An integrated power supply system and a dispatch method thereof

Abstract

The application discloses an integrated power supply system and a scheduling method thereof, and belongs to the technical field of power supply systems.The method comprises the following steps: collecting micro-grid operation state data in real time; establishing a storage battery stress factor table, identifying effective charging and discharging cycles, and calculating a life loss value, which is converted into a life loss cost; taking the minimization of total operation cost as a target, constructing an optimization model containing the life loss cost, and adopting a model prediction control framework and an improved particle swarm optimization algorithm to generate an optimal scheduling instruction set in each scheduling period.The application improves the economy and reliability of the whole life cycle of the system through real-time quantitative evaluation and optimal scheduling of the life of storage batteries.

Description

Technical Field

[0001] This invention belongs to the field of power system technology, specifically relating to an integrated power system and its scheduling method. Background Technology

[0002] To ensure continuous power supply to critical equipment and save energy, hybrid microgrids are often constructed, primarily powered by solar and wind energy, supplemented by diesel generators. In this system, the intelligent integrated power supply, as the core control and energy dispatch unit of the microgrid, is a crucial component for maintaining stable power supply from the grid.

[0003] Currently, existing power equipment energy management methods mostly focus on real-time power balance and short-term operational economy (such as minimizing diesel consumption), but generally neglect the refined management of integrated power supplies. Due to the randomness of renewable energy input and the volatility of load demand, batteries are in non-standard random charging and discharging conditions for extended periods. Compared with regular standard cycles, this random operating condition significantly leads to instability in the power supply of integrated power supplies, affecting the safe and stable operation of the power grid.

[0004] Current technologies lack effective models and methods to quantify in real-time the "instantaneous loss" caused by a single charge-discharge operation to the lifespan of an integrated power supply battery. Although offline assessment methods for battery health exist, they cannot meet the real-time requirements of online scheduling. More importantly, current technologies do not incorporate "lifespan loss" as a key optimization variable into energy scheduling decisions in real time. This may lead to system operation strategies that, in order to save a small amount of diesel fuel in the short term, frequently subject the battery to damaging shallow charge-discharge cycles. From a life-cycle perspective, this comes at the heavy cost of a sharp decline in battery lifespan, resulting in even worse economic efficiency. Summary of the Invention

[0005] To address the aforementioned problems, this invention provides an integrated power system scheduling method to resolve the issues present in the background art.

[0006] To achieve the aforementioned objectives, in a first aspect, the present invention proposes an integrated power system scheduling method, comprising: Step S1: Collect microgrid operation status data in real time. The operation status data includes photovoltaic power generation, wind power generation, load demand power, diesel generator status, power equipment battery operation status, state of charge, temperature, voltage and current. Step S2: Establish a stress factor table for the battery. Based on the battery's current data, identify and analyze the effective charge-discharge cycles of the battery within a preset time window. Calculate the loss value within the effective charge-discharge cycle time period based on the stress factor table, convert the loss value into battery life loss cost, and update the stress factor table every preset time interval. Step S3: With minimizing the total operating cost as the objective function, establish an optimization model that includes lifetime loss costs. The total operating cost includes diesel generator fuel costs, battery lifetime loss costs, electricity purchase costs from the main grid, and renewable energy abandonment penalty costs. Using a model predictive control framework, within each optimization cycle, based on the microgrid's real-time operating status data and short-term forecast information, use an improved particle swarm optimization algorithm to solve the optimization model and obtain the optimal scheduling instruction set.

[0007] Furthermore, calculating the loss value over the effective charge-discharge cycle period includes the following steps: Extreme value analysis is performed on the current time series data of the battery to identify all local maxima and local minima, forming a peak-valley feature point sequence. Based on the peak-valley feature point sequence, complete cycles are identified, and the depth value of the complete cycle is calculated. The depth value of the identified complete cycle is matched and queried with a pre-established battery stress factor table to obtain the loss weight value corresponding to each depth range. The loss weight values ​​of each complete cycle are accumulated to obtain the loss value in the current analysis time period.

[0008] Furthermore, identifying the complete loop based on the peak-valley feature point sequence includes the following steps: Four consecutive feature points are selected at once from the peak-valley feature point sequence. The first difference between the first feature point and the second feature point, the second difference between the second feature point and the third feature point, and the third difference between the third feature point and the fourth feature point are calculated respectively. When the second difference is less than or equal to the first difference and less than or equal to the third difference, it is determined that the second feature point and the third feature point form a complete cycle. The second difference is used as the depth value of the complete cycle. The feature points that form the complete cycle are removed from the peak-valley feature point sequence. The complete cycle is identified again for the removed peak-valley feature point sequence until no new complete cycle can be identified.

[0009] Furthermore, updating the stress factor table includes the following steps: Establish a battery loss prediction model. The model takes the historical operating status data of the battery as input and outputs the predicted loss value of the battery over a future period of time. The predicted loss value for a future time period is obtained based on the prediction model. The calculated loss value for the corresponding time period is obtained. The calculated loss value and the predicted loss value are compared to obtain the deviation value between the two. If the deviation value exceeds the preset threshold, the stress factor table is adjusted based on the deviation value and the historical operating conditions of the battery.

[0010] Furthermore, establishing a multi-objective optimization model that includes lifetime loss costs includes the following steps: Define decision variables, including diesel engine output power, battery charging and discharging power, renewable energy dispatch ratio, and load reduction. Set the objective function to minimize total operating cost. Set constraints, including power balance constraints, equipment operation constraints, and battery life constraints. The battery life loss cost in the total cost is the battery life loss amount multiplied by the battery unit loss cost.

[0011] Furthermore, solving the multi-objective optimization model to obtain the optimal scheduling instruction set includes the following steps: Based on real-time operation status data and preset scenario classification rules, the current operation scenario type is identified. Based on the operation scenario type, historical high-quality scheduling schemes and weather forecast information, multiple candidate scheduling schemes are generated to form a candidate scheduling scheme set. Based on the operation scenario type, the corresponding algorithm parameter combination is called from the preset parameter configuration library. The algorithm parameters include inertia weight, acceleration factor and constraint penalty coefficient. The candidate scheduling scheme is updated based on the combination of algorithm parameters. Knowledge-guided mutation operation is performed on the updated candidate scheduling scheme. The candidate scheduling scheme set after mutation is sorted using the fast non-dominated sorting method. The candidate scheduling scheme set is calculated using the congestion distance calculation method. This step is repeated until the iteration termination condition is met. The iteration termination condition is that the maximum number of iterations is reached or the improvement of the total running cost of the candidate scheduling scheme in 10 consecutive iterations is less than a preset threshold. The optimal scheduling scheme is selected from the final set of candidate scheduling schemes, and the optimal scheduling instruction set is generated based on the optimal scheduling scheme.

[0012] Furthermore, updating the candidate scheduling scheme based on the combination of algorithm parameters includes the following steps: For each candidate scheduling scheme, an adjustment instruction is calculated. The adjustment instruction is a weighted synthesis of three parts: inertia, individual cognition, and social learning. The corresponding inertia weight, individual cognition weight, and social learning weight are obtained from the preset parameter configuration library according to the current running scenario type. Each candidate scheduling scheme is added to the adjustment instruction calculated in the previous step to obtain the updated candidate scheduling scheme. Constraint checks and amplitude limiting are then performed on the updated candidate scheduling scheme.

[0013] Furthermore, performing knowledge-guided mutation operations on the updated candidate scheduling scheme includes the following steps: For each candidate scheduling scheme, check whether the net charge and discharge amount of the battery within any consecutive time period is greater than the preset threshold. If so, use the power scaling method to perform deep variation correction. Identify the start time of the diesel engine in each candidate scheduling scheme. When the continuous running time is detected to be less than the preset time threshold, extend the running time to the minimum start time or cancel the start command. Identify the time periods in each candidate scheduling scheme where the power generation of renewable energy exceeds the load demand, check the charging status of the batteries during these periods, and if the batteries are not charged, use excess renewable energy to charge them; perform safety and feasibility verification on the candidate scheduling schemes after the mutation operation, roll back the candidate scheduling schemes that fail the verification to the state before the mutation for the candidate scheduling schemes that succeed in the verification, evaluate the change in the total operating cost before and after the candidate scheduling scheme mutation, if the total operating cost decreases, record the mutation type, mutation magnitude and mutation effect, and establish a mutation knowledge base.

[0014] Furthermore, the process of calculating the candidate scheduling scheme set using the congestion distance calculation method includes the following steps: A non-dominated sorting algorithm is used to obtain multiple non-dominated layer sequences. Each non-dominated layer has multiple candidate scheduling schemes. For each candidate scheduling scheme in a non-dominated layer, the candidate scheduling schemes are sorted according to each sub-cost. The congestion distance between two candidate scheduling schemes located at the boundary is set to infinity. For candidate scheduling schemes located at non-boundaries, the difference between its sub-cost and that of its two adjacent candidate scheduling schemes is calculated. After normalizing the difference, the sum of the normalized difference is calculated. The sum of the difference is used as the first distance of the candidate scheduling scheme. The result of adding the first distances corresponding to all sub-costs is used as the final congestion distance. The total number of candidate scheduling schemes is preset. First, candidate scheduling schemes are selected and added to the candidate scheduling scheme set in descending order of non-dominated layer. Within the same non-dominated layer, candidate scheduling schemes are selected in descending order of congestion distance until the total number of candidate scheduling schemes reaches the preset value.

[0015] Secondly, this application provides an integrated power supply system, the system comprising: The data acquisition module collects real-time microgrid operating status data, including photovoltaic power generation, wind power generation, load demand power, diesel generator status, power equipment battery operating status, state of charge, temperature, voltage and current. The loss calculation module establishes a stress factor table for the battery, identifies and analyzes the effective charge-discharge cycles of the battery within a preset time window based on the battery's current data, calculates the loss value within the effective charge-discharge cycle time period based on the stress factor table, converts the loss value into battery life loss cost, and updates the stress factor table every preset time interval. The optimization scheduling module takes minimizing the total operating cost as the objective function and establishes an optimization model that includes the cost of lifespan loss. The total operating cost includes the fuel cost of diesel generators, the cost of battery lifespan loss, the cost of purchasing electricity from the main grid, and the cost of renewable energy abandonment penalty. A model predictive control framework is adopted. In each optimization cycle, based on the real-time operating status data of the microgrid and short-term forecast information, an improved particle swarm optimization algorithm is used to solve the optimization model to obtain the optimal scheduling instruction set.

[0016] The beneficial effects of this invention are as follows: This invention accurately identifies and quantifies battery life loss under random charge-discharge conditions by establishing a stress factor table and an improved Rainflow counting method, transforming abstract life decay into calculable loss costs. Battery life loss costs are incorporated into the optimization objective, forming the total operating cost along with diesel fuel costs, electricity purchase costs, and energy curtailment penalty costs, achieving a balance between short-term operation and long-term equipment health. An improved particle swarm optimization algorithm, combined with scene recognition, knowledge-guided mutation operations, and non-dominated sorting, enhances the feasibility and diversity of scheduling schemes, adapting to the complex and ever-changing microgrid operating environment. By optimizing battery charging and discharging strategies and diesel engine start-stop logic, renewable energy curtailment is reduced, clean energy utilization is improved, and dependence on traditional diesel power generation is decreased. Furthermore, by establishing a loss prediction model and a mutation knowledge base, the stress factor table and scheduling strategy are dynamically updated, improving the system's accuracy and robustness in actual operation. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of one embodiment of an integrated power system scheduling method according to the present application. Figure 2 This is a schematic diagram of one embodiment of an integrated power supply system according to the present application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0019] It is understood that the terms "first," "second," etc., used in this application may be used herein to describe various elements, but unless otherwise specified, these elements are not limited by these terms. These terms are used only to distinguish one element from another. For example, without departing from the scope of this application, a first script may be referred to as a second script, and similarly, a second script may be referred to as a first script.

[0020] like Figure 1As shown, an integrated power system scheduling method includes: Step S1: Collect microgrid operation status data in real time. The operation status data includes photovoltaic power generation, wind power generation, load demand, diesel generator status, power equipment battery operation status, state of charge, temperature, voltage and current.

[0021] Specifically, in independent microgrids, the batteries of power supply equipment are key devices for ensuring system stability and improving the absorption of new energy sources. However, they are costly and have limited lifespan. Existing microgrid energy management strategies mainly focus on real-time power balancing and operating cost optimization, but neglect the refined management of battery lifespan. In wind-solar-diesel-storage microgrids, due to the randomness of energy input and the volatility of load, batteries are in non-standard random charge-discharge conditions for extended periods. This condition significantly accelerates battery aging. To enable real-time assessment and priority consideration of battery lifespan health during energy dispatching in microgrids, and to achieve optimal system lifecycle costs, this application is proposed.

[0022] First, to achieve optimized decision-making, microgrid operating status data is collected. This data includes photovoltaic power generation, wind power generation, load demand, diesel generator status, operating status of power equipment batteries, state of charge, temperature, voltage, and current. It is important to note that if the diesel generator is running, its power generation also needs to be collected. The operating status of the power equipment batteries refers to whether they are charging, discharging, or idle. The state of charge refers to the energy storage level of the batteries. Data such as battery temperature, voltage, and current provide a data foundation for establishing predictive models for the batteries.

[0023] Step S2: Establish a stress factor table for the battery. Based on the battery's current data, identify and analyze the effective charge-discharge cycles of the battery within a preset time window. Calculate the loss value within the effective charge-discharge cycle time period based on the stress factor table, convert the loss value into battery life loss cost, and update the stress factor table every preset time interval.

[0024] Specifically, to accurately calculate battery life loss, a stress factor table is pre-established. This table stores the correspondence between different charge-discharge cycle ranges and battery capacity decay rates. It can be established based on accelerated aging experiments. Based on battery current data, the effective charge-discharge cycles of the battery within a preset time window are identified and analyzed. The loss value within the effective charge-discharge cycle period is calculated based on the stress factor table. The specific method for calculating the loss value will be explained in detail later. The loss value is then converted into battery life loss cost. Since the usage of batteries varies, the stress factor table will also change depending on the situation. Therefore, the stress factor table is updated at preset time intervals to improve the accuracy of loss value calculation.

[0025] Step S3: With minimizing the total operating cost as the objective function, establish an optimization model that includes lifetime loss costs. The total operating cost includes diesel generator fuel costs, battery lifetime loss costs, electricity purchase costs from the main grid, and renewable energy abandonment penalty costs. Using a model predictive control framework, within each optimization cycle, based on the microgrid's real-time operating status data and short-term forecast information, use an improved particle swarm optimization algorithm to solve the optimization model and obtain the optimal scheduling instruction set.

[0026] Specifically, to consider battery lifespan degradation during energy dispatch, an optimization model incorporating lifespan degradation costs is established with the objective function of minimizing total operating costs. The total operating cost includes diesel generator fuel costs, battery lifespan degradation costs, electricity purchase costs from the main grid, and renewable energy abandonment penalty costs. A model predictive control framework is adopted. Within each optimization cycle, based on real-time microgrid operating status data and short-term forecast information, the operating status data (including photovoltaic power generation, wind power generation, remaining battery capacity, battery temperature, diesel generator status, and load demand) represents the current system state. The short-term forecast information includes photovoltaic power generation forecasts, wind power generation forecasts, load forecasts, and battery... Battery status prediction and short-term prediction information can be obtained based on pre-established prediction models. For example, photovoltaic power generation prediction can be obtained through a pre-established photovoltaic power generation prediction model. Other short-term prediction models can also obtain corresponding prediction data through pre-established prediction models. The status data and short-term prediction information are input into the optimization model, and the multi-objective optimization model is solved using an improved particle swarm optimization algorithm to obtain the optimal scheduling instruction set. The improved particle swarm optimization algorithm is an optimization algorithm that fully considers the application scenario of actual microgrid scheduling. The specific solution process will be explained in detail in the candidate section. The optimal scheduling instruction set includes diesel engine start-stop instructions, battery charging and discharging instructions, renewable energy generation restriction instructions, and load reduction instructions.

[0027] In this embodiment, calculating the loss value within the effective charge-discharge cycle time period includes the following steps: Extreme value analysis is performed on the current time series data of the battery to identify all local maxima and local minima, forming a peak-valley feature point sequence. Based on the peak-valley feature point sequence, complete cycles are identified, and the depth value of the complete cycle is calculated. The depth value of the identified complete cycle is matched and queried with a pre-established battery stress factor table to obtain the loss weight value corresponding to each depth range. The loss weight values ​​of each complete cycle are accumulated to obtain the loss value in the current analysis time period.

[0028] Specifically, to extract key inflection points from continuous current data, simplify data complexity, and lay the foundation for complete stress cycle identification, extreme value analysis is performed on the battery current time series data to identify all local maxima and local minima. Local maxima are also called peak points (a point higher than its immediate and adjacent points), and local minima are also called valley points (a point lower than its immediate and adjacent points). Based on the peak-valley feature point sequence, complete cycles are identified, and the depth of the complete cycle is calculated. The specific method for identifying complete cycles and calculating the depth of the complete cycle will be explained in detail in the candidate section. Then, the depth values ​​of the identified complete cycles are matched with the pre-established battery stress factor table to convert the physical cycle range values ​​into standardized life loss weights, and establish a quantitative relationship between cycle stress and battery aging. In other words, based on the battery electrochemical aging mechanism, through a large number of accelerated aging experiments in the early stage, a table of correspondence between different cycle depths and battery capacity decay rates is established. Using the table lookup method or interpolation method, the actual identified cycle range is mapped to standardized loss weights. Then, the loss weight values ​​of each complete cycle are accumulated to obtain the loss value in the current analysis period.

[0029] In this embodiment, identifying a complete loop based on peak-valley feature point sequences includes the following steps: Four consecutive feature points are selected at once from the peak-valley feature point sequence. The first difference between the first feature point and the second feature point, the second difference between the second feature point and the third feature point, and the third difference between the third feature point and the fourth feature point are calculated respectively. When the second difference is less than or equal to the first difference and less than or equal to the third difference, it is determined that the second feature point and the third feature point form a complete cycle. The second difference is used as the depth value of the complete cycle. The feature points that form the complete cycle are removed from the peak-valley feature point sequence. The complete cycle is identified again for the removed peak-valley feature point sequence until no new complete cycle can be identified.

[0030] Specifically, assuming the obtained peak-valley feature point sequence is -50, -10, -40, +30, 10, 25, +5, for four consecutive peak-valley points A: -50, B: -10, C: -40, D: +30, the difference between two adjacent peak-valley points is calculated, yielding 40, 30, and 70. It is then determined whether the fluctuation in the middle is smaller than the fluctuations on either side. In this example, 30 is less than both 40 and 70, indicating that a complete charge-discharge cycle is formed from the second feature point to the third feature point. Physically, this means the battery experienced a cycle from a relatively high discharge current of -10 to a deeper discharge current of -40, and then back again. This 30A range of cycles causes a quantifiable lifespan loss to the battery, thus illustrating this... The loop is complete. The second and third feature points are removed from the peak-valley feature point sequence. Then, the loop continues to identify the peak-valley feature point sequence after removal. The current peak-valley feature point sequence is -50, +30, 10, 25, +5. For the four consecutive peak-valley points -50, +30, 10, and 25, the first difference between the first and second feature points, the second difference between the second and third feature points, and the third difference between the third and fourth feature points are calculated, yielding three differences: 80, 20, and 15. Since 20 is less than 80 but greater than 15, it indicates that the fluctuation from 30 to 10 is larger than the fluctuation from 10 to 25. This means that the current change is still ongoing and a clear "peak-valley" pattern has not yet formed. The complete cycle of "valley-peak" or "valley-peak-valley" fluctuations may be part of a larger cycle, or more data points may be needed to determine the cycle structure. At this point, these points are temporarily retained. The corresponding physical meaning is that the battery's current change pattern is not yet clear enough to determine whether an independent stress cycle has been completed. Therefore, subsequent data is analyzed, specifically the following four data points: +30, 10, 25, and +5. The calculated differences are 20, 15, and 20. Since the second difference is less than or equal to the first difference and less than or equal to the third difference, it indicates that a complete charge-discharge cycle is formed from the second characteristic point 10 to the third characteristic point 25. The difference of 15 is taken as the depth value of this complete cycle. At this point, 10 and 25 are removed from the peak-valley feature point sequence, resulting in -50, +30, and +5. If there are subsequent data values, the remaining peak-valley feature points can be analyzed using the same method. If not, two complete cycles are identified from the current data, corresponding to depth values ​​of 30 and 15. The stress factor table is then consulted. Assuming the stress factor table shows a loss weight of 0.01 for 0-20, 0.05 for 20-50, 0.12 for 50-100, and 0.2 for values ​​above 100, then the weight for 30 is 0.05, and the weight for 15 is 0.01, resulting in a calculated loss value of 0.5 + 0.1 = 0.06, representing that under the operating conditions during this period, the battery consumed 0.06 standard cycles of its lifespan. This loss value can be used to calculate lifespan loss costs and incorporated into the optimization model. The above-mentioned calculation of battery lifespan loss under random operating conditions based on the improved Rainflow counting method scientifically quantifies the abstract concept of lifespan, ensuring a more accurate assessment of battery lifespan loss under random and intermittent operating conditions, and providing a more reliable basis for subsequent intelligent scheduling.

[0031] In this embodiment, updating the stress factor table includes the following steps: Establish a battery loss prediction model. The model takes the historical operating status data of the battery as input and outputs the predicted loss value of the battery over a future period of time. The predicted loss value for a future time period is obtained based on the prediction model. The calculated loss value for the corresponding time period is obtained. The calculated loss value and the predicted loss value are compared to obtain the deviation value between the two. If the deviation value exceeds the preset threshold, the stress factor table is adjusted based on the deviation value and the historical operating conditions of the battery.

[0032] Specifically, the aforementioned method for calculating the loss value is a high-efficiency algorithm based on real-time data. It can quickly identify local peaks and valleys in the battery's charge and discharge current sequence and accurately calculate the depth and number of each stress cycle. This method relies on current actual operating data and can respond in real-time to instantaneous changes in the battery, such as sudden charge and discharge events, thus providing timely lifespan loss assessment. However, the calculation of the loss value is directly related to the stress factor table. Since the actual usage of batteries is complex and variable, the corresponding stress factor table may differ from the previously obtained experimental table as the battery is used. To accurately calculate battery loss, a battery loss prediction model is established based on historical battery operating data, including battery temperature and historical charge and discharge current data, to predict the loss value for a future preset time period. The system obtains predicted loss values ​​for future time periods based on a prediction model, calculates the corresponding loss values ​​for those time periods, and compares the calculated loss values ​​with the predicted loss values ​​to obtain the deviation value. If the deviation value exceeds a preset threshold, it indicates that the calculated loss value is inaccurate, further indicating a deviation in the stress factor table. Therefore, the stress factor table is adjusted based on the deviation value. If the calculated loss value is less than the predicted loss value, and the system determines that the battery is under harsh conditions of high temperature, high humidity, and frequent charging and discharging for a long time, the actual aging rate of the battery is much faster than the preset aging rate. Based on this result, the weight coefficients in the stress factor table are increased. If the calculated loss value is greater than the predicted loss value, the current battery operating environment may be relatively stable, and the charging and discharging current may also be stable for a long time, resulting in slower aging. Based on this result, the weight coefficients in the stress factor table are decreased.

[0033] The above method eliminates the need for mechanically applying fixed values ​​to the stress factor table used to calculate loss values. Instead, it uses the actual health status of the battery as perceived by the prediction model and adjusts the stress factor table based on the prediction results, ensuring that the calculated loss values ​​are always highly accurate.

[0034] In this embodiment, establishing a multi-objective optimization model that includes lifetime loss costs includes the following steps: Define decision variables, including diesel engine output power, battery charging and discharging power, renewable energy dispatch ratio, and load reduction. Set the objective function to minimize total operating cost. Set constraints, including power balance constraints, equipment operation constraints, and battery life constraints. The battery life loss cost in the total cost is the battery life loss amount multiplied by the battery unit loss cost.

[0035] Specifically, decision variables refer to the specific power and status commands that need to be issued to each device in each decision cycle (e.g., every 5 minutes). Diesel engine output power refers to the amount of power ordered to be generated by the diesel engine in the next scheduling cycle; battery charging / discharging power refers to the power at which the battery is charged or discharged in the next scheduling cycle; renewable energy dispatch ratio refers to the proportion of power generated by photovoltaic and wind power generation to be used to supply the load, and the proportion to be abandoned; load shedding refers to the amount of power to be cut off for non-critical loads in extreme situations (e.g., when energy is severely insufficient and batteries need protection). Decision variables are adjustable, facilitating the finding of the optimal dispatch command to minimize the total cost of ownership while satisfying all constraints. Constraints include power balance constraints, equipment operation constraints, and battery life constraints. The power balance constraint is defined as: solar power generation + wind power generation + diesel power generation + battery discharge power = load power + battery charging power + abandoned power + load shedding power. Equipment operation constraints include minimum and maximum output power limits for diesel generators, upper and lower limits for battery energy storage, battery charging / discharging power limits, and renewable energy dispatch limits.

[0036] In this embodiment, solving the multi-objective optimization model to obtain the optimal scheduling instruction set includes the following steps: Based on real-time operation status data and preset scenario classification rules, the current operation scenario type is identified. Based on the operation scenario type, historical high-quality scheduling schemes and weather forecast information, multiple candidate scheduling schemes are generated to form a candidate scheduling scheme set. Based on the operation scenario type, the corresponding algorithm parameter combination is called from the preset parameter configuration library. The algorithm parameters include inertia weight, acceleration factor and constraint penalty coefficient. The candidate scheduling scheme is updated based on the combination of algorithm parameters. Knowledge-guided mutation operation is performed on the updated candidate scheduling scheme. The candidate scheduling scheme set after mutation is sorted using the fast non-dominated sorting method. The candidate scheduling scheme set is calculated using the congestion distance calculation method. This step is repeated until the iteration termination condition is met. The iteration termination condition is that the maximum number of iterations is reached or the improvement of the total running cost of the candidate scheduling scheme in 10 consecutive iterations is less than a preset threshold. The optimal scheduling scheme is selected from the final set of candidate scheduling schemes, and the optimal scheduling instruction set is generated based on the optimal scheduling scheme.

[0037] Specifically, to improve convergence speed and initial solution quality, differentiated initialization strategies are adopted based on different microgrid operation scenarios. Real-time microgrid operation status data is collected, and the current operation scenario type is identified based on pre-defined scenario classification rules. These rules specify that renewable energy output exceeds 120% of load demand (sufficient wind and solar power scenario), renewable energy output is less than 60% of load demand (insufficient wind and solar power scenario), load demand exceeds 85% of rated capacity (peak-hour scenario), and critical equipment malfunctions (emergency backup scenario). Multiple candidate scheduling schemes are generated based on the operation scenario type, historical high-quality scheduling schemes, and weather forecast information. For example, in low wind and solar power scenarios, the initial scheme primarily based on battery charging is generated; in low wind and solar power scenarios, the initial scheme primarily based on diesel generator power supply is generated. Furthermore, knowledge of historical high-quality schemes is incorporated to avoid generating obviously unreasonable initial solutions.

[0038] A parameter configuration library was established based on a large amount of experimental data, with optimized parameter combinations for each type of operating scenario threshold. Among them, the inertia weight controls the balance between global exploration and local development capabilities, the acceleration factor affects the speed at which particles move to the optimal position, and the constraint penalty coefficient ensures the feasibility of the solution. The parameter values ​​have been verified by a large number of experiments to ensure optimal performance in each scenario.

[0039] By executing the standard particle swarm optimization algorithm, candidate scheduling schemes are updated based on combinations of algorithm parameters. The specific update process will be explained in detail later. The knowledge-guided mutation operation performed on the updated candidate scheduling schemes refers to designing specialized mutation rules based on practical experience in microgrid operation. Mutations on battery charge / discharge depth avoid overcharging and over-discharging, mutations on diesel generator operating time reduce frequent start-stop cycles, and mutations on renewable energy consumption improve the utilization rate of clean energy. These mutation operations, based on the standard algorithm, are guided by domain knowledge; the specific mutation operation process will be explained in detail later.

[0040] To provide a rich variety of high-quality scheduling schemes, a fast non-dominated sorting method is used to sort the candidate scheduling scheme set after mutation operations. The congestion distance calculation method is used to calculate the candidate scheduling scheme set. The fast non-dominated sorting method effectively identifies high-quality candidate scheduling schemes, and the congestion distance is used to ensure the uniform distribution of candidate scheduling schemes in the candidate scheduling scheme set.

[0041] Repeatedly execute the mutation operation of the candidate scheduling scheme and the operation of generating the candidate scheduling scheme set until the iteration termination condition is met. The iteration termination condition is that the maximum number of iterations is reached or the improvement of the total running cost of the candidate scheduling scheme by 10 consecutive iterations is less than a preset threshold. Finally, select the optimal scheduling scheme from the final obtained candidate scheduling scheme set and generate the optimal scheduling instruction set based on the optimal scheduling scheme.

[0042] The above method provides effective technical support for the optimized scheduling decisions of intelligent integrated power supply equipment by deeply optimizing the method specifically for the characteristics of microgrid scheduling problems.

[0043] In this embodiment, updating the candidate scheduling scheme based on the combination of algorithm parameters includes the following steps: For each candidate scheduling scheme, an adjustment instruction is calculated. The adjustment instruction is a weighted synthesis of three parts: inertia, individual cognition, and social learning. The corresponding inertia weight, individual cognition weight, and social learning weight are obtained from the preset parameter configuration library according to the current running scenario type. Each candidate scheduling scheme is added to the adjustment instruction calculated in the previous step to obtain the updated candidate scheduling scheme. Constraint checks and amplitude limiting are then performed on the updated candidate scheduling scheme.

[0044] Specifically, to efficiently optimize each candidate scheduling scheme, an adjustment instruction is calculated for each scheme. This instruction is a weighted composite of three parts: an inertial component, an individual cognitive component, and a social learning component. The inertial component inherits the previous adjustment instruction from the candidate scheme to maintain search continuity, and its weight is the inertial weight. The social learning component refers to the direction and distance of adjustment towards the current globally optimal candidate scheduling scheme, and its weight is the social learning weight. The individual cognitive component refers to the direction and distance of adjustment towards the historical optimal position of the current candidate scheduling scheme, and its weight is the individual cognitive weight. To generate a new generation of candidate scheduling schemes based on the adjustment instructions, each candidate scheduling scheme is added to the adjustment instruction calculated in the previous step to obtain an updated candidate scheduling scheme. By updating the candidate scheduling schemes, incremental improvements are achieved, ensuring the smoothness and physical feasibility of the scheme updates.

[0045] To ensure that the updated candidate scheduling schemes meet basic operational requirements, constraint checks are performed on them. These checks include: battery power constraint processing: ensuring that the charging and discharging power of the candidate scheduling schemes does not exceed the maximum allowable power; diesel generator output constraint processing: ensuring that the diesel generator output is between the minimum output and the rated power; and power balance constraint processing: ensuring that the power generation of each candidate scheduling scheme is basically balanced with the load demand. Furthermore, the updated candidate scheduling schemes undergo amplitude limiting processing to ensure that the adjustment range is within a reasonable range. Specifically, this includes: setting a maximum adjustment range; when the magnitude of the adjustment command exceeds the maximum adjustment range, scaling the adjustment command proportionally; and dynamically adjusting the maximum adjustment range according to the operational scenario type, setting a larger value during the exploration phase and a smaller value during the fine-tuning optimization phase.

[0046] The above method deeply integrates the traditional particle swarm optimization algorithm with the specific characteristics of microgrid scheduling, solving the problem of insufficient interpretability and practicality of abstract algorithms in practical engineering applications, and providing reliable technical support for the optimization decision-making of intelligent integrated power equipment.

[0047] In this embodiment, performing a knowledge-guided mutation operation on the updated candidate scheduling scheme includes the following steps: For each candidate scheduling scheme, check whether the net charge and discharge amount of the battery within any consecutive time period is greater than the preset threshold. If so, use the power scaling method to perform deep variation correction. Identify the start time of the diesel engine in each candidate scheduling scheme. When the continuous running time is detected to be less than the preset time threshold, extend the running time to the minimum start time or cancel the start command. Identify the time periods in each candidate scheduling scheme where the power generation of renewable energy exceeds the load demand, check the charging status of the batteries during these periods, and if the batteries are not charged, use excess renewable energy to charge them; perform safety and feasibility verification on the candidate scheduling schemes after the mutation operation, roll back the candidate scheduling schemes that fail the verification to the state before the mutation for the candidate scheduling schemes that succeed in the verification, evaluate the change in the total operating cost before and after the mutation for the candidate scheduling schemes that succeed in the verification, if the total operating cost decreases, record the mutation type, mutation magnitude and mutation effect, and establish a mutation knowledge base.

[0048] Specifically, to prevent overcharging and discharging of batteries and extend their lifespan, based on the electrochemical characteristics of batteries, deep charging and discharging accelerates battery aging. A sliding time window detection method is used to identify scheduling periods that violate the depth of charge / discharge limits. A proportional scaling method is employed to maintain the time distribution characteristics of the original scheduling scheme, imposing safety limits only on the amplitude. This ensures the safety of candidate scheduling schemes while significantly extending battery lifespan. Through preventative protection, premature battery failure due to unreasonable scheduling schemes is avoided, reducing the overall lifespan maintenance cost.

[0049] To reduce frequent start-stop operations of diesel generators and lower equipment wear and operating costs, based on the mechanical characteristics of diesel generators, frequent start-stop operations increase equipment wear and maintenance costs. By identifying short-term operation commands, two strategies are adopted to correct the situation: extending the operation time or canceling the start-stop operation. The extension strategy is applicable to critical power supply periods, while the cancellation strategy is applicable to alternative power supply periods, ensuring that system reliability is not affected. Through the above optimized operation strategies, the operating efficiency of diesel generators is improved while reducing equipment wear and maintenance costs.

[0050] To improve the utilization rate of renewable energy and reduce curtailment, based on the principle of prioritizing energy use, energy storage charging should be prioritized during periods of renewable energy surplus. Furthermore, by monitoring the difference between wind and solar power output and load demand in real time, the charging power of the batteries can be dynamically adjusted. A minimum limit can be set to ensure that the charging power does not exceed the equipment capacity or the total amount of available renewable energy, thereby achieving full and rational utilization of energy.

[0051] To ensure the safety and feasibility of the mutated scheduling scheme, the candidate scheduling schemes after the mutation operation are verified for safety and feasibility. For candidate scheduling schemes that fail the verification, they are rolled back to the state before mutation. For candidate scheduling schemes that succeed in the verification, the change in the total operating cost before and after the mutation is evaluated. If the total operating cost decreases, the mutation type, mutation magnitude and mutation effect are recorded, and a mutation knowledge base is established. By establishing the mutation knowledge base, experience is provided for subsequent mutation operations, and the self-learning optimization of the mutation strategy is realized.

[0052] In this embodiment, the process of calculating the candidate scheduling scheme set using the congestion distance calculation method includes the following steps: A non-dominated sorting algorithm is used to obtain multiple non-dominated layer sequences. Each non-dominated layer has multiple candidate scheduling schemes. For each candidate scheduling scheme in a non-dominated layer, the candidate scheduling schemes are sorted according to each sub-cost. The congestion distance between two candidate scheduling schemes located at the boundary is set to infinity. For candidate scheduling schemes located at non-boundaries, the difference between its sub-cost and that of its two adjacent candidate scheduling schemes is calculated. After normalizing the difference, the sum of the normalized difference is calculated. The sum of the difference is used as the first distance of the candidate scheduling scheme. The result of adding the first distances corresponding to all sub-costs is used as the final congestion distance. The total number of candidate scheduling schemes is preset. First, candidate scheduling schemes are selected and added to the candidate scheduling scheme set in descending order of non-dominated layer. Within the same non-dominated layer, candidate scheduling schemes are selected in descending order of congestion distance until the total number of candidate scheduling schemes reaches the preset value.

[0053] Specifically, assuming there are 7 candidate scheduling schemes in a non-dominated layer, taking only fuel cost and battery life loss cost as examples, scheme A: (80, 60), scheme B: (70, 75), scheme C: (90, 50), scheme D: (85, 65), scheme E: (75, 80), and scheme F: (95, 45), sorted by fuel cost, we get: scheme B: 70, scheme E: 75, scheme A: 80, scheme D: 85, scheme C: 90, scheme F: 95. Sorted by battery life loss cost, we get: scheme F: 45, scheme C: 50, scheme A: 60, scheme D: 65, scheme B: 75, scheme E: 80. The range of cost 1 is 95-70=25, and the range of cost 2 is 80-45=35. For the sorted sequence of cost 1 [B, E, A, D, C, ...], ... [F], Schemes B and F are boundary schemes with congestion distances set to infinity. The first distance for scheme E is: (difference between B and E) / 25 + (difference between E and A) / 25 = 10 / 24 = 0.4. The first distances for schemes A, D, and C are calculated in the same way, and are all 0.4. Similarly, the first distances for each scheduling scheme are calculated based on battery life loss cost, which are: Scheme F: Infinity, Scheme C: 0.429, Scheme A: 0.429, Scheme D: 0.429, Scheme B: 0.429, Scheme E: Infinity. The sum of the first distances corresponding to all sub-costs is taken as the most used congestion distance for each scheme. The final congestion distance results are: Scheme A: 0.829, Scheme B: Infinity, Scheme C: 0.829, Scheme D: 0.829, Scheme E: Infinity, and Scheme F: Infinity.

[0054] The total number of candidate scheduling schemes is preset, for example, 50. First, candidate scheduling schemes are selected and added to the candidate scheduling scheme set in descending order of non-dominated layer. Within the same non-dominated layer, candidate scheduling schemes are selected in descending order of congestion distance until the total number of candidate scheduling schemes reaches the preset value.

[0055] Furthermore, obtaining multiple non-dominated layer sequences using a non-dominated sorting algorithm includes the following steps: For any two candidate scheduling schemes A and B, the rule for determining the dominance relationship is as follows: if all objective function values ​​of scheme A are no worse than those of scheme B, and at least one objective function value is strictly better than that of scheme B, then scheme A is said to dominate scheme B. Record the set of other schemes dominated by each candidate scheduling scheme, as well as the number of schemes dominated by other schemes. Select schemes with zero dominated numbers from all candidate scheduling schemes to form the first layer of non-dominated solution set. Decrement the number of dominated numbers of the schemes dominated by the selected non-dominated solution set by one. Select schemes with zero dominated numbers from the remaining schemes to form the next layer of non-dominated solution set. Repeat the above process until all candidate scheduling schemes are assigned to the corresponding non-dominated layer.

[0056] Specifically, assuming there are 5 candidate scheduling schemes, we need to minimize two objectives: fuel cost (F1) and lifetime loss cost (F2). The scheme data are as follows: Scheme A: F1=100, F2=50, Scheme B: F1=80, F2=70, Scheme C: F1=120, F2=40. Option D: F1=90, F2=80; Option E: F1=110, F2=60. Calculate the number of items controlled: Option A: Controlled by B (B's F1=80<100, F2=70>50, not satisfying the control condition), controlled by C (C's F2=40<50, but F1=120>100, not satisfying the control condition), controlled by D (D's F1=90<100, F2=80>50, not satisfying the control condition), controlled by E (E's F1=110>100, F2=60>50, not satisfying the control condition), number of items controlled = 0; Option B: Compared with A (F1=8... 0 < 100, F2 = 70 > 50, do not dominate A), compared with C (F1 = 80 < 120, F2 = 70 > 40, do not dominate C), compared with D (F1 = 80 < 90, F2 = 70 < 80, dominate D), compared with E (F1 = 80 < 110, F2 = 70 > 60, do not dominate E), the number of dominated items = 0; Option C: compared with A (F1 = 120 > 100, F2 = 40 < 50, do not dominate A), compared with B (F1 = 120 > 80, F2 = 4... 0 < 70, not dominating B), compared with D (F1 = 120 > 90, F2 = 40 < 80, not dominating D), compared with E (F1 = 120 > 110, F2 = 40 < 60, dominating E), the number of dominated is 0; Option D: dominated by B, compared with A (F1 = 90 < 100, F2 = 80 > 50, not dominating), compared with C (F1 = 90 < 120, F2 = 80 > 40, not dominating), compared with E (F1 = 90 < 110, F2 = 80 > 60, ... No domination), number of dominated options = 1; Option E: Dominated by C, compared by A (F1=110>100, F2=60>50, no domination), compared by B (F1=110>80, F2=60<70, no domination), compared by D (F1=110>90, F2=60<80, no domination), number of dominated options = 1. After the above steps, the first non-dominated layer is obtained: Options A, B, C (number of dominated options = 0). The second non-dominated layer is: Options D, E (number of dominated options = 1).

[0057] The above describes an integrated power system scheduling method according to an embodiment of this application. The following describes an integrated power system according to an embodiment of this application. Please refer to [link / reference]. Figure 2 One embodiment of an integrated power supply system in this application includes: The data acquisition module collects real-time microgrid operating status data, including photovoltaic power generation, wind power generation, load demand power, diesel generator status, power equipment battery operating status, state of charge, temperature, voltage and current. The loss calculation module establishes a stress factor table for the battery, identifies and analyzes the effective charge-discharge cycles of the battery within a preset time window based on the battery's current data, calculates the loss value within the effective charge-discharge cycle time period based on the stress factor table, converts the loss value into battery life loss cost, and updates the stress factor table every preset time interval. The optimization scheduling module takes minimizing the total operating cost as the objective function and establishes an optimization model that includes the cost of lifespan loss. The total operating cost includes the fuel cost of diesel generators, the cost of battery lifespan loss, the cost of purchasing electricity from the main grid, and the cost of renewable energy abandonment penalty. A model predictive control framework is adopted. In each optimization cycle, based on the real-time operating status data of the microgrid and short-term forecast information, an improved particle swarm optimization algorithm is used to solve the optimization model to obtain the optimal scheduling instruction set.

[0058] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0059] The above embodiments merely illustrate several implementation methods of the present invention, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this patent should be determined by the appended claims. The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An integrated power system scheduling method, characterized in that, include: Step S1: Collect microgrid operation status data in real time. The operation status data includes photovoltaic power generation, wind power generation, load demand power, diesel generator status, power equipment battery operation status, state of charge, temperature, voltage and current. Step S2: Establish a stress factor table for the battery. Based on the battery's current data, identify and analyze the effective charge-discharge cycles of the battery within a preset time window. Calculate the loss value within the effective charge-discharge cycle time period based on the stress factor table, convert the loss value into battery life loss cost, and update the stress factor table every preset time interval. Step S3: Using minimizing the total operating cost as the objective function, establish an optimization model that includes lifetime loss costs. The total operating cost includes diesel generator fuel costs, battery lifetime loss costs, electricity purchase costs from the main grid, and renewable energy abandonment penalty costs. Employing a model predictive control framework, within each optimization cycle, based on real-time microgrid operating status data and short-term forecast information, use an improved particle swarm optimization algorithm to solve the optimization model and obtain the optimal scheduling instruction set. This includes: identifying the current operating scenario type based on real-time operating status data and preset scenario classification rules; generating a candidate scheduling scheme set based on the operating scenario type, historical high-quality scheduling schemes, and weather forecast information; calling the corresponding algorithm parameter combination from a preset parameter configuration library based on the operating scenario type, including inertia weight, acceleration factor, and constraint penalty coefficient; and updating the candidate scheduling schemes based on the algorithm parameter combination, including: calculating adjustment instructions for each candidate scheduling scheme. The adjustment instruction is a weighted synthesis of three parts: inertia, individual cognition, and social learning. The corresponding inertia weight, individual cognition weight, and social learning weight are obtained from a preset parameter configuration library based on the current operating scenario type. Each candidate scheduling scheme is added to the adjustment instruction calculated in the previous step to obtain an updated candidate scheduling scheme. Constraint checks and amplitude limiting are performed on the updated candidate scheduling schemes. Knowledge-guided mutation operations are performed on the updated candidate scheduling schemes. The candidate scheduling scheme set after mutation is sorted using a fast non-dominated sorting method. The candidate scheduling scheme set is calculated using a congestion distance calculation method. This step is repeated until the iteration termination condition is met. The iteration termination condition is that the maximum number of iterations is reached or the improvement of the total operating cost of the candidate scheduling scheme in 10 consecutive iterations is less than a preset threshold. The optimal scheduling scheme is selected from the final obtained candidate scheduling scheme set, and the optimal scheduling instruction set is generated based on the optimal scheduling scheme.

2. The method according to claim 1, characterized in that, Calculating the loss value over the effective charge-discharge cycle period includes the following steps: Extreme value analysis is performed on the current time series data of the battery to identify all local maxima and local minima, forming a peak-valley feature point sequence. Based on the peak-valley feature point sequence, complete cycles are identified, and the depth value of the complete cycle is calculated. The depth value of the identified complete cycle is matched and queried with a pre-established battery stress factor table to obtain the loss weight value corresponding to each depth range. The loss weight values ​​of each complete cycle are accumulated to obtain the loss value in the current analysis time period.

3. The method according to claim 2, characterized in that, The complete loop identification based on peak-valley feature point sequence includes the following steps: Four consecutive feature points are selected at once from the peak-valley feature point sequence. The first difference between the first feature point and the second feature point, the second difference between the second feature point and the third feature point, and the third difference between the third feature point and the fourth feature point are calculated respectively. When the second difference is less than or equal to the first difference and less than or equal to the third difference, it is determined that the second feature point and the third feature point form a complete cycle. The second difference is used as the depth value of the complete cycle. The feature points that form the complete cycle are removed from the peak-valley feature point sequence. The complete cycle is identified again for the removed peak-valley feature point sequence until no new complete cycle can be identified.

4. The method according to claim 1, characterized in that, Updating the stress factor table includes the following steps: Establish a battery loss prediction model. The model takes the historical operating status data of the battery as input and outputs the predicted loss value of the battery over a future period of time. The predicted loss value for a future time period is obtained based on the prediction model. The calculated loss value for the corresponding time period is obtained. The calculated loss value and the predicted loss value are compared to obtain the deviation value between the two. If the deviation value exceeds the preset threshold, the stress factor table is adjusted based on the deviation value and the battery's historical operating conditions.

5. The method according to claim 1, characterized in that, Establishing a multi-objective optimization model that includes lifetime loss costs includes the following steps: Define decision variables, including diesel engine output power, battery charging and discharging power, renewable energy dispatch ratio, and load reduction. Set the objective function to minimize total operating cost. Set constraints, including power balance constraints, equipment operation constraints, and battery life constraints. The battery life loss cost in the total cost is the battery life loss amount multiplied by the battery unit loss cost.

6. The method according to claim 1, characterized in that, Performing a knowledge-guided mutation operation on the updated candidate scheduling scheme includes the following steps: For each candidate scheduling scheme, check whether the net charge and discharge amount of the battery within any consecutive time period is greater than the preset threshold. If so, use the power scaling method to perform deep variation correction. Identify the start time of the diesel engine in each candidate scheduling scheme. When the continuous running time is detected to be less than the preset time threshold, extend the running time to the minimum start time or cancel the start command. Identify the time periods in each candidate scheduling scheme where the power generation of renewable energy exceeds the load demand, check the charging status of the battery during these time periods, and if the battery is not charged, use the excess renewable energy to charge the battery. The candidate scheduling schemes after mutation are verified for safety and feasibility. For candidate scheduling schemes that fail the verification, they are rolled back to their state before mutation. For candidate scheduling schemes that succeed in the verification, the change in the total operating cost before and after the mutation is evaluated. If the total operating cost decreases, the mutation type, mutation magnitude and mutation effect are recorded, and a mutation knowledge base is established.

7. The method according to claim 1, characterized in that, The process of calculating the candidate scheduling scheme set using the congestion distance method includes the following steps: A non-dominated sorting algorithm is used to obtain multiple non-dominated layer sequences. Each non-dominated layer has multiple candidate scheduling schemes. For each candidate scheduling scheme in a non-dominated layer, the candidate scheduling schemes are sorted according to each sub-cost. The congestion distance between two candidate scheduling schemes located at the boundary is set to infinity. For candidate scheduling schemes located at non-boundaries, the difference between its sub-cost and that of its two adjacent candidate scheduling schemes is calculated. After normalizing the difference, the sum of the normalized difference is calculated. The sum of the difference is used as the first distance of the candidate scheduling scheme. The result of adding the first distances corresponding to all sub-costs is used as the final congestion distance. The total number of candidate scheduling schemes is preset. First, candidate scheduling schemes are selected and added to the candidate scheduling scheme set in descending order of non-dominated layer. Within the same non-dominated layer, candidate scheduling schemes are selected in descending order of congestion distance until the total number of candidate scheduling schemes reaches the preset value.

8. An integrated power supply system for executing the integrated power supply system scheduling method as described in claims 1-7, characterized in that, The system includes the following modules: The data acquisition module collects real-time microgrid operating status data, including photovoltaic power generation, wind power generation, load demand power, diesel generator status, power equipment battery operating status, state of charge, temperature, voltage and current. The loss calculation module establishes a stress factor table for the battery, identifies and analyzes the effective charge-discharge cycles of the battery within a preset time window based on the battery's current data, calculates the loss value within the effective charge-discharge cycle time period based on the stress factor table, converts the loss value into battery life loss cost, and updates the stress factor table every preset time interval. The optimization scheduling module, with the objective function of minimizing total operating cost, establishes an optimization model that includes lifetime depreciation costs. Total operating cost includes diesel generator fuel costs, battery lifetime depreciation costs, electricity purchase costs from the main grid, and renewable energy abandonment penalty costs. Employing a model predictive control framework, within each optimization cycle, based on real-time microgrid operating status data and short-term forecast information, an improved particle swarm optimization algorithm is used to solve the optimization model to obtain the optimal scheduling instruction set. This includes: identifying the current operating scenario type based on real-time operating status data and preset scenario classification rules; generating multiple candidate scheduling schemes based on the operating scenario type, historical high-quality scheduling schemes, and weather forecast information; calling corresponding algorithm parameter combinations from a preset parameter configuration library based on the operating scenario type; algorithm parameters including inertia weight, acceleration factor, and constraint penalty coefficient; and updating the candidate scheduling schemes based on the algorithm parameter combinations, including: calculating adjustment instructions for each candidate scheduling scheme. The adjustment instruction is a weighted synthesis of three parts: inertia, individual cognition, and social learning. The corresponding inertia weights, individual cognition weights, and social learning weights are obtained from a preset parameter configuration library based on the current operating scenario type. Each candidate scheduling scheme is added to the adjustment instruction calculated in the previous step to obtain an updated candidate scheduling scheme. Constraint checks and amplitude limiting are performed on the updated candidate scheduling schemes. Knowledge-guided mutation operations are performed on the updated candidate scheduling schemes. The candidate scheduling scheme set after mutation is sorted using a fast non-dominated sorting method. The candidate scheduling scheme set is calculated using a congestion distance calculation method. This step is repeated until the iteration termination condition is met. The iteration termination condition is that the maximum number of iterations is reached or the improvement in the total operating cost of the candidate scheduling scheme after 10 consecutive iterations is less than a preset threshold. The optimal scheduling scheme is selected from the final obtained candidate scheduling scheme set, and an optimal scheduling instruction set is generated based on the optimal scheduling scheme.

Images