Micro-grid energy storage multi-stage economic dispatching method, system and device based on differential evolution and storage medium

Abstract

The invention discloses a micro-grid energy storage multi-stage economic dispatching method, system and device based on differential evolution and a storage medium, solves the problem of how to improve micro-grid energy storage dispatching efficiency, and belongs to the field of power system operation and control. Comprising the following steps: generating a forced charging baseline instruction of which the system net power accords with a photovoltaic surplus period threshold; setting a power upper limit to initialize population individuals of a differential evolution (DE) algorithm, and generating a test vector; superposing the forced charging baseline instruction and the test vector to synthesize a preliminary attempt power, correcting a capacity boundary exceeding a capacity constraint, and outputting a first correction power; superposing the first corrected power and the system net power to synthesize a system net load, and outputting second corrected power when the corrected system net load is greater than a power boundary of a threshold value; constructing a high-fidelity objective function to execute DE optimization, and outputting an optimal individual to perform microgrid energy storage scheduling control; according to the method, the nonlinear battery life loss and the refined peak-valley arbitrage economy are taken into consideration under the condition that the physical operation safety of the bottom layer is guaranteed.

Description

Technical Field

[0001] This invention belongs to the field of power system operation and control technology, and particularly relates to energy storage optimization scheduling technology in microgrid energy management systems. Specifically, this invention relates to a multi-stage economic scheduling method, system, device and storage medium for microgrid energy storage based on differential evolution, which combines multi-source data prediction and heuristic optimization algorithms. Background Technology

[0002] Energy storage systems are a core regulating resource for microgrids to smooth power fluctuations and achieve peak-valley arbitrage to reduce electricity costs. However, existing energy storage dispatch and control technologies still face many challenges in practical applications, mainly in the following three aspects: 1. The rule-based logic control strategy has a rigidity problem: Some existing low-cost systems still rely on manually set fixed charging and discharging periods (such as timer control). This method is too mechanical and cannot adapt to the temporary adjustment of time-of-use electricity prices or the complex multi-peak electricity prices in the spot market. Especially when there is a time sequence conflict between "photovoltaic consumption" and "off-peak electricity price", or when there is a sudden change in photovoltaic output, rule control is difficult to make the optimal trade-off, which can easily lead to curtailment of solar power or illegal backfeeding.

[0003] 2. Mathematical programming-based methods can lead to distortion in physical modeling: While mixed-integer linear programming is currently the mainstream method in academia and industry, in order to meet the standard mathematical format requirements of solvers, this method forces the simplification of complex nonlinear electrochemical energy storage models into linear models; this leads to two serious consequences: First, the battery life is overdrawn. Because the nonlinear decay of battery cycle life with discharge depth and rate is ignored, the calculated "theoretical optimal solution" often accelerates battery aging in actual implementation. Secondly, the action oscillates, often assuming that the battery charging and discharging efficiency is constant and ignoring the high loss characteristics in the low power range, resulting in a large number of invalid "glitch" actions (frequent small start-stops) in the strategy, which is not worth the effort.

[0004] 3. Traditional heuristic algorithms suffer from a contradiction between "safety and optimization capability": While traditional heuristic algorithms such as genetic algorithms and particle swarm optimization can handle nonlinear models, they perform poorly when dealing with strongly constrained problems. Due to the randomness of their search mechanism, the generated strategies are prone to violating the hard constraints of the power grid (such as instantaneous transformer overload or battery SOC exceeding limits). Existing solutions typically add a huge penalty term to the objective function, but this leads to difficulties in algorithm convergence and cannot fundamentally eliminate the safety risks of physical limits exceeding limits. Summary of the Invention

[0005] To address the challenge of simultaneously ensuring the safety of underlying physical operations (such as preventing transformer overload and battery overcharging / over-discharging) while also balancing nonlinear battery lifespan degradation with refined peak-valley arbitrage economics in existing microgrid energy storage scheduling technologies, this invention proposes a multi-stage economic scheduling method and device for microgrid energy storage based on differential evolution. By constructing a three-layer collaborative architecture of "forced charging baseline command generation + differential evolution economic optimization + physical constraint refinement and correction," it provides a scheduling scheme that integrates deterministic physical constraints and heuristic global optimization. This scheme overcomes the shortcomings of traditional mathematical programming models' oversimplification of physical characteristics and the poor engineering feasibility of traditional heuristic algorithms' output strategies. It also overcomes the physical distortion of mathematical programming models and the poor security of heuristic algorithms in existing technologies. Without relying on expensive commercial mathematical programming solvers, it achieves the following objectives: Physical safety assurance: Through physical rules and simulation verification, strict measures are taken to prevent transformer overload, battery overcharging and over-discharging, and illegal reverse power transmission; High-fidelity model: Preserves the non-linear characteristics of battery aging and efficiency, avoiding equipment lifespan overdraft and operational oscillation caused by model simplification; Optimal Global Benefits: While ensuring safety, achieve the overall global optimization of microgrid photovoltaic absorption rate, peak-valley arbitrage revenue and equipment life extension, and output a smooth and executable scheduling strategy.

[0006] The objective of this invention is specifically achieved through the following technical solutions: In a first aspect, this invention discloses a multi-stage economic dispatch method for microgrid energy storage based on differential evolution, including: S1, Net power of the system for predicting time series data for future periods, obtained from the target platform; S2, construct a set of thresholds for system net power that meet photovoltaic surplus periods; combine the minimum value of the equipment's maximum charging capacity, photovoltaic overflow, and anti-reverse current safety margin to generate the corresponding forced charging baseline command; S3, initializes the population individuals of the differential evolution algorithm (DE) by setting the power upper limit according to the forced charging baseline command; using The strategy generates a mutation vector for each individual; and a binomial crossover operation is used to generate a test vector of the mutation vector. S4, superimpose the forced charging baseline command and the test vector to synthesize the initial attempt power; calculate the battery state of charge (SOC) of the initial attempt power hour by hour; if the SOC at the next moment exceeds the capacity constraint, correct the capacity boundary of the initial attempt power and forcibly cut off the initial attempt power until the capacity constraint is met, and output the first corrected power. S5, the first corrected power is superimposed with the system net power to form the system net load; if the system net load is greater than the transformer overload threshold, the power boundary of the system net load is corrected until the transformer overload threshold is met, and the second corrected power is output. S6. Calculate the electricity purchase cost and battery aging penalty from the second corrected power, and construct a high-fidelity objective function for output cost by combining the SOC consistency penalty. If the cost of the current test vector corresponding to the second corrected power is less than the cost of the original population individual, then adopt a greedy selection strategy to update the population individual and replace the original individual; otherwise, retain the original individual and obtain the locally optimal individual. S7 executes DE in cycles from S3 to S6 until the termination condition is met, then outputs the globally optimal individual value. After numerical discretization and dead-zone filtering, power control execution commands are generated to control the energy storage scheduling of the microgrid.

[0007] Secondly, this invention discloses a multi-stage economic dispatch system for microgrid energy storage based on differential evolution, comprising: The data acquisition and feature preprocessing module is used to calculate the net power of the system for predicting future time series obtained from the target platform. The dynamic baseline generation module is used to construct a set of thresholds for when the system net power meets the photovoltaic surplus period; and to generate a set of forced charging baseline instructions corresponding to the minimum value of the equipment's maximum charging capacity, photovoltaic overflow amount, and anti-reverse current safety margin. The differential evolution module is used to initialize the population of individuals in the differential evolution algorithm (DE) based on the power limit set by the forced charging baseline command; it employs... The strategy generates a mutation vector for each individual; and a binomial crossover operation is used to generate a test vector of the mutation vector. The capacity constraint refinement module is used to superimpose the forced charging baseline command and the test vector to synthesize the initial trial power; it calculates the battery state of charge (SOC) of the initial trial power hourly; if the SOC at the next moment exceeds the capacity constraint, it corrects the capacity boundary of the initial trial power and forcibly cuts off the initial trial power until the capacity constraint is met, and outputs the first corrected power. The power boundary constraint refinement module is used to combine the first corrected power with the system net power to synthesize the system net load; if the system net load is greater than the transformer overload threshold, the power boundary of the system net load is corrected until the transformer overload threshold is met, and the second corrected power is output. The optimization module is used to calculate the electricity purchase cost and battery aging penalty from the second corrected power, and to construct a high-fidelity objective function for output cost by combining the SOC consistency penalty. If the cost of the current test vector corresponding to the second corrected power is less than the cost of the original population individual, a greedy selection strategy is used to update the population individual and replace the original individual; otherwise, the original individual is retained to obtain the local optimal individual. The scheduling decision and execution module cyclically calls the adaptive differential evolution module, capacity constraint refinement module, power boundary constraint refinement module, and optimization module to execute DE until the termination condition is met. After that, it outputs the global optimal individual, which is then generated after numerical discretization and dead-zone filtering to control the energy storage scheduling of the microgrid.

[0008] Thirdly, the present invention discloses a microgrid energy storage multi-stage economic dispatch device based on differential evolution, characterized in that it includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the microgrid energy storage multi-stage economic dispatch method based on differential evolution as described in the first aspect of the present invention.

[0009] Fourthly, the present invention discloses a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the multi-stage economic dispatch method for microgrid energy storage based on differential evolution as described in the first aspect of the present invention.

[0010] The beneficial effects of this invention are: 1. This invention fundamentally solves the problem of balancing optimization and safety in heuristic algorithms, ensuring absolute safety in physical operation. Existing genetic algorithms or particle swarm optimization algorithms typically use a "penalty function method" to handle constraints, which can easily lead to strategies violating physical limitations (such as transformer overload or battery overcharge) due to improper weight settings. This invention innovatively designs a "physical constraint refinement module" and a "power boundary constraint refinement module," embedding them into the iterative loop of the differential evolution algorithm. This "embedded simulation" mechanism is equivalent to installing a real-time safety valve, which can automatically force the randomly generated strategy to be corrected to a physically feasible solution. This means that no matter how randomly the algorithm searches, the output strategy will 100% satisfy the transformer capacity limit, anti-reverse current limit, and battery SOC boundary limit, eliminating safety accidents such as equipment overload caused by the randomness of the algorithm.

[0011] 2. This invention overcomes the "lifespan overdraft" problem caused by linearization in mathematical programming models, significantly extending the lifespan of energy storage devices. Existing mixed-integer linear programming (MILP) methods, in order to meet solver requirements, are forced to ignore the nonlinear characteristics of battery aging, resulting in the calculated "theoretical optimal solution" often at the cost of reduced battery life. This invention eliminates the need for linearization simplification of the model, directly introducing a nonlinear aging cost that increases exponentially with the charge / discharge rate into the objective function. This allows the scheduling strategy to keenly perceive the high losses caused by high-rate charge / discharge, automatically avoiding large current surges and ineffective frequent throughput, thereby maximizing the cycle life of the battery pack while achieving economic arbitrage.

[0012] 3. This invention addresses the risk of curtailment caused by rigid rule-based control strategies, achieving a dynamic balance between photovoltaic (PV) consumption and economic arbitrage. To address the inability of traditional rule-based control to adapt to PV fluctuations, this invention adopts a layered architecture of "rigid baseline + flexible optimization." First, a dynamic baseline generation module prioritizes the rigid task of "PV consumption" based on real-time net load forecasts, ensuring that existing PV power is not wasted or illegally fed back through mandatory charging baseline commands. Then, a differential evolution algorithm is used to search for economic arbitrage opportunities in the remaining space. This design ensures both high-proportion renewable energy consumption and full utilization of peak-valley price differences in energy storage, avoiding the unbalanced effects of single-rule control.

[0013] 4. This invention eliminates the "glitch" phenomenon in the strategy, improving the engineering usability of control commands. Existing methods often generate strategies with numerous tiny, invalid actions (glitches), easily leading to frequent start-stop oscillations in the PCS (converter). This invention integrates numerical discretization and dead-zone filtering mechanisms in the backend, effectively filtering out meaningless, minute power fluctuations and outputting smooth, stable control commands. This not only reduces mechanical wear and switching losses in hardware but also allows the device to be directly deployed in low-cost embedded controllers without relying on expensive commercial solvers, demonstrating significant engineering application value. Attached Figure Description

[0014] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments.

[0015] Figure 1 This is a schematic diagram of the process of the multi-stage economic dispatch method for microgrid energy storage based on differential evolution provided in the embodiments of the present invention.

[0016] Figure 2 This is a schematic diagram illustrating the revenue details provided in an embodiment of the present invention. Detailed Implementation

[0017] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0018] Example 1 Embodiment 1 of this invention provides a multi-stage economic dispatch method for microgrid energy storage based on differential evolution, including: S1, Net power of the system for predicting time series data for future periods, obtained from the target platform; S2, construct a set of thresholds for system net power that meet photovoltaic surplus periods; combine the minimum value of the equipment's maximum charging capacity, photovoltaic overflow, and anti-reverse current safety margin to generate the corresponding forced charging baseline command; S3, initializes the population individuals of the differential evolution algorithm (DE) by setting the power upper limit according to the forced charging baseline command; using The strategy generates a mutation vector for each individual; and a binomial crossover operation is used to generate a test vector of the mutation vector. S4, superimpose the forced charging baseline command and the test vector to synthesize the initial attempt power; calculate the battery state of charge (SOC) of the initial attempt power hour by hour; if the SOC at the next moment exceeds the capacity constraint, correct the capacity boundary of the initial attempt power and forcibly cut off the initial attempt power until the capacity constraint is met, and output the first corrected power. S5, the first corrected power is superimposed with the system net power to form the system net load; if the system net load is greater than the transformer overload threshold, the power boundary of the system net load is corrected until the transformer overload threshold is met, and the second corrected power is output. S6. Calculate the electricity purchase cost and battery aging penalty from the second corrected power, and construct a high-fidelity objective function for output cost by combining the SOC consistency penalty. If the cost of the current test vector corresponding to the second corrected power is less than the cost of the original population individual, then adopt a greedy selection strategy to update the population individual and replace the original individual; otherwise, retain the original individual and obtain the locally optimal individual. S7 executes DE in cycles from S3 to S6 until the termination condition is met, then outputs the globally optimal individual value. After numerical discretization and dead-zone filtering, power control execution commands are generated to control the energy storage scheduling of the microgrid.

[0019] like Figure 1 As shown, the scheduling process in Embodiment 1 of the present invention begins with the top-level data intelligence platform. This platform, as the system's perception center, is responsible for pushing input vectors, including load forecast data, photovoltaic forecast data, and electricity price / parameter data, to the scheduling device in real time.

[0020] The method flow is described in conjunction with the system disclosed in this invention: First, the data acquisition and feature preprocessing module cleans the input data and loads physical parameters; Next, the dynamic baseline generation module locks in "rigid tasks" based on physical rules and prioritizes calculating the mandatory charging baseline command that meets the photovoltaic power consumption requirements; Subsequently, the process enters the core Differential Evolution (DE) global optimization loop. In this loop, after population initialization, the Differential Evolution module generates experimental vectors through mutation and crossover, and embeds capacity constraint refinement and power boundary constraint refinement modules to perform real-time physical simulation and forced correction on each experimental vector, ensuring that only physically feasible solutions can enter the fitness evaluation stage. After multiple iterations and meeting the termination condition, the globally optimal individual is output as the globally optimal strategy.

[0021] Finally, the strategy data flows to the scheduling decision and execution module at the bottom, where numerical discretization and dead-zone filtering are performed to generate hardware-executable power control commands, which are then uploaded to the platform to achieve closed-loop control of the microgrid.

[0022] In S1, the target platform acquires the predicted time series for future periods, including: the load forecast series after multi-source data fusion, the photovoltaic forecast series, and the time-of-use electricity price vector; such as obtaining the next 24 hours (i.e., ...) from the data intelligence platform via a communication interface. Time points, time resolution The time series data is used as the prediction time series; where, The load forecast sequence is generated based on historical electricity consumption habits and production plans, and is used to predict the basic power consumption on the user side. The basic power consumption on the user side does not include energy storage charging and discharging and photovoltaic output. The photovoltaic forecast sequence, generated based on meteorological irradiance and temperature forecasts, is used to predict distributed photovoltaic power generation. The time-of-use electricity price vector, as the core weight for economic optimization, includes an array of electricity prices for peak, mid-peak, flat, and valley periods.

[0023] In S1, the method for calculating the system's net power is as follows: ; In the formula, t represents the net power of the system without energy storage intervention, and t is a time variable; To predict the load forecast sequence in the time series, This is a photovoltaic prediction sequence in the prediction time series.

[0024] In S2, the constructed system net power conforms to the photovoltaic surplus period threshold set. for: This indicates the net power of the system after traversing the predicted time series, such as traversing all 96 time points of the day, where all photovoltaic output exceeds the load and there is a risk of backfeeding.

[0025] In S2, the forced charging baseline command corresponding to the generated set is: ; In the formula, Forced charging baseline command, the forced charging baseline command targets a set At every moment, prevent photovoltaic power from being illegally fed back into the upper-level power grid. This will serve as an unalterable "foundation" for the next stage. This ensures that regardless of subsequent economic considerations, the system will always prioritize meeting the crucial safety requirement of "anti-backflow." To the device's maximum charging capacity, For modulus, for The modulus value represents the amount of photovoltaic spillover; To prevent backflow safety margin, it is usually taken as The transformer capacity.

[0026] In S3, the power limit is set. for: ; ; In the formula, This represents the upper limit of power, indicating the maximum charging margin. This is the lower bound of power, representing the maximum discharge capacity. The power is dynamically set based on the remaining available power to ensure that the initial solution does not violate the power limit.

[0027] Population initialization, such as: generation Individuals (suggestions) ), each individual It is a 96-dimensional vector.

[0028] In S3, the following is adopted The strategy generates the mutation vector for each individual as follows: ; In the formula, For the mutation vector, i This provides an index of the individuals in the population currently undergoing mutation operations. G Let the current iteration algebra be... The scaling factor controls the search step size, such as... ; Let r1 be the vector of the individual with index r1 in the Gth generation population. Let r2 be the vector of the individual with index r2 in the Gth generation population. Let r be the vector of the individual with index r3 in the Gth generation population. Indexes of randomly selected distinct populations.

[0029] In S3, the experimental vector for generating the mutation vector using the binomial crossover operation is: ; In the formula, For the i-th test vector of generation G The component values ​​in the j-th dimension, For the i-th mutation vector of generation G The component values ​​in the j-th dimension, Let be the numerical value of the j-th dimension of the target vector of the i-th individual in the G-th generation. Let be a random number generated for the j-th dimension, which follows a uniform distribution in the interval [0, 1]. j For the dimension index of the vector, The crossover probability constant is... This is an integer index randomly selected within the dimension range.

[0030] In S4, the initial attempt at synthesis power for: .

[0031] In S4, the method for hourly extrapolating the battery's state of charge (SOC) for the initial power test is as follows: ; In the formula, for At this moment, the battery state of charge is simulated for the initial power test. for At this moment, the battery state of charge is simulated for the initial power test. For the current charge / discharge conversion efficiency, during charging During discharge ; For battery charging efficiency, For the battery's discharge efficiency, This refers to the battery's rated capacity.

[0032] In S4, if the SOC at the next time step exceeds the capacity constraint, the capacity boundary of the initial attempt power is modified, and the method for forcibly cutting off the initial attempt power is as follows: If the SOC at the next moment, i.e. SOC(t+1) Exceeding capacity constraints If necessary, the initial attempt power should be adjusted. The capacity boundary is used to obtain the corrected power. Use corrected power For the initial attempt power Forced truncation is performed; the calculation method for the corrected power is as follows: ; In the formula, For scheduling time intervals, This represents the maximum allowable state of charge of the battery. This is the minimum allowable state of charge of the battery.

[0033] In S5, the system net load The calculation method is as follows: ; In the formula, The first corrected power includes the corrected power of the output that does not meet the capacity constraint. and initial attempt power that meets capacity constraints .

[0034] In S5, if the system net load exceeds the transformer overload threshold, the methods for correcting the power boundary of the system net load include: If the system net load is greater than the positive overload threshold If the system load is forced to decrease, the energy storage charging power will be forcibly reduced or the discharging power will be increased until the system net load is no greater than the positive overload threshold, thus obtaining the first corrected system net load. as well as The corresponding first target power ;in, This is the positive overload threshold. If the system net load is less than the reverse overload threshold Then, the charging power is forcibly increased to absorb excess photovoltaic power until the system net load is not less than the reverse overload threshold, thus obtaining the second corrected system net load. as well as The corresponding second target power ;in, This is the reverse overload threshold; By conforming to the transformer overload threshold The corresponding first correction power as well as , forming the second corrected power .

[0035] This step serves as the last line of defense before output and is also a "fitness filter" in the differential evolution process. It can accept any given policy input (whether randomly initialized or generated through evolution) and output a modified policy that satisfies all physical constraints through a serial simulation mechanism.

[0036] In S6, the electricity purchase cost item The calculation method is as follows: ; In the formula, t is the time variable, and T is the maximum time value. For the second corrected power, To predict the time-of-use electricity price vector in a time series, This is the scheduling time interval.

[0037] In S6, the battery aging penalty item The calculation method is as follows: ; In the formula, This is the basic linear aging depreciation factor for the battery. This is a nonlinear aging acceleration factor that varies with charge / discharge rate. It is a natural constant. This is a nonlinear exponential factor for battery aging.

[0038] Among them, the battery aging penalty term uses an exponential function to simulate the accelerated aging characteristics of the battery at high rates, forcing the algorithm to automatically avoid frequent charging and discharging with high losses, that is, to eliminate "glitch".

[0039] In S6, the SOC consistency penalty term is calculated as follows: ; In the formula, This represents the actual state of charge of the battery at the end of the scheduling cycle. This represents the desired state of charge at the end of the scheduling cycle.

[0040] In S6, the constructed high-fidelity objective function is: ; In the formula, The cost output by the high-fidelity objective function. For electricity purchase costs, This is a penalty item for battery aging. This is a SOC consistency penalty item. As the weight of the battery aging penalty item, This represents the weight of the SOC consistency penalty item.

[0041] In S6, the method for updating individuals in the population using a greedy selection strategy is as follows: ; In the formula, For the updated number generation Population individuals, Let be the i-th test vector of the G-th generation, i.e., the current test vector; For the current experimental vector, the i-th individual in the original population of the G-th generation is... The cost value calculated for a high-fidelity objective function; like The cost is no greater than The cost, then use replace Enter the next generation as the locally optimal individual; otherwise, retain. As a locally optimal individual.

[0042] In S7, the method for numerical discretization is as follows: After the DE iteration loop meets the termination condition, the individual with the minimum high-fidelity objective function cost value is selected from the final generation population as the globally optimal individual. The power instructions in continuous floating-point form contained in the globally optimal individual are rounded to obtain the globally optimal discretized control instructions adapted to the data format of the energy storage converter PCS register. .

[0043] In S7, the dead-time filtering method is as follows: ; In the formula, For power control execution instructions after dead-time filtering, This is the globally optimal discretized control command. for The modulus, This is the action dead zone threshold.

[0044] The steps of dead-time filtering, in order to protect the hardware, include setting... (For example, 2% of rated power) effectively filters out minor noise generated during calculations, avoiding frequent start-stop oscillations of the energy storage converter near zero power.

[0045] The final 96-point instruction sequence is packaged and uploaded to the platform via MQTT / HTTP protocol, where it is uniformly scheduled and distributed to the underlying execution mechanism.

[0046] Example 2 Embodiment 2 of the present invention provides a multi-stage economic dispatch system for microgrid energy storage based on differential evolution, comprising: The data acquisition and feature preprocessing module is used to calculate the net power of the system for predicting future time series obtained from the target platform. The dynamic baseline generation module is used to construct a set of thresholds for when the system net power meets the photovoltaic surplus period; and to generate a set of forced charging baseline instructions corresponding to the minimum value of the equipment's maximum charging capacity, photovoltaic overflow amount, and anti-reverse current safety margin. The dynamic baseline generation module is executed only once before the start of the scheduling cycle. Its core purpose is to use physical rules to prioritize "rigid tasks," thereby compressing the search space of subsequent optimization algorithms and ensuring priority consumption of photovoltaic power.

[0047] The differential evolution module is used to initialize the population of individuals in the differential evolution algorithm (DE) based on the power limit set by the forced charging baseline command; it employs... The strategy generates a mutation vector for each individual; and a binomial crossover operation is used to generate a test vector of the mutation vector. The capacity constraint refinement module is used to superimpose the forced charging baseline command and the test vector to synthesize the initial trial power; it calculates the battery state of charge (SOC) of the initial trial power hourly; if the SOC at the next moment exceeds the capacity constraint, it corrects the capacity boundary of the initial trial power and forcibly cuts off the initial trial power until the capacity constraint is met, and outputs the first corrected power. The power boundary constraint refinement module is used to combine the first corrected power with the system net power to synthesize the system net load; if the system net load is greater than the transformer overload threshold, the power boundary of the system net load is corrected until the transformer overload threshold is met, and the second corrected power is output. The optimization module is used to calculate the electricity purchase cost and battery aging penalty from the second corrected power, and to construct a high-fidelity objective function for output cost by combining the SOC consistency penalty. If the cost of the current test vector corresponding to the second corrected power is less than the cost of the original population individual, a greedy selection strategy is used to update the population individual and replace the original individual; otherwise, the original individual is retained to obtain the local optimal individual. The scheduling decision and execution module cyclically calls the adaptive differential evolution module, capacity constraint refinement module, power boundary constraint refinement module, and optimization module to execute DE until the termination condition is met. After that, it outputs the global optimal individual, which is then generated after numerical discretization and dead-zone filtering to control the energy storage scheduling of the microgrid.

[0048] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system described above can be referred to the corresponding process in the aforementioned method embodiment one, and will not be repeated here.

[0049] Example 3 Embodiment 3 of the present invention provides a microgrid energy storage multi-stage economic dispatch device based on differential evolution, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the microgrid energy storage multi-stage economic dispatch method based on differential evolution as described in Embodiment 1.

[0050] Example 4 Embodiment 4 of the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the multi-stage economic dispatch method for microgrid energy storage based on differential evolution as described in Embodiment 1.

[0051] Verification experiment: To verify the effectiveness of the technical solution of the present invention, a verification experiment is provided for illustration: Example 1: A photovoltaic-storage microgrid project for a large automotive service company in Changsha: 1. Application Scenarios and Challenges: This embodiment is applied to a large automobile sales and service company in Changsha. The load on this industrial park is highly volatile, including not only regular office power consumption but also high-power electric vehicle fast charging stations. The main pain points are: when the repair workshop is shut down at noon, the photovoltaic system is prone to backfeeding violations; and when the showroom lighting is combined with temporary customer fast charging in the evening, the total load can easily exceed the transformer capacity limit.

[0052] 2. Data Acquisition and Algorithm Application: This embodiment adopts a two-tier architecture of "IoT management platform + data intelligence platform". The underlying scheduling device first uploads the operation data to the enterprise's IoT management platform through the IoT communication link; the data intelligence platform synchronizes the data of the management platform in real time through the API interface, performs core strategy calculations, and sends the generated scheduling instructions back to the IoT management platform, and finally distributes them to the field.

[0053] To address the aforementioned pain points, the data intelligence platform automatically implemented the following adjustments: (1) Regarding the risk of photovoltaic back-transmission: The "dynamic baseline module" in the algorithm identifies the photovoltaic surplus period based on the synchronous prediction data, generates a forced charging instruction and pushes it to the management platform, thus locking the photovoltaic consumption task.

[0054] (2) Regarding the risk of transformer overload: When the algorithm attempts to profit from discharging during the evening peak, the built-in "physical refining module" detects the impact of fast charging load. The algorithm immediately triggers a correction mechanism, forcibly reducing the energy storage discharge power and packaging the safety strategy to send to the management platform for execution.

[0055] 3. Implementation Results: A comparative analysis of the economic benefits of this invention under a typical daily operating scenario: (1) Safety backup: The system successfully coped with the daily impact of multiple electric vehicle fast charging, and the number of transformer overloads and reverse power transmission violations were both zero, verifying the effectiveness of the "physical constraint refinement module" in ensuring the safe operation of the power grid.

[0056] (2) Extended lifespan: The algorithm effectively filters out invalid "glitch" actions caused by load fluctuations through dead zone filtering, reducing the average daily cycle loss of the battery by about 5%, and significantly extending the service life of the energy storage equipment.

[0057] (3) Cost Reduction and Efficiency Improvement: As can be seen from the data comparison in Figure 2, the present invention significantly improves the overall revenue of the microgrid. The total revenue of the actual operation throughout the day was RMB 983.06, while the total revenue after adopting the technical solution of the present invention reached RMB 1098.37, with a daily net revenue increase of RMB 115.31 and a daily revenue growth rate of approximately 11.7%. Specifically, during the "peak period" when the electricity price is highest, the discharge revenue increased from RMB 1216.14 to RMB 1317.72; while during the "valley period" when the electricity price is lower, the charging cost decreased from RMB 1025.89 to RMB 834.18. In summary, under the premise of ensuring the maintenance power load and fast charging safety, the algorithm maximizes the utilization of the energy storage peak-valley arbitrage space, resulting in a significant reduction in the monthly comprehensive electricity cost of the park.

[0058] In summary, this invention provides a multi-stage economic dispatch method for microgrid energy storage based on differential evolution, the core technical points of which are as follows: 1. A three-tiered scheduling architecture of "rigid baseline + flexible optimization + physical correction" was established. Unlike traditional single optimization, this invention first generates an insurmountable mandatory charging baseline through physical rules to lock in the rigid task of photovoltaic power consumption; on this baseline, the differential evolution algorithm is used to search for economic arbitrage space; finally, a physical constraint refinement module is used for safety fallback, realizing the hierarchical decoupling of physical safety and economic benefits.

[0059] 2. A closed-loop iterative optimization mechanism based on embedded physical simulation was constructed. This invention improves the open-loop structure of the traditional differential evolution algorithm by directly embedding the physical constraint refinement module into the algorithm's iterative loop. After each population generation of trial vectors, the simulation model is first subjected to forced correction (including SOC boundary correction and transformer overload correction) to obtain a physically feasible solution before fitness evaluation. This mechanism ensures that the algorithm's evolutionary direction always remains within the physically safe boundary.

[0060] 3. A high-fidelity objective function based on nonlinear lifetime loss is proposed. This overcomes the distortion problem of linear programming models by introducing a nonlinear aging coefficient that grows exponentially with the charge / discharge rate into the objective function, combined with a SOC consistency penalty term. Coupled with a dead-zone filtering strategy in the backend, the algorithm can automatically avoid high-loss "glitch" actions, outputting smooth control commands that extend equipment lifespan.

[0061] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A multi-stage economic dispatch method for microgrid energy storage based on differential evolution, characterized in that, include: S1, Net power of the system for predicting time series data for future periods, obtained from the target platform; S2, construct a set of thresholds for system net power that meet photovoltaic surplus periods; combine the minimum value of the equipment's maximum charging capacity, photovoltaic overflow, and anti-reverse current safety margin to generate the corresponding forced charging baseline command; S3, initializes the population individuals of the differential evolution algorithm (DE) by setting the power upper limit according to the forced charging baseline command; using The strategy generates a mutation vector for each individual; and a binomial crossover operation is used to generate a test vector of the mutation vector. S4, superimpose the forced charging baseline command and the test vector to synthesize the initial attempt power; calculate the battery state of charge (SOC) of the initial attempt power hour by hour; if the SOC at the next moment exceeds the capacity constraint, correct the capacity boundary of the initial attempt power and forcibly cut off the initial attempt power until the capacity constraint is met, and output the first corrected power. S5, the first corrected power is superimposed with the system net power to form the system net load; if the system net load is greater than the transformer overload threshold, the power boundary of the system net load is corrected until the transformer overload threshold is met, and the second corrected power is output. S6. Calculate the electricity purchase cost and battery aging penalty from the second corrected power, and construct a high-fidelity objective function for output cost by combining the SOC consistency penalty. If the cost of the current test vector corresponding to the second corrected power is less than the cost of the original population individual, then adopt a greedy selection strategy to update the population individual and replace the original individual; otherwise, retain the original individual and obtain the locally optimal individual. S7 executes DE in cycles from S3 to S6 until the termination condition is met, then outputs the globally optimal individual value. After numerical discretization and dead-zone filtering, power control execution commands are generated to control the energy storage scheduling of the microgrid.

2. The multi-stage economic dispatch method for microgrid energy storage based on differential evolution as described in claim 1, characterized in that, In S1, the target platform obtains the predicted time series for future periods, including: the load forecast series after multi-source data fusion, the photovoltaic forecast series, and the time-of-use electricity price vector; among which, The load forecast sequence is generated based on historical electricity consumption habits and production plans, and is used to predict the basic power consumption on the user side. The basic power consumption on the user side does not include energy storage charging and discharging and photovoltaic output. The photovoltaic forecast sequence, generated based on meteorological irradiance and temperature forecasts, is used to predict distributed photovoltaic power generation. The time-of-use electricity price vector, as the core weight for economic optimization, includes an array of electricity prices for peak, mid-peak, flat, and valley periods.

3. The multi-stage economic dispatch method for microgrid energy storage based on differential evolution as described in claim 2, characterized in that, In S1, the method for calculating the system's net power is as follows: ； In the formula, t represents the net power of the system without energy storage intervention, and t is a time variable; To predict the load forecast sequence in the time series, This is a photovoltaic prediction sequence in the prediction time series.

4. The multi-stage economic dispatch method for microgrid energy storage based on differential evolution as described in claim 3, characterized in that, In S2, the constructed system net power conforms to the photovoltaic surplus period threshold set. for: This indicates the net power of the system after traversing the predicted time series, where all photovoltaic outputs exceed the load and there is a risk of backfeeding.

5. The multi-stage economic dispatch method for microgrid energy storage based on differential evolution as described in claim 4, characterized in that, In S2, the forced charging baseline command corresponding to the generated set is: ； In the formula, This is a mandatory charging baseline command. To the device's maximum charging capacity, For modulus, for The modulus value represents the amount of photovoltaic spillover; To prevent backflow safety margin.

6. The multi-stage economic dispatch method for microgrid energy storage based on differential evolution as described in claim 5, characterized in that, In S3, the power limit is set. for: ； ； In the formula, This represents the upper limit of power, indicating the maximum charging margin. This is the lower bound of power, representing the maximum discharge capacity.

7. The multi-stage economic dispatch method for microgrid energy storage based on differential evolution as described in claim 6, characterized in that, In S3, the following is adopted The strategy generates the mutation vector for each individual as follows: ； In the formula, For the mutation vector, i This provides an index of the individuals in the population currently undergoing mutation operations. G Let the current iteration algebra be... This is the scaling factor, which controls the search step size. Let r1 be the vector of the individual with index r1 in the Gth generation population. Let r2 be the vector of the individual with index r2 in the Gth generation population. Let r be the vector of the individual with index r3 in the Gth generation population. Indexes of randomly selected distinct populations.

8. The multi-stage economic dispatch method for microgrid energy storage based on differential evolution as described in claim 7, characterized in that, In S3, the experimental vector for generating the mutation vector using the binomial crossover operation is: ； In the formula, For the i-th test vector of generation G The component values ​​in the j-th dimension, For the i-th mutation vector of generation G The component values ​​in the j-th dimension, Let be the numerical value of the j-th dimension of the target vector of the i-th individual in the G-th generation. Let be a random number generated for the j-th dimension, which follows a uniform distribution in the interval [0, 1]. j For the dimension index of the vector, The crossover probability constant is... This is an integer index randomly selected within the dimension range.

9. The multi-stage economic dispatch method for microgrid energy storage based on differential evolution as described in claim 8, characterized in that, In S4, the initial attempt at synthesis power for: 。 10. The multi-stage economic dispatch method for microgrid energy storage based on differential evolution as described in claim 9, characterized in that, In S4, the method for hourly extrapolating the battery's state of charge (SOC) for the initial power test is as follows: ； In the formula, for At this moment, the battery state of charge is simulated for the initial power test. for At this moment, the battery state of charge is simulated for the initial power test. For the current charge / discharge conversion efficiency, during charging During discharge ; For battery charging efficiency, For the battery's discharge efficiency, This refers to the battery's rated capacity.

11. The multi-stage economic dispatch method for microgrid energy storage based on differential evolution as described in claim 10, characterized in that, In S4, if the SOC at the next time step exceeds the capacity constraint, the capacity boundary of the initial attempt power is modified, and the method for forcibly cutting off the initial attempt power is as follows: If the SOC at the next moment, i.e. SOC(t+1) Exceeding capacity constraints If necessary, the initial attempt power should be adjusted. The capacity boundary is used to obtain the corrected power. Use corrected power For the initial attempt power Forced truncation is performed; the calculation method for the corrected power is as follows: ； In the formula, For scheduling time intervals, This represents the maximum allowable state of charge of the battery. This is the minimum allowable state of charge of the battery.

12. The multi-stage economic dispatch method for microgrid energy storage based on differential evolution as described in claim 11, characterized in that, In S5, the system net load The calculation method is as follows: ； In the formula, The first corrected power includes the corrected power of the output that does not meet the capacity constraint. and initial attempt power that meets capacity constraints .

13. The multi-stage economic dispatch method for microgrid energy storage based on differential evolution as described in claim 12, characterized in that, In S5, if the system net load exceeds the transformer overload threshold, the methods for correcting the power boundary of the system net load include: If the system net load is greater than the positive overload threshold If the system load is forced to decrease, the energy storage charging power will be forcibly reduced or the discharging power will be increased until the system net load is no greater than the positive overload threshold, thus obtaining the first corrected system net load. as well as The corresponding first target power ;in, This is the positive overload threshold. If the system net load is less than the reverse overload threshold Then, the charging power is forcibly increased to absorb excess photovoltaic power until the system net load is not less than the reverse overload threshold, thus obtaining the second corrected system net load. as well as The corresponding second target power ;in, This is the reverse overload threshold; By conforming to the transformer overload threshold The corresponding first correction power as well as , forming the second corrected power .

14. The multi-stage economic dispatch method for microgrid energy storage based on differential evolution as described in claim 13, characterized in that, In S6, the electricity purchase cost item The calculation method is as follows: ； In the formula, t is the time variable, and T is the maximum time value. For the second corrected power, To predict the time-of-use electricity price vector in a time series, This is the scheduling time interval.

15. The multi-stage economic dispatch method for microgrid energy storage based on differential evolution as described in claim 14, characterized in that, In S6, the battery aging penalty item The calculation method is as follows: ； In the formula, This is the basic linear aging depreciation factor for the battery. This is a nonlinear aging acceleration factor that varies with charge / discharge rate. It is a natural constant. This is a nonlinear exponential factor for battery aging.

16. The multi-stage economic dispatch method for microgrid energy storage based on differential evolution as described in claim 15, characterized in that, In S6, the SOC consistency penalty term is calculated as follows: ； In the formula, This represents the actual state of charge of the battery at the end of the scheduling cycle. This represents the desired state of charge at the end of the scheduling cycle.

17. The multi-stage economic dispatch method for microgrid energy storage based on differential evolution as described in claim 16, characterized in that, In S6, the constructed high-fidelity objective function is: ； In the formula, The cost output by the high-fidelity objective function. For electricity purchase costs, This is a penalty item for battery aging. This is a SOC consistency penalty item. As the weight of the battery aging penalty item, This represents the weight of the SOC consistency penalty item.

18. The multi-stage economic dispatch method for microgrid energy storage based on differential evolution as described in claim 17, characterized in that, In S6, the method for updating individuals in the population using a greedy selection strategy is as follows: ； In the formula, For the updated number generation Population individuals, Let be the i-th test vector of the G-th generation, i.e., the current test vector; For the current experimental vector, the i-th individual in the original population of the G-th generation is... The cost value calculated for a high-fidelity objective function; like The cost is no greater than The cost, then use replace Enter the next generation as the locally optimal individual; otherwise, retain. As a locally optimal individual.

19. The multi-stage economic dispatch method for microgrid energy storage based on differential evolution as described in claim 18, characterized in that, In S7, the method for numerical discretization is as follows: After the DE iteration loop meets the termination condition, the individual with the minimum high-fidelity objective function cost value is selected from the final generation population as the globally optimal individual. The power instructions in continuous floating-point form contained in the globally optimal individual are rounded to obtain the globally optimal discretized control instructions adapted to the data format of the energy storage converter PCS register. .

20. The multi-stage economic dispatch method for microgrid energy storage based on differential evolution as described in claim 19, characterized in that, In S7, the dead-time filtering method is as follows: ； In the formula, For power control execution instructions after dead-time filtering, This is the globally optimal discretized control command. for The modulus, This is the action dead zone threshold.

21. A multi-stage economic dispatch system for microgrid energy storage based on differential evolution, characterized in that, include: The data acquisition and feature preprocessing module is used to calculate the net power of the system for predicting future time series obtained from the target platform. The dynamic baseline generation module is used to construct a set of thresholds for when the system net power meets the photovoltaic surplus period; and to generate a set of forced charging baseline instructions corresponding to the minimum value of the equipment's maximum charging capacity, photovoltaic overflow amount, and anti-reverse current safety margin. The differential evolution module is used to initialize the population of individuals in the differential evolution algorithm (DE) based on the power limit set by the forced charging baseline command; it employs... The strategy generates a mutation vector for each individual; and a binomial crossover operation is used to generate a test vector of the mutation vector. The capacity constraint refinement module is used to superimpose the forced charging baseline command and the test vector to synthesize the initial trial power; it calculates the battery state of charge (SOC) of the initial trial power hourly; if the SOC at the next moment exceeds the capacity constraint, it corrects the capacity boundary of the initial trial power and forcibly cuts off the initial trial power until the capacity constraint is met, and outputs the first corrected power. The power boundary constraint refinement module is used to combine the first corrected power with the system net power to synthesize the system net load; if the system net load is greater than the transformer overload threshold, the power boundary of the system net load is corrected until the transformer overload threshold is met, and the second corrected power is output. The optimization module is used to calculate the electricity purchase cost and battery aging penalty from the second corrected power, and to construct a high-fidelity objective function for output cost by combining the SOC consistency penalty. If the cost of the current test vector corresponding to the second corrected power is less than the cost of the original population individual, a greedy selection strategy is used to update the population individual and replace the original individual; otherwise, the original individual is retained to obtain the local optimal individual. The scheduling decision and execution module cyclically calls the adaptive differential evolution module, capacity constraint refinement module, power boundary constraint refinement module, and optimization module to execute DE until the termination condition is met. After that, it outputs the global optimal individual, which is then generated after numerical discretization and dead-zone filtering to control the energy storage scheduling of the microgrid.

22. A multi-stage economic dispatch device for microgrid energy storage based on differential evolution, characterized in that, It includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of the microgrid energy storage multi-stage economic dispatch method based on differential evolution as described in any one of claims 1 to 20.

23. A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the multi-stage economic dispatch method for microgrid energy storage based on differential evolution as described in any one of claims 1 to 20.

Images