Rule learning-based distributed photovoltaic variable universe fuzzy control method, system and device, and medium

Abstract

The invention discloses a distributed photovoltaic variable universe fuzzy control method, system and device based on rule learning and a medium, and belongs to the technical field of power system control, and the method comprises the steps: obtaining the operation data of a power distribution network containing distributed photovoltaic, and carrying out the preprocessing of the operation data; analyzing by adopting a machine learning algorithm, and extracting a control rule; mapping an input variable into a fuzzy set through a membership function, determining an output fuzzy set according to a fuzzy rule base and a fuzzy reasoning method, and converting the output fuzzy set into a power control instruction through a defuzzification method; dynamically adjusting the input variable discourse domain and the output variable discourse domain of the fuzzy controller through the discourse domain adjustment coefficient; and adjusting the distributed photovoltaic power output according to the power control instruction. According to the method, the variable universe fuzzy control strategy of the distributed photovoltaic system in the power distribution network is obtained through analogue simulation and rule learning, and it can be ensured that the characteristics of the power distribution network reach the optimal level in different operation states.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power system control, in particular to a distributed photovoltaic variable domain fuzzy control method, system, equipment and medium based on rule learning. BACKGROUND

[0002] With the wide application of renewable energy, the penetration rate of distributed photovoltaic in distribution network is gradually increasing. However, due to the influence of factors such as light intensity, the output power of photovoltaic system has great volatility, which poses a challenge to the frequency and voltage stability of distribution network.

[0003] Traditional distributed photovoltaic often adopts maximum power point tracking (MPPT) control, droop control, etc. The former can only passively change the output size with the change of light intensity, and cannot provide adjustment for the frequency and voltage of distribution network. The latter mainly focuses on the adjustment of photovoltaic system through fixed parameter control strategy, without fully considering the dynamic changes of photovoltaic system under different working conditions. Therefore, in order to cope with different scene requirements and improve the frequency and voltage safety and stability level of distribution network, a flexible distributed photovoltaic control method should be studied. SUMMARY

[0004] In view of the above problems, the present application provides a distributed photovoltaic variable domain fuzzy control method, system, equipment and medium based on rule learning.

[0005] Therefore, the technical problem solved by the present application is that the output power of distributed photovoltaic system is greatly fluctuant due to factors such as light intensity, and traditional control methods such as MPPT control can only passively follow the change of light, and cannot provide adjustment for the frequency and voltage of distribution network. Droop control adopts fixed parameter strategy, and cannot fully consider the dynamic changes of photovoltaic system under different working conditions.

[0006] To solve the above technical problems, the present application provides the following technical scheme: a distributed photovoltaic variable domain fuzzy control method based on rule learning, comprising, obtaining the operation data of distribution network containing distributed photovoltaic, the operation data including the light intensity, temperature, output power of distributed photovoltaic and the frequency and voltage parameters of distribution network, recording the operation scene causing the stability problem of distribution network to form a fault data set, and preprocessing the operation data; analyzing the preprocessed operation data by using machine learning algorithm, and extracting control rules describing the relationship between input variables and output variables; constructing a fuzzy control strategy based on the control rules, mapping the input variables to a fuzzy set by using membership function, determining the fuzzy set of output according to fuzzy rule base and fuzzy reasoning method, and converting the fuzzy set of output to power control instruction by using defuzzification method; The universe of discourse of the input and output variables of the fuzzy controller is dynamically adjusted by a universe of discourse adjustment coefficient, which is determined based on the numerical range of the input variables. The power output of the distributed photovoltaic system is adjusted according to the power control command.

[0007] As a preferred embodiment of the rule-based learning-based distributed photovoltaic variable universe fuzzy control method of the present invention, the preprocessing of the running data includes smoothing the running data using a filtering method; After smoothing, anomaly detection methods are used to identify and remove data points that do not conform to the expected pattern. The data after removing outlier data points is normalized to standardize data of different dimensions to a uniform numerical range.

[0008] As a preferred embodiment of the rule-based learning-based distributed photovoltaic variable universe fuzzy control method of the present invention, the step of analyzing the preprocessed running data using a machine learning algorithm includes constructing multiple prediction models using the machine learning algorithm, and each prediction model is trained based on a subset of the running data. By integrating the prediction results of multiple prediction models, the influence weight of each input variable on the output variable is calculated; Identify key input variables based on the influence weights; Based on the mapping relationship between the key input variables and output variables, control rules are extracted.

[0009] As a preferred embodiment of the rule-based distributed photovoltaic variable universe fuzzy control method described in this invention, the step of constructing a fuzzy control strategy based on the control rules includes selecting input variables and mapping the precise values ​​of the input variables to corresponding fuzzy sets through membership functions. A fuzzy rule base is constructed based on the control rules, and the fuzzy rule base describes the mapping relationship between the fuzzy sets of input variables and the fuzzy sets of output variables; The fuzzy inference method is used to infer the fuzzy set of the input variables and the fuzzy rule base to determine the fuzzy set of the output variables; The fuzzy set of output variables is converted into precise control command values ​​using a defuzzification method.

[0010] As a preferred embodiment of the rule-based learning-based distributed photovoltaic variable universe fuzzy control method of the present invention, wherein: the dynamic adjustment of the input variable universe and output variable universe of the fuzzy controller by adjusting the universe adjustment coefficient includes identifying the data in the fault dataset and the data in the running data that deviates from the average value; The domain adjustment coefficient of the input variable and the domain adjustment coefficient of the output variable are set for the data in the fault data set and the data deviating from the average value; The domain adjustment coefficient of the input variable is a function of the value of the corresponding input variable, and the domain adjustment coefficient of the output variable is a function of the value of the input variable reflecting the system operation state; The input variable domain and the output variable domain of the fuzzy controller are adjusted according to the domain adjustment coefficient to obtain an adjusted membership function; Fuzzy reasoning is performed based on the adjusted membership function and a fuzzy rule base to generate an adjusted power control instruction.

[0011] The beneficial effects of the preferred technical scheme are as follows: The input / output domain of the fuzzy controller is dynamically adjusted according to the input variable value and the system operation state through the domain adjustment coefficient, the original range of the domain is maintained to avoid system oscillation when the input variable deviation is large, and the domain is reduced to improve the control precision when the input variable deviation is small, thereby solving the problem that the fixed domain fuzzy control cannot simultaneously achieve high precision and stability in the multi-working condition operation of the photovoltaic system.

[0012] As a preferred scheme of the rule learning-based distributed photovoltaic variable domain fuzzy control method, the membership function is a triangular membership function, the fuzzy reasoning method is the maximum-minimum composition method, and the defuzzification method is the barycenter method. The fuzzy rule base is constructed by two-by-two combination of the input variables, each input variable combination corresponds to a fuzzy rule, and the fuzzy rule describes the control strategy for adjusting the distributed photovoltaic power output under different input conditions.

[0013] As a preferred scheme of the rule learning-based distributed photovoltaic variable domain fuzzy control method, the data in the fault data set and the data deviating from the average value are identified by setting a frequency threshold and a voltage threshold, and the data is determined as fault data when the frequency of the power distribution network deviates from the rated value by more than the frequency threshold or the voltage of the power distribution network deviates from the rated value by more than the voltage threshold. The data deviating from the average value is identified by calculating the deviation of the data point from the average value, and the data is determined as the data deviating from the average value when the deviation of the data point is greater than a preset deviation threshold. The domain adjustment coefficient of the input variable is a function of the input variable deviation, the input variable domain is maintained in the original domain range when the input variable deviation is greater than a first preset deviation, and the input variable domain is smaller than the original domain range when the input variable deviation is less than a second preset deviation.

[0014] The beneficial effects of the preferred technical solutions are that the fault data are identified through the frequency threshold and the voltage threshold, the data deviating from the average value are identified through the deviation threshold, the domain adjustment coefficient is set according to the input variable deviation, the original domain is maintained to ensure stability when the deviation exceeds the first preset deviation, and the domain is reduced to improve the precision when the deviation is less than the second preset deviation, thereby realizing the differential control of the normal working condition and the abnormal working condition, and ensuring that the distributed photovoltaic system can effectively maintain the frequency and voltage stability of the power distribution network under various operating conditions.

[0015] The application provides a distributed photovoltaic variable domain fuzzy control system based on rule learning.

[0016] To solve the above technical problems, the application provides the following technical scheme: a distributed photovoltaic variable domain fuzzy control system based on rule learning, comprising: a data acquisition module configured to acquire operating data of a power distribution network containing distributed photovoltaic, wherein the operating data comprises the intensity of illumination, temperature and output power of the distributed photovoltaic, and the frequency and voltage parameters of the power distribution network, and records an operating scene causing a power distribution network stability problem to form a fault data set; a data preprocessing module configured to preprocess the operating data; a rule learning module configured to analyze the preprocessed operating data by using a machine learning algorithm, and extract a control rule describing the relationship between input variables and output variables; a fuzzy control module configured to construct a fuzzy control strategy based on the control rule, map the input variables into a fuzzy set by using a membership function, determine the fuzzy set of the output according to a fuzzy rule base and a fuzzy reasoning method, and convert the fuzzy set of the output into a power control instruction by using a defuzzification method; a variable domain adjustment module configured to dynamically adjust the input variable domain and the output variable domain of the fuzzy controller by using a domain adjustment coefficient, wherein the domain adjustment coefficient is determined according to the numerical range of the input variable; a control execution module configured to adjust the power output of the distributed photovoltaic according to the power control instruction.

[0017] The application provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the distributed photovoltaic variable domain fuzzy control method based on rule learning when executing the computer program.

[0018] The application provides a computer readable storage medium, which stores a computer program, and the computer program implements the steps of the distributed photovoltaic variable domain fuzzy control method based on rule learning when executed by a processor.

[0019] The beneficial effects of the present application: through simulation, the operation data of the distributed photovoltaic is obtained, and after data cleaning, the operation rules of the distributed photovoltaic are extracted as a training set, and a fuzzy control method is further formed to guide the flexible output adjustment of the distributed photovoltaic in the actual distribution network, and the frequency and voltage characteristics of the distribution network are optimized.

[0020] Further, the present application considers a variable universe distributed photovoltaic control method, which can more accurately and flexibly control the abnormal operation state of the distributed photovoltaic, and ensure efficient operation under various environmental conditions. BRIEF DESCRIPTION OF DRAWINGS

[0021] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0022] Figure 1 A distributed photovoltaic fuzzy control system structure diagram of a variable universe fuzzy control method of distributed photovoltaic based on rule learning provided by an embodiment of the present application.

[0023] Figure 2 A variable universe photovoltaic fuzzy control system structure diagram of a variable universe fuzzy control method of distributed photovoltaic based on rule learning provided by an embodiment of the present application.

[0024] Figure 3 A 3D schematic diagram of distributed photovoltaic fuzzy rules and membership degrees (γ1) of a variable universe fuzzy control method of distributed photovoltaic based on rule learning provided by an embodiment of the present application.

[0025] Figure 4 A 3D schematic diagram of distributed photovoltaic fuzzy rules and membership degrees (γ2) of a variable universe fuzzy control method of distributed photovoltaic based on rule learning provided by an embodiment of the present application.

[0026] Figure 5 A 3D schematic diagram of distributed photovoltaic fuzzy rules and membership degrees (γ3) of a variable universe fuzzy control method of distributed photovoltaic based on rule learning provided by an embodiment of the present application.

[0027] Figure 6 A 3D schematic diagram of distributed photovoltaic fuzzy rules and membership degrees (γ4) of a variable universe fuzzy control method of distributed photovoltaic based on rule learning provided by an embodiment of the present application. DETAILED DESCRIPTION

[0028] In order to make the present application more apparent and easy to understand, the specific embodiments of the present application are described in detail below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should fall within the scope of protection of the present application.

[0029] Embodiment 1, refer to Figure 1 For an embodiment of the present application, the embodiment provides a distributed photovoltaic variable domain fuzzy control method based on rule learning, comprising: Step 1: Obtain the operation data of the distribution network containing distributed photovoltaic, the operation data including the light intensity, temperature, output power of the distributed photovoltaic and the frequency and voltage parameters of the distribution network, record the operation scene causing the stability problem of the distribution network to form a fault data set, and pretreat the operation data; Step 2: Analyze the pretreated operation data by using a machine learning algorithm, and extract control rules describing the relationship between input variables and output variables; Step 3: Construct a fuzzy control strategy based on the control rules, map the input variables to a fuzzy set through a membership function, determine the output fuzzy set according to a fuzzy rule base and a fuzzy reasoning method, and convert the output fuzzy set to a power control instruction through a defuzzification method; Step 4: Dynamically adjust the input variable domain and the output variable domain of the fuzzy controller through a domain adjustment coefficient, the domain adjustment coefficient being determined according to the numerical range of the input variable; Step 5: Adjust the power output of the distributed photovoltaic according to the power control instruction.

[0030] In the actual operation of the distributed photovoltaic connected to the distribution network, the fluctuation of the light intensity causes the photovoltaic output power to have strong randomness and intermittence. When multiple photovoltaic systems simultaneously fluctuate, it will cause the dramatic change of the frequency and voltage of the distribution network. The traditional maximum power point tracking control can only passively adjust the output following the light change and cannot actively participate in the adjustment of the distribution network. The droop control with fixed parameters is prone to cause lag in adjustment response or excessive adjustment to cause oscillation when the light intensity changes greatly or the load of the distribution network suddenly changes, because the control parameters cannot adapt to different working conditions. The embodiment records the fault scene to form a fault data set, and includes the abnormal working condition into the training sample, so that the subsequent variable domain mechanism can set a special domain adjustment strategy for the fault scene.

[0031] The technological advancements of this embodiment lie in: automatically identifying key variables and extracting control rules from massive amounts of operational data through machine learning algorithms, overcoming the limitations of traditional methods that rely on expert experience to set fixed parameters; more importantly, by adjusting the universe of discourse coefficients, the fuzzy controller achieves adaptability to different operating conditions. When the system deviation is large, a wide universe of discourse is maintained to avoid oscillations; when the system deviation is small, the universe of discourse is narrowed to improve accuracy, resolving the contradiction between accuracy and stability in fixed-universe fuzzy control. In particular, a dedicated universe of discourse adjustment strategy is set for fault datasets, enabling the controller to maintain optimal control performance under various operating conditions such as normal operation of the photovoltaic system, abnormal frequency or voltage, and severe fluctuations in illumination, achieving the technical effect of a single controller adapting to multiple operating conditions.

[0032] Example 2, refer to Figures 1-6 As one embodiment of the present invention, based on the previous embodiment, a distributed photovoltaic variable universe fuzzy control method based on rule learning is provided, comprising: Step 1: Obtaining operational data of the distribution network containing distributed photovoltaic power, and preprocessing the operational data, includes the following steps A1-A3: A1: The running data is smoothed using a filtering method; A2: After smoothing, use anomaly detection methods to identify and remove data points that do not conform to the expected pattern; A3: Normalize the data after removing outlier data points to standardize data of different dimensions to a uniform numerical range.

[0033] In this embodiment of the application, in step A1, the smoothing process is performed by: smoothing the running data using a mean filtering method, the calculation formula for the mean filtering is: in, For the first Smoothed values ​​for each data point; This refers to the size of the filtering window; For the first The first data window One original data point; Represents the first step of mean filtering. Data from each window start, Represents the first step of mean filtering. Window data to Finish.

[0034] In an optional implementation, in step A1, the smoothing process can be performed by: smoothing the running data using a median filter, selecting a filter window size of [missing value]. The data points within the window The data points are sorted in ascending order of their numerical values. When the number is odd, the middle data point is used as the smoothing value. When the number is even, the average of the two middle data points is taken as the smoothing value.

[0035] In another alternative implementation, in step A1, the smoothing process can also be performed by: using a Gaussian filter to smooth the running data, assigning different weights to each data point within the filter window according to the Gaussian function and performing a weighted average, with the data points at the center position having a larger weight and the data points at the edge position having a smaller weight.

[0036] Step 2: Machine learning algorithms are used to analyze the preprocessed runtime data and extract control rules describing the relationship between input and output variables, including the following steps B1-B4: B1: Multiple prediction models are constructed using the machine learning algorithm described above, and each prediction model is trained based on a subset of the running data; B2: By integrating the prediction results of multiple prediction models, calculate the influence weight of each input variable on the output variable; B3: Identify key input variables based on the aforementioned influence weights; B4: Extract control rules based on the mapping relationship between the key input variables and output variables.

[0037] In this embodiment, step 2 involves the machine learning algorithm analyzing the preprocessed running data using a random forest algorithm. Specifically, this involves randomly selecting multiple subsets from the running dataset, each subset containing partial sample data. A decision tree is constructed for each subset. At each node of the decision tree, a subset of input variables is randomly selected for feature splitting. Parameters such as light intensity, temperature, distribution network frequency, and voltage are used as input variables, and distributed photovoltaic power output is used as the target variable. A complete decision tree is constructed through recursive splitting. Integrating the prediction results of multiple decision trees, the prediction formula for the random forest is: in, This represents the final prediction result of the random forest. For the first A decision tree for input The prediction results The total number of decision trees is denoted by . The importance score of each input variable to the distributed photovoltaic power output is calculated, key variables with significant impact on system stability are identified, and finally, irradiance, distribution network frequency deviation, and frequency change rate are selected as key indicators. Control rules in the form of "if irradiance is high and distribution network frequency is low, then increase active power output" are extracted.

[0038] In an optional embodiment, in step 2, the machine learning algorithm can analyze the pre-processed operation data by using a gradient boosting decision tree algorithm, and gradually build multiple decision trees through iteration, each new decision tree being used to correct the prediction error of the previous decision tree. In each iteration, the prediction residual of the current model for the training sample is calculated, a new decision tree is built to fit these residuals, and the new decision tree is added to the model in a weighted manner. The contribution weight of each decision tree is controlled by setting the learning rate, and overfitting is prevented by setting the maximum depth of the tree and the minimum sample size of the leaf node. The cumulative split gain of each input variable in all decision trees is calculated, the variables with larger cumulative gain are identified as key input variables, and control rules are extracted based on these key variables.

[0039] In another optional embodiment, in step 2, the machine learning algorithm can also analyze the pre-processed operation data by using a support vector machine algorithm, mapping the input variables to a high-dimensional feature space, and finding the optimal classification or regression hyperplane in this space. The non-linear relationship between the input variables and the output variables is handled by selecting an appropriate kernel function, common kernel functions including radial basis kernel function, polynomial kernel function, etc. The complexity of the model and the training error are balanced by adjusting the penalty parameter, and the properties of the mapped feature space are controlled by adjusting the kernel function parameter. The support vector weight coefficients corresponding to each input variable are calculated, the variables with larger absolute values of the weight coefficients are identified as key input variables, and control rules between the input variables and the output variables are extracted based on the support vectors and the weight coefficients.

[0040] In step 3: based on the control rules, a fuzzy control strategy is constructed, the input variables are mapped to fuzzy sets by membership functions, the output fuzzy sets are determined according to the fuzzy rule base and fuzzy reasoning method, and the output fuzzy sets are converted to power control instructions by defuzzification method, including the following steps C1-C4: C1: select input variables, and map the exact values of the input variables to corresponding fuzzy sets by membership functions; C2: construct a fuzzy rule base according to the control rules, the fuzzy rule base describing the mapping relationship between the fuzzy sets of the input variables and the fuzzy sets of the output variables; C3: use a fuzzy reasoning method to reason according to the fuzzy sets of the input variables and the fuzzy rule base, and determine the fuzzy sets of the output variables; C4: use a defuzzification method to convert the fuzzy sets of the output variables to exact control instruction values.

[0041] In the embodiment of the application, in step 3, the membership function maps the exact value of the input variable to the fuzzy set by using a triangular membership function, and the calculation formula of the triangular membership function is: wherein, is the membership degree of the input variable ; is the parameter of the triangular membership function, is the left end point of the triangle, is the top point of the triangle, is the right end point of the triangle. The light intensity, temperature of the distributed photovoltaic, and the frequency and voltage of the power distribution network are selected as the input variables, and multiple triangular membership functions are set for each input variable, corresponding to different fuzzy language values of low, medium and high, respectively.

[0042] In an optional embodiment, in step 3, the membership function can be obtained by: using a trapezoidal membership function to map the exact value of the input variable to a fuzzy set, the trapezoidal membership function is defined by four parameters, including the left lower bottom end point, the left upper bottom end point, the right upper bottom end point, and the right lower bottom end point. When the numerical value of the input variable is between the left upper bottom end point and the right upper bottom end point, the membership degree is 1; when the numerical value of the input variable is between the left lower bottom end point and the left upper bottom end point, the membership degree increases linearly with the numerical value; when the numerical value of the input variable is between the right upper bottom end point and the right lower bottom end point, the membership degree decreases linearly with the numerical value; when the numerical value of the input variable is outside the trapezoidal interval, the membership degree is 0. The trapezoidal membership function is suitable for fuzzy language values with stable intervals, such as the input variable range corresponding to the normal state.

[0043] In another optional embodiment, in step 3, the membership function can also be obtained by: using a Gaussian membership function to map the exact value of the input variable to a fuzzy set, the Gaussian membership function is defined by two parameters, including the center point and the standard deviation, and the membership degree curve is in the shape of a bell. The membership degree of the Gaussian membership function is 1 at the center point, and the membership degree decreases smoothly according to the Gaussian distribution law as the input variable deviates from the center point. The Gaussian membership function has the characteristics of continuous derivability, and is suitable for control scenarios that require smooth changes in output, which can avoid the sudden change phenomenon at the junction of adjacent fuzzy sets.

[0044] In the embodiment of the application, in step C3, the fuzzy reasoning method is obtained by: using the max-min composition method to reason according to the fuzzy set of the input variable and the fuzzy rule base to determine the fuzzy set of the output variable, and the calculation formula of the max-min composition method is: wherein, and are the membership functions of the input and output, is the output membership function obtained by reasoning. For each rule in the fuzzy rule base, the minimum value of the membership degrees of the input variables in the antecedent of the rule is calculated first, and then the maximum value of the minimum values corresponding to all rules is taken to obtain the fuzzy set of the output variable.

[0045] In an optional embodiment, in step C3, the fuzzy inference method can determine the fuzzy set of the output variable by inferring from the fuzzy set of the input variable and the fuzzy rule base using a max-product composition method. The max-product composition method replaces the minimum operation in the max-min composition method with a product operation. For each rule in the fuzzy rule base, the product of the membership degrees of the input variables in the antecedent of the rule is calculated first, and then the maximum value of the products corresponding to all rules is taken to obtain the fuzzy set of the output variable. The max-product composition method can more fully consider the influence of the membership degrees of the input variables on the inference result, and is suitable for control scenarios that require the inference result to be sensitive to input changes.

[0046] In another optional embodiment, in step C3, the fuzzy inference method can determine the exact value of the output variable by inferring from the fuzzy set of the input variable and the fuzzy rule base using a Sugeno fuzzy inference method. The rule consequent of the Sugeno fuzzy inference method is a linear function or a constant of the input variable. For each rule, the minimum or product of the membership degrees of the input variables in the antecedent of the rule is calculated first as the activation strength of the rule, then the activation strength is multiplied by the function value of the rule consequent, and finally the weighted results of all rules are summed and normalized to directly obtain the exact value of the output variable without an additional defuzzification step. The Sugeno fuzzy inference method has higher computational efficiency and is suitable for control scenarios with higher real-time requirements.

[0047] In step 4: dynamically adjusting the input variable domain and the output variable domain of the fuzzy controller by adjusting the domain adjustment coefficient, wherein the domain adjustment coefficient is determined according to the numerical range of the input variable, comprising steps D1-D5: D1: identifying the data in the fault data set and the data deviating from the average value in the operating data; D2: setting the input variable domain adjustment coefficient and the output variable domain adjustment coefficient for the data in the fault data set and the data deviating from the average value; D3: the input variable domain adjustment coefficient is a function of the numerical value of the corresponding input variable, and the output variable domain adjustment coefficient is a function of the numerical value of the input variable reflecting the system operating state; D4: adjusting the input variable domain and the output variable domain of the fuzzy controller according to the domain adjustment coefficient to obtain an adjusted membership function; D5: generating an adjusted power control instruction based on the adjusted membership function and the fuzzy rule base.

[0048] Further, the membership function is a triangular membership function, the fuzzy inference method is a max-min composition method, and the defuzzification method is a centroid method. The fuzzy rule base is constructed by input variables in pairs, and each pair of input variables corresponds to a fuzzy rule. The fuzzy rule describes the control strategy for adjusting the distributed photovoltaic power output under different input conditions.

[0049] The data in the fault data set and the data deviating from the average value are identified by setting a frequency threshold and a voltage threshold. When the frequency of the power distribution network deviates from the rated value by more than the frequency threshold, or the voltage of the power distribution network deviates from the rated value by more than the voltage threshold, it is determined as fault data. The data deviating from the average value is identified by calculating the deviation of the data point from the average value. When the deviation of the data point exceeds a preset deviation threshold, it is determined as data deviating from the average value. The universe of discourse adjustment coefficient of the input variable is a function of the input variable deviation. When the input variable deviation exceeds a first preset deviation, the universe of discourse adjustment coefficient of the input variable maintains the input variable universe of discourse within the original universe of discourse range. When the input variable deviation is less than a second preset deviation, the universe of discourse adjustment coefficient of the input variable makes the input variable universe of discourse smaller than the original universe of discourse range.

[0050] It should be noted that the first preset deviation and the second preset deviation are used to determine the degree of deviation of the input variable, so as to determine the value of the universe of discourse adjustment coefficient. The first preset deviation corresponds to a larger deviation range of the input variable, and the second preset deviation corresponds to a smaller deviation range of the input variable. When the input variable deviation exceeds the first preset deviation, the universe of discourse adjustment coefficient maintains the universe of discourse within the original universe of discourse range. When the input variable deviation is less than the second preset deviation, the universe of discourse adjustment coefficient makes the universe of discourse smaller. The specific values of the first preset deviation and the second preset deviation are set according to the operating characteristics and control accuracy requirements of the actual power distribution network.

[0051] Embodiment 3 is an embodiment of the present application, which provides a distributed photovoltaic variable universe of discourse fuzzy control system based on rule learning, comprising: A data acquisition module is configured to acquire operating data of a power distribution network containing distributed photovoltaic. The operating data includes the intensity of light, temperature, output power of the distributed photovoltaic, and frequency and voltage parameters of the power distribution network, and records an operating scenario that causes a power distribution network stability problem to form a fault data set. A data preprocessing module is configured to preprocess the operating data. A rule learning module is configured to analyze the preprocessed operating data using a machine learning algorithm, and extract control rules describing the relationship between input variables and output variables. a fuzzy control module, configured to construct a fuzzy control strategy based on the control rules, map input variables to fuzzy sets through a membership function, determine an output fuzzy set according to a fuzzy rule base and a fuzzy inference method, and convert the output fuzzy set into a power control instruction through a defuzzification method; a variable domain adjustment module, configured to dynamically adjust input variable domains and output variable domains of the fuzzy controller through a variable domain adjustment coefficient, the variable domain adjustment coefficient being determined according to a numerical range of the input variable; a control execution module, configured to adjust a power output of the distributed photovoltaic according to the power control instruction.

[0052] In order to obtain operation data while ensuring the safety of the power distribution network and equipment, modeling and simulation of the power distribution network and the distributed photovoltaic are performed through Matlab / Simulink to simulate the operation of the actual power distribution network. The light intensity, power distribution network frequency, voltage value and other indicators are selected as independent variables, and the power output of the distributed photovoltaic is selected as the dependent variable. Through massive operation simulation, the results of each data in different scenarios are monitored in real time, and the data are transmitted to the storage system for recording to form a complete data set. In the simulation process, the distributed photovoltaic operation scenarios that cause stability problems of the power distribution network are recorded separately. We set thresholds for frequency and voltage to identify abnormal operation states. For example, when the frequency deviates from the rated value by ±0.2 Hz or the voltage deviates from the rated value by ±5%, a fault set is formed. In this way, the data collection not only covers the normal operation state, but also includes potential risk scenarios, providing a comprehensive information base for subsequent analysis and optimization.

[0053] In order to ensure the accuracy and reliability of subsequent analysis, data cleaning and preprocessing are needed to improve data quality. First, the mean filter method is used to smooth the data curve, reduce random fluctuations in the data, and highlight the main trend of the data. The formula of the mean filter is: wherein, is the smoothed value of the i th data point; is the size of the filter window; is the i th original data point in the i th data window; indicates that the i th window data of the mean filter starts from indicates that the i th window data of the mean filter ends at

[0054] ​​​​​​​Next, outlier detection techniques are used to identify and remove data points that do not conform to the expected pattern, avoiding accidental errors in measurement. Outlier detection can be achieved through standard deviation-based detection methods, setting a threshold value when the data point deviation exceeds the threshold value, the data point is considered an outlier and is removed. For a given data set, we calculate the deviation of each data point from the mean: where, is the deviation of the th data point; is the original data point; is the mean of the data set; is the standard deviation of the data set.

[0055] Further, in order to adapt to the learning and training of the model, data normalization is needed to improve the consistency of the data. The minimum-maximum scaling method is used to standardize data of different dimensions: where, is the normalized data point; , are the minimum and maximum values of the data set, respectively. Through this method, the data is scaled to the range [0, 1], laying a solid foundation for efficient learning and training of the model.

[0056] The above processed data is analyzed to obtain the relationship between the variables. The cleaned data set is analyzed using the random forest algorithm. Specifically, first, a decision tree is constructed. Each decision tree is trained from a random subset of the data set, with light intensity, power distribution network frequency, voltage, and other power system parameters as input variables , and distributed photovoltaic power output as the target variable, at each node, a random feature subset is selected for splitting. Second, integrate multiple tree predictions. Using the random forest algorithm, the prediction results of multiple decision trees are integrated to improve the accuracy of the model, and the prediction formula of the random forest is: where, is the final prediction result of the random forest, is the prediction result of the th decision tree for input , is the total number of decision trees. Third, variable importance evaluation. Calculate the importance score of each input variable on the distributed photovoltaic power output, identify key variables that have a significant impact on system stability, and finally select light intensity, power distribution network frequency deviation, and frequency change rate as key indicators.

[0057] Finally, rule extraction. After completing the data analysis, we summarize and extract "if-then" rules for the above key indicators and power output variables. These rules are used to describe how to adjust the operation of the photovoltaic system under different conditions to optimize the operation characteristics of the distribution network.

[0058] Specifically, we can infer that if the light intensity is high, increasing the active power output of the distributed photovoltaic system helps optimize the frequency characteristics of the distribution network. Each rule can be expressed as: where, is the variable, is the threshold, is the operation taken.

[0059] Through these steps, we can effectively learn and summarize the coupling relationship between variables and develop operation strategies to optimize system performance.

[0060] According to the correlation analysis of the above variables, a fuzzy control strategy is constructed to adjust the power output of the distributed photovoltaic system under different operating conditions, optimizing the operation characteristics of the distribution network.

[0061] First, input variable fuzzification is performed, selecting the light intensity, temperature of the distributed photovoltaic system, and the frequency and voltage of the distribution network as input variables. The precise input values are mapped to fuzzy sets using a triangular membership function.

[0062] where, is the membership degree of the input variable ; , , are the parameters of the triangular membership function, and are the minimum and maximum intervals of the membership degree, and the membership degree is 0, is the triangular membership interval, and the membership degree is non-zero only when the above input variables are in this interval.

[0063] Second, the fuzzy rule base is constructed. According to the rule learning results, a fuzzy rule base between two independent variables is constructed, such as if the light intensity is high and the distribution network frequency is low, then the active power output is increased: where, is the light intensity of the distributed photovoltaic system; is the frequency of the distribution network; is the active power output of the distributed photovoltaic system.

[0064] Furthermore, a fuzzy inference engine is designed. This engine performs inference based on the input fuzzy set and rule base, and uses the max-min synthesis method to determine the output fuzzy set. in, and Membership functions for input and output.

[0065] Next, deblurring is performed, using the centroid method to deblur the fuzzy output set obtained by the inference engine and transform it into specific control commands.

[0066] in, This is the output value after deblurring. To output the membership function.

[0067] Finally, application control commands for distributed photovoltaic (PV) systems are generated. These defuzzified control commands are then applied to the distributed PV systems in the actual distribution network. During operation, the power output of the PV systems is dynamically adjusted based on the measured operating status to optimize the frequency and voltage characteristics of the distribution network. A 3D schematic diagram of the four fuzzy controllers is shown below. Figures 3-6 As shown, these represent frequency deviations. With the Distributed power supply Light intensity at any time Strategies for Effective Output under Co-operation Frequency deviation With the Distributed power supply Rate of change of light intensity over time Strategies for Effective Output under Co-operation Rate of change of frequency With the Distributed power supply Light intensity at any time Strategies for Effective Output under Co-operation Rate of change of frequency With the Distributed power supply Rate of change of light intensity over time Strategies for Effective Output under Co-operation The structure diagram of the fuzzy control system is as follows: Figure 1 As shown, This serves as a reference for the current active power output of the distributed power source.

[0068] The improvement of this invention lies in the fact that, compared with traditional fixed parameter control, this method can dynamically adjust the control strategy according to the real-time operating status.

[0069] The above fuzzy control rules are global optimal solutions for all operating conditions, but they will not be accurate when some independent variables deviate significantly from the average value in the data set or are in the fault set. Therefore, the fuzzy control rules are adjusted by variable domain adjustment, such as Figure 2 .

[0070] First, the domain of the input variables is analyzed and adjusted, the fault set and the data set deviating greatly from the average value are analyzed, the variables that need to be adjusted are identified, and the domain of the input variables is dynamically adjusted. For example, the domain of the light intensity is adjusted according to the time period of the day.

[0071] wherein, is the original variable, is the adjusted variable, is the time.

[0072] Second, the fuzzy rule base is updated in real time. The variable domain of the fuzzy controller can be realized by the scaling factor. Taking the fuzzy controller I in Figure 2 as an example, if the input frequency deviation is , and the domain of the output is , then the new domain obtained by the scaling factor is and to reflect the control difference under different operating conditions. In turn, , represent the scaling factors of the input frequency change rate and the output , respectively.

[0073] In the design of the scaling factor, the scaling factor of the input is generally a function of the input itself, while the scaling factor of the output usually depends on the input that best reflects the operating condition, to more accurately and flexibly match multiple operating conditions. In this paper, since the variable domain control introduces a frequency regulation dead zone to improve the transient frequency characteristics of the system, the scaling factor of each output is a function of the corresponding frequency input ( or ). That is, Taking the input end scaling factor as an example: if the system frequency deviation is large, , the domain is almost unchanged, and the original fuzzy rules are used for control; if the frequency deviation is small, the domain should be smaller, making the fuzzy rules denser and providing higher control precision, i.e. The domain should be smaller. This design method solves the contradiction between precision and domain size when the fuzzy controller handles different working conditions. By adjusting the domain size, the controller can maintain high precision while avoiding system oscillation caused by excessive adjustment, such as Figure 2 .

[0074] wherein, is the frequency deviation of the distribution network; is the frequency change rate of the distribution network.

[0075] Next, the dynamically adjusted max-min composition method is used to process the fuzzy set under variable domain wherein, is the updated input membership function, is the output membership function.

[0076] Finally, a system feedback mechanism is established to continuously monitor the control effect, and the variable domain parameter settings are optimized according to the feedback information to ensure efficient operation of the distributed photovoltaic system under various environmental conditions. Through these steps, the variable domain fuzzy control strategy can more flexibly respond to dynamically changing system environments, ensuring efficient and stable operation of the distributed photovoltaic system under different operating conditions.

[0077] The method of this embodiment introduces a variable domain mechanism into distributed photovoltaic control, dynamically adjusts the input and output domains of the fuzzy controller, and solves the problem of insufficient precision of traditional fixed domain fuzzy control when handling multiple working conditions of power systems. Especially for the intermittent and volatile characteristics of photovoltaic systems, a stretching factor based on the size of the frequency deviation is designed to achieve a balance between control precision and stability.

[0078] This embodiment also provides an electronic device suitable for a variable domain fuzzy control method for distributed photovoltaic systems based on rule learning, comprising: a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions to realize a variable domain fuzzy control method for distributed photovoltaic systems based on rule learning as proposed in the above embodiment.

[0079] This embodiment also provides a storage medium having a computer program stored thereon, which is executed by a processor to realize a variable domain fuzzy control method for distributed photovoltaic systems based on rule learning as proposed in the above embodiment.

[0080] The storage medium proposed in the embodiment belongs to the same inventive concept as the distributed photovoltaic variable domain fuzzy control method based on rule learning proposed in the above embodiment, and the technical details not described in detail in the embodiment can be seen from the above embodiment, and the embodiment has the same beneficial effects as the above embodiment.

[0081] Through the above description of the embodiments, those skilled in the art can clearly understand that the present application can be realized by means of software and necessary general hardware, and of course can also be realized by hardware, but in many cases the former is a better embodiment. Based on such understanding, the technical solutions of the present application or the part that contributes to the prior art can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a floppy disk, a read-only memory (ROM), a random access memory (RAM), a FLASH, a hard disk or an optical disk, etc., including a number of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods of various embodiments of the present application.

[0082] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and are not limiting, and although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the present application, and they should be covered in the scope of the claims of the present application.

Claims

1. A distributed photovoltaic variable universe fuzzy control method based on rule learning, characterized in that: The application relates to a power output control method for distributed photovoltaic (PV) in a power distribution network. The method comprises the following steps: acquiring operation data of a power distribution network containing distributed PV, the operation data including illumination intensity, temperature, output power of the distributed PV, and frequency and voltage parameters of the power distribution network, recording an operation scene causing a power distribution network stability problem to form a fault data set, and pre-processing the operation data; adopting a machine learning algorithm to analyze the pre-processed operation data, and extracting control rules describing the relationship between input variables and output variables; constructing a fuzzy control strategy based on the control rules, mapping input variables into fuzzy sets through membership functions, determining the fuzzy set of output variables according to a fuzzy rule base and a fuzzy reasoning method, and converting the fuzzy set of output variables into a power control instruction through a defuzzification method; dynamically adjusting the input variable domain and the output variable domain of the fuzzy controller through a domain adjustment coefficient, the domain adjustment coefficient being determined according to the numerical range of the input variables; adjusting the power output of the distributed PV according to the power control instruction. 2.The distributed photovoltaic variable universe fuzzy control method based on rule learning of claim 1, wherein: The pre-processing of the operation data comprises the following steps: adopting a filtering method to perform smoothing processing on the operation data; after the smoothing processing, identifying and eliminating data points not conforming to an expected mode through an abnormal data identification method; performing normalization processing on the data after the elimination of abnormal data points, and standardizing data of different dimensions to a unified numerical interval.

3. The distributed photovoltaic variable universe fuzzy control method based on rule learning according to claim 2, characterized in that: The analysis of the pre-processed operation data through the machine learning algorithm comprises the following steps: adopting the machine learning algorithm to construct multiple prediction models, each prediction model being trained based on a subset of operation data; calculating the influence weight of each input variable on the output variable by integrating the prediction results of the multiple prediction models; identifying key input variables according to the influence weight; extracting control rules based on the mapping relationship between the key input variables and the output variables.

4. The distributed photovoltaic variable universe fuzzy control method based on rule learning according to claim 3, characterized in that: The construction of the fuzzy control strategy based on the control rules comprises the following steps: selecting input variables, and mapping the exact values of the input variables into corresponding fuzzy sets through membership functions; constructing a fuzzy rule base according to the control rules, the fuzzy rule base describing the mapping relationship between the fuzzy set of input variables and the fuzzy set of output variables; adopting a fuzzy reasoning method to reason according to the fuzzy set of input variables and the fuzzy rule base, and determining the fuzzy set of output variables; adopting a defuzzification method to convert the fuzzy set of output variables into an exact control instruction value.

5. The distributed photovoltaic variable universe fuzzy control method based on rule learning according to claim 4, characterized in that: The dynamic adjustment of the input variable domain and the output variable domain of the fuzzy controller through the domain adjustment coefficient comprises the following steps: identifying data in the fault data set and data deviating from the average value in the operation data; setting the domain adjustment coefficient of input variables and the domain adjustment coefficient of output variables for the data in the fault data set and the data deviating from the average value; the domain adjustment coefficient of the input variables being a function of the numerical value of the corresponding input variables, and the domain adjustment coefficient of the output variables being a function of the numerical value of the input variables reflecting the system operation state; adjusting the input variable domain and the output variable domain of the fuzzy controller according to the domain adjustment coefficient, and obtaining an adjusted membership function. The fuzzy inference is performed based on the adjusted membership function and the fuzzy rule base to generate an adjusted power control instruction.

6. The distributed photovoltaic variable universe fuzzy control method based on rule learning according to claim 5, characterized in that: The membership function is a triangular membership function, the fuzzy inference method is a maximum-minimum composition method, and the defuzzification method is a barycenter method. The fuzzy rule base is constructed by combining each two input variables, each combination of input variables corresponds to a fuzzy rule, and the fuzzy rule describes a control strategy for adjusting the distributed photovoltaic power output under different input conditions.

7. The distributed photovoltaic variable universe fuzzy control method based on rule learning according to claim 6, characterized in that: The data in the fault data set and the data deviating from the average value are identified by setting a frequency threshold and a voltage threshold. When the frequency of the power distribution network deviates from the rated value by more than the frequency threshold, or the voltage of the power distribution network deviates from the rated value by more than the voltage threshold, the data is determined as fault data. The data deviating from the average value is identified by calculating the deviation of the data point from the average value. When the deviation of the data point exceeds a preset deviation threshold, the data is determined as data deviating from the average value. The domain adjustment coefficient of the input variable is a function of the input variable deviation. When the input variable deviation exceeds a first preset deviation, the domain adjustment coefficient of the input variable maintains the input variable domain within the original domain range. When the input variable deviation is less than a second preset deviation, the domain adjustment coefficient of the input variable makes the input variable domain smaller than the original domain range.

8. A distributed photovoltaic variable universe fuzzy control system based on rule learning, applying a distributed photovoltaic variable universe fuzzy control method based on rule learning as claimed in any one of claims 1 to 7, characterized in that, It comprises: a data acquisition module for acquiring operation data of a power distribution network containing distributed photovoltaics, the operation data including the intensity of light, temperature, output power of the distributed photovoltaics, and frequency and voltage parameters of the power distribution network, and recording an operation scenario causing a power distribution network stability problem to form a fault data set; a data preprocessing module for preprocessing the operation data; a rule learning module for analyzing the preprocessed operation data using a machine learning algorithm to extract control rules describing the relationship between input variables and output variables; a fuzzy control module for constructing a fuzzy control strategy based on the control rules, mapping input variables into a fuzzy set through a membership function, determining the output fuzzy set according to a fuzzy rule base and a fuzzy inference method, and converting the output fuzzy set into a power control instruction through a defuzzification method; a variable domain adjustment module for dynamically adjusting the input variable domain and output variable domain of the fuzzy controller through a domain adjustment coefficient, the domain adjustment coefficient being determined according to the numerical range of the input variable; a control execution module for adjusting the power output of the distributed photovoltaics according to the power control instruction. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-7. The processor executes the computer program to implement the steps of the variable domain fuzzy control method for distributed photovoltaics based on rule learning according to any one of claims 1 to 7.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the variable domain fuzzy control method for distributed photovoltaics based on rule learning according to any one of claims 1 to 7.

Images