Photovoltaic on-site consumption collaborative control method and system based on intelligent fusion terminal

Abstract

The application discloses a photovoltaic on-site consumption collaborative control method and system based on an intelligent fusion terminal, and comprises the following steps: initializing control parameters, generating initial samples, obtaining photovoltaic on-site consumption power fitness corresponding to each sample, and constructing a sample database DB; samples in the database DB are sorted according to advantages and disadvantages, a parent iteration population is constructed, and a child population is generated through crossing and variation; a plurality of radial basis function proxy models are constructed, photovoltaic on-site consumption fitness of each individual in the child population is respectively predicted, and a global optimal control parameter solution is obtained; whether the optimal control parameter solution is updated is judged, if the optimal control parameter solution is not updated, a credible full crossing operation is performed; if the optimal control parameter solution is updated, whether a maximum fitness evaluation number termination condition is satisfied is iteratively judged, a control instruction is issued under the optimal control parameter solution, and control parameter iteration optimization is realized.

Description

Technical Field

[0001] This invention belongs to the field of power distribution system operation and control technology, specifically relating to a method and system for coordinated control of local photovoltaic consumption based on an intelligent integrated terminal. Background Technology

[0002] The statements herein provide only background information in relation to this invention and do not necessarily constitute prior art.

[0003] With the rapid development of distributed photovoltaic (PV) technology, the penetration rate of distributed PV in low-voltage distribution areas continues to increase. However, this has led to increasingly prominent problems such as insufficient local PV consumption, voltage exceeding limits at distribution nodes, and reverse overload on power lines. These issues seriously affect the safe, stable, and economical operation of the distribution network and limit the large-scale development of distributed PV. Maximizing local PV consumption has become a key requirement for ensuring the safe, stable, and economical operation of distribution areas and improving PV utilization efficiency.

[0004] In-depth analysis reveals that maximizing local photovoltaic (PV) power consumption is a typical single-objective, black-box, expensive optimization problem: its control objective is singular and clear, namely, maximizing the total active power consumed locally by PV. However, the fitness evaluation process for control parameters relies on power flow calculations of the distribution network or on-site measured data. Each evaluation is time-consuming and computationally expensive, making it impossible to support large-scale iterative optimization using traditional control methods. This type of control problem can be described as follows:

[0005] ;

[0006] in, Let t be the local photovoltaic power consumption in time period t, where T is the total number of optimization time periods. The input variables are the control parameters of the adjustable equipment in the distribution network, and the output is the local photovoltaic power consumption. There is no explicit analytical formula, and the input variables have clear upper and lower limit constraints.

[0007] Existing photovoltaic (PV) grid integration control methods for low-voltage distribution areas still have many shortcomings: First, they are not fully adapted to the edge computing resource constraints of new intelligent integrated terminals, resulting in insufficient lightweight control methods and difficulty in achieving local closed-loop control; second, the prediction accuracy and robustness of control parameters are limited, easily leading to ineffective control due to prediction deviations, thus affecting the local PV grid integration effect; third, the control parameter selection strategy is unreasonable, with low efficiency in selecting high-value control parameter samples and high control costs; fourth, the control methods are prone to getting trapped in local optima, resulting in insufficient global control capabilities; and fifth, they are not closely integrated with the local control capabilities of terminals, leading to poor implementation of control strategies and difficulty in meeting the "efficient, accurate, and lightweight" control requirements in practical engineering applications.

[0008] Furthermore, when distributed photovoltaic (PV) output surges and local load consumption is insufficient, the voltage at distribution network nodes can easily approach or reach its upper limit. If optimal coordinated control strategies are not implemented in a timely manner, voltage over-limit faults can occur, resulting in wasted PV energy. Traditional control methods struggle to achieve optimal coordinated regulation across multiple devices, failing to balance voltage safety with local PV consumption. Therefore, developing a PV consumption coordinated control method and system that adapts to the edge computing power of new intelligent fusion terminals, has low evaluation costs, high control precision, and can be executed locally in a closed loop, has significant engineering application value and practical significance. Summary of the Invention

[0009] To address the shortcomings of existing technologies, the purpose of this invention is to provide a collaborative control method and system for local photovoltaic (PV) consumption based on intelligent fusion terminals. This method overcomes the technical deficiencies of existing control methods, such as poor adaptability, high control costs, susceptibility to local optima, and inability to simultaneously ensure voltage safety and PV consumption. It can quickly determine the optimal control parameters with a limited number of fitness assessments, maximizing local PV consumption in the distribution area while ensuring qualified voltage and safe operation of the distribution network. It also adapts to the edge computing requirements of new intelligent fusion terminals. This method is suitable for low-voltage distribution areas with high-penetration distributed PV access, achieving a synergistic guarantee of maximizing local PV consumption and ensuring the safe and stable operation of the distribution network.

[0010] To achieve the above objectives, the present invention is implemented through the following technical solution:

[0011] In a first aspect, the present invention provides a photovoltaic on-site consumption collaborative control method based on an intelligent fusion terminal, comprising the following steps:

[0012] Initialize control parameters, generate initial samples, obtain the photovoltaic local consumption power adaptability corresponding to each sample, and construct sample database DB;

[0013] The samples in the database DB are sorted according to their quality, a parent iterative population is constructed, and crossover and mutation are performed to generate the offspring population.

[0014] Multiple radial basis function surrogate models are constructed to predict the local photovoltaic consumption fitness of each individual in the offspring population, and the global optimal control parameter solution is obtained.

[0015] If the optimal control parameter solution has not been updated, a reliable full crossover operation is performed. If it has been updated, the process iteratively checks whether the termination condition of the maximum fitness evaluation count is met, and then issues control commands with the optimal control parameter solution to achieve iterative optimization of the control parameters.

[0016] Furthermore, maximizing the local consumption of distributed photovoltaic power belongs to a single-objective black-box expensive optimization problem. The objective function is to maximize the total active power of distributed photovoltaic power in the local area, and its mathematical expression is: ,in Let t be the local photovoltaic power consumption in time period t. To optimize the total number of time periods, the fitness assessment relies on power flow calculations in the distribution network or actual measurements from smart fusion terminals. Each assessment takes no less than 100ms, and the number of assessments is strictly limited by edge computing resources. This control method significantly reduces the actual number of assessments through a multi-model proxy prediction strategy, adapting to edge computing power constraints.

[0017] Furthermore, the control parameters include the problem dimension d, population size N, crossover probability CR, mutation factor F, and its historical memory bank. and The maximum fitness evaluation count (MaxFes) needs to be set in conjunction with the edge computing power limitations of the intelligent fusion terminal to ensure that the operating efficiency of the control method matches the terminal's computing power, and to ensure that the control method operates in a lightweight and efficient manner; the historical memory bank is also included. and Initialized to empty, this is used to store the valid values ​​of crossover probability and mutation factor during iteration. The problem dimension corresponds to the number of control parameters, specifically including the reactive power output of the photovoltaic inverter, energy storage charging and discharging power, capacitor switching status, and voltage regulation commands of the adjustable equipment controlled by the intelligent fusion terminal. All control parameters must be taken within the operating range of the corresponding equipment to ensure the control strategy can be implemented effectively.

[0018] Furthermore, the initial samples were generated using a Latin hypercube experimental design, with the number of samples set to 30-50 depending on the problem dimension. Each sample corresponds to a complete distribution network control strategy (i.e., a set of control parameter values). Through distribution network power flow calculations or measured data from intelligent integrated terminals, the photovoltaic local consumption power fitness corresponding to each sample was obtained. The samples and their fitness values ​​were stored in the real evaluation historical sample database DB, completing the initial construction of the sample library and providing basic data support for subsequent control parameter optimization and iterative control.

[0019] Furthermore, after constructing the sample database DB, a check can be performed to determine whether the termination condition of the maximum fitness evaluation count is met. If the maximum fitness evaluation count MaxFes has been reached, the optimal control parameter solution that maximizes local photovoltaic power consumption in the database DB is output. If not, all samples in the current database DB are sorted from best to worst in terms of local photovoltaic power consumption fitness, and the top N samples are selected to form the parent iteration population. Meanwhile, N second-best samples are selected in order of quality and archived into database A, so as to retain and reuse high-quality control parameter samples and provide high-quality sample support for subsequent iterative control.

[0020] Furthermore, a mutation operation is performed on the parent population to generate an experimental population V. Then, a binomial crossover operation is performed on both the parent population and the experimental population to generate a offspring population U. By combining superior control parameters from different individuals through crossover, population diversity is improved. Mutation and crossover expand the search range for control parameters, increasing the probability of capturing optimal control parameters. Each parent individual is configured with a specific mutation factor F.

[0021] Furthermore, the steps for constructing multiple radial basis function surrogate models are as follows:

[0022] Multiple training sample sets with different distributions are extracted from the database DB, and various radial basis function (RBF) surrogate models are constructed respectively. The multiple models run in parallel to predict the photovoltaic on-site consumption fitness of each individual in the offspring population. Through multi-model collaborative prediction, the reliability and accuracy of fitness prediction are improved by taking into account both global control parameter search and local accurate prediction, thereby reducing the actual evaluation cost.

[0023] Taking the construction of four radial basis function surrogate models as an example, the four models are: 1) : Obtained by training on all samples in database DB ( ), used for fitness prediction of the global control parameter search range; 2) From the database DB The optimal sample and its vicinity The samples are trained to focus on accurate predictions near the optimal control parameters; 3) : The top of the best samples in database DB Training with 100 training samples to enhance the prediction weights of high-quality control parameter samples; 4) :Depend on The system is trained using recently evaluated samples from the database to adapt to the dynamic changes in the real-time operating status of the distribution network; the parameter values ​​are set according to the problem dimension, and are used for problems with fewer than 100 dimensions. In 100-dimensional problems .

[0024] Furthermore, based on the predicted mean of multiple proxy models and prediction variance Calculate the improvement probability of each offspring relative to its parent. ; Screening Improvement Probability Individuals exceeding a set value (optional 0.95) are archived to the database. This forms a highly reliable sample set of potential improved control parameters; simultaneously, it calculates the solution of each offspring relative to the current globally optimal control parameters. Improvement probability Select the one with the highest probability of improvement. offspring individuals The actual photovoltaic absorption adaptability is calculated using the objective function (distribution network power flow calculation or terminal measurement). The individual and its true fitness value are archived to the database DB to ensure that each real assessment focuses on high-value control parameter samples.

[0025] This process employs a dual-criteria hierarchical sampling strategy to efficiently screen high-value control parameter samples and reduce the cost of ineffective evaluations: the first criterion is based on the improvement probability of offspring individuals relative to their parent individuals. Based on the standard, samples of high-confidence potential control parameters with an improvement probability greater than 0.95 are selected and stored in the database. This reduces ineffective evaluations; the second layer uses the solution of the offspring individual relative to the current global optimal control parameters. Improvement probability To ensure optimal performance, a single, standardized entity is selected for real-world evaluation. This ensures that each real-world evaluation drives the control parameters towards the optimal direction to the greatest extent possible. This maximizes the value of each real-world evaluation while minimizing control costs.

[0026] Furthermore, according to Further update the database ,make sure The control performance of offspring individuals, calculated in real-world scenarios, is superior to that of their parents. Invalid control parameter samples with low fitness are removed, and new, highly reliable samples are added to ensure the database's effectiveness. The quality of the data; and the historical memory database for updating the crossover probability CR and the mutation factor F. and Simultaneously, the control parameters are dynamically optimized based on historical iteration data to improve the convergence performance of the control method and update the current globally optimal control parameter solution. .

[0027] The historical memory bank MCR and MP update method for crossover probability CR and mutation factor F are as follows: Set up a historical memory with H entries, the value of index k is from 1 to H, indicating the position of the element to be updated in the historical memory, k is initialized to 1, and after each iteration, k=k+1. When k>H, the value of k is reset to 1; and The updates are calculated using both weighted average and weighted Lehmer mean to ensure the rationality and effectiveness of the control parameter updates.

[0028] Furthermore, it is determined whether the global optimal control parameter solution has been updated. If the global optimal control parameter solution has not been updated (i.e., the optimal control effect has stagnated), then a union set is selected from the database DB. (Global optimal control parameter solution) With database A local radial basis function surrogate model is constructed using the sample closest to the set of individuals in the set (based on Euclidean distance). For globally optimal control parameter solutions With database For each individual, a random dimension reliable full cross operation is performed, and the control parameters of the two are exchanged one dimension at a time without repetition, generating multiple candidate control parameter sets; the cross dimension j is not repeatedly selected to ensure that each dimension participates in the cross.

[0029] Furthermore, the random dimension reliable full cross-validation is activated when the globally optimal control parameter solution stagnates. It performs dimension-wise, non-repeating cross-validation between high-confidence potential improved control parameter samples and the globally optimal control parameter solution, generating multiple candidate control parameters. These parameters are then processed through a local surrogate model. Candidate control parameters are pre-screened, and those with the best predicted fitness are selected for real-world evaluation. This effectively breaks through local optima traps and improves the global search capability and control performance of the control method. During the crossover process, each dimension crossover is completed, utilizing... Predict the fitness of candidate control parameters, update the current optimal candidate control parameters, and ensure the efficiency of the crossover process.

[0030] Furthermore, a local radial basis function surrogate model is used. Predict the photovoltaic local consumption fitness of all candidate control parameters after crossover, select the candidate control parameter with the best predicted fitness, perform the actual fitness calculation, archive the candidate control parameter and its actual fitness value to the database DB, and update the current globally optimal control parameter solution. Then, return to iterate and determine whether the termination condition for the maximum fitness evaluation count is met; if the global optimal control parameter solution has been updated, return directly to iterate and determine whether the termination condition for the maximum fitness evaluation count is met, until the termination condition is met.

[0031] Furthermore, the control method strictly meets the constraints for safe operation of the distribution network, specifically including: upper and lower limits of node voltage safety. (in Let be the voltage of the i-th node in the t-th time period. , These constraints include lower and upper limits for node voltage, inverter reactive power regulation range constraints, line capacity limits, power factor limits, and three-phase voltage imbalance limits. These constraints ensure that the distribution network does not experience voltage exceedances, line overloads, or unqualified power factors during control, guaranteeing the safe and stable operation of the distribution network. These constraints also limit the population generation process, ensuring that the generated population is within a specified space. Furthermore, these constraints restrict the triggering conditions and the final generated control commands, ensuring that while achieving optimal power absorption, the issued commands remain within the existing State Grid standards.

[0032] Furthermore, the trigger conditions for iterative optimization of control parameters are set: when the intelligent fusion terminal and power quality monitoring equipment collect the voltage of any node in the distribution network in real time and meet the requirements... (in To preset the warning threshold (typically 0.05kV, set according to distribution network operation specifications), when a voltage is deemed "too high," this control method is immediately triggered to initiate parameter iterative optimization. When the node voltage is monitored to have fallen back to a safe range (…), the control method continues to optimize the parameters iteratively. After that, the control method stops running and the equipment returns to normal operation mode; if the voltage is detected to be too high again, the above control method process is repeated to achieve dynamic coordinated control.

[0033] Secondly, the present invention also provides a photovoltaic on-site consumption collaborative control system based on an intelligent fusion terminal, used to implement the above control method, comprising:

[0034] The first module is used to initialize control parameters, generate initial samples, obtain the photovoltaic local consumption power adaptability corresponding to each sample, and build a sample database DB.

[0035] The second module is used to sort the samples in the database DB according to their quality, construct the parent iterative population, and perform crossover and mutation to generate the offspring population.

[0036] The third module is used to construct multiple radial basis function surrogate models to predict the photovoltaic on-site consumption fitness of each individual in the offspring population, and obtain the global optimal control parameter solution.

[0037] The fourth module is used to determine whether the optimal control parameter solution has been updated. If it has not been updated, a reliable full crossover operation is performed. If it has been updated, the module iteratively determines whether the termination condition of the maximum fitness evaluation count is met, and then issues control commands with the optimal control parameter solution to achieve iterative optimization of control parameters.

[0038] The beneficial effects of the present invention are as follows:

[0039] The photovoltaic local consumption collaborative control method of the present invention can accurately maximize the local consumption of photovoltaic power: the control method takes the maximization of the total active power of local photovoltaic power consumption as the core objective, and through iterative optimization of control parameters, it enables the electricity generated by distributed photovoltaic power to be directly consumed in the local distribution network as much as possible, thereby reducing the curtailment rate, improving the utilization rate of photovoltaic power, and avoiding voltage over-limit problems caused by photovoltaic power overflow to the grid.

[0040] The photovoltaic local consumption collaborative control method of the present invention is adapted to edge computing power constraints: the control method is lightweight and significantly reduces the number of real fitness assessments through strategies such as multi-model proxy prediction and hierarchical sampling. It is adapted to the edge computing resource constraints of new intelligent fusion terminals, realizes local closed-loop execution of control commands, does not rely on cloud computing power, and has a fast response speed.

[0041] The photovoltaic local consumption coordinated control method of the present invention has high control accuracy and strong robustness: it adopts multiple radial basis function surrogate models for coordinated prediction and combines a dual-criteria hierarchical sampling strategy to improve the prediction accuracy and reliability of control parameters, avoid ineffective control caused by prediction deviation, and ensure stable control effect.

[0042] The photovoltaic local consumption collaborative control method of the present invention breaks through the local optimum trap: by performing a reliable full cross operation with random dimensions, high-quality candidate control parameters are generated when the global optimal control parameter solution stagnates, effectively breaking through the local optimum, improving the global search capability of the control method, and ensuring that the optimal control parameter solution that maximizes the local consumption of photovoltaic power is found.

[0043] The photovoltaic local consumption coordinated control method of the present invention ensures the safe and stable operation of the distribution network: the control method strictly follows the safety operation constraints of the distribution network, and through precise optimization of control parameters, ensures that the node voltage does not exceed the limit, the line is not overloaded, and the transformer operates safely, thereby achieving coordinated protection of photovoltaic local consumption and distribution network safety operation.

[0044] The photovoltaic local consumption collaborative control method of the present invention has strong engineering applicability: the control parameters are all actual operating parameters of adjustable equipment in the distribution network, and the control commands can be directly sent to each device through the new intelligent integrated terminal without the need for additional hardware. The control strategy is simple, feasible, and adaptable to the actual application needs of low-voltage distribution area projects. Attached Figure Description

[0045] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0046] Figure 1 This is a flowchart of a photovoltaic on-site consumption collaborative control method based on an intelligent fusion terminal according to one or more embodiments of the present invention;

[0047] Figure 2 This is a framework diagram of a photovoltaic on-site consumption collaborative control system based on an intelligent fusion terminal according to one or more embodiments of the present invention. Detailed Implementation

[0048] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0049] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, unless otherwise expressly indicated by the invention, the singular form is intended to include the plural form as well. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0050] As described in the background section, there are shortcomings in the existing technology. In order to solve the above-mentioned technical problems, this invention proposes a method and system for coordinated control of on-site photovoltaic power consumption based on intelligent fusion terminals.

[0051] Example 1:

[0052] In a typical embodiment of the present invention, such as Figure 1 As shown, a collaborative control method for local photovoltaic consumption based on intelligent fusion terminals is proposed. This embodiment is based on the edge computing scenario of intelligent fusion terminals and is aimed at low-voltage distribution transformer areas with high penetration of distributed photovoltaic access, so as to achieve collaborative control of maximizing local photovoltaic consumption and safe and stable operation of distribution network.

[0053] The specific implementation steps of this method are as follows:

[0054] Step 1: Initialize control parameters;

[0055] The problem dimension d (corresponding to the number of control parameters, i.e., the total dimension of photovoltaic inverter reactive power output, energy storage charging and discharging power, capacitor switching status, and voltage regulation commands; in this embodiment, d=8), population size N (set to 80 based on edge computing power), crossover probability CR (initially set to 0.7), mutation factor F (initially set to 0.5), and its historical memory bank are set. and (H=10, H is the number of historical memory entries, index k=1), set the maximum fitness evaluation count MaxFES (set to 150 times according to edge computing power limit) to ensure that the control parameters are adapted to the operation capability of the new intelligent fusion terminal.

[0056] Step 2: Sample library initialization;

[0057] Initial samples were generated using a Latin hypercube experimental design, with a sample size of 40. Each sample corresponds to a complete distribution network control strategy (i.e., a set of control parameter values). The photovoltaic local consumption power fitness corresponding to each sample was obtained through distribution network power flow calculation (using PSCAD / EMTDC simulation software). The samples and their fitness values ​​were stored in the real evaluation historical sample database DB to complete the initialization and calibration of the sample database, ensuring that the samples cover the entire operating range of the control parameters.

[0058] Step 3: Determine the termination condition;

[0059] The real-time statistical algorithm counts the number of actual fitness evaluations performed. If the maximum fitness evaluation count (MaxFES, 150 times) has been reached, the sample with the highest fitness in the local photovoltaic consumption database (DB) is output as the optimal control parameter solution. If the maximum fitness evaluation count has not been reached, the algorithm proceeds to the next iteration.

[0060] Step 4: Parent population construction;

[0061] All samples in the database DB are sorted from best to worst based on their fitness for local photovoltaic power consumption, and the top 80 samples are selected as the parent iteration population. This ensures the quality of the parent population; at the same time, 80 second-best samples are selected in order of quality and archived in database A, retaining high-quality control parameter samples to provide support for subsequent iterative control.

[0062] Step 5: Mutation and crossover operations;

[0063] Perform a DE / current-to-pbest / 1 mutation operation on the parent population to generate the experimental population. The mutation formula is:

[0064] ;

[0065] in, Indicates the first generation of the parent population A random elite individual among elite individuals. Each parent individual here... Configure specific F is the mutation factor, and each parent individual is configured with a specific F. i,G p r1,G p r2,G Let i and y represent the i-th individual in the G-th generation population and two randomly selected individuals, respectively.

[0066] Then, a binomial crossover operation is performed on the parent population and its experimental population to produce the offspring population. The crossover formula is as follows:

[0067] .

[0068] Step 6: Multi-agent model construction and prediction;

[0069] Four training sample sets with different distributions are extracted from the database DB: the first set consists of all samples in DB, which is used to train the global proxy model. The second group is the first one in the database DB. The optimal sample and its vicinity Samples are used to train the local agent model. The third group consists of the top-performing samples in the database. Training samples are used to train the elite agent model. The fourth group is The most recently evaluated samples from the database are used to train and update the agent model. After the four radial basis function surrogate models are trained in parallel, the photovoltaic on-site consumption fitness of each individual in the offspring population is predicted, and the predicted mean for each individual is calculated. and prediction variance .

[0070] ;

[0071] .

[0072] Step 7: Hierarchical sampling filtering;

[0073] Based on the predicted mean of the four proxy models and prediction variance Calculate the improvement probability of each offspring relative to its parent. ,

[0074] ;

[0075] in, Represents i parent individuals True adaptability.

[0076] Screening Improvement Probability Individuals with a value greater than 0.95 are stored in the database. This forms a highly reliable sample set of potential improved control parameters; simultaneously, it calculates the solution of each offspring relative to the current globally optimal control parameters. Improvement probability ,

[0077] ;

[0078] in, This represents the current globally optimal solution.

[0079] Select the offspring with the highest probability of improvement. The true fitness assessment is performed through power flow calculation in the distribution network, and the individual and its true fitness value are determined. Archived to database DB.

[0080] Step 8: Update parameters and optimal solution;

[0081] Update database ,make sure The offspring individuals, after actual calculation, are superior to the parents, so the new high-confidence samples selected in step 7 are adjusted; at the same time, the historical memory databases of crossover probability CR and mutation factor F are updated. and , The weighted average is used for calculation, and the formula is as follows:

[0082] ;

[0083] ;

[0084] The weighted Lehmer mean is used for calculation, and the formula is as follows:

[0085] ;

[0086] .

[0087] Among them, S CR,k S F,k Let represent the set of "successful crossover rate CR" and "successful scaling factor F" stored at the kth position in the historical memory bank, respectively. Only CR and F that have achieved successful evolution in the iteration are stored in the corresponding sets for subsequent iterations to generate CR and F, so that the algorithm is more inclined to CR and F that have performed better in history.

[0088] The SHADE algorithm sets up a history memory with H entries, where H is the number of history memory entries. The index k ranges from 1 to H, representing the position of the element to be updated in the history memory. k is initialized to 1, and after each iteration, k = k + 1. When k > H, the value of k is reset to 1.

[0089] ;

[0090] in, This represents the improvement in fitness of offspring individuals relative to their parents, compared to... and ,like Then update the globally optimal control parameter solution. Record the corresponding control parameters.

[0091] Step 9: Trusted full cross-operation;

[0092] Determine if the global optimal control parameter solution has been updated. If the global optimal control parameter solution has not been updated for six consecutive generations (i.e., the optimal control effect has stagnated), then select the union set from the database DB. (Global optimal control parameter solution) With database A local radial basis function surrogate model is constructed using the 15 closest samples (from a set of individuals) by Euclidean distance. ; Targeting the solution of globally optimal control parameters With database For each individual in the process, a random-dimensional, reliable full crossover operation is performed, exchanging the control parameters of both individuals dimension by dimension without repetition, generating multiple candidate control parameters; a local proxy model is then used. Predict the fitness of all candidate control parameters, select the candidate control parameter with the best predicted fitness, perform a true fitness evaluation, archive the candidate control parameter and its true fitness value to the database DB, and update the current globally optimal control parameter solution. .

[0093] Step 10: Iteration loop;

[0094] Return to step 3, re-evaluate the termination condition, and continue until the maximum fitness evaluation count of 150 is reached. Output the optimal control parameter solution with the highest fitness for local photovoltaic consumption in the database DB, convert it into control commands, and send them to various adjustable devices (photovoltaic inverters, energy storage systems, capacitors, high overload on-load tap-changing distribution transformers, etc.) through the new intelligent integrated terminal for execution, so as to achieve precise control of the voltage of the distribution network nodes and maximize local photovoltaic consumption.

[0095] In this embodiment, when the new intelligent fusion terminal monitors that the voltage at any node of the distribution network meets the requirements... ≥ - ( It is 1.07kV. When the voltage is set to 0.05kV, the above control method is immediately triggered; when the node voltage is detected to drop back to... < Afterwards, the control method stops operating, and the equipment resumes normal operation mode; if the voltage is detected to be too high again, the above control method process is repeated to achieve dynamic coordinated control and ensure the coordinated realization of safe and stable operation of the distribution network and maximization of local photovoltaic consumption.

[0096] Example 2:

[0097] In another typical embodiment of the present invention, a photovoltaic on-site consumption collaborative control system based on an intelligent fusion terminal is proposed, which is used to implement the above control method, and includes:

[0098] The first module is used to initialize control parameters, generate initial samples, obtain the photovoltaic local consumption power adaptability corresponding to each sample, and build a sample database DB.

[0099] The second module is used to sort the samples in the database DB according to their quality, construct the parent iterative population, and perform crossover and mutation to generate the offspring population.

[0100] The third module is used to construct multiple radial basis function surrogate models to predict the photovoltaic on-site consumption fitness of each individual in the offspring population, and obtain the global optimal control parameter solution.

[0101] The fourth module is used to determine whether the optimal control parameter solution has been updated. If it has not been updated, a reliable full crossover operation is performed. If it has been updated, the module iteratively determines whether the termination condition of the maximum fitness evaluation count is met, and then issues control commands with the optimal control parameter solution to achieve iterative optimization of control parameters.

[0102] Specifically, it can be set up as a physical layer, a sensing layer, an optimization layer, and a control layer, with each layer working together to achieve the maximization of local photovoltaic consumption and the coordinated guarantee of safe and stable operation of the distribution network;

[0103] Physical layer: including distributed photovoltaic, energy storage system, flexible load, high overload on-load tap-changing distribution transformer, and new intelligent integrated terminal, used to carry out the operation and control command execution of distribution network, and provide hardware support for the implementation of control methods;

[0104] The perception layer includes an energy metering module, a data acquisition module, an energy quality monitoring module, and a transformer status monitoring module. It is used to collect power distribution network operation data and equipment status data, providing data support for iterative optimization of control parameters and fitness assessment.

[0105] Optimization layer: This is a single-objective, expensive optimization used to maximize the local consumption of photovoltaic power. It is the core module for implementing the above control methods and specifically includes the first to fourth modules mentioned above. It provides the core direction for the iterative optimization of control parameters and ensures that the optimization of control parameters always revolves around maximizing the local consumption of photovoltaic power.

[0106] The control layer includes an inverter power control module, an on-load tap changer control module, a capacitor switching control module, and an edge autonomous regulation module. It is used to receive the optimal control parameter solution, execute the corresponding control commands, realize the coordinated control of multiple devices in the distribution network, and ensure that the control method is implemented.

[0107] Furthermore, the intelligent fusion terminal is used to receive the optimal control parameter solution and send control commands to various adjustable devices (photovoltaic inverters, energy storage systems, capacitors, high overload on-load tap-changing distribution transformers, etc.) to realize the local execution of control commands, adapt to edge computing resource constraints, and ensure the efficient implementation of control methods. At the same time, the intelligent fusion terminal is also used to collect real-time operation data of the distribution network and feed it back to the optimization layer and control layer to realize the dynamic adjustment of control parameters.

[0108] Furthermore, the data collected by the sensing layer includes distribution network node voltage data, line power data, photovoltaic output data, energy storage operation data, transformer status data, and power quality data. All data are used for iterative optimization and fitness assessment of control parameters to ensure the rationality and effectiveness of control parameters and provide accurate data support for maximizing the control of local photovoltaic consumption.

[0109] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A collaborative control method for on-site photovoltaic power consumption based on an intelligent integrated terminal, characterized in that, Includes the following steps: Initialize control parameters, generate initial samples, obtain the photovoltaic local consumption power adaptability corresponding to each sample, and construct sample database DB; The samples in the database DB are sorted according to their quality, a parent iterative population is constructed, and crossover and mutation are performed to generate the offspring population. Multiple radial basis function surrogate models are constructed to predict the local photovoltaic consumption fitness of each individual in the offspring population, and the global optimal control parameter solution is obtained. If the optimal control parameter solution has not been updated, a reliable full crossover operation is performed. If it has been updated, the process iteratively checks whether the termination condition of the maximum fitness evaluation count is met, and then issues control commands with the optimal control parameter solution to achieve iterative optimization of the control parameters.

2. The photovoltaic local consumption coordinated control method as described in claim 1, characterized in that, Control parameters include problem dimension d, population size N, crossover probability CR, mutation factor F, and historical memory bank. and The maximum fitness evaluation count (MaxFes) is set in conjunction with the edge computing power limitations of the intelligent fusion terminal; the historical memory bank is also included. and Initialize to empty; this is used to store the valid values ​​of crossover probability and mutation factor during the iteration process.

3. The photovoltaic on-site consumption coordinated control method as described in claim 1, characterized in that, By using power flow calculations of the distribution network or measured data from intelligent integrated terminals, the local photovoltaic power consumption fitness corresponding to each sample is obtained. The samples and their fitness values ​​are then stored in the sample database DB to complete the initial construction of the sample database.

4. The photovoltaic local consumption coordinated control method as described in claim 1, characterized in that, in After constructing the sample database DB, a check is first performed to determine whether the termination condition of the maximum fitness evaluation count is met. If the maximum fitness evaluation count MaxFes has been reached, the optimal control parameter solution that maximizes the local consumption of photovoltaic power is output from the database DB; otherwise, the subsequent steps are performed.

5. The photovoltaic local consumption coordinated control method as described in claim 1, characterized in that, The steps for constructing multiple radial basis function surrogate models are as follows: Multiple training sample sets with different distributions were extracted from the database DB, and various radial basis function surrogate models were constructed. The multiple models were run in parallel to predict the on-site photovoltaic consumption fitness of each individual in the offspring population.

6. The photovoltaic local consumption coordinated control method as described in claim 1, characterized in that, Based on the predicted mean and predicted variance of various surrogate models, the improvement probability of each offspring relative to its parent is calculated. ;filter Individuals exceeding the set value are archived to the database. This forms a highly reliable sample set of potential improved control parameters; simultaneously, it calculates the solution of each offspring relative to the current globally optimal control parameters. Improvement probability Select the one with the highest probability of improvement. offspring individuals The true photovoltaic grid integration adaptability is calculated using the objective function. The individual and its true fitness value are archived to the database DB.

7. The photovoltaic local consumption coordinated control method as described in claim 6, characterized in that, according to Further update the database Simultaneously update the historical memory database of crossover probability CR and mutation factor F. and Simultaneously, the control parameters are dynamically optimized based on historical iteration data, and the current globally optimal control parameter solution is updated. .

8. The photovoltaic on-site consumption coordinated control method as described in claim 1, characterized in that, If the global optimal control parameter solution has not been updated, then the sample with the closest Euclidean distance to the union set is selected from the database (DB), and a local radial basis function surrogate model is constructed, which is oriented towards the global optimal control parameter solution and the database. For each individual, a random dimension reliable full cross operation is performed, and the control parameters of the two are exchanged one dimension at a time without repetition, generating multiple sets of candidate control parameters; the cross dimension is not repeatedly selected to ensure that each dimension participates in the cross.

9. The photovoltaic local consumption coordinated control method as described in claim 8, characterized in that, The local radial basis function surrogate model is used to predict the photovoltaic local consumption fitness of all candidate control parameters after the crossover. The candidate control parameter with the best predicted fitness is selected, and the true fitness is calculated. The candidate control parameter and its true fitness value are archived to the database DB. After updating the current global optimal control parameter solution, the process returns to iteratively determine whether the termination condition of the maximum fitness evaluation number is met.

10. A photovoltaic on-site consumption collaborative control system based on an intelligent integrated terminal, characterized in that it includes: The first module is used to initialize control parameters, generate initial samples, obtain the photovoltaic local consumption power adaptability corresponding to each sample, and build a sample database DB. The second module is used to sort the samples in the database DB according to their quality, construct the parent iterative population, and perform crossover and mutation to generate the offspring population. The third module is used to construct multiple radial basis function surrogate models to predict the photovoltaic on-site consumption fitness of each individual in the offspring population, and obtain the global optimal control parameter solution. The fourth module is used to determine whether the optimal control parameter solution has been updated. If it has not been updated, a reliable full crossover operation is performed. If it has been updated, the module iteratively determines whether the termination condition of the maximum fitness evaluation count is met, and then issues control commands with the optimal control parameter solution to achieve iterative optimization of control parameters.

Images