Fuzzy control-based three-phase load adaptive commutation optimization method for low-voltage transformer area

Abstract

The invention relates to the technical field of power distribution networks, in particular to a low-voltage transformer area three-phase load adaptive commutation optimization method based on fuzzy control, and the method comprises the steps: obtaining single-phase load and distributed photovoltaic real-time operation data in a low-voltage transformer area; predicting future output data of the distributed photovoltaic system; establishing a three-phase total power model fusing photovoltaic predicted output and load data, and calculating a three-phase unbalance degree in a future time period; evaluating a candidate commutation scheme by adopting a fuzzy control rule, wherein the input variables of the candidate commutation scheme comprise the unbalance degree improvement degree and the commutation action cost; and selecting an optimal scheme according to an evaluation result and executing a commutation operation. According to the invention, the three-phase load balance degree of the transformer area can be effectively improved, the line loss is reduced, and the economy and reliability of power distribution network operation are improved.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of power distribution networks, in particular to a low-voltage transformer area three-phase load adaptive phase switching optimization method based on fuzzy control. BACKGROUND

[0002] In the operation of a low-voltage power distribution network, the problem of three-phase load imbalance has long existed, and especially with the widespread access of distributed photovoltaics, the randomness and intermittency of the output of the distributed photovoltaics further aggravate the degree of three-phase current imbalance. Traditional phase switching control methods mostly rely on historical load data for static optimization and fail to effectively consider the dynamic influence of real-time output fluctuations of distributed photovoltaics on three-phase power distribution, resulting in limited actual adjustment effect. Although existing technologies attempt to introduce phase switching strategies, they generally lack quantitative analysis of photovoltaic output prediction and the influence of uncertainty, and fail to establish a decision-making mechanism that cooperates with the cost of phase switching actions, making it difficult to achieve fine adaptive adjustment. SUMMARY

[0003] In order to solve or at least partially solve the above technical problems, the embodiments of the present application provide a low-voltage transformer area three-phase load adaptive phase switching optimization method based on fuzzy control.

[0004] In a first aspect, the present application provides a low-voltage transformer area three-phase load adaptive phase switching optimization method based on fuzzy control, comprising the following steps:

[0005] S1, acquiring real-time operation data of each single-phase load user and distributed photovoltaic user in a low-voltage transformer area;

[0006] S2, predicting future output of the distributed photovoltaic user according to real-time operation data of the distributed photovoltaic user, to obtain predicted photovoltaic output data;

[0007] S3, establishing a three-phase total power model that integrates the predicted photovoltaic output data and real-time operation data of the single-phase load user, for calculating three-phase imbalance degree in a future period;

[0008] S4, according to the calculation result of the three-phase total power model, using fuzzy control rules to evaluate candidate schemes that need phase switching operation, the input variables of the fuzzy control rules at least including the degree of imbalance improvement caused by phase switching and the cost of phase switching action;

[0009] S5, according to the evaluation result of the fuzzy control rules, selecting the optimal phase switching scheme and performing phase switching operation.

[0010] Optionally, the three-phase total power model in S3 is represented by the following formula:

[0011]

[0012] wherein P k,t denotes the total power of phase k at time t, k = A, B, C; r denotes the total number of load categories; N k,y denotes the number of load users of the y-th category connected to phase k; K y,t denotes the typical load value of the y-th category at time t; s denotes the total number of distributed photovoltaic users; γ k,z denotes the connection state variable of the z-th distributed photovoltaic user to phase k, taking the value of 1 when connected and 0 when not connected; denotes the predicted output data of the z-th distributed photovoltaic user at time t.

[0013] Optionally, the fuzzy control rules in S4 are constructed in the following manner:

[0014] S401, define the fuzzy input variable imbalance improvement rate ΔU, which represents the degree of decrease in predicted three-phase imbalance after candidate phase switching operation;

[0015] S402, define the fuzzy input variable phase switching action cost F, which represents the number of switch actions required for phase switching operation and the risk of user power outage;

[0016] S403, define the fuzzy output variable phase switching priority index R, which represents the recommended priority of the phase switching scheme;

[0017] S404, establish a fuzzy rule base based on the imbalance improvement rate ΔU and the phase switching action cost F, and obtain the evaluation value of the phase switching priority index R through fuzzy reasoning.

[0018] Optionally, the fuzzy reasoning process in S404 includes the following steps:

[0019] S4041, calculate the membership function value of the imbalance improvement rate ΔU, which has the function form:

[0020]

[0021] wherein x1 is the actual imbalance improvement rate calculation value, and a and b are parameters of the membership function;

[0022] S4042, calculate the membership function value of the phase switching action cost F, which has the function form:

[0023]

[0024] wherein x2 is the actual phase switching action cost calculation value, c is the mean parameter, and d is the variance parameter;

[0025] S4043, de-fuzzify the phase switching priority index R by weighted average method:

[0026]

[0027] wherein ω i represents the activation strength of the i-th activated fuzzy rule, which is obtained by taking the minimum of the membership functions of the unbalance degree improvement rate ΔU and the commutation action cost F; μ i represents the barycentric position of the output fuzzy set corresponding to the i-th activated fuzzy rule; n represents the total number of activated fuzzy rules.

[0028] Optionally, the predicted output data of the distributed photovoltaic user in S2 is obtained by the following method:

[0029] S201, establishing a distributed photovoltaic short-term output prediction model considering weather type and historical output characteristics;

[0030] S202, taking the light intensity, ambient temperature in the real-time operation data and the historical output data of the distributed photovoltaic user as input features;

[0031] S203, outputting the predicted photovoltaic output data of multiple time sections in the future by the distributed photovoltaic short-term output prediction model.

[0032] Optionally, the distributed photovoltaic short-term output prediction model in S201 adopts the following improved prediction formula:

[0033]

[0034] wherein η z represents the conversion efficiency of the z-th distributed photovoltaic component; A z represents the effective light receiving area of the z-th distributed photovoltaic component; I t represents the predicted light intensity at t time; α represents the temperature coefficient of the photovoltaic component; T t represents the ambient temperature at t time; T ref represents the reference test temperature of the photovoltaic component; f(WT t ) represents an output correction function based on the weather type WT t , which is used to correct the predicted output value according to different weather conditions.

[0035] Optionally, the distributed photovoltaic short-term output prediction model further includes adaptive compensation of output prediction error, specifically including the following steps:

[0036] S211, real-time monitoring and recording the historical prediction error sequence e z,1 of each distributed photovoltaic user; z,2 ; z, 3...ez,t ;

[0037] S212, establishing a prediction error band model based on time series analysis:

[0038]

[0039] wherein, represents the prediction error band width of the zth distributed photovoltaic user at time t; σ(·) represents the standard deviation calculation function of the historical prediction error sequence; represents the prediction error sequence of the zth user from time t-N to t-1; represents the inverse function of the standard normal distribution; α represents the significance level;

[0040] S213, incorporating the prediction error band width into the three-phase total power model to form a three-phase power interval prediction considering the uncertainty of distributed photovoltaic output:

[0041]

[0042] wherein, and respectively represent the predicted lower limit and upper limit of the total power of the kth phase at time t;

[0043] The imbalance improvement rate ΔU in the fuzzy control rule is robustly evaluated based on the three-phase power interval prediction.

[0044] Optionally, the fuzzy control rule in S4 further includes the following steps when evaluating the candidate commutation scheme:

[0045] S411, establishing a risk assessment model considering the prediction error distribution characteristics:

[0046]

[0047] wherein, ρ represents the overall predictability index of the distributed photovoltaic cluster output; σ z represents the standard deviation of the historical prediction error of the zth distributed photovoltaic user; μ z represents the average value of the historical output of the zth distributed photovoltaic user; S represents the total number of distributed photovoltaic users;

[0048] S412, taking the predictability index ρ as the third input variable of the fuzzy control rule, for adjusting the evaluation weight of the commutation priority index R; when the value of ρ is lower than the index threshold, the confidence in the prediction effect is reduced, and a more conservative commutation strategy is preferred; when the value of ρ is higher than the index threshold, the confidence in the prediction effect is improved, and a more aggressive commutation strategy is preferred.

[0049] Optionally, the method further comprises:

[0050] S421, record the actual effect data of the historical commutation operation, including the deviation of the actual unbalance improvement value and the predicted unbalance improvement value;

[0051] S422, establish a fuzzy rule weight dynamic adjustment strategy based on reinforcement learning, when the actual improvement effect is continuously better than the predicted effect, the triggering weight of the corresponding fuzzy rule is enhanced, and when the actual improvement effect is continuously lower than the predicted effect, the triggering weight of the corresponding fuzzy rule is reduced.

[0052] Optionally, the method further comprises:

[0053] S431, collect and analyze abnormal event data in the historical commutation operation process, including switch failure, phase-to-phase short circuit and user complaint events;

[0054] S432, construct a commutation operation risk probability prediction model:

[0055]

[0056] Wherein, P risk represents the expected risk probability of the candidate commutation operation; N fault represents the number of switch failures in the history of the line; N short represents the number of phase-to-phase short circuits in the history of the line; N complaint represents the number of user complaints in the history of the line; N total represents the total number of historical commutation operations of the line; ε1, ε2, ε3 represent weight coefficients;

[0057] S433, the risk probability P risk as the fourth input variable of the fuzzy control rule, when P risk exceeds the preset threshold, the high-risk commutation operation is prohibited to be executed, and a warning information is generated.

[0058] The low-voltage area three-phase load adaptive commutation optimization method based on fuzzy control provided by the application has the following beneficial effects:

[0059] This invention comprehensively assesses the system's operating status by collecting real-time operational data from single-phase load users and distributed photovoltaic (PV) users within a low-voltage distribution area, providing an accurate data foundation for subsequent analysis. By predicting the future output of distributed PV users, it allows for advance understanding of PV power generation trends, providing forward-looking data support for calculating three-phase imbalance. By establishing a three-phase total power model that integrates predicted PV output data and real-time load data, it can more accurately calculate the three-phase imbalance in future periods, overcoming the limitations of traditional methods that ignore the impact of PV or rely solely on historical data. Fuzzy control rules are used to evaluate commutation schemes, comprehensively considering multiple factors such as the degree of imbalance improvement and commutation operation costs, making the decision-making process more comprehensive and rational. By selecting the optimal commutation scheme and executing the operation, dynamic balance regulation of the three-phase load in the distribution area can be achieved, effectively improving power quality. Because it fully considers the impact of distributed PV output, this method maintains good regulation effects even in distribution areas with high PV penetration. By minimizing commutation operation costs, it reduces the number of switching operations and equipment wear, improving the economic efficiency of system operation. The entire method is dynamically optimized based on real-time data, adapting to changes in transformer load and photovoltaic output, thus exhibiting good adaptability and practicality. By improving three-phase imbalance, line losses and transformer losses can be reduced, increasing the utilization efficiency of power distribution equipment. Simultaneously, reducing unnecessary commutation operations also helps improve power supply reliability and user satisfaction. Attached Figure Description

[0060] Figure 1 A schematic flowchart of an adaptive commutation optimization method for three-phase loads in low-voltage distribution areas based on fuzzy control is provided in an embodiment of the present invention.

[0061] Figure 2 This is a schematic diagram of a process for constructing fuzzy control rules according to an embodiment of the present invention;

[0062] Figure 3 This is a schematic diagram illustrating a process for obtaining predicted power output data of distributed photovoltaic users, as provided in an embodiment of the present invention. Detailed Implementation

[0063] In order to make the purposes, technical solutions and advantages of the present application clearer, the following further describes the specific embodiments of the present application with reference to the drawings. It can be understood that the specific embodiments described herein are only used to explain the present application, but not to limit the present application. In addition, it should be noted that, for the convenience of description, only the parts related to the present application are shown in the drawings, but not all the contents. Before discussing the exemplary embodiments in more detail, it should be mentioned that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although the flowcharts describe the operations (or steps) as sequential processes, many of the operations can be implemented in parallel, concurrently or simultaneously. In addition, the order of the operations can be rearranged. The processes can be terminated when the operations are completed, but can also have additional steps not included in the drawings. The processes can correspond to methods, functions, procedures, subroutines, etc.

[0064] The technical solutions in the embodiments of the present application will be described clearly below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art belong to the scope of protection of the present application.

[0065] Figure 1 A flowchart of a low-voltage area three-phase load adaptive phase switching optimization method based on fuzzy control provided by the embodiment of the present application, the method comprising the following steps:

[0066] S1, obtaining real-time running data of each single-phase load user and distributed photovoltaic user in the low-voltage area;

[0067] S2, predicting the future output of the distributed photovoltaic user according to the real-time running data of the distributed photovoltaic user, obtaining predicted photovoltaic output data;

[0068] S3, establishing a three-phase total power model integrating the predicted photovoltaic output data and the real-time running data of the single-phase load user, for calculating the three-phase unbalance degree in the future period;

[0069] S4, according to the calculation result of the three-phase total power model, using fuzzy control rules to evaluate the candidate scheme requiring phase switching operation, the input variables of the fuzzy control rules at least including the unbalance degree improvement degree generated by phase switching and the phase switching action cost;

[0070] S5, according to the evaluation result of the fuzzy control rules, selecting the optimal phase switching scheme and executing the phase switching operation.

[0071] The implementation of the present scheme first starts from the comprehensive collection of low-voltage area operation data. Through the installation of a comprehensive monitoring unit on the low-voltage side of the distribution transformer, real-time operation data of all single-phase load users and distributed photovoltaic users are synchronously collected at fixed time intervals. These data mainly include voltage, current and active power data of each user access point. The data collection unit uploads the real-time operation data to the area intelligent management terminal through power line carrier communication or wireless communication. The intelligent management terminal preprocesses the received raw data, including data verification, outlier rejection and missing data interpolation, to ensure the integrity and reliability of the data used in subsequent analysis.

[0072] After completing data collection and preprocessing, special processing is performed for the characteristics of distributed photovoltaic users. Due to the influence of factors such as light intensity and environmental temperature, photovoltaic output has obvious intermittent and fluctuating characteristics, and the output of future time periods needs to be predicted. The prediction process uses a method based on historical data analysis and weather information fusion, which generates predicted photovoltaic output data for multiple time sections in the future by analyzing the historical output pattern of photovoltaic users and combining short-term weather forecast information. These prediction data provide important input for subsequent three-phase imbalance analysis.

[0073] Based on the collected real-time load data and predicted photovoltaic output data, a three-phase total power calculation model is constructed. This model calculates the total power values of phase A, phase B and phase C in the future time period. For each phase, the total power is composed of the sum of the powers of all single-phase load users connected to that phase and the sum of the predicted outputs of all distributed photovoltaic users connected to that phase. Through this separate phase calculation method, the power distribution of each phase in the future time period can be accurately reflected, providing a basis for three-phase imbalance degree calculation.

[0074] After completing the three-phase total power calculation, the three-phase imbalance of the system is further analyzed. The imbalance degree calculation uses the general definition method, which calculates the deviation degree of each phase power from the average power based on the three-phase total power values. Through this calculation method, the imbalance degree of the system in the three-phase power distribution can be quantitatively evaluated, providing a basis for subsequent phase change decision.

[0075] Based on the calculation results of the three-phase imbalance degree, a fuzzy control method is used to evaluate the possible phase change schemes. The evaluation process considers two main factors: one is the improvement degree of the three-phase imbalance degree after the implementation of the phase change scheme, i.e. the technical effect of the scheme; the other is the cost required for implementing the phase change scheme, including the number of switch operations and the possible impact on user short-time power outage. By establishing a fuzzy reasoning system, the two factors are comprehensively weighed to generate a priority index for each candidate phase change scheme. The higher the priority index, the better the comprehensive performance of the scheme in terms of technical effect and implementation cost.

[0076] According to the fuzzy evaluation results, the scheme with the highest priority index is selected from all candidate schemes as the final execution scheme. The phase switching operation is performed through an intelligent phase switching switch, which can realize the switching of user phase without interrupting power supply. During the execution process, the system monitors the actual effect after the phase switching operation in real time and feeds back the monitoring results to the intelligent management terminal. This feedback mechanism enables the system to continuously learn and optimize the phase switching decision, thereby improving the accuracy of subsequent decisions.

[0077] The entire implementation process forms a continuously optimized circulation system. The system regularly collects the latest operation data, updates the photovoltaic output prediction, recalculates the three-phase imbalance degree, and evaluates new phase switching schemes. Through this dynamic adjustment method, the system can adapt to changes in load and photovoltaic output, and always maintain the best balance state of three-phase load. The implementation of this method does not require modification of existing power distribution equipment, but only needs to use existing data acquisition devices and intelligent phase switching switches, and has good practicability and economy.

[0078] Through the above implementation mode, the three-phase imbalance condition of the low-voltage transformer area can be effectively improved, the power loss and equipment overload risk caused by unbalanced operation can be reduced, and the utilization efficiency and power supply reliability of the power distribution equipment can be improved. At the same time, since the evaluation method considering the technical effect and implementation cost is adopted, unnecessary phase switching operation is avoided, the service life of the equipment is prolonged, and the maintenance cost is reduced.

[0079] In some embodiments, the three-phase total power model in S3 is represented by the following formula:

[0080]

[0081] where P k,t represents the total power of phase k at time t, k = A, B, C; r represents the total number of load categories; N k,y represents the number of users of the yth category of load connected to phase k; K y,t represents the typical load value of the yth category of load at time t; s represents the total number of distributed photovoltaic users; γ k,z represents the connection state variable of the zth distributed photovoltaic user to phase k, taking the value of 1 when connected and 0 when not connected; represents the predicted output data of the zth distributed photovoltaic user at time t.

[0082] In the specific implementation of the three-phase total power model, the calculation timing of the model and the source of the input data need to be determined first. The model starts calculation at a fixed time period, usually with the same interval time as data collection. The input data required for calculation includes two parts: one is the preprocessed real-time load data, and the other is the predicted output data of distributed photovoltaic.

[0083] For load data processing, the system first identifies the category of each single-phase load user. Load classification is based on the user's historical electricity consumption characteristics and is achieved using cluster analysis. Each category of user has a typical load curve, which reflects the electricity consumption characteristics of this type of user at different times of the day. The system maintains a classification database, recording the typical load values ​​of each category at different times. When calculating the total power of a phase, it is necessary to count the number of users of each category connected to that phase, multiply the number by the typical load value of the corresponding category, and finally sum them up to obtain the total load power of that phase.

[0084] Processing distributed photovoltaic (PV) data is more complex. Each distributed PV user requires individual processing because their installed capacity, azimuth, and tilt angle may differ. The system obtains each user's connection phase information by querying PV user profiles. For PV users connected to a specific phase, their predicted power output data is directly included in the total power of that phase. The predicted power output data comes from a dedicated PV power prediction module, which comprehensively considers factors such as weather conditions, seasonal variations, and historical power output patterns.

[0085] When calculating the total power of a specific phase, a weighted summation method is used, as shown in the formula above. The power contribution values ​​of each type of load and each photovoltaic user are calculated separately and then added together. The calculation of the load portion requires the use of classification statistics, while the photovoltaic portion needs to be processed for each user individually. This phase-by-phase calculation method can accurately reflect the power composition structure of each phase, providing detailed data support for subsequent imbalance analysis.

[0086] After the calculations are completed, the system generates total power curves for each phase at different time points. These curves visually demonstrate the power variation trends of each phase over a future period, providing an important reference for commutation decisions. By comparing the differences in the power curves of each phase, the three-phase imbalance problem in the system and its severity can be identified.

[0087] The implementation of this model relies on accurate user profile data and reliable prediction algorithms. The system needs to periodically update user access phase information to ensure the accuracy of the model input. Simultaneously, by comparing with on-site monitoring data, the prediction model is continuously calibrated and optimized to improve the reliability of the calculation results. This implementation method ensures both accuracy and practicality, meeting the needs of real-world engineering applications.

[0088] Figure 2 This is a schematic diagram illustrating a process for constructing fuzzy control rules according to an embodiment of the present invention. In some implementations, the fuzzy control rules in step S4 are constructed in the following manner:

[0089] S401. Define the fuzzy input variable imbalance improvement rate ΔU, which characterizes the degree of reduction in the predicted three-phase imbalance after the candidate commutation operation.

[0090] S402, define the fuzzy input variable commutation action cost F, which represents the number of switch actions required for commutation operation and the risk of user power failure;

[0091] S403, define the fuzzy output variable commutation priority index R, which represents the recommended priority of the commutation scheme;

[0092] S404, establish a fuzzy rule base based on the unbalance improvement rate ΔU and the commutation action cost F, and obtain the evaluation value of the commutation priority index R through fuzzy reasoning.

[0093] In the process of constructing fuzzy control rules, the input and output variables of the system need to be clearly defined first. The input variables include the unbalance improvement rate and the commutation action cost, and the output variable is the commutation priority index. The definition of these variables needs to be reasonably set in combination with actual operation data and system characteristics.

[0094] The calculation of unbalance improvement rate is based on the change of three-phase unbalance before and after commutation. The system compares the unbalance values before and after the implementation of the commutation scheme to calculate the specific improvement degree. This index reflects the contribution of the commutation scheme in terms of technical effect, and the higher the improvement rate, the better the effect of the scheme on improving three-phase balance.

[0095] The evaluation of commutation action cost needs to consider multiple actual factors. Among them, the number of switches that need to be operated, the range of users that may be affected during operation, and the wear and tear of the equipment itself. The system will quantify the execution cost of each candidate scheme according to historical operation data and equipment state information. The lower the cost value, the less resource investment required to implement the scheme.

[0096] The commutation priority index, as the output variable, comprehensively reflects the overall priority of the scheme. The calculation of this index needs to consider both technical effect and implementation cost, and through certain rules, the two input variables are mapped to the final priority index value.

[0097] After defining the variables, appropriate fuzzy rule base needs to be established. The rule base contains a series of conditional statements that describe the corresponding output results under different input conditions. These rules are based on expert experience and historical operation data, and can reflect the actual needs and control goals of system operation.

[0098] The establishment process of the rule base needs to fully consider various possible input combinations. For different value combinations of unbalance improvement rate and commutation action cost, there should be corresponding rules to guide them. The setting of rules should ensure that all possible operating conditions are covered, and the system can make reasonable decisions in all situations.

[0099] In the implementation process, the system will call these rules to evaluate the candidate scheme in real time. The evaluation result is output in the form of commutation priority index, which provides the basis for subsequent scheme selection. In this way, the system can realize intelligent decision-making, taking into account the implementation cost while ensuring the technical effect, and achieving the optimal comprehensive operation effect.

[0100] In some embodiments, the fuzzy inference process in S404 includes the following steps:

[0101] S4041, the membership function value of the imbalance improvement rate ΔU is calculated, and the function form is:

[0102]

[0103] Wherein, x1 is the actual imbalance improvement rate calculation value, a and b are the parameters of the membership function;

[0104] S4042, the membership function value of the commutation action cost F is calculated, and the function form is:

[0105]

[0106] Wherein, x2 is the actual commutation action cost calculation value, c is the mean parameter, and d is the variance parameter;

[0107] S4043, the commutation priority index R is calculated by the weighted average method:

[0108]

[0109] Wherein, ω i represents the activation strength of the i-th activated fuzzy rule, which is obtained by taking the minimum operation of the membership functions of the imbalance improvement rate ΔU and the commutation action cost F; μ i represents the barycenter position of the output fuzzy set corresponding to the i-th activated fuzzy rule; n represents the total number of activated fuzzy rules.

[0110] In the specific implementation process of fuzzy inference, the system first obtains the actual calculation values of the two input variables. For the imbalance improvement rate, the actual calculation value comes from the output result of the commutation scheme evaluation module; for the commutation action cost, the actual calculation value comes from the calculation result of the cost evaluation module. These actual calculation values are input data for the fuzzy inference process.

[0111] Next, the system calculates the function value of the membership function corresponding to each input variable. For the membership function value of the imbalance improvement rate, a piecewise linear function μ ΔU(x1) is calculated. The function contains three key parameters: lower threshold a, transition threshold b and actual upper limit x1. When the actual calculation value x1 is less than or equal to the lower threshold, the function output is zero; when the actual calculation value x1 is between the lower threshold and the transition threshold, the function output increases linearly; when the actual calculation value x1 is greater than the transition threshold, the function output remains at the maximum value. This function form can reasonably reflect the gradual change process from invalid to effective.

[0112] For the commutation action cost, a Gaussian function is adopted F (x2) is calculated. The function contains two key parameters: mean parameter c (center value) and variance parameter d (width parameter). The mean parameter c represents the most ideal cost value, and the variance parameter d controls the decay speed of the function. The closer the actual calculation value is to the center value, the larger the function output is; the farther the distance is, the smaller the function output is. This function form can reflect the reasonable range of cost control.

[0113] After completing the membership calculation, the system starts to execute fuzzy reasoning. The reasoning process is based on the pre-established rule base, and each rule defines the corresponding relationship between input conditions and output results. The system will traverse all rules to find those rules whose input conditions match the current actual situation. For each activated rule, the system will calculate its activation strength, which is the smaller one of the membership values of the input conditions corresponding to the rule.

[0114] Finally, the defuzzification calculation is performed. The system adopts the weighted average method to comprehensively process the output values of all activated rules. The output value of each rule is weighted by its activation strength, and the final commutation priority index accurate value is calculated by weighted average. This accurate value is a specific numerical value, which directly reflects the priority of the commutation scheme, and the higher the value is, the more preferred the scheme is.

[0115] The whole reasoning process can quickly process a large number of candidate schemes to provide accurate decision basis for subsequent scheme selection. Through this implementation, the scientificity of decision is guaranteed, and the running efficiency of the system is improved.

[0116] Figure 3 A process schematic diagram for obtaining predicted output data of a distributed photovoltaic user is provided for the embodiments of the present application. In some embodiments, the predicted output data of the distributed photovoltaic user in S2 is obtained by the following method:

[0117] S201, a distributed photovoltaic short-term output prediction model considering weather type and historical output characteristics is established;

[0118] S202, the light intensity, environmental temperature and historical output data of the distributed photovoltaic user in the real-time running data are taken as input features;

[0119] S203, output predicted photovoltaic output data of multiple time sections in the future by the distributed photovoltaic short-term output prediction model.

[0120] In the process of obtaining distributed photovoltaic prediction output data, a prediction model suitable for local characteristics needs to be established first. The model fully considers the change characteristics of weather types and historical output rules, and can adapt to prediction needs under different meteorological conditions. The input data of the prediction model mainly includes real-time collected illumination intensity, environmental temperature, and historical output records of photovoltaic users. After preprocessing, these data form a standardized feature vector.

[0121] During the model running process, first, the input data is analyzed for features, and feature indicators with strong correlation with photovoltaic output are extracted. These feature indicators include instantaneous illumination intensity, environmental temperature change trend, historical contemporaneous output data, etc. Through comprehensive analysis of these feature indicators, the model can capture the change rules and influencing factors of photovoltaic output.

[0122] Based on feature analysis, the model performs output prediction calculation. The prediction calculation adopts a multi-time section output mode, which can simultaneously generate prediction values at multiple future time points. The prediction of each time section considers the continuity and correlation in time series, ensuring the rationality and smoothness of the prediction results in the time dimension.

[0123] The prediction process also takes into account the influence of weather type conversion. When the weather condition changes, the model adjusts the prediction algorithm parameters to adapt to the new meteorological conditions. This adaptive mechanism can improve prediction accuracy, especially in cases of weather changes such as turning cloudy or sunny, while still maintaining good prediction results.

[0124] After prediction is completed, the system outputs prediction data in a standard format. These data are arranged in time series, and each time point contains a predicted output value and a corresponding confidence index. The prediction data is transmitted to the three-phase total power calculation module through a data interface, providing input for subsequent power balance calculation.

[0125] The entire prediction process is executed periodically, usually with the same period as data collection. The system saves historical prediction records and actual output data, and continuously optimizes prediction algorithm parameters through comparative analysis. This continuous improvement mechanism enables the prediction model to gradually adapt to local climate characteristics and photovoltaic device characteristics, improving long-term prediction accuracy.

[0126] In some embodiments, the distributed photovoltaic short-term output prediction model in S201 uses the following improved prediction formula:

[0127]

[0128] where ηz ηzdenotes the conversion efficiency of the z-th distributed photovoltaic module; A z Azdenotes the effective light-receiving area of the z-th distributed photovoltaic module; I t Itdenotes the predicted light intensity at time t; a denotes the temperature coefficient of the photovoltaic module; T t Tdenotes the ambient temperature at time t; T ref Tdenotes the reference test temperature of the photovoltaic module; f(WT t ) denotes the output correction function based on the weather type WT t , which is used to correct the predicted output value according to different weather conditions.

[0129] In the implementation of the distributed photovoltaic short-term output prediction model, it is necessary to accurately obtain the characteristic parameters of the photovoltaic module first. These parameters include the inherent characteristics of the photovoltaic module such as conversion efficiency, effective light-receiving area and temperature coefficient. The conversion efficiency reflects the ability of the photovoltaic module to convert light energy into electrical energy, the effective light-receiving area determines the total amount of received light radiation, and the temperature coefficient represents the sensitivity of the output power to temperature changes.

[0130] Real-time environmental data, including light intensity and ambient temperature, are required during the prediction process. Light intensity data is obtained through irradiance sensors, and ambient temperature data is collected through temperature sensors. These real-time data and the characteristic parameters of the photovoltaic module are used as inputs to the prediction model.

[0131] When the model is calculated, the theoretical output value is first calculated based on ideal conditions. The theoretical calculation uses physical formulas, taking into account the product relationship between light intensity and effective light-receiving area, and introducing a temperature correction factor. The temperature correction factor is calculated by the difference between the temperature coefficient and the ambient temperature relative to the reference test temperature, and is used to compensate for the impact of temperature changes on the output power.

[0132] The most important improvement is the introduction of a weather type correction function. This function adjusts the theoretical output value according to the actual weather conditions. Different weather types correspond to different correction coefficients, such as sunny, cloudy, rainy and other weather conditions, which use different correction parameters. The determination of the correction coefficient is based on statistical analysis of historical data, establishing the corresponding relationship between weather type and output characteristics.

[0133] During the operation of the model, real-time weather information is obtained through a meteorological data interface. The system identifies the current weather type and calls the corresponding correction function parameters. The correction function adjusts the theoretical output value proportionally, making the prediction results more consistent with the output characteristics under actual weather conditions.

[0134] The prediction result is checked for rationality before output. The system compares the current prediction value with historical data and recent output trends to ensure the rationality of the prediction result. If an abnormal value is found, the system will recalculate the prediction.

[0135] The entire prediction process uses a rolling update method to regularly update the prediction results with the latest data. Through continuous data accumulation and model optimization, the prediction accuracy is gradually improved. This implementation ensures the accuracy of the prediction results and has good practicality, meeting the needs of engineering applications.

[0136] In some embodiments, the distributed photovoltaic short-term output prediction model further includes adaptive compensation of output prediction error, specifically including the following steps:

[0137] S211, real-time monitoring and recording the historical prediction error sequence e of each distributed photovoltaic user z,1 ,e z,2 ,e z, 3...e z,t ;

[0138] S212, establish a prediction error band model based on time series analysis:

[0139]

[0140] wherein, represents the prediction error bandwidth of the zth distributed photovoltaic user at time t; σ(·) represents the standard deviation calculation function of the historical prediction error sequence; represents the prediction error sequence of the zth user from t-N to t-1; represents the inverse function of the standard normal distribution; α represents the significance level;

[0141] S213, incorporate the prediction error bandwidth into the three-phase total power model to form a three-phase power interval prediction considering the uncertainty of distributed photovoltaic output:

[0142]

[0143] wherein,

[0144] and respectively represent the predicted lower limit and upper limit of the total power of the kth phase at time t;

[0145] The imbalance improvement rate ΔU in the fuzzy control rule is robustly evaluated based on the three-phase power interval prediction.

[0146] In the implementation of distributed photovoltaic output prediction error compensation, it is necessary to first establish a monitoring record of historical prediction errors. The system continuously collects actual output data of each photovoltaic user and compares it with the prediction value at the corresponding time point to calculate the prediction error value. These error values are saved in chronological order to form a prediction error sequence e z,1 z,2 z,3 z,t .

[0147] The error sequence analysis uses a sliding time window method to select error data in the recent period as the analysis sample. The window length is set according to the actual situation, and a time span that can reflect the error change characteristics is usually selected. The statistical characteristic values of the error data in the window are calculated, mainly including the distribution range and fluctuation degree of the error.

[0148] The prediction error band calculation uses statistical principles based on the standard deviation and probability distribution characteristics of the error sequence. The standard deviation calculation function quantifies the dispersion of the error sequence, reflecting the stability of the prediction result. The probability distribution parameters are used to determine the confidence range of the error, and the significance level parameters control the width of the confidence interval. The error band width value at a certain confidence level is calculated by the inverse function.

[0149] The error band width represents the uncertainty range of the prediction result, and the upper and lower limits of the possible fluctuation of the prediction value are represented in the form of an interval. The error band width value is superimposed on the original prediction value to form a prediction output interval. This interval reflects the possible range of photovoltaic output considering the prediction error.

[0150] The three-phase power interval prediction model is extended based on this. The model considers both the load prediction value and the photovoltaic prediction output interval, and calculates the upper and lower limit values of the total power of each phase through interval operation rules. In the calculation process, the load part uses a fixed value, and the photovoltaic part uses an interval value, and finally the prediction range of the three-phase power is obtained.

[0151] The interval prediction result provides more comprehensive operation information, not only giving the expected value of the power, but also giving the possible fluctuation range. This range information provides a risk assessment basis for subsequent phase change decisions, enabling the system to make more robust decisions considering uncertainty.

[0152] The system regularly updates the error sequence data and dynamically adjusts the error band width calculation parameters. By continuously monitoring the actual distribution of prediction errors, the error band model can adapt to changes in prediction accuracy and maintain the effectiveness of compensation. This implementation improves the system's ability to respond to prediction uncertainty and enhances the reliability of operation decisions.

[0153] ​​​It is found in the implementation process that the prediction error band model used in the above embodiments has certain limitations. The original model is based on the standard deviation of the historical error sequence and the normal distribution assumption, but in actual application, the photovoltaic output prediction error often presents asymmetric distribution characteristics, and the error distribution has obvious dynamic characteristics with time. Therefore, the error band calculation model can be improved as follows:

[0154] A dynamic error band model based on quantile regression is established:

[0155]

[0156] wherein, and denote the upper and lower limits of the prediction error band of the zth photovoltaic user at time t; denotes the quantile function, τ upper and τ lower denote the upper and lower quantile levels; denotes the historical prediction error sequence.

[0157] The improved embodiment first uses a sliding time window method to obtain prediction error data in a recent period of time. The window length is dynamically adjusted according to the seasonal characteristics of the photovoltaic output, a shorter window is used in summer to adapt to the rapid change of light conditions, and a longer window is used in winter to maintain statistical stability. The error data in the window is arranged in ascending order, and the upper and lower boundaries of the error distribution are directly obtained by quantile calculation.

[0158] The quantile level is dynamically set according to the prediction time scale and the weather type. For short-term prediction, a narrower quantile interval is used to reflect the higher prediction accuracy requirement; for medium and long-term prediction, a wider quantile interval is used to adapt to greater uncertainty. Under different weather conditions, the system can adjust the quantile level, a narrower interval is used on sunny days, and a wider interval is used on rainy days.

[0159] The error band model outputs an asymmetric prediction interval, which better reflects the actual characteristics of the error distribution. The upper error band focuses on the risk of high power prediction, and the lower error band focuses on the risk of low power prediction. This asymmetric structure can more accurately depict the uncertainty characteristics of photovoltaic output prediction.

[0160] The improved error band is incorporated into the three-phase power interval prediction model to form an asymmetric power prediction interval:

[0161]

[0162] wherein the power lower limit considers the most conservative prediction, and the power upper limit considers the most optimistic prediction.

[0163] This implementation can more accurately reflect the uncertainty range of photovoltaic power output prediction, and provide more reliable risk assessment basis for subsequent phase switching decision. By dynamically adjusting the quantile, the system can adapt to the prediction error characteristics under different seasons and weather conditions, improving the accuracy and practicality of interval prediction. The improved model is especially suitable for handling the common asymmetric error distribution problem in photovoltaic power output prediction, making the risk assessment results more close to the actual operation.

[0164] In some embodiments, the fuzzy control rule in S4 further includes the following steps when evaluating the candidate phase switching scheme:

[0165] S411, a risk assessment model considering the distribution characteristics of prediction error is established:

[0166]

[0167] wherein p represents the overall predictability index of distributed photovoltaic cluster output; σ z represents the standard deviation of the historical prediction error of the zth distributed photovoltaic user; μ z represents the average value of the historical output of the zth distributed photovoltaic user; S represents the total number of distributed photovoltaic users;

[0168] S412, the predictability index p is taken as the third input variable of the fuzzy control rule, which is used to adjust the evaluation weight of the phase switching priority index R; when the p value is lower than the index threshold, the confidence in the prediction effect is reduced, and a more conservative phase switching strategy is preferred; when the p value is higher than the index threshold, the confidence in the prediction effect is improved, and a more aggressive phase switching strategy is preferred.

[0169] In the implementation process of distributed photovoltaic predictability assessment, the system first needs to calculate the individual predictability index of each photovoltaic user. This calculation is based on historical operation data, including the actual output value and the corresponding prediction value of each user. For each user, the system calculates the standard deviation of its historical prediction error, which reflects the fluctuation degree of the prediction result; at the same time, the average value of the historical output is calculated, which reflects the typical power generation level of the user.

[0170] The individual predictability index is characterized by the ratio of the error standard deviation to the average output. The smaller this ratio, the higher the prediction accuracy of the user relative to its power generation scale, and the better the predictability; the larger the ratio, the greater the uncertainty of the prediction result relative to the power generation scale. The system takes the average of this ratio of all photovoltaic users to obtain the overall predictability index p.

[0171] The overall predictability indicator ρ ranges from 0 to 1. When ρ is close to 0, it indicates that the predictability of the entire photovoltaic cluster is high, and the prediction error is relatively small compared to the power generation. When ρ is close to 1, it indicates that the prediction error is relatively large compared to the power generation, and the overall predictability is poor. This indicator comprehensively reflects the influence of meteorological conditions, equipment performance, prediction models, and other factors on prediction accuracy.

[0172] In the application of fuzzy control rules, the overall predictability indicator ρ participates in the decision-making process as the third input variable. The system dynamically adjusts the confidence weight of the prediction result according to the value of ρ. When ρ is low, the system gives higher weight to the prediction result, and tends to adopt an aggressive adjustment strategy based on prediction data; when ρ is high, the system reduces the dependence on the prediction result, and instead adopts a more conservative adjustment strategy.

[0173] This scheme enables the system to select the most suitable control strategy according to the actual accuracy level of photovoltaic prediction. When the prediction accuracy is high, the prediction data plays a guiding role, and when the prediction accuracy is low, a more robust control method is relied on, ensuring the adjustment effect and avoiding control errors caused by prediction errors.

[0174] The system regularly updates the calculation of the predictability indicator, usually with the same update period as the prediction model. By continuously monitoring the trend of the indicator, the system can timely discover changes in prediction accuracy and adjust the parameter settings of the control strategy accordingly. This implementation improves the adaptability of the system to changes in prediction uncertainty and enhances the robustness of operation control.

[0175] In some scenarios, the original formula evaluates predictability by calculating the average of the ratio of the standard deviation of each photovoltaic user's prediction error to the average of the output, but this method does not fully consider the influence of different user installation capacity differences on the overall predictability of the system. In actual application, the prediction accuracy of large-capacity photovoltaic users has a more significant impact on the system, and needs to be given higher weight.

[0176] Therefore, the predictability indicator calculation formula can be improved as follows:

[0177]

[0178] where ∈ z is the weight coefficient of the zth photovoltaic user, determined by the installation capacity C z and the historical maximum output P max,z of the user:

[0179] ∈ z = C z × P max,z

[0180] The improved embodiment first collects the installed capacity data and historical operation data of each photovoltaic user. The installed capacity data is obtained from the user archive, and the historical maximum output is obtained by analyzing the long-term operation record. The system establishes an independent weight coefficient archive for each user and regularly updates and maintains it.

[0181] The weight coefficient calculation comprehensively considers both installed capacity and actual output capacity. Installed capacity reflects the theoretical power generation potential of the user, and historical maximum output represents the actual power generation capacity. The weight coefficient obtained by multiplying these two factors can more comprehensively reflect the influence of the user on the overall predictability of the system.

[0182] In calculating the overall predictability index, the weighted average method is used instead of the original simple arithmetic average. The prediction accuracy index of each user is multiplied by its weight coefficient and then summed, and finally divided by the sum of the weight coefficients. This calculation method makes the prediction accuracy of large-capacity users have a greater impact on the overall index, which is more in line with the actual operation requirements.

[0183] The improved predictability index still ranges from 0 to 1, but has a more explicit physical meaning. When the index value is close to 1, it indicates that the overall predictability of the system is good, especially the prediction accuracy of large-capacity users is high; when the index value is close to 0, it indicates that the prediction accuracy of the system is poor, or the prediction uncertainty of large-capacity users is large.

[0184] The system regularly updates the weight coefficients and predictability index, usually updating the weight coefficients once a month and calculating the predictability index every day. By continuously monitoring the trend of the index, the system can timely discover the systematic changes in prediction accuracy and adjust the control strategy accordingly.

[0185] The improved predictability index is applied to the fuzzy control rules as the third input variable in the decision-making process. The system dynamically adjusts the confidence weight of the prediction result according to the weighted predictability index value. When the index value is high, the system gives higher confidence to the prediction result of large-capacity users; when the index value is low, the system reduces the dependence on the prediction result.

[0186] This improved embodiment can more accurately evaluate the overall predictability level of the system, making the subsequent commutation decision more scientific and reasonable. By considering the influence of user capacity differences, the system can better grasp the prediction accuracy of key users and improve the accuracy and reliability of operation decisions. The improved index is especially suitable for areas with large differences in photovoltaic user capacity, and can more realistically reflect the actual predictability of system operation.

[0187] In some embodiments, the method further comprises:

[0188] S421、record the actual effect data of the historical commutation operation, including the deviation of the actual unbalance improvement value and the predicted unbalance improvement value;

[0189] S422、establish a fuzzy rule weight dynamic adjustment strategy based on reinforcement learning, when the actual improvement effect is consistently better than the predicted effect, increase the triggering weight of the corresponding fuzzy rule, when the actual improvement effect is consistently lower than the predicted effect, decrease the triggering weight of the corresponding fuzzy rule.

[0190] In the implementation process of self-learning optimization, the system first establishes a historical operation database for recording the actual effect data of each commutation operation. The database contains basic information of commutation schemes, predicted improvement effect (predicted unbalance improvement value), actual execution result (actual unbalance improvement value), and operation time, etc. After each commutation operation is completed, the system stores the relevant data into the database to form a complete historical record.

[0191] After the data recording is completed, the system starts the effect evaluation program. This program compares the actual achieved unbalance improvement value with the predicted value and calculates the deviation between them. The deviation calculation uses relative value method, which considers both the absolute difference and the proportional relationship relative to the predicted value. The evaluation results are divided into three levels: actual effect better than prediction, actual effect consistent with prediction, and actual effect lower than prediction.

[0192] Based on the evaluation results, the system adjusts the fuzzy rule weight. The adjustment process uses incremental learning method, which gradually corrects the rule weight according to the effect trend of recent multiple operations. When the actual effect is consistently better than the predicted effect, the weight of the corresponding rule will be appropriately increased; when the actual effect is consistently lower than the predicted effect, the weight of the corresponding rule will be appropriately decreased. The weight adjustment amplitude is determined comprehensively according to the deviation degree and the duration.

[0193] During the adjustment process, the system will consider the effect differences under different operating conditions. The same rule may perform differently under different load levels and different photovoltaic output conditions, so the weight adjustment will be combined with specific operating scenarios. The system establishes the correspondence between operating condition characteristics and rule effect by analyzing the rule performance under different operating conditions in historical data.

[0194] After the weight adjustment is completed, the system will verify the effectiveness of the new weight. Through simulation testing and historical data playback, it checks whether the decision effect under the new weight has been improved. After verification, the new weight takes effect and is applied to the subsequent decision-making process. The entire adjustment process is carried out periodically to ensure that the rule weight can adapt to changes in system operating state in a timely manner.

[0195] In this way, the system can continuously optimize the decision-making quality and improve the success rate and effect stability of commutation operation. With the passage of time, the system accumulates more and more experience, and the decision-making accuracy gradually improves, eventually forming an intelligent decision-making system with adaptive ability.

[0196] In some embodiments, the method further comprises:

[0197] S431, collect and analyze abnormal event data in the historical commutation operation process, including switch failure, phase-to-phase short circuit and user complaint event;

[0198] S432, build a commutation operation risk probability prediction model:

[0199]

[0200] Wherein, P risk represents the expected risk probability of the candidate commutation operation; N fault represents the number of switch failures in the history of the line; N short represents the number of phase-to-phase short circuits in the history of the line; N complaint represents the number of user complaints in the history of the line; N total represents the total number of historical commutation operations of the line; ε1, ε2, ε3 represent weight coefficients;

[0201] S433, the risk probability P risk as the fourth input variable of the fuzzy control rule, when P risk exceeds the preset threshold, the high-risk commutation operation is prohibited, and a warning information is generated.

[0202] In the specific implementation of commutation operation risk assessment, the system first establishes an abnormal event database, which continuously records various abnormal situations occurring in each commutation operation. Abnormal events mainly include three types of switch failure, phase-to-phase short circuit and user complaint, each of which records detailed occurrence time, line identification and event nature. The database is regularly updated and maintained to ensure the integrity and accuracy of the historical data.

[0203] The risk probability calculation adopts a weighted average method, and the number of occurrences of three types of abnormal events in the history of a specific line is counted respectively. The number of switch failures reflects the reliability of the equipment, the number of phase-to-phase short circuits reflects the safety level of the operation, and the number of user complaints reflects the quality of power supply service. These three types of number data and the total number of historical commutation operations of the line form a proportional relationship, forming a basic risk assessment index.

[0204] Weight coefficients are introduced in the calculation process, and different weight coefficients correspond to different types of abnormal events. The setting of weight coefficients is based on the severity of the consequences of the event. Generally, the weight of the phase-to-phase short circuit is higher, the weight of the switch failure is second, and the weight of the user complaint is relatively low. The weight value is determined through expert evaluation and historical data analysis, and is adjusted and optimized regularly according to the actual operation.

[0205] The calculation of the risk probability value integrates the weighted influence of the three types of abnormal events, and the final result is a value between zero and one. The higher the value, the greater the risk of the line during the current commutation operation. The system sets a risk threshold, which triggers the protection mechanism when the calculated value exceeds the threshold.

[0206] The risk probability value is the fourth input variable of the fuzzy control rule and participates in the decision-making process. When evaluating candidate schemes, the system considers three dimensions: technical effect, implementation cost, and risk probability. When the risk probability value of a certain scheme is too high, even if its technical effect and implementation cost indicators are good, the system will reduce the priority of the scheme.

[0207] For schemes with risk probability exceeding the limit, the system excludes them and generates a warning message. The warning message contains detailed risk analysis results, explaining the main risk sources and possible consequences. The operation and maintenance personnel can manually intervene according to the warning message to decide whether to take additional measures or adjust the operation scheme.

[0208] The system regularly updates the risk assessment model parameters and adjusts the weight coefficients and risk threshold according to the latest operation data. Through continuous learning and optimization, the model can accurately reflect the actual operation state of the line equipment, improving the accuracy and timeliness of risk assessment. This implementation effectively reduces the operation risk of commutation and improves the safety and reliability of the system.

[0209] In the implementation process, it is found that the risk assessment model used in the above implementation method calculates the risk probability by weighted average, but the weight coefficient uses a fixed value, which does not fully consider the dynamic change characteristics of different abnormal events with factors such as season and equipment aging. In actual operation, the switch failure rate may increase with the increase of equipment service life, the phase-to-phase short circuit probability may be related to climate conditions, and user complaints often have seasonal characteristics.

[0210] The risk probability calculation formula is improved as follows:

[0211]

[0212] wherein, N fault , N short , and N complaint .

[0213] wherein the dynamic weight coefficient is determined by a time function:

[0214] ε j (t) = ε j,base + α j × f j (t) + β j × g j (age)

[0215] The improved embodiment first establishes the time dimension analysis function of the abnormal event database. The system not only records the type and number of abnormal events, but also records the time stamp, environmental temperature, humidity and other working condition data of each event occurrence, as well as the basic information such as the device operation time. These data provide a basis support for dynamic weight calculation.

[0216] Dynamic weight calculation includes three components: basic weight value ε j,base , time adjustment factor f j (t) and device aging factor g j (age). The basic weight value maintains the basic proportional relationship of the original weight system; the time adjustment factor is based on historical data analysis of the occurrence regularity of various types of abnormal events under different seasons and weather conditions; the device aging factor considers the influence curve of device service life on failure rate.

[0217] The calculation of the time adjustment factor uses the periodic function analysis method, and the adjustment coefficient is α j . The system analyzes the monthly distribution regularity of various types of abnormal events in historical data, and establishes an adjustment coefficient curve with a period of one year. For example, the risk coefficient of interphase short circuit is appropriately increased during the summer high temperature period, and the user complaint risk coefficient is correspondingly increased during the winter heating load concentration period. These adjustment coefficients are obtained through statistical analysis and regression calculation of historical data.

[0218] The calculation of the device aging factor is based on the device reliability theory, and the adjustment coefficient is β j . The system establishes a device aging model for each line, and calculates the risk addition coefficient under the current device state according to the device operation time, maintenance record and average life data of similar devices. The longer the operation time and the worse the maintenance record of the line, the larger the device aging factor value and the higher the risk addition coefficient.

[0219] The improved risk probability value can more accurately reflect the real-time risk situation. The system dynamically adjusts the weight coefficients of various types of abnormalities according to the current time, environmental conditions and device state, so that the risk assessment result is closer to the actual operation situation. The risk threshold is also set as a dynamic value, which automatically adjusts the warning standard according to the overall operating environment.

[0220] This improved implementation enables the risk assessment model to be self-adaptive, adjusting the evaluation parameters according to time changes, environmental conditions and equipment status. By establishing a more refined risk evaluation system, the system can more accurately identify high-risk operations and provide more reliable safety protection for commutation decision-making. The improved model is particularly suitable for distribution areas with complex operating environments and large differences in equipment conditions, effectively improving the safety and reliability of system operation.

[0221] The above are only preferred embodiments of the present application and the technical principles applied. The present application is not limited to the specific embodiments described herein, and various obvious changes, readjustments and substitutions made by those skilled in the art will not deviate from the protection scope of the present application. Therefore, although the present application has been described in more detail through the above embodiments, the present application is not limited to the above embodiments, and can include more other equivalent embodiments without deviating from the concept of the present application, and the scope of the present application is determined by the scope of the claims.

Claims

1. A method for adaptive commutation optimization of three-phase loads in low-voltage distribution areas based on fuzzy control, characterized in that, Includes the following steps: S1. Obtain real-time operating data of each single-phase load user and distributed photovoltaic user in the low-voltage distribution area; S2. Based on the real-time operating data of the distributed photovoltaic users, predict the future output of the distributed photovoltaic users to obtain predicted photovoltaic output data; S3. Establish a three-phase total power model that integrates the predicted photovoltaic output data and the real-time operating data of the single-phase load users, for calculating the three-phase imbalance in future time periods; S4. Based on the calculation results of the three-phase total power model, fuzzy control rules are used to evaluate the candidate schemes that require commutation operation. The input variables of the fuzzy control rules include at least the degree of improvement of the imbalance caused by commutation and the cost of commutation operation. S5. Based on the evaluation results of the fuzzy control rules, select the optimal commutation scheme and execute the commutation operation.

2. The method according to claim 1, characterized in that, The three-phase total power model in S3 is expressed by the following formula: Among them, P k,t Represents the total power of phase k at time t, where k = A, B, C; r represents the total number of load categories; N k,y K represents the number of users of type y who are connected to phase k; y,t γ represents the typical load value of type y load at time t; s represents the total number of distributed photovoltaic users; k,z The connection state variable represents whether the z-th distributed photovoltaic user is connected to phase k. It takes the value 1 when connected and 0 when not connected. This represents the predicted power output data of the z-th distributed photovoltaic user at time t.

3. The method according to claim 1, characterized in that, The fuzzy control rules in S4 are constructed as follows: S401. Define the fuzzy input variable imbalance improvement rate ΔU, which characterizes the degree of reduction in the predicted three-phase imbalance after the candidate commutation operation. S402. Define a fuzzy input variable, commutation cost F, which represents the number of switching actions required for commutation and the risk of power outage for users. S403. Define the fuzzy output variable commutation priority index R, which represents the recommended priority of commutation schemes; S404. Establish a fuzzy rule base based on the imbalance improvement rate ΔU and the commutation action cost F, and obtain the evaluation value of the commutation priority index R through fuzzy inference.

4. The method according to claim 3, characterized in that, The fuzzy inference process in S404 includes the following steps: S4041. Calculate the membership function value of the imbalance improvement rate ΔU, whose functional form is: Where x1 is the calculated value of the actual imbalance improvement rate, and a and b are the parameters of the membership function; S4042. Calculate the membership function value of the commutation operation cost F, the function form of which is: Where x2 is the calculated cost of the actual commutation action, c is the mean parameter, and d is the variance parameter; S4043. The commutation priority index R is calculated using a weighted average method to defuzzify it. Where, ω i The activation intensity of the i-th activated fuzzy rule is represented by μ, which is obtained by taking the smaller of the membership functions of the imbalance improvement rate ΔU and the commutation cost F. i This indicates the centroid position of the output fuzzy set corresponding to the i-th activated fuzzy rule; n represents the total number of activated fuzzy rules.

5. The method according to claim 2, characterized in that, The predicted power output data of distributed photovoltaic users in S2 is obtained through the following methods: S201. Establish a short-term output prediction model for distributed photovoltaic power that takes into account weather type and historical output characteristics; S202, take the light intensity, ambient temperature and historical power output data of the distributed photovoltaic users in the real-time operation data as input features; S203. Output the predicted photovoltaic output data for multiple future time sections through the distributed photovoltaic short-term output prediction model.

6. The method according to claim 5, characterized in that, The distributed photovoltaic short-term output prediction model in S201 adopts the following improved prediction formula: Where, η z A represents the conversion efficiency of the z-th distributed photovoltaic module; z I represents the effective light-receiving area of ​​the z-th distributed photovoltaic module; t The predicted light intensity at time t is represented by α; the temperature coefficient of the photovoltaic module is represented by T. t T represents the ambient temperature at time t; ref Indicates the reference test temperature for photovoltaic modules; f(WT) t ) indicates based on weather type WT t The output correction function is used to correct the predicted output value according to different weather conditions.

7. The method according to claim 6, characterized in that, The distributed photovoltaic short-term output prediction model also includes adaptive compensation for output prediction errors, specifically including the following steps: S211. Monitor and record the historical prediction error sequence of each distributed photovoltaic user in real time. z,1 ,e z,2 ,e z,3 ...e z,t ; S212. Establish a prediction error band model based on time series analysis: in, σ(·) represents the prediction error bandwidth of the z-th distributed photovoltaic user at time t; σ(·) represents the standard deviation calculation function of the historical prediction error sequence. Let represent the prediction error sequence of the z-th user from time tN to time t-1; This represents the inverse function of the standard normal distribution; α represents the significance level. S213. Incorporate the prediction error bandwidth into the three-phase total power model to form a three-phase power range prediction that considers the uncertainty of distributed photovoltaic output: in, and These represent the lower and upper limits of the predicted total power of phase k at time t, respectively; The imbalance improvement rate ΔU in the fuzzy control rule is robustly evaluated based on the three-phase power range prediction.

8. The method according to claim 1, characterized in that, The fuzzy control rule in S4, when evaluating candidate commutation schemes, also includes the following steps: S411. Establish a risk assessment model that considers the characteristics of the prediction error distribution: Where ρ represents the overall predictability index of the output of the distributed photovoltaic cluster; σ z μ represents the standard deviation of the historical prediction error for the z-th distributed photovoltaic user. z The z-th distributed photovoltaic (PV) user represents the historical average power output; S represents the total number of distributed PV users. S412. The predictability index ρ is used as the third input variable of the fuzzy control rule to adjust the evaluation weight of the commutation priority index R. When the value of ρ is lower than the index threshold, the confidence in the prediction effect is reduced, and a more conservative commutation strategy is preferred. When the value of ρ is higher than the index threshold, the confidence in the prediction effect is increased, and a more aggressive commutation strategy is preferred.

9. The method according to claim 3, characterized in that, The method further includes: S421. Record the actual effect data of historical commutation operations, including the deviation between the actual improvement value of imbalance and the predicted improvement value of imbalance; S422. Establish a dynamic adjustment strategy for fuzzy rule weights based on reinforcement learning. When the actual improvement effect is consistently better than the predicted effect, increase the trigger weight of the corresponding fuzzy rule. When the actual improvement effect is consistently lower than the predicted effect, decrease the trigger weight of the corresponding fuzzy rule.

10. The method according to claim 9, characterized in that, The method further includes: S431. Collect and analyze abnormal event data during historical commutation operations, including switch failures, phase-to-phase short circuits, and user complaint events. S432. Construct a probability prediction model for commutation operation risk: Among them, P risk N represents the expected risk probability of this candidate phase-change operation; fault This indicates the number of switch failures in the history of this line; N short N represents the number of phase-to-phase short circuits in the history of this line; complaint This indicates the number of user complaints for this line in history; N total This represents the total number of historical commutation operations for this line; ε1, ε2, and ε3 represent weighting coefficients. S433, the risk probability P risk As the fourth input variable of the fuzzy control rule, when P risk If the preset threshold is exceeded, the high-risk commutation operation will be prohibited, and a warning message will be generated.

Images