Inverter power supply network construction type control method based on artificial neural network

Abstract

The invention discloses an inverter power supply network construction type control method based on an artificial neural network, and relates to the technical field of inverter power supply network construction type control. Collecting an electrical parameter sequence set of an inverter power supply grid-connected point; constructing a power supply network construction control analyzer based on the deep feedforward neural network, and outputting a first network construction control parameter; building a power distribution network simulation topology model based on PSCAD to perform parameter optimization, and generating a second network construction control parameter; and analyzing and determining the fluctuation degree of the operation state, carrying out weighted fusion on the two types of control parameters according to a set dynamic fitting strategy, obtaining an adaptive networking control strategy, and implementing regulation and control. According to the invention, on-line intelligent optimization of control parameters is realized through a two-way parallel mechanism of quick response of the neural network and accurate verification of the simulation model in combination with a self-adaptive fitting strategy of operation state perception, and the stability, adaptability and control precision of the inverter power supply under complex working conditions are effectively improved.

Description

TECHNICAL FIELD

[0001] The application relates to the network configuration type control technology of an inverter power supply, in particular to an inverter power supply network configuration type control method based on an artificial neural network. BACKGROUND

[0002] With the increasing proportion of renewable energy grid connection, the proportion of inverter power supply in the power system is increasing. The traditional follow-network control strategy is prone to stability problems in the weak grid or high proportion of new energy access scenarios. The network configuration type control technology can enhance the inertia and damping support capacity of the grid by simulating the operating characteristics of a synchronous generator.

[0003] At present, the network configuration type control method is mostly based on a fixed control structure or a classical proportional integral derivative control, and the parameters thereof usually depend on artificial setting or offline optimization, and it is difficult to adapt to the dynamic changes of the grid operating state. In the complex working conditions of time-varying grid parameters, load fluctuations or fault disturbances, the adaptive ability and robustness of the traditional control method are limited, and control instability or performance degradation is easily caused. At the same time, the existing method is sensitive to operating state fluctuations, and over-regulation or oscillation phenomenon is easily caused when the electrical parameters change. SUMMARY

[0004] The application provides an inverter power supply network configuration type control method based on an artificial neural network to solve the technical problems of the prior art, such as fixed network configuration type control parameters, insufficient adaptive ability, dependence on artificial experience, and limited stability in complex working conditions.

[0005] The technical solution of the application for solving the above technical problems is as follows: The application provides an inverter power supply network configuration type control method based on an artificial neural network, comprising: Collecting an electrical parameter sequence set of an inverter power supply grid connection point in a historical time zone; Building a power supply network control analyzer based on a deep feedforward neural network, performing control analysis in a preset time zone according to the electrical parameter sequence set, and outputting first network configuration control parameters; Building a power distribution network simulation topology model based on PSCAD, performing network configuration control parameter optimization according to the electrical parameter sequence set, and generating second network configuration control parameters; Determining the operating state fluctuation degree based on the electrical parameter sequence set analysis, setting a dynamic fitting strategy according to the operating state fluctuation degree to control and fit the first network configuration control parameters and the second network configuration control parameters, obtaining an adaptive network configuration control strategy, and regulating and controlling the inverter power supply in the preset time zone.

[0006] The application has the following beneficial effects: Compared with the prior art, the application firstly realizes effective fusion of data driving and physical model through a double-path parallel optimization mechanism of a deep feedforward neural network and a PSCAD simulation model, significantly improves the accuracy and reliability of the control strategy. Secondly, the running state fluctuation degree evaluation and dynamic fitting strategy are introduced, so that the control parameters can be adaptively adjusted according to the real-time running state of the power grid, and the adaptability under complex working conditions is enhanced. Thirdly, the intelligent parameter optimization and fitting method is adopted, which replaces the traditional control parameter setting method relying on artificial experience, greatly improves the generation efficiency and optimization accuracy of the control strategy. Finally, through the double-path parallel mechanism of neural network rapid response and simulation model accurate verification, the control framework close to the engineering practice is constructed, which has good engineering applicability and can provide effective support for the stable operation of the power grid under high proportion of new energy access. BRIEF DESCRIPTION OF DRAWINGS

[0007] Figure 1 A flowchart of an inverter power supply network type control method based on an artificial neural network is provided. Figure 2 A circuit control principle schematic diagram of an inverter power supply network type control method based on an artificial neural network is provided. DETAILED DESCRIPTION

[0008] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, not all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the application.

[0009] In the description of the application, the terms "first", "second" are used only for the purpose of description, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features. Therefore, the features defined with "first", "second" can explicitly or implicitly include one or more of the features. In the description of the application, the meaning of "multiple" is two or more, unless otherwise specifically limited.

[0010] In the description of the present application, the term "for example" is used to mean "serving as an instance, illustration, or example, with the aim of clarifying the description". Any embodiment described as "for example" in the present application is not necessarily to be construed as preferred or advantageous over other embodiments. The following description is presented to enable any person skilled in the art to make and use the present application. In the following description, for the purpose of explanation, details are set forth. It is apparent to those skilled in the art that the present application can be practiced without the use of these specific details. In other instances, well-known structures and processes are not elaborated in order not to obscure the description of the present application with unnecessary details. Therefore, the present application is not intended to be limited to the embodiments shown, but is to be accorded the widest scope consistent with the principles and features disclosed.

[0011] Embodiment one, as shown in Figure 1 , Figure 2 The present application provides a control method for grid-connected inverter based on artificial neural network, comprising: S10: collecting a set of electrical parameter sequences of the grid-connected point of the inverter in a historical time zone; Specifically, the electrical parameters of the grid-connected point of the inverter at a plurality of continuous time nodes in the historical time zone are collected to obtain a plurality of electrical parameter sequences as the set of electrical parameter sequences, wherein the electrical parameters include three-phase voltage instantaneous value, three-phase current instantaneous value, d-axis voltage, q-axis voltage, d-axis current and q-axis current.

[0012] The grid-connected point of the inverter is the physical interface point for the inverter to exchange and connect with the public grid, and it is the key position for monitoring and controlling the operation state of the inverter. The electrical parameters of the grid-connected point of the inverter directly reflect the interaction characteristics between the inverter and the grid.

[0013] Collecting the electrical parameters of the grid-connected point of the inverter in the historical time zone means that a historical time period is set according to the control requirements, such as 3 minutes, and a plurality of time nodes are selected in a continuous and equal interval manner in the historical time zone. The selection frequency of the time nodes is set according to the required control accuracy and data amount, such as collecting 1 point per second or 1 point per 10 milliseconds. Then, a plurality of sets of electrical parameter instantaneous values of the grid-connected point of the inverter are synchronously collected at each selected time node. The collected electrical parameters specifically include three-phase voltage instantaneous value, three-phase current instantaneous value, d-axis voltage, q-axis voltage, d-axis current and q-axis current. Together, they constitute a set of electrical parameter sequences reflecting the operation state of the grid-connected point of the inverter, providing a data basis for subsequent analysis and control.

[0014] Further, the operation state fluctuation degree is analyzed and determined based on the set of electrical parameter sequences, comprising: obtaining a set of electrical parameter sequences, wherein the set of electrical parameter sequences comprises a three-phase voltage instantaneous value sequence, a three-phase current instantaneous value sequence, a d-axis voltage sequence, a q-axis voltage sequence, a d-axis current sequence, and a q-axis current sequence; performing parameter fluctuation calculation on the three-phase voltage instantaneous value sequence, the three-phase current instantaneous value sequence, the d-axis voltage sequence, the q-axis voltage sequence, the d-axis current sequence, and the q-axis current sequence respectively to obtain a plurality of electrical parameter fluctuation coefficients, wherein the electrical parameter fluctuation coefficient is a ratio of a standard deviation to a mean value of a parameter in an electrical parameter sequence; performing parameter fluctuation calculation on the three-phase voltage instantaneous value sequence, the three-phase current instantaneous value sequence, the d-axis voltage sequence, the q-axis voltage sequence, the d-axis current sequence, and the q-axis current sequence respectively to obtain a plurality of electrical parameter fluctuation coefficients, wherein the electrical parameter fluctuation coefficient is a ratio of a standard deviation to a mean value of a parameter in an electrical parameter sequence;

[0015] Specifically, the running state fluctuation degree is determined based on the set of collected electrical parameter sequences. The running state fluctuation degree is used to represent the stability and change intensity of the electrical running state of the grid-connected point of the inverter-type power supply in a specific time period. This index is obtained by calculating the statistical fluctuation characteristics of each key electrical parameter sequence and comprehensively considering the influence weight of different parameters on stability. The fundamental purpose of determining the running state fluctuation degree is to realize quantitative perception and accurate evaluation of the grid running condition, thereby providing a scientific basis for intelligent adjustment of the control strategy.

[0016] First, a complete set of electrical parameter sequences is obtained. The set of electrical parameter sequences includes a three-phase voltage instantaneous value sequence, a three-phase current instantaneous value sequence, a d-axis voltage sequence, a q-axis voltage sequence, a d-axis current sequence, and a q-axis current sequence. Each sequence is composed of electrical parameter instantaneous values arranged in chronological order within a historical time zone.

[0017] Second, parameter fluctuation calculation is performed on each electrical parameter sequence. The fluctuation coefficient of each independent electrical parameter sequence is calculated. The specific calculation method of the electrical parameter fluctuation coefficient is as follows: take the standard deviation of all values in the electrical parameter sequence, and then divide it by the arithmetic mean value, i.e., obtain the ratio of the standard deviation to the mean value. This ratio, as a dimensionless index, can effectively eliminate the influence of the dimension difference of different electrical parameters and accurately reflect the relative fluctuation degree of each electrical parameter in the time dimension. Through this step, a plurality of electrical parameter fluctuation coefficients corresponding to three-phase voltage, three-phase current, d-axis voltage, q-axis voltage, d-axis current, and q-axis current can be obtained.

[0018] Finally, a weighted summation is performed to calculate the operational fluctuation. Based on the correlation between each electrical parameter and the stability of the inverter power supply control, an appropriate weight is assigned to the fluctuation coefficient of each electrical parameter. The higher the correlation, the greater the weight assigned to the fluctuation coefficient. Specifically, each weight is pre-set according to the parameter's impact on the stability of the inverter power supply control. For example, the d-axis voltage and q-axis voltage, which have the most direct impact on voltage stability, can be assigned a higher weight, such as 0.25; three-phase voltage and three-phase current parameters are assigned a medium weight, such as 0.15; and d-axis current and q-axis current are assigned a relatively low weight, such as 0.05. The sum of all weight values ​​is 1 to ensure the normalization of the calculation results.

[0019] Then, by multiplying all electrical parameter fluctuation coefficients by their corresponding weights and summing the products, the operating state fluctuation is obtained, which comprehensively characterizes the stability of the operating state. This operating state fluctuation can fully reflect the operating state stability of the inverter power supply at the grid connection point, providing an important basis for the dynamic adjustment of subsequent control strategies.

[0020] S20: Construct a power grid control analyzer based on a deep feedforward neural network, perform control analysis within a preset time zone according to the electrical parameter sequence set, and output the first grid control parameters; Specifically, a power grid control analyzer is constructed based on a deep feedforward neural network, including: Based on the historical operation logs of the inverter power supply, several sample electrical parameter sequence sets and several sample network control parameters are collected as sample training datasets. The network control parameters include d-axis reference voltage, q-axis reference voltage, and reference phase. The sample training dataset is divided into P equal parts, and the first training set is constructed by randomly selecting P samples with replacement from the P samples training data. The first training set is obtained by iteratively selecting P samples, where P is an integer greater than 20. The deep feedforward neural network is trained under supervision using the P training sets until it converges, resulting in P power grid control analysis units, which are then combined to obtain a power grid control analyzer.

[0021] A power grid control analyzer is constructed based on a deep feedforward neural network. The deep feedforward neural network is an artificial neural network model containing an input layer, multiple hidden layers, and an output layer. Information flows unidirectionally from the input layer to the output layer, and the layers are fully connected via weight matrices. Complex function mappings are achieved through nonlinear activation functions. The power grid control analyzer is an intelligent analyzer trained from this deep feedforward neural network. It automatically analyzes and generates initial grid control parameters adapted to the current operating state based on a set of real-time acquired electrical parameter sequences.

[0022] First, a sample training dataset is constructed. Based on the historical operation logs of the inverter power supply, a large number of historical operation data samples are collected. Each sample contains a set of sample electrical parameter sequences and its corresponding sample network control parameters. The sample electrical parameter sequence sets include time-series data of parameters such as three-phase voltage and current, and d / q-axis voltage and current. The sample network control parameters are control commands that have been verified as excellent at the corresponding historical moments, including d-axis reference voltage, q-axis reference voltage, and reference phase. These paired input-output data together constitute the sample training dataset for training the power supply network control analyzer.

[0023] Secondly, multiple differentiated training sets are constructed. The complete sample training dataset is divided into P subsets, where P is an integer greater than 20. Then, P random samplings with replacement are performed, each sampling randomly selecting one subset from the P subsets to form a training set. This results in P training sets with subtle differences in data distribution. By introducing randomness at the data level, the model units obtained based on different training sets exhibit predictive diversity.

[0024] Specifically, each power grid control analysis unit mainly consists of an input layer, a feature abstraction layer, and a control parameter output layer. The input layer receives a standardized sequence vector of electrical parameters, containing the instantaneous values ​​of three-phase voltages, three-phase currents, and the sequence data of d-axis voltage, q-axis voltage, d-axis current, and q-axis current within the historical time zone. The feature abstraction layer employs a deep fully connected neural network structure with multiple hidden layers. The number of neurons in the hidden layers is configured according to the dimensions of the input features. Each neural network layer uses the ReLU activation function to introduce nonlinear transformation capabilities, and Dropout layers are embedded between network layers with a dropout rate set between 0.3 and 0.5 to effectively suppress model overfitting and improve its generalization performance. The output layer uses a linear activation function to map the final abstract features to three consecutive control parameter values, serving as the d-axis reference voltage, q-axis reference voltage, and reference phase, respectively.

[0025] During training, key hyperparameters included a learning rate of 0.0005, 200 training epochs, and a batch size of 32. The learning rate was set to balance training stability and convergence accuracy; the number of training epochs ensured the model fully learned the control patterns in the data; and the batch size balanced training efficiency with gradient stability. Specifically, supervised learning was used, and the resulting P training sets were divided into training, validation, and test sets in a 7:2:1 ratio.

[0026] Furthermore, P identical deep feedforward neural networks are independently trained under supervised supervision. The network weight parameters are iteratively optimized using backpropagation and the Adam optimizer. The mean square error loss function is used to measure the deviation between the predicted control parameters and the actual optimal control parameters. The training process is monitored using a validation set. Training is terminated when the validation set loss function value no longer decreases after multiple rounds and the root mean square error of the model prediction reaches a predetermined threshold (e.g., below 0.05), resulting in a converged power grid control analysis unit. This unit effectively captures the complex nonlinear relationship between the electrical operating state and the optimal control command, achieving accurate prediction of grid control parameters. Finally, these P trained power grid control analysis units are combined to form a complete power grid control analyzer. In practical applications, this analyzer can integrate the outputs of multiple units to improve the accuracy and reliability of control.

[0027] Further, based on the electrical parameter sequence set, control analysis is performed within a preset time zone to output the first network control parameters, including: The ratio of the calculated operational fluctuation to the preset standard fluctuation is set as the unit selection compensation coefficient; The number of adaptation units selected, K, is obtained by taking the integer part of the product of the unit selection compensation coefficient and the initial number of units selected, where the initial number of units selected is 5, and K is greater than or equal to 2 and less than or equal to P. K units are randomly selected from the P power grid control analysis units of the power grid control analyzer. Control analysis is performed within a preset time zone based on the electrical parameter sequence set. The first grid control parameters are obtained by mean fitting of the K output results.

[0028] First, the calculation unit selects a compensation coefficient. This coefficient is the ratio of the operating state fluctuation to the preset standard state fluctuation. The preset standard state fluctuation is a benchmark value obtained based on historical stable operating data, used to characterize the fluctuation level of the power grid control under ideal and stable operating conditions. By selecting the compensation coefficient through the calculation unit, the degree of deviation of the current operating state from the standard state can be quantified.

[0029] Secondly, the number of adaptive units selected, K, is determined. The calculated unit selection compensation coefficient is multiplied by the initial number of units selected, and the product is rounded up to obtain the final value of K. Specifically, the initial number of units selected is preset to 5. Simultaneously, the value of K is constrained to a range greater than or equal to 2 and less than or equal to the total number of power grid control analysis units P, allowing the number of units involved in the analysis to adapt to the stability of the power grid control state: when the operating state fluctuates significantly, the unit selection compensation coefficient increases, and the value of K increases accordingly, integrating the results of more analysis units to improve the robustness of the decision; when the operating state is stable, fewer units are used to improve computational efficiency.

[0030] Finally, integrated analysis is performed and the results are output. From the P power grid control analysis units included in the power grid control analyzer, K power grid control analysis units are randomly selected. These K power grid control analysis units perform parallel control analysis on the electrical parameter sequence set within the current preset time zone, with each power grid control analysis unit independently outputting a set of grid control parameters. Then, the control parameter results output by the K power grid control analysis units are averaged, that is, the average of all d-axis reference voltage values, the average of all q-axis reference voltage values, and the average of all reference phase values ​​are taken. The three average values ​​obtained are used as the integrated, final first grid control parameters.

[0031] In summary, by using this random selection and mean fitting mechanism, multiple differentiated power grid control analysis units can be comprehensively utilized to balance individual deviations, thereby outputting more stable and reliable control commands.

[0032] S30: Build a power distribution network simulation topology model based on PSCAD, optimize the network control parameters according to the electrical parameter sequence set, and generate the second network control parameters; First, a power distribution network simulation topology model is built based on PSCAD, including: In the PSCAD simulation environment, inverter power supply components, line impedance components, load components, and switching components are selected and configured according to the topology of the physical power distribution network. Based on the actual electrical connection relationship of the physical power distribution network, the inverter power supply element, line impedance element, load element and switching element are electrically connected to form a closed-loop simulation network; Configure a controllable network-based control interface for the inverter power element in the simulation network, wherein the control interface allows external input of control parameters and drives the operation of the inverter power element.

[0033] PSCAD, short for Power Systems Computer Aided Design, is a professional electromagnetic transient simulation software widely used in power system simulation and analysis. This software accurately simulates the electromagnetic characteristics of various components in a power system under transient and steady-state conditions, providing a high-precision digital simulation environment for power analysis and control strategy verification. A distribution network simulation topology model is built based on PSCAD. This model is a digital twin that accurately reflects the electrical structure and dynamic characteristics of the actual distribution network. It includes not only the static topological connections of the network but also embeds dynamic mathematical models of various components. It is used to reproduce and predict the behavior of the distribution network under different operating conditions and disturbances in a virtual environment, providing a reliable simulation platform for offline testing and optimization of network control parameters. Specifically, the steps for building this distribution network simulation topology model are as follows: First, component selection and parameter configuration are performed. In the PSCAD simulation software environment, based on the actual topology of the target physical distribution network, corresponding components are selected and configured from the software component library. These include inverter power supply components for simulating distributed generation, line impedance components for simulating transmission line characteristics, load components for simulating electrical loads, and switching components for simulating grid topology changes or fault operations. After selection, the electrical parameters of each component are configured in detail according to the nameplate parameters or field measurement data of the corresponding equipment in the actual physical distribution network. Configuration includes setting the rated capacity and DC voltage of the inverter power supply components, setting the resistance and reactance values ​​of the line impedance components, and setting the active and reactive power of the load components. Detailed parameter configuration ensures that the constructed distribution network simulation topology model accurately reflects the static operating characteristics and dynamic response characteristics of the actual physical distribution network.

[0034] Secondly, electrical connections and network construction are performed. Following the actual electrical connections of the physical power distribution network, the configured components are connected. The connection process must conform to the circuit connection rules described by Kirchhoff's laws. For example, the output terminal of the inverter power supply component is connected to the load component through line impedance components, and switching components are connected at appropriate locations to simulate normal operation or fault isolation scenarios. Ultimately, a closed-loop simulation network with a complete topology is formed, capable of realizing power transmission and distribution.

[0035] Finally, control interface configuration is required. To enable real-time control of inverter power sources in the distribution network simulation topology model by external control strategies, a controllable network-type control interface needs to be configured for the inverter power source components in the simulation network. The network-type control interface, as a functional module in the distribution network simulation topology model, has the ability to receive external input control parameters. After receiving the control parameters, this interface can directly drive the operation of the inverter power source components, changing the amplitude and phase of their output voltage, thereby simulating the actual operating behavior of the inverter power source under a specific control strategy.

[0036] Further, based on the electrical parameter sequence set, network control parameters are optimized to generate second network control parameters, including: Obtain the grid control parameter adjustment threshold for the inverter power supply, and randomly select several initial grid control parameters within the grid control parameter adjustment threshold. Several network control schemes are obtained by combining the electrical parameter sequence set and the several initial network control parameters respectively; Within the power distribution network simulation topology model, network control simulation is performed according to the several network control schemes, and several sets of operational stability indices are output, including transient stability indices, voltage stability indices, and frequency stability indices. Based on the aforementioned initial network control parameters and several sets of operational stability indicators, the network control parameters are optimized to generate the second network control parameters.

[0037] Specifically, optimizing network control parameters based on the electrical parameter sequence set is a process of searching for the optimal combination of control parameters that enables the distribution network simulation topology model to achieve the best overall stability, based on optimization algorithms. This ultimately generates a second set of network control parameters. These second set of control parameters are a set of control commands selected based on clear stability indicators after testing and evaluating a large number of candidate parameter combinations in the distribution network simulation topology model built in PSCAD. Their key feature is that they have been verified by the distribution network simulation topology model, possessing high reliability and scenario adaptability, and can serve as an important supplement and verification to the data-driven first set of network control parameters.

[0038] First, the adjustment thresholds for the grid control parameters of the inverter power supply are obtained. These thresholds define the safe upper and lower limits for the adjustment of three control parameters: d-axis reference voltage, q-axis reference voltage, and reference phase. The settings are based on factors including the inverter's hardware tolerance, grid operation procedures, and stability requirements. For example, the d-axis reference voltage threshold can be set to 0.8 to 1.2 times the rated value, the q-axis reference voltage threshold to -0.3 to +0.3 times the rated voltage, and the reference phase threshold to -π / 6 to +π / 6 radians. Within the feasible region defined by these grid control parameter adjustment thresholds, several initial grid control parameters are selected using a random sampling method. These initial grid control parameters are then used as the starting point for the optimization algorithm to achieve a broad exploration of the parameter space.

[0039] Secondly, the real-time collected electrical parameter sequences are combined with each initial network control parameter to form several complete network control schemes. Each network control scheme contains a specific set of control parameters and the corresponding real-time operating state of the power grid, which can be used as an independent test case and input into the distribution network simulation topology model to evaluate the effect of applying this set of control parameters under that specific operating state.

[0040] Then, several network control schemes are sequentially input into a pre-built distribution network simulation topology model for network control simulation. After each simulation run, a set of operational stability indices is recorded and output. This set of operational stability indices includes stability evaluation criteria across multiple dimensions, specifically transient stability indices reflecting power and angle fluctuations, voltage stability indices characterizing voltage deviation and recovery capability, and frequency stability indices measuring frequency offset. Through simulation, the control effectiveness of each scheme in the simulated environment can be quantitatively evaluated.

[0041] Finally, based on several initial network construction control parameters and their corresponding sets of operational stability indices, the network construction control parameters are optimized. Specifically, this optimization process aims to improve the stability of network construction control. By analyzing the mapping relationship between different parameter combinations and stability indices, the optimal combination of network construction control parameters that achieves the best overall stability of network construction control is found from all candidate schemes, and this combination is determined as the final second network construction control parameter.

[0042] Specifically, based on the aforementioned initial network control parameters and several sets of operational stability indicators, network control parameters are optimized to generate second network control parameters, including: Several system stability coefficients are obtained based on the aforementioned set of operational stability indicators; The initial network control parameters are set as the initial solution, and the initial network control parameters are arranged in descending order of system stability coefficient to generate an initial solution sequence. The first N solutions of the initial solution sequence are selected as optimal solutions, and the last M solutions are selected as inferior solutions. The sum of M and N is the number of initial solutions, M is L times N, and L is greater than or equal to 10. Randomly equal-cluster the M inferior solutions with the N superior solutions as the center to obtain N solution sets. Within the N solution sets, the inferior solutions in the solution sets are adjusted according to the preset optimization step size with the superior solutions as the adjustment direction to obtain N updated solution sets. Identify the N updated solution sets. Within the same updated solution set, if the system stability coefficient of a suboptimal solution is greater than or equal to the system stability coefficient of a superior solution, then replace the superior solution with the suboptimal solution. Continue iterative optimization until the preset number of optimization convergences is reached, output N current updated solution sets, and select the optimal solution with the largest system stability coefficient in the N current updated solution sets as the second network control parameter.

[0043] Specifically, the ratio of the preset standard state fluctuation to the operating state fluctuation is set as the convergence number compensation coefficient, and the product of the convergence number compensation coefficient and the preset standard optimization convergence number is rounded to obtain the preset optimization convergence number.

[0044] First, based on several sets of operational stability indicators, a comprehensive evaluation is performed on each set of operational stability indicators to calculate the system stability coefficient. The system stability coefficient = α × normalized value of transient stability indicator + β × normalized value of voltage stability indicator + γ × normalized value of frequency stability indicator, where α, β, and γ are weighting coefficients, and α + β + γ = 1. Each weight is pre-set according to the relative importance of each stability type in network control. For example, in a scenario where voltage support is the primary objective, the voltage stability weight β can be set to 0.5, the transient stability weight α to 0.3, and the frequency stability weight γ to 0.2. The normalized value of each indicator is obtained through extreme value normalization, specifically mapping the original value of each stability indicator to the [0, 1] interval. The calculation formula is: Normalized value of indicator = (Original value - Historical minimum value of the indicator) / (Historical maximum value of the indicator - Historical minimum value of the indicator). This system stability coefficient is used to quantitatively characterize the overall stability of the distribution network simulation topology model under the corresponding network control scheme; a higher value indicates better stability.

[0045] Secondly, the initial network control parameters are considered as initial solutions to the optimization problem. Based on the calculated system stability coefficients, several initial network control parameters are arranged in descending order to form an initial solution sequence. Subsequently, the top N solutions in this sequence are marked as optimal solutions, and the bottom M solutions are marked as inferior solutions. The sum of parameters M and N equals the total number of initial solutions, and the value of M is set to L times N, where L is an integer greater than or equal to 10. This allocation method ensures that the number of inferior solutions far exceeds the number of optimal solutions, providing ample adjustment space for subsequent optimization operations.

[0046] Then, centering on the N optimal solutions, the M inferior solutions are distributed around the N optimal solutions using a random equal-value clustering method, forming N solution sets. Within each solution set, using the optimal solutions as the adjustment direction, the parameter values ​​of all inferior solutions within that solution set are updated and adjusted according to a preset optimization step size, thus obtaining N updated solution sets. The optimization step size control parameter is a key parameter for the adjustment magnitude in each iteration, and is set according to the range of the mesh construction control parameter adjustment threshold, such as 1% to 5% of the d-axis reference voltage adjustment threshold, 1% to 5% of the q-axis reference voltage adjustment threshold, and 1% to 5% of the reference phase adjustment threshold. This setting method ensures that the parameter adjustment process has sufficient finesse for local search while maintaining a relatively fast convergence speed.

[0047] Furthermore, identify N updated solution sets. Within the same solution set, if the system stability coefficient corresponding to an adjusted inferior solution is found to be greater than or equal to the system stability coefficient of the current superior solution, then replace the original superior solution with the inferior solution to ensure that the position of the superior solution can move towards a better direction.

[0048] Finally, the clustering, adjustment, and replacement steps described above are repeated until the preset number of optimization convergences is reached. After the iteration process, N current updated solution sets are output, and the optimal solution with the largest system stability coefficient is selected from these N solution sets and determined as the final second network control parameter. The ratio of the preset standard state volatility to the operating state volatility is set as the convergence number compensation coefficient. The product of the convergence number compensation coefficient and the preset standard optimization convergence number is rounded to obtain the preset optimization convergence number. Specifically, the preset standard state volatility is a benchmark value obtained based on historical stable operating data, representing the volatility level of the network control under ideal stable operating conditions. The operating state volatility is a quantitative indicator reflecting the current stability of the network control, calculated through real-time electrical parameter sequences. The ratio of these two values ​​is defined as the convergence number compensation coefficient. This coefficient directly reflects the degree of deviation of the current operating state from the standard state. When the operating state volatility is greater than the preset standard state volatility, it indicates that the network control is in a state of significant disturbance or instability; in this case, the convergence number compensation coefficient is greater than 1. By multiplying the convergence compensation coefficient by a preset standard optimization convergence number, such as 50, and rounding the product, a preset optimization convergence number suitable for the current operating conditions can be obtained. When the operating state is stable, the number of iterations can be automatically reduced to improve computational efficiency; when the operating state experiences significant fluctuations, the number of iterations is automatically increased to provide more sufficient search time for the optimization algorithm, thereby ensuring that high-precision second network control parameters can still be obtained under complex operating conditions.

[0049] Therefore, the obtained second network control parameters are control commands that enable the system stability to reach local or global optimality after multiple rounds of iterative optimization.

[0050] S40: Based on the analysis of the electrical parameter sequence set, determine the operating state fluctuation. According to the operating state fluctuation, set a dynamic fitting strategy to control and fit the first grid control parameter and the second grid control parameter to obtain an adapted grid control strategy, and regulate the inverter power supply in the preset time zone.

[0051] Specifically, the fluctuation of operating status is determined based on the analysis of electrical parameter sequence sets, as shown in the calculation method in step S10.

[0052] Further, based on the operational state fluctuation, a dynamic fitting strategy is set to control and fit the first network construction control parameters and the second network construction control parameters to obtain an adapted network construction control strategy, including: The predicted optimization accuracy is obtained by matching the preset number of optimization convergences, wherein the optimization accuracy is positively correlated with the number of optimization convergences. The second initial weight is optimized and adjusted based on the prediction optimization accuracy to obtain the second adaptation weight. The second adaptation weight is the confidence weight of the second network control parameter and is positively correlated with the prediction optimization accuracy. The second initial weight is 0.4, and the second adaptation weight is greater than or equal to 0.3 and less than or equal to 0.6. The first adaptation weight is obtained by subtracting the second adaptation weight from 1, wherein the first adaptation weight is the confidence weight of the first network control parameter; Based on the first adaptation weight and the second adaptation weight, the first network construction control parameters and the second network construction control parameters are fitted to obtain the adaptive network construction control strategy.

[0053] First, based on the preset number of convergence iterations used in the final optimization process, a table comparing the number of convergence iterations with the optimization accuracy is constructed using historical data for matching analysis. The more simulation iterations, the more thorough the search, and the more reliable the results are generally. Therefore, the predicted optimization accuracy is positively correlated with the preset number of convergence iterations.

[0054] Secondly, based on the initial weight of 0.4, the predicted optimization accuracy obtained in the previous step is optimized and adjusted to obtain the final second adaptation weight. The specific adjustment method is as follows: calculate the ratio of the predicted optimization accuracy to the preset standard optimization accuracy, set this ratio as the adjustment coefficient, and multiply it by the initial weight to obtain the second adaptation weight. The preset standard optimization accuracy is a benchmark value, set according to the correspondence between the number of convergence attempts and the accuracy of the obtained parameters in historical simulation data. For example, the accuracy corresponding to the minimum number of convergence attempts required to achieve satisfactory optimization results is set to 0.8 using statistical analysis. This second adaptation weight represents the confidence level of the second network control parameters in the final decision. Its value is positively correlated with the predicted optimization accuracy and is constrained to the range of 0.3 to 0.6. When the simulation optimization process is considered reliable, the weight of the second network control parameters can reach up to 60%; when the reliability is low, its weight will not be lower than 30%.

[0055] For example, the standard optimization accuracy is set to 0.8, and the second initial weight is 0.4. If the current predicted optimization accuracy is 0.7, then the adjustment coefficient is 0.7 / 0.8 = 0.875, and the second adaptation weight is 0.875 × 0.4 = 0.35. This second adaptation weight is constrained to the range of 0.3 to 0.6, ensuring that the confidence level of the second network control parameters can dynamically and reasonably fluctuate according to the predicted reliability of the simulation optimization process.

[0056] Furthermore, by subtracting the second adaptation weight from 1, the first adaptation weight is obtained, which represents the confidence weight of the first network control parameter. This ensures that the sum of the weights of the two control parameters is always 1, thus forming a complete decision-making basis.

[0057] Finally, using the calculated first and second adaptation weights, the first and second network construction control parameters are weighted and fused respectively, resulting in a final network construction control strategy that combines the advantages of data-driven rapid response and physical model verification, and is adapted to the current operating state. Specifically, the final network construction control strategy = first adaptation weight × first network construction control parameter + second adaptation weight × second network construction control parameter.

[0058] In summary, the embodiments of this application have at least the following technical effects: Compared to existing technologies, this application firstly achieves deep integration of data-driven and physical models by constructing a dual-path parallel optimization mechanism of a deep feedforward neural network and a PSCAD simulation model, significantly improving the accuracy and reliability of the control strategy. Secondly, by introducing operational state fluctuation assessment and dynamic fitting strategies, the control parameters can be adaptively adjusted according to the real-time operating state of the power grid, effectively enhancing the adaptability under complex operating conditions. Finally, an intelligent parameter optimization and fitting method is adopted to replace the traditional control parameter tuning method that relies on manual experience, greatly improving the generation efficiency and optimization accuracy of the control strategy.

[0059] In summary, by employing a dual-parallel mechanism of rapid neural network response and precise verification through simulation models, combined with an adaptive fitting strategy based on operational status awareness, online intelligent optimization of control parameters is achieved. This effectively improves the stability, adaptability, and control accuracy of inverter power supplies under complex operating conditions, demonstrating good engineering applicability and providing reliable technical support for the stable operation of the power grid under conditions of high-proportion renewable energy access.

[0060] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0061] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0062] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.

Claims

1. A grid-type control method for inverter power supplies based on artificial neural networks, characterized in that the method... include: Collect the electrical parameter sequence set of the inverter power supply grid connection point within the historical time zone; A power grid control analyzer is constructed based on a deep feedforward neural network. It performs control analysis within a preset time zone according to the electrical parameter sequence set and outputs the first grid control parameters. A power distribution network simulation topology model is built based on PSCAD. The network construction control parameters are optimized according to the electrical parameter sequence set to generate the second network construction control parameters. Based on the analysis of the electrical parameter sequence set, the operating state fluctuation is determined. According to the operating state fluctuation, a dynamic fitting strategy is set to control and fit the first grid control parameter and the second grid control parameter to obtain an adapted grid control strategy, which is used to regulate the inverter power supply in the preset time zone.

2. The inverter-type power supply grid-based control method based on artificial neural networks according to claim 1, characterized in that, Electrical parameters of the inverter power supply grid connection point are collected at several consecutive time nodes within a historical time zone to obtain multiple electrical parameter sequences, which are used as an electrical parameter sequence set. The electrical parameters include instantaneous values ​​of three-phase voltage, instantaneous values ​​of three-phase current, d-axis voltage, q-axis voltage, d-axis current, and q-axis current.

3. The inverter-type power supply grid-based control method based on artificial neural networks according to claim 2, characterized in that, Based on the analysis of the electrical parameter sequence set, the operating status fluctuation is determined, including: Obtain an electrical parameter sequence set, wherein the electrical parameter sequence set includes a three-phase voltage instantaneous value sequence, a three-phase current instantaneous value sequence, a d-axis voltage sequence, a q-axis voltage sequence, a d-axis current sequence, and a q-axis current sequence; The parameter fluctuations of the three-phase voltage instantaneous value sequence, the three-phase current instantaneous value sequence, the d-axis voltage sequence, the q-axis voltage sequence, the d-axis current sequence, and the q-axis current sequence are calculated respectively to obtain multiple electrical parameter fluctuation coefficients. The electrical parameter fluctuation coefficient is the ratio of the standard deviation to the mean of the parameter in the electrical parameter sequence. Weights are assigned based on the correlation between parameters and the stability of inverter power supply control. The fluctuation coefficients of the multiple electrical parameters are weighted and summed to obtain the operating state fluctuation. The weights are positively correlated with the correlation.

4. The inverter-type power supply grid-based control method based on artificial neural networks according to claim 1, characterized in that, A power grid control analyzer based on a deep feedforward neural network is constructed, including: Based on the historical operation logs of the inverter power supply, several sample electrical parameter sequence sets and several sample network control parameters are collected as sample training datasets. The network control parameters include d-axis reference voltage, q-axis reference voltage, and reference phase. The sample training dataset is divided into P equal parts, and the first training set is constructed by randomly selecting P samples with replacement from the P samples training data. The first training set is obtained by iteratively selecting P samples, where P is an integer greater than 20. The deep feedforward neural network is trained under supervision using the P training sets until it converges, resulting in P power grid control analysis units, which are then combined to obtain a power grid control analyzer.

5. The inverter-type power supply grid-based control method based on artificial neural networks according to claim 4, characterized in that, Based on the electrical parameter sequence set, control analysis is performed within a preset time zone, and the first network control parameters are output, including: The ratio of the calculated operational fluctuation to the preset standard fluctuation is set as the unit selection compensation coefficient; The number of adaptation units selected, K, is obtained by taking the integer part of the product of the unit selection compensation coefficient and the initial number of units selected, where the initial number of units selected is 5, and K is greater than or equal to 2 and less than or equal to P. K units are randomly selected from the P power grid control analysis units of the power grid control analyzer. Control analysis is performed within a preset time zone based on the electrical parameter sequence set. The first grid control parameters are obtained by mean fitting of the K output results.

6. The inverter-type power supply grid-based control method based on artificial neural networks according to claim 1, characterized in that, A power distribution network simulation topology model was built based on PSCAD, including: In the PSCAD simulation environment, inverter power supply components, line impedance components, load components, and switching components are selected and configured according to the topology of the physical power distribution network. Based on the actual electrical connection relationship of the physical power distribution network, the inverter power supply element, line impedance element, load element and switching element are electrically connected to form a closed-loop simulation network; Configure a controllable network-based control interface for the inverter power element in the simulation network, wherein the control interface allows external input of control parameters and drives the operation of the inverter power element.

7. The inverter-type power supply grid-based control method based on artificial neural networks according to claim 6, characterized in that, Based on the electrical parameter sequence set, network control parameters are optimized to generate second network control parameters, including: Obtain the grid control parameter adjustment threshold for the inverter power supply, and randomly select several initial grid control parameters within the grid control parameter adjustment threshold. Several network control schemes are obtained by combining the electrical parameter sequence set and the several initial network control parameters respectively; Within the power distribution network simulation topology model, network control simulation is performed according to the several network control schemes, and several sets of operational stability indices are output, including transient stability indices, voltage stability indices, and frequency stability indices. Based on the aforementioned initial network control parameters and several sets of operational stability indicators, the network control parameters are optimized to generate the second network control parameters.

8. The inverter-type power supply grid-based control method based on artificial neural networks according to claim 7, characterized in that, Based on the aforementioned initial network control parameters and several sets of operational stability indicators, network control parameters are optimized to generate second network control parameters, including: Several system stability coefficients are obtained based on the aforementioned set of operational stability indicators; The initial network control parameters are set as the initial solution, and the initial network control parameters are arranged in descending order of system stability coefficient to generate an initial solution sequence. The first N solutions of the initial solution sequence are selected as optimal solutions, and the last M solutions are selected as inferior solutions. The sum of M and N is the number of initial solutions, M is L times N, and L is greater than or equal to 10. Randomly equal-cluster the M inferior solutions with the N superior solutions as the center to obtain N solution sets. Within the N solution sets, the inferior solutions in the solution sets are adjusted according to the preset optimization step size with the superior solutions as the adjustment direction to obtain N updated solution sets. Identify the N updated solution sets. Within the same updated solution set, if the system stability coefficient of a suboptimal solution is greater than or equal to the system stability coefficient of a superior solution, then replace the superior solution with the suboptimal solution. Continue iterative optimization until the preset number of optimization convergences is reached, output N current updated solution sets, and select the optimal solution with the largest system stability coefficient in the N current updated solution sets as the second network control parameter.

9. The inverter-type power supply grid-based control method based on artificial neural networks according to claim 8, characterized in that, The ratio of the preset standard state fluctuation to the operating state fluctuation is set as the convergence number compensation coefficient. The product of the convergence number compensation coefficient and the preset standard optimization convergence number is rounded to obtain the preset optimization convergence number.

10. The inverter-type power supply grid-based control method based on artificial neural networks according to claim 9, characterized in that, Based on the operational state fluctuation, a dynamic fitting strategy is set to control and fit the first and second network construction control parameters to obtain an adapted network construction control strategy, including: The predicted optimization accuracy is obtained by matching the preset number of optimization convergences, wherein the optimization accuracy is positively correlated with the number of optimization convergences. The second initial weight is optimized and adjusted based on the prediction optimization accuracy to obtain the second adaptation weight. The second adaptation weight is the confidence weight of the second network control parameter and is positively correlated with the prediction optimization accuracy. The second initial weight is 0.4, and the second adaptation weight is greater than or equal to 0.3 and less than or equal to 0.

6. The first adaptation weight is obtained by subtracting the second adaptation weight from 1, wherein the first adaptation weight is the confidence weight of the first network control parameter; Based on the first adaptation weight and the second adaptation weight, the first network construction control parameters and the second network construction control parameters are fitted to obtain the adaptive network construction control strategy.

Images