaic power generation to be increased to full capacity within the maximum output range of the photovoltaic system, making full use of clean energy and reducing the operating time of diesel engines;

[0250] Second priority: Control the discharge of energy storage, quickly release active power to support the grid, make up for short-term active power deficit, and control the response time to the millisecond level;

[0251] Third priority: If energy storage and photovoltaic output cannot meet frequency regulation requirements, immediately instruct the diesel engine to quickly increase its output, utilize the stable output characteristics of the diesel engine to make up for the remaining active power gap, and ensure rapid frequency recovery.

[0252] By relying on MPC rolling optimization to dynamically set the energy storage SOC segment adaptive recovery coefficient, the SOC is maintained in the optimal range, avoiding SOC imbalance caused by passive frequency regulation and ensuring continuous frequency regulation capability.

[0253] MPC is dynamically set by segmented adaptive recovery of energy storage SOC:

[0254] Set the energy storage SOC segmented adaptive recovery coefficient, including the charging coefficient Kc and the discharging coefficient Kd; this recovery coefficient adopts a piecewise nonlinear function, with a smooth transition through an S-curve to avoid sudden changes in recovery power; the parameters are set based on engineering experience of lithium iron phosphate battery BMS (Battery Management System): reference recovery coefficient K E =0.8 dimensionless, representing the ratio of maximum recovery power to rated power), steepness coefficient n=10 (controlling the slope of the S-curve).

[0255] The SOC threshold is defined as follows: (Lower limit danger value) (Upper limit danger value) (Lower warning value) (Lower value) (Higher warning level) (One-order threshold is too high).

[0256] The expression for the charging coefficient Kc is:

[0257] ;

[0258] The expression for the discharge coefficient Kd is:

[0259] .

[0260] Energy storage SOC segmented adaptive recovery MPC dynamic optimization:

[0261] To address the issues of energy storage SOC imbalance and insufficient continuous frequency regulation capability caused by traditional passive frequency regulation, this paper utilizes MPC rolling optimization to dynamically set segmented adaptive recovery coefficients for energy storage SOC.

[0262] 1. Divide the SOC into optimal operating range, warning range, and danger range, and adaptively adjust the SOC recovery speed and target value based on the current SOC range and future frequency regulation demand prediction;

[0263] 2. When the SOC is in the optimal range, maintain stable charging and discharging, and only make small adjustments according to the frequency regulation requirements; when the SOC is close to the lower / upper warning value, the MPC will issue a charging / discharging recovery command first, and use the frequency regulation interval to bring the SOC back to the optimal range.

[0264] 3. By segmented adaptive control, overcharging and over-discharging of the SOC are prevented, ensuring that the energy storage always has sufficient frequency regulation reserve capacity and guaranteeing the system's continuous frequency regulation capability.

[0265] The specific process for dynamic adjustment includes:

[0266] S1: SOC interval division:

[0267] Lower danger zone: Warning; Lower limit zone: Normal area Warning upper limit zone: Dangerous upper limit zone: ;

[0268] S2: Dynamic adjustment of the recovery coefficient:

[0269] In each rolling optimization cycle, the MPC controller calculates the recovery coefficients Kc and Kd (discharge coefficients) in real time based on the current SOC range and future frequency regulation requirements.

[0270] When the SOC is in the danger lower limit area, the MPC forcibly issues a charging command, Kc=1, Kd=0, and stops discharging;

[0271] When the SOC is in the lower warning limit zone, MPC prioritizes charging recovery, Kc gradually increases from 0.2 to 0.8 according to the S-shaped curve, and Kd is limited to below 0.2;

[0272] When the SOC is in the normal range, Kc and Kd are adjusted in a balanced manner according to the frequency modulation requirements to maintain SOC stability.

[0273] When the SOC is in the warning upper limit area, MPC prioritizes discharge recovery, Kd increases, and Kc decreases;

[0274] When the SOC is in the dangerous upper limit area, the MPC forcibly issues a discharge command, Kd=1, Kc=0, and stops charging;

[0275] S3: Generation of recovery commands:

[0276] MPC multiplies the recovery factor by the frequency modulation power requirement to obtain the final charge and discharge power command, thereby actively pulling the SOC back to the normal range while meeting the frequency modulation requirements.

[0277] The iterative mechanism of the planning and operation closed loop is as follows: summarize the operation data every 24 hours, revise the cost coefficient in step one based on the accumulated operation data every 90 days and rerun the optimization model in step one, and use the newly obtained energy storage rated power and capacity as the updated boundary constraints for MPC regulation in steps two and four.

[0278] The MPC is used to achieve a two-way closed loop between upper-layer capacity planning and lower-layer real-time frequency modulation commands. The specific internal operation is as follows:

[0279] (1) Parameter transfer from upper layer to lower layer, that is, parameter transfer from offline to online:

[0280] ① The planning parameters obtained from the upper-level optimization, such as the rated energy storage power Pess, the rated capacity Eess, the optimal photovoltaic load reduction rate range [0, dmax], and the economic output range of the diesel engine, are stored in the parameter table of the central controller.

[0281] ② When the lower-level MPC starts, it reads these boundary constraints from the parameter table as hard constraints for rolling optimization;

[0282] (2) Real-time feedback from the lower to the upper layers, i.e., correction from online to offline:

[0283] During real-time operation of MPC, the central controller continuously records the following operational data, with a storage period of 1 day:

[0284] ① Energy storage charge / discharge depth, cycle count, number of times SOC exceeds limit, and duration;

[0285] ② Actual fuel consumption, number of start-stop cycles, and number of hill climbs of the diesel engine;

[0286] ③ The actual number of times and amount of photovoltaic load shedding;

[0287] ④ The number of times the frequency deviation exceeded the limit and the extent of the deviation;

[0288] ⑤ The actual cost of frequency modulation penalties and the subsidies received.

[0289] The specific steps of closed-loop iteration are as follows:

[0290] (1) Data collection and cleaning: Every 24 hours, the controller will summarize the operating data, remove outliers, and calculate the daily average indicators;

[0291] (2) Cost parameter adjustment: Every quarter, the cost coefficients in the upper objective function are adjusted using accumulated operating data.

[0292] If the actual energy storage life loss is greater than the design value (e.g., the actual number of cycles exceeds the expectation), then the energy storage life loss cost coefficient should be increased.

[0293] If the actual frequency modulation penalty cost is higher than the preset value, then increase the penalty cost coefficient c. pun ;

[0294] If the actual diesel engine fuel consumption is higher than the theoretical value, then the k, m, and n coefficients in the diesel engine cost function are adjusted.

[0295] (3) Rerun the upper-level optimization: Substitute the corrected cost coefficient into the upper-level energy storage capacity optimization model, run the improved GSO algorithm again, and obtain the new optimal energy storage power and capacity;

[0296] (4) Parameter update and distribution: Update the new energy storage capacity configuration parameters to the parameter table and distribute them to the lower-level MPC controller; if the difference between the old and new parameters exceeds the threshold (e.g., capacity change > 5%), issue a prompt and suggest that the system expand or reduce the energy storage capacity.

[0297] (5) Rolling execution: The above correction and optimization process is automatically executed once a quarter, forming a long-term closed loop of "planning → operation → feedback → correction → replanning", so that the energy storage capacity configuration always adapts to the changes in actual operating conditions.

[0298] Step 5: Adaptive operation between grid and off-grid, real-time monitoring of grid status, participation in system frequency regulation in grid-connected mode; islanding mode maintains voltage and frequency stability to ensure power supply to critical loads; the aforementioned adaptive operation between grid and off-grid includes:

[0299] Grid-connected mode operation:

[0300] Real-time monitoring of grid connection status, frequency and voltage signals; when the system is stably connected to the grid, it participates in the primary / secondary frequency regulation of the grid throughout the entire process.

[0301] 1. Receive frequency regulation commands from the power grid dispatch center, coordinate with the MPC control framework, quickly respond to power grid frequency fluctuations, and allocate photovoltaic, energy storage, and diesel power output according to predetermined strategies;

[0302] 2. Simultaneously calculate the benefits and costs of frequency regulation to maximize the economic benefits of the system while meeting the power grid frequency regulation assessment requirements and avoiding penalty costs.

[0303] Island mode operation:

[0304] When off-grid signals such as grid faults or power outages are detected, the system switches to islanded operation mode within milliseconds, with the core objective shifting to maintaining voltage and frequency stability in the islanded system.

[0305] 1. Suspend grid frequency regulation related commands, MPC quickly switches control logic, and prioritizes ensuring continuous power supply to critical loads;

[0306] 2. Real-time balancing of photovoltaic output, load demand, energy storage, and diesel engine output within the island; short-term power fluctuations are suppressed through rapid charging and discharging of energy storage, while the diesel engine provides stable base power;

[0307] 3. Strictly control the frequency and voltage deviation of the islanded system within the allowable range to avoid load power outages and equipment damage. After the power grid returns to normal, seamlessly switch back to grid-connected mode to achieve a smooth switching between grid connection and off-grid operation.

[0308] The above embodiments are only for illustrating the technical concept and features of the present invention, and are intended to enable those skilled in the art to understand the content of the present invention and implement it accordingly. They should not be construed as limiting the scope of protection of the present invention. All equivalent transformations or modifications made in accordance with the spirit and essence of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A scheduling method for grid energy storage based on MPC-integrated planning and operation of photovoltaic-storage-diesel collaborative grid energy storage, characterized in that, Comprising the following steps: Step 1: Construct an energy storage capacity optimization model with the minimum annual frequency regulation operation cost; Step 2: Construct a short-term photovoltaic load reduction frequency regulation optimization model established with the economy and frequency stability of the system during actual operation as objectives in the lower layer; Step 3: Adopt the improved golden section search optimization algorithm (GSO) combined with fuzzy membership degree decision to realize global optimization; Step 4: Based on MPC-based integrated regulation of photovoltaic-energy storage-diesel collaborative planning-operation, build a multi-time scale MPC regulation framework, integrate ultra-short-term prediction of photovoltaic output, load prediction and power grid frequency regulation signal prediction to construct a system state prediction model, and perform rolling prediction on frequency regulation demand and energy output change in the next 1 to 24 hours, so as to realize closed-loop collaboration between planning and operation; Step 5: Perform grid-connected and off-grid adaptive operation, monitor the power grid state in real time, participate in system frequency regulation in grid-connected mode; maintain voltage and frequency stability and ensure power supply for important loads in island mode.

2. The scheduling method for integrated grid energy storage based on MPC-driven photovoltaic-storage-diesel collaborative planning and operation, as described in claim 1, is characterized in that: In the improved golden section search optimization algorithm GSO described in Step 3, a variable step size operator and an adaptive mutation strategy are introduced to balance global search and local development and avoid premature convergence; The variable step size operator is: ； wherein, T is the variable step size operator, exp is a natural exponential function; q is the current number of iterations, and qmax=100 is the maximum number of iterations; The step size update formula is: ; wherein, T is the variable step size operator, Sstep'(q) represents the step size of the q-th iteration, and Sstep'(q+1) is the updated step size; TSstep'(q) represents the product of the variable step size operator and the current step size, which is used to attenuate the step size to balance global search and local development; cos and sin represent cosine and sine respectively; c1 and c2 are learning factors, with c1=c2=1.5; r1 and r2 are uniformly distributed random numbers within [0, 1]; Xpbest is the historical optimal position of the i-th particle; Xgbest is the global optimal position; Xi is the current position of the i-th particle at the q-th iteration; The particle position update formula is: ； In the above formula, Xi(q) is the current position of the i-th particle; Xi(q+1) is the updated position; The adaptive mutation probability is: ； pm is the adaptive mutation probability, which is used to control the probability of particle mutation; qmax is the maximum number of iterations; that is, the mutation probability is low at the initial stage of iteration, and the mutation probability increases at the later stage of iteration; for each particle, a random number randϵ[0,1] is generated, if rand < pm, the position of the particle is randomly reinitialized within the solution space to maintain population diversity.

3. The scheduling method for integrated grid energy storage planning and operation based on MPC for photovoltaic-storage-diesel co-planning, as described in claim 1, is characterized in that: The fuzzy membership decision described in Step 3 constructs lower-layer bi-objective optimization to generate a Pareto optimal solution set, and selects the global optimal compromise solution through a fuzzy membership function; the membership degree of the j-th objective of the i-th solution is: For the Pareto solution set, calculate the membership degree of each solution for each objective: ； wherein, Fi,j is the function value of the i-th solution under the j-th objective, Ij and Mj are the minimum value and maximum value of the objective respectively; λij is the membership degree of the i-th solution to the j-th objective, and a value closer to 1 indicates that the solution is more optimal in this objective; Comprehensive membership degree: ； Where np is the number of solutions in the Pareto solution set; λi is the comprehensive membership value of the i-th solution; the solution with the largest λi is selected as the global optimal solution, and the photovoltaic load reduction rate d and diesel engine output Pgs are determined, providing an optimization reference for subsequent MPC real-time control.

4. The scheduling method for integrated grid energy storage based on MPC-driven photovoltaic-storage-diesel collaborative planning and operation, as described in claim 1, is characterized in that: The integrated control described in step four includes a segmented nonlinear adaptive recovery mechanism for the energy storage SOC. This mechanism uses a charging coefficient Kc and a discharging coefficient Kd based on an S-shaped curve to dynamically adjust the recovery intensity according to the current SOC's location in the danger zone, warning zone, or normal zone. The multi-timescale MPC control framework includes a prediction module, a rolling optimization module, and a feedback correction module; The prediction module integrates three types of ultra-short-term prediction data: Photovoltaic output ultra-short-term forecast: Based on historical irradiance and meteorological data, a time series forecast model is used to generate photovoltaic output forecasts for the next 1 to 24 hours in a rolling manner, with a resolution of 5 minutes; Load forecasting: Based on historical load patterns and temperature and holiday information, generate load forecast curves for the next 1 to 24 hours; Power grid frequency regulation signal prediction: Receive historical sequences of AGC commands from the power grid dispatch center and combine them with load fluctuation trends to predict frequency regulation demand for the next 1 to 24 hours: power increase / decrease commands; And / or, the rolling optimization module constructs a system state prediction model, adopts a model predictive control algorithm, takes the frequency regulation demand and energy output changes in the next 1 to 24 hours as input, takes the minimum system operating cost and the minimum frequency deviation as objectives, solves the optimal control sequence in real time, and executes only the first control command at the current moment; And / or, the feedback correction module compares the actual sampled values ​​with the predicted values, corrects the prediction model state at the next moment, and forms a closed-loop control. The actual sampled values ​​include: energy storage SOC, frequency deviation, actual diesel engine output, and actual photovoltaic load reduction.

5. The scheduling method for integrated grid energy storage based on MPC-driven photovoltaic-storage-diesel collaborative planning and operation, as described in claim 1, is characterized in that: The system state prediction model described in step four uses a linear discrete state-space model as the prediction model, and its structure includes: ； Where x(k), u(k), w(k), and v(k) are the state variable, control variable, process noise, and measurement noise, respectively; x(k+1) represents the state variable vector of the system at the next moment, which is one of the core outputs of the MPC prediction model. A, B, and C are the system matrices, which are identified based on the actual parameters of the microgrid (inertia constant, damping coefficient, energy storage charging and discharging efficiency, etc.).

6. The scheduling method for integrated grid energy storage based on MPC-driven photovoltaic-storage-diesel collaborative planning and operation, as described in claim 1, is characterized in that: Step four describes the MPC execution of a two-layer rolling optimization process, which includes the outer layer SOC performing trajectory optimization every hour and the inner layer power allocation performing rolling optimization every 5 minutes. The trajectory optimization includes the following steps: (1) Read the actual SOCS(k) and diesel engine output Pgs(k) at the current time k; (2) Obtain photovoltaic output, load, and frequency modulation signal prediction data for the next 6 hours from the prediction module; (3) With the objective of minimizing the system operating cost over the next 6 hours, solve the quadratic programming problem to obtain the optimal SOC reference trajectory for the next 6 hours. } and the corresponding charge and discharge power sequence; (4) The first control command u(k) is sent to each execution unit. The first control command u(k) includes the energy storage charging and discharging power, photovoltaic load reduction rate, and diesel engine adjustment amount. (5) Discard the remaining control sequence and wait for the next sampling period to re-optimize; The aforementioned scrolling optimization includes the following steps: (1) Based on the current measured SOC value, ultra-short-term prediction (future 15 minutes, resolution 5 minutes) and actual frequency deviation; (2) Allocate energy storage charging and discharging power, photovoltaic load reduction and diesel engine adjustment in real time according to the scenario-specific control strategy; (3) Generate a power instruction sequence within 5 minutes, and execute only the first instruction; (4) Collect feedback quantities (frequency, SOC, power) and correct the inner prediction model.

7. The scheduling method for integrated grid energy storage based on MPC-driven photovoltaic-storage-diesel collaborative planning and operation, as described in claim 1, is characterized in that: Step four describes the MPC scenario-specific frequency over-limit control strategy: High frequency exceeding limits: MPC prioritizes energy storage charging, reduces diesel engine output, and reduces photovoltaic load when necessary, thus proactively suppressing frequency rise; Low frequency exceeds limit: MPC prioritizes photovoltaic power generation and energy storage discharge support, with the insufficient part quickly supplemented by diesel engine; By relying on MPC rolling optimization to dynamically set the energy storage SOC segment adaptive recovery coefficient, the SOC is maintained in the optimal range, avoiding SOC imbalance caused by passive frequency regulation and ensuring continuous frequency regulation capability.

8. The scheduling method for integrated grid energy storage based on MPC-driven photovoltaic-storage-diesel collaborative planning and operation, as described in claim 1, is characterized in that: The MPC mentioned in step four is dynamically set by the segmented adaptive recovery of the energy storage SOC: Set the energy storage SOC segmented adaptive recovery coefficient, including the charging coefficient Kc and the discharging coefficient Kd; this recovery coefficient adopts a piecewise nonlinear function, with a smooth transition through an S-curve to avoid sudden changes in recovery power; the parameters are set based on engineering experience of lithium iron phosphate battery BMS (Battery Management System): reference recovery coefficient K E =0.8 dimensionless, representing the ratio of maximum recovery power to rated power), steepness coefficient n=10 (controlling the slope of the S-curve). The SOC threshold is defined as follows: Upper limit danger value , lower value , , ; The expression for the charging coefficient Kc is: ； The expression for the discharge coefficient Kd is: 。 9. The scheduling method for integrated grid energy storage based on MPC-driven photovoltaic-storage-diesel collaborative planning and operation, as described in claim 1, is characterized in that: The specific process of dynamic adjustment described in step four includes: S1: SOC interval division: Lower danger zone: Warning; Lower limit zone: Normal area Warning upper limit zone: Dangerous upper limit zone: ; S2: Dynamic adjustment of the recovery coefficient: In each rolling optimization cycle, the MPC controller calculates the recovery coefficients Kc and Kd (discharge coefficients) in real time based on the current SOC range and future frequency regulation requirements. When the SOC is in the danger lower limit area, the MPC forcibly issues a charging command, Kc=1, Kd=0, and stops discharging; When the SOC is in the lower warning limit zone, MPC prioritizes charging recovery, Kc gradually increases from 0.2 to 0.8 according to the S-shaped curve, and Kd is limited to below 0.2; When the SOC is in the normal range, Kc and Kd are adjusted in a balanced manner according to the frequency modulation requirements to maintain SOC stability. When the SOC is in the warning upper limit area, MPC prioritizes discharge recovery, Kd increases, and Kc decreases; When the SOC is in the dangerous upper limit area, the MPC forcibly issues a discharge command, Kd=1, Kc=0, and stops charging; S3: Generation of recovery commands: MPC multiplies the recovery factor by the frequency modulation power requirement to obtain the final charge and discharge power command, thereby actively pulling the SOC back to the normal range while meeting the frequency modulation requirements.

10. The scheduling method for integrated grid energy storage based on MPC-driven photovoltaic-storage-diesel collaborative planning and operation, as described in claim 1, is characterized in that: The iterative mechanism of the planning and operation closed loop described in step four is as follows: summarize the operation data every 24 hours, revise the cost coefficient in step one based on the accumulated operation data every 90 days and rerun the optimization model in step one, and use the newly obtained energy storage rated power and capacity as the updated boundary constraints for MPC regulation in steps two and four. The MPC is used to achieve a two-way closed loop between upper-layer capacity planning and lower-layer real-time frequency modulation commands. The specific internal operation is as follows: (1) Parameter transfer from upper layer to lower layer, that is, parameter transfer from offline to online: ① The planning parameters obtained from the upper-level optimization, such as the rated energy storage power Pess, the rated capacity Eess, the optimal photovoltaic load reduction rate range [0, dmax], and the economic output range of the diesel engine, are stored in the parameter table of the central controller. ② When the lower-level MPC starts, it reads these boundary constraints from the parameter table as hard constraints for rolling optimization; (2) Real-time feedback from the lower to the upper layers, i.e., correction from online to offline: During real-time operation of MPC, the central controller continuously records the following operational data, with a storage period of 1 day: ① Energy storage charge / discharge depth, cycle count, number of times SOC exceeds limit, and duration; ② Actual fuel consumption, number of start-stop cycles, and number of hill climbs of the diesel engine; ③ The actual number of times and amount of photovoltaic load shedding; ④ The number of times the frequency deviation exceeded the limit and the extent of the deviation; ⑤ The actual cost of frequency modulation penalties and the subsidies received.

Images