Voltage fitting method and device without electrical model, equipment and medium

Abstract

The invention provides a voltage fitting method without an electrical model. The method comprises the steps that an active power matrix and a reactive power matrix of a low-voltage power distribution network are acquired, and an input data set is constructed; inputting the input data set into a width learning model to obtain an output voltage of the low-voltage power distribution network and taking the output voltage as an output data set of the width learning model; acquiring an original input data set of the low-voltage power distribution network, performing dynamic principal component analysis on the original input data set to obtain a dynamic latent variable data set, and inputting the dynamic latent variable data set instead of the original input data set into the width learning model for training; extracting a dynamic feature weight in the dynamic latent variable data set, and replacing a random weight of the width learning model with the dynamic feature weight to obtain a width learning voltage fitting model; and performing parameter optimization according to a preset particle swarm algorithm to obtain an optimal parameter of the voltage fitting model. According to the technical scheme of the embodiment, the precision of voltage fitting in the low-voltage power distribution network can be remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of low-voltage distribution network processing, and in particular to a voltage fitting method, device, equipment and medium without an electrical model. Background Art

[0002] In the operation of low-voltage distribution networks, planning and voltage control are key technical supports for system stability and efficient operation. The core of this approach lies in accurately calculating voltage based on the distribution network topology and line parameters to support power flow analysis, optimization, and scheduling. Data loss directly impairs the accuracy of distribution network model construction, primarily due to the ambiguity of network topology and the unknown nature of line parameters. These factors make traditional power flow calculation and voltage analysis methods based on network parameters difficult to effectively apply. With the increasing complexity of distribution networks, particularly with the diversification of load types and the increasing proportion of distributed energy resources, constructing detailed and accurate electrical models has become increasingly difficult. This approach is not only time-consuming but also requires high equipment investment and maintenance costs, limiting its widespread application in practical low-voltage distribution networks. Summary of the Invention

[0003] The present invention aims to solve at least one of the technical problems existing in the prior art. To this end, the present invention proposes a voltage method, device, electronic device, and storage medium without an electrical model, which can significantly improve the accuracy of voltage fitting in a low-voltage distribution network.

[0004] In a first aspect, an embodiment of the present invention provides a voltage fitting method without an electrical model, comprising:

[0005] Obtaining an active power matrix and a reactive power matrix of a low-voltage distribution network, and constructing an input data set according to the active power matrix and the reactive power matrix;

[0006] Inputting the input data set into a width learning model to obtain the output voltage of the low-voltage distribution network, and using the output voltage as the output data set of the width learning model;

[0007] Obtaining an original input data set of the low-voltage distribution network, performing dynamic principal component analysis on the original input data set to obtain a dynamic latent variable data set, and inputting the dynamic latent variable data set into the width learning model for training instead of the original input data set;

[0008] Extracting dynamic feature weights from the dynamic latent variable dataset, and replacing random weights of the width learning model with the dynamic feature weights to obtain a width learning voltage fitting model of the low-voltage distribution network;

[0009] Parameter optimization is performed on the width learning voltage fitting model according to a preset particle swarm algorithm to obtain optimal parameters of the voltage fitting model.

[0010] In some embodiments of the present invention, obtaining the active power matrix and the reactive power matrix of the low-voltage distribution network includes:

[0011] Obtain a subnode set, an active output matrix, a three-phase current effective value square matrix, a three-phase resistance, and a three-phase active net load matrix of the low-voltage distribution network, and calculate the active power matrix based on the subnode set, the active output matrix, the three-phase current effective value square matrix, the three-phase resistance, and the three-phase active net load matrix;

[0012] Obtaining a reactive output matrix, a three-phase reactance, and a three-phase reactive net load matrix of the low-voltage distribution network, and calculating the reactive power matrix based on the set of child nodes, the reactive output matrix, the three-phase reactance, the three-phase current, and the reactive net load matrix;

[0013] Obtain the branch voltage, three-phase impedance matrix and the conjugate matrix of the three-phase impedance matrix of the low-voltage distribution network, and calculate the voltage effective value square matrix according to the branch voltage, the three-phase resistance, the three-phase active power matrix, the three-phase reactive power matrix, the three-phase impedance matrix, the three-phase current and the conjugate matrix.

[0014] In some embodiments of the present invention, obtaining the width learning voltage fitting model of the low-voltage distribution network includes:

[0015] Acquire power data and voltage data of smart meters in the low-voltage distribution network, use the power data as an input data set of the width learning model, and use the voltage data as an output data set of the width learning model;

[0016] Inputting the input data set into the width learning model to obtain feature nodes of the width learning model, wherein the width learning model includes an input layer, a feature layer, an enhancement layer, and an output layer;

[0017] Obtaining an activation function of a feature node of the width learning model, a mapping weight of a feature layer, and a feature layer bias, and constructing a feature node matrix of the width learning model according to the input layer data, the activation function, the mapping weight, and the first feature bias, wherein the feature node represents the feature layer of the width learning model;

[0018] Obtain the enhancement node deviation of the enhancement layer and the activation function of the enhancement node of the feature layer, and construct an enhancement node matrix of the width learning model according to the activation function and the feature node, wherein the enhancement node matrix represents the enhancement layer of the width learning model.

[0019] In some embodiments of the present invention, obtaining a dynamic latent variable dataset includes:

[0020] Obtaining the input data matrix of the original input data set, the mapping weight matrix of the dynamic latent variables, the autoregressive coefficients, and the error term information;

[0021] A dynamic latent variable matrix is ​​obtained according to the input data matrix, the mapping weight matrix, the autoregressive coefficients and the error term information.

[0022] In some embodiments of the present invention, after obtaining the dynamic latent variable matrix, the method further includes:

[0023] Constructing a dynamic time series feature weight matrix according to the autoregressive coefficient, the dynamic latent variable matrix and the input matrix;

[0024] The mapping weight is obtained according to the dynamic time series feature weight matrix.

[0025] In some embodiments of the present invention, the dynamic latent variable matrix is ​​expressed by the following formula:

[0026]

[0027] Among them, X DL，t is the dynamic latent variable matrix, (X t ) T is the input data matrix at the first preset moment, is the mapping weight matrix.

[0028] In some embodiments of the present invention, the performing parameter optimization on the voltage fitting model according to a preset particle swarm algorithm includes:

[0029] Obtaining the number of dynamic latent variables in the dynamic latent variable data set and the number of enhanced nodes of the enhanced node;

[0030] Calculating a root mean square error after fitting the width learning model according to the number of dynamic latent variables, the number of groups of feature nodes, and the number of enhanced nodes;

[0031] Obtain the constraints of the width learning model, use the first parameter as a control variable, the root mean square error as an objective function, and construct an optimization model of the width learning model in combination with the constraints.

[0032] In a second aspect, an embodiment of the present invention provides a voltage fitting device without an electrical model, comprising at least one control processor and a memory for communicating with the at least one control processor; the memory stores instructions that can be executed by the at least one control processor, and the instructions are executed by the at least one control processor so that the at least one control processor can execute the voltage fitting method without an electrical model as described in the first aspect above.

[0033] In a third aspect, an embodiment of the present invention provides an electronic device, comprising the voltage fitting device without an electrical model as described in the second aspect above.

[0034] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium storing computer-executable instructions, wherein the computer-executable instructions are used to execute the voltage fitting method without an electrical model as described in the first aspect above.

[0035] The voltage fitting method without an electrical model according to an embodiment of the present invention has at least the following beneficial effects:

[0036] The active power matrix and reactive power matrix of the low-voltage distribution network are obtained, and an input data set is constructed based on the active power matrix and the reactive power matrix; the input data set is input into the width learning model to obtain the output voltage of the low-voltage distribution network, and the output voltage is used as the output data set of the width learning model; the original input data set of the low-voltage distribution network is obtained, and a dynamic principal component analysis is performed on the original input data set to obtain a dynamic latent variable data set, and the dynamic latent variable data set is replaced by the original input data set and input into the width learning model for training; the dynamic feature weights in the dynamic latent variable data set are extracted, and the dynamic feature weights are replaced by the random weights of the width learning model to obtain a width learning voltage fitting model of the low-voltage distribution network; the parameters of the width learning voltage fitting model are optimized according to the preset particle swarm algorithm to obtain the optimal parameters of the voltage fitting model. According to the technical solution of this embodiment, the accuracy of voltage fitting in the low-voltage distribution network can be significantly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 is a flow chart of a voltage fitting method without an electrical model provided by one embodiment of the present invention;

[0038] Figure 2 A flowchart of obtaining the active power matrix and reactive power matrix of a low-voltage distribution network provided by an embodiment of the present invention;

[0039] Figure 3 This is a flow chart of obtaining a width learning voltage fitting model for a low-voltage distribution network provided by an embodiment of the present invention;

[0040] Figure 4is a flow chart of obtaining a dynamic latent variable data set provided by one embodiment of the present invention;

[0041] Figure 5 This is a flow chart after obtaining a dynamic latent variable matrix provided by one embodiment of the present invention;

[0042] Figure 6 This is a flow chart of optimizing parameters of a voltage fitting model according to a preset particle swarm algorithm provided by an embodiment of the present invention;

[0043] Figure 7 It is a structural diagram of a voltage fitting device without an electrical model provided by another embodiment of the present invention. DETAILED DESCRIPTION

[0044] The following describes embodiments of the present invention in detail. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended only to explain the present invention and are not to be construed as limiting the present invention.

[0045] In the description of the present invention, it should be understood that descriptions involving orientations, such as up, down, front, back, left, right, etc., indicating orientations or positional relationships, are based on the orientations or positional relationships shown in the accompanying drawings. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation. Therefore, they cannot be understood as limitations on the present invention.

[0046] In the description of the present invention, "several" means one or more, "many" means more than two, "greater than," "less than," and "exceed" are understood to exclude the number itself, while "above," "below," and "within" are understood to include the number itself. The use of "first" and "second" in the description is solely for the purpose of distinguishing technical features and should not be construed as indicating or implying relative importance, implicitly specifying the number of the indicated technical features, or implicitly specifying the order of the indicated technical features.

[0047] In the description of the present invention, unless otherwise clearly defined, terms such as setting, installing, and connecting should be understood in a broad sense, and technicians in the relevant technical field can reasonably determine the specific meanings of the above terms in the present invention based on the specific content of the technical solution.

[0048] An embodiment of the present invention provides a voltage fitting method without an electrical model, comprising obtaining an active power matrix and a reactive power matrix of a low-voltage distribution network, constructing an input data set based on the active power matrix and the reactive power matrix; inputting the input data set into a width learning model to obtain the output voltage of the low-voltage distribution network, and using the output voltage as the output data set of the width learning model; obtaining the original input data set of the low-voltage distribution network, performing a dynamic principal component analysis on the original input data set to obtain a dynamic latent variable data set, and inputting the dynamic latent variable data set into the width learning model for training instead of the original input data set; extracting dynamic feature weights from the dynamic latent variable data set, and replacing the random weights of the width learning model with the dynamic feature weights to obtain a width learning voltage fitting model for the low-voltage distribution network; and performing parameter optimization on the width learning voltage fitting model using a preset particle swarm algorithm to obtain the optimal parameters of the voltage fitting model. According to the technical solution of this embodiment, width learning is used to train historical operating data of the low-voltage distribution network, with node power as input and node voltage as output. Width learning is suitable for small-scale data sets and has a short training time, overcoming the limitations of traditional models that rely on large data and are difficult to adjust. This embodiment also uses dynamic principal component analysis to transform the original high-dimensional dataset into low-dimensional dynamic latent variables, effectively capturing time-varying features and filtering out noise, providing high-quality data support for voltage fitting. Further improvements are made to width learning, incorporating dynamic time series feature weights and introducing an optimized particle swarm algorithm for parameter optimization, significantly improving the accuracy of voltage fitting.

[0049] The control method of the embodiment of the present invention is further described below based on the accompanying drawings.

[0050] Reference Figure 1 , Figure 1 A flowchart of a voltage fitting method without an electrical model is provided in an embodiment of the present invention. The voltage fitting method without an electrical model includes but is not limited to the following steps:

[0051] Step S11, obtaining the active power matrix and reactive power matrix of the low-voltage distribution network, and constructing an input data set according to the active power matrix and the reactive power matrix;

[0052] It's important to note that the active power, reactive power, and voltage RMS square matrices contain information about power transmission and voltage status at each node and line in the low-voltage distribution network. By analyzing these matrix data, the width learning model can comprehensively understand the grid's operating status and accurately identify potential fault areas. The voltage RMS square matrix directly reflects the voltage level at each node in the grid. By analyzing this matrix, the width learning model can accurately assess the voltage stability of the low-voltage distribution network and promptly identify problems such as voltage overshoot and voltage fluctuation.

[0053] Step S12, inputting the input data set into the width learning model to obtain the output voltage of the low-voltage distribution network, and using the output voltage as the output data set of the width learning model;

[0054] It should be noted that the voltage in the low-voltage distribution network is affected by multiple factors and is highly nonlinear. This model can accurately capture the intrinsic relationship between these factors and the output voltage by learning the input data set, thereby achieving efficient and accurate fitting of the low-voltage distribution network voltage. Compared with traditional methods, it can improve the accuracy and speed of voltage fitting.

[0055] Furthermore, this embodiment employs a wide learning approach, using node power as input and fitting the output node voltage, achieving voltage fitting without the need for a physical electrical model. Wide learning extracts the core features of power data through a feature layer, applies nonlinear transformations to enrich the representation in an enhancement layer, and rapidly calculates output weights using a pseudo-inverse matrix. This effectively captures the nonlinear relationship between power and voltage, significantly reducing data requirements and accelerating model training.

[0056] Step S13, obtaining an original input data set of the low-voltage distribution network, performing dynamic principal component analysis on the original input data set to obtain a dynamic latent variable data set, and inputting the dynamic latent variable data set into the width learning model for training instead of the original input data set;

[0057] It should be noted that dynamic principal component analysis is used to process the low-voltage distribution network dataset and extract dynamic latent variables as input features. In view of the time series dynamics and nonlinear coupling characteristics in the data, dynamic principal component analysis is combined with an autoregressive model to characterize the time-varying laws of the latent variables, weaken redundant interference, and provide high-quality and representative data support for width learning, thereby improving the voltage fitting accuracy. Furthermore, this embodiment replaces the randomly initialized weights with the mapping weights of dynamic principal component analysis through width learning weight design to construct a weight matrix that focuses on dynamic time series characteristics. This improvement highlights the key impact of input data on voltage, reduces the interference of secondary information, and enhances the ability of the width learning model to parse and capture complex time series data.

[0058] It should be noted that in low-voltage distribution networks, voltage changes dynamically with factors such as load fluctuations and the access of distributed power sources. Dynamic latent variable datasets can capture the dynamic characteristics of the power grid, while dynamic feature weights can weight features according to their importance at different moments. The voltage fitting model constructed based on this can accurately describe voltage variations under different operating conditions and effectively track dynamic voltage fluctuations. Furthermore, the width learning model itself has powerful nonlinear mapping capabilities. Combined with the dynamic latent variable dataset and dynamic feature weights obtained through dynamic principal component analysis, it can better explore the complex relationship between voltage and other related variables, thereby improving the voltage fitting accuracy and more accurately reflecting the actual distribution and variation of voltage in low-voltage distribution networks.

[0059] Step S14, extracting dynamic feature weights from the dynamic latent variable data set, and replacing the random weights of the width learning model with the dynamic feature weights to obtain a width learning voltage fitting model for the low-voltage distribution network;

[0060] It's important to note that dynamic latent variables are an optimized representation of raw data features. They can more effectively reflect the data's inherent structure and changing trends. Compared to raw data, dynamic latent variable datasets can provide more representative and discriminative features for tasks like model training and pattern recognition, thereby improving model accuracy and performance.

[0061] It should be noted that this embodiment introduces an attention mechanism to improve the weight design of width learning. By enabling the width learning fitting model to differentially process input data, it highlights key features and reduces the interference of secondary information. In dynamic principal component analysis, the mapping weights reflect the importance of the variable's projection in the principal component space and can characterize the role of dynamic latent variables in time series features.

[0062] Step S15 , performing parameter optimization on the width learning voltage fitting model according to a preset particle swarm algorithm to obtain the optimal parameters of the voltage fitting model.

[0063] It should be noted that this embodiment uses parameters as control variables, the root mean square error of voltage fitting as the objective function, and, combined with parameter constraints, constructs an optimization model. An improved particle swarm algorithm is then employed for parameter optimization. By simulating the collaborative movement of a swarm of particles within the search space and iteratively updating the particle speed and position, the optimal parameter combination is efficiently searched. To further enhance the optimization performance and avoid being trapped in local optima, this embodiment introduces a simulated annealing algorithm to improve the particle swarm algorithm and ensure its global search capabilities.

[0064] It should be noted that this embodiment uses an improved particle swarm optimization algorithm to determine the parameters for dynamic principal component analysis and width learning, optimizing model performance using root mean square error as the objective function. The improved particle swarm optimization algorithm searches for the optimal parameter combination through particle collaboration and introduces a simulated annealing algorithm to avoid local optima and ensure global optimization.

[0065] It should be noted that this embodiment uses node power as input and node voltage as output, and adopts width learning to train the historical operating data of the low-voltage distribution network. Among them, width learning is suitable for small-scale data sets and has a short training time, overcoming the limitations of traditional models such as reliance on big data and difficulty in adjustment. At the same time, this embodiment converts the original high-dimensional data set into low-dimensional dynamic latent variables through dynamic principal component analysis, effectively capturing time-varying characteristics and filtering out noise, providing high-quality data support for voltage fitting. Width learning is further improved by integrating dynamic time series feature weights and introducing an optimized particle swarm algorithm for parameter optimization, thereby significantly improving the accuracy of voltage fitting.

[0066] In addition, in one embodiment, referring to Figure 2 ,exist Figure 1 Step S11 of the illustrated embodiment also includes but is not limited to the following steps:

[0067] Step S21, obtaining a subnode set, an active output matrix, a three-phase current effective value square matrix, a three-phase resistance, and a three-phase active net load matrix of the low-voltage distribution network, and calculating an active power matrix based on the subnode set, the active output matrix, the three-phase current effective value square matrix, the three-phase resistance, and the three-phase active net load matrix;

[0068] Step S22, obtaining the reactive output matrix, three-phase reactance and three-phase reactive net load matrix of the low-voltage distribution network, and calculating the reactive power matrix based on the child node set, reactive output matrix, three-phase reactance, three-phase current and reactive net load matrix;

[0069] Step S23: Obtain the branch voltage, three-phase impedance matrix, and conjugate matrix of the three-phase impedance matrix of the low-voltage distribution network, and calculate the voltage effective value square matrix based on the branch voltage, three-phase resistance, three-phase active power matrix, three-phase reactive power matrix, three-phase impedance matrix, three-phase current, and conjugate matrix.

[0070] It should be noted that the calculation of the active power matrix is ​​expressed by the following first formula:

[0071]

[0072] Among them, S n is the set of child nodes, P nk is the active power output matrix, r mn is the square matrix of the three-phase current effective values, i mn is the three-phase resistance, P L，n is the three-phase active net load matrix;

[0073] The reactive power matrix is ​​calculated as follows:

[0074]

[0075] Among them, P nk is the reactive power matrix, x mn is the three-phase reactance, Q L，n is the reactive net load matrix;

[0076] The square matrix of the effective voltage value is calculated and expressed by the following third formula:

[0077]

[0078] Among them, u m is the branch voltage, Z mn is the three-phase impedance matrix, is the conjugate matrix of the three-phase impedance matrix.

[0079] The three-phase matrix is ​​of the form

[0080] In addition, in one embodiment, referring to Figure 3 ,exist Figure 1 Step S14 of the illustrated embodiment also includes but is not limited to the following steps:

[0081] Step S31, obtaining power data and voltage data of smart meters in the low-voltage distribution network, using the power data as an input data set of the width learning model, and using the voltage data as an output data set of the width learning model;

[0082] Step S32: inputting the input data set into the width learning model to obtain feature nodes of the width learning model, wherein the width learning model includes an input layer, a feature layer, an enhancement layer, and an output layer;

[0083] Step S33, obtaining the activation function of the feature node of the width learning model, the mapping weight of the feature layer, and the feature layer bias, and constructing a feature node matrix of the width learning model according to the input layer data, the activation function, the mapping weight, and the first feature bias, wherein the feature node represents the feature layer of the width learning model;

[0084] Step S34, obtaining the enhancement node bias of the enhancement layer and the activation function of the enhancement node of the feature layer, and constructing the enhancement node matrix of the width learning model according to the activation function and the feature node, wherein the enhancement node matrix represents the enhancement layer of the width learning model.

[0085] It should be noted that width learning consists of an input layer, a feature layer, an enhancement layer, and an output layer. According to the technical solution of this embodiment, the input layer data X is set to the first power parameter P and the second power parameter Q, and the output layer data Y corresponds to the node voltage U to achieve the goal of voltage fitting. The feature layer F contains multiple groups of feature nodes, which are generated by mapping the input data and are used to extract key feature information of the data. The mapping relationship between the input data X and the feature node F is expressed by the following fourth formula:

[0086]

[0087] Among them, F nF represents the matrix of the nFth group of feature nodes; φ represents the activation function of the feature node; W F represents the mapping weight of the feature layer; β nF Indicates F nF The deviation of ; NF represents the number of groups of feature nodes.

[0088] The enhancement layer H is a crucial component of the width learning model. It consists of multiple enhancement nodes, generated by applying nonlinear transformations to all feature nodes in the feature layer to further enrich the data representation capabilities. Specifically, the generation of the enhancement nodes follows a clear mathematical transformation relationship, as shown in the fifth formula below. This nonlinear transformation not only enhances the model's ability to capture complex data patterns but also provides more robust feature support for accurate computation in the subsequent output layer.

[0089]

[0090] Where H nH represents the nHth enhancement node matrix; ξ represents the activation function of the enhancement node; W H represents the mapping weight from F to H; nH Indicates H nH The first characteristic deviation of ; NH represents the number of enhanced nodes.

[0091] The output layer, Y, serves as the final step in the width learning model. Its output nodes are generated through a collaborative mapping of feature nodes and enhancement nodes, achieving a complete transformation from input data to target output. Specifically, this mapping relationship is clearly defined through a mathematical expression, as shown in the sixth formula below. This design leverages the fundamental information extracted from feature nodes and the nonlinear scaling provided by enhancement nodes, ensuring that the output layer efficiently and accurately reflects node voltages, thereby improving model accuracy.

[0092] Y=[F|H]W FH ;

[0093] Among them, W FHis the first mapping weight matrix of F and H to Y obtained by the pseudo-inverse method.

[0094] According to the descriptions of the fourth to sixth formulas, width learning shows the following three significant advantages in model-free voltage fitting of low-voltage distribution networks.

[0095] First, width learning, as an efficient and concise neural network architecture, is centered around implementing nonlinear data transformations by introducing a layer of randomly initialized augmented nodes between the input and output layers. Unlike traditional methods, this model does not rely on deep structures or layer-by-layer iterative gradient optimization. Instead, it uses a pseudo-inverse matrix calculation of the output layer weights to complete global parameter updates in a single pass. This design significantly simplifies the training process, effectively improving computational efficiency and training speed.

[0096] Secondly, wide learning employs a single hidden layer design, significantly reducing the number of parameters to be optimized compared to the multi-layer networks of deep learning. Through the synergistic effect of feature nodes and enhancement nodes, the model expands the data's representation capabilities and enhances its ability to analyze data. As a result, this method maintains excellent fitting performance even with small datasets, making it particularly suitable for scenarios with limited data resources.

[0097] Finally, width learning requires minimal hardware resources and does not rely on large-scale GPU computing. This feature enables it to efficiently complete voltage fitting tasks in low-voltage distribution network environments with limited computing power, significantly improving the model's applicability and practical value in real-world deployments.

[0098] In addition, in one embodiment, referring to Figure 4 ,exist Figure 1 Step S13 of the illustrated embodiment also includes but is not limited to the following steps:

[0099] Step S41, obtaining the input data matrix of the original input data set, the mapping weight matrix of the dynamic latent variable, the autoregressive coefficient, and the error term information;

[0100] Step S42: obtaining a dynamic latent variable matrix according to the input data matrix, the mapping weight matrix, the autoregressive coefficients and the error term information.

[0101] It should be noted that the accuracy of voltage fitting by width learning is highly dependent on the representativeness and quality of the data set. To this end, the present invention uses a dynamic principal component analysis method to extract features from the low-voltage distribution network data set to generate new dynamic latent variable components and support voltage fitting. The data set contains load data, photovoltaic output data and control data for each phase of each node. There is a nonlinear cross-coupling relationship between these variables, and due to the influence of mutual impedance, self-impedance and electrical distance, the correlation between nodes of different phases and with a long electrical distance is low. At the same time, the data has significant time series dynamic characteristics, such as trend, periodicity and random fluctuations, and is accompanied by time series autocorrelation effects. Based on this, dynamic principal component analysis captures these time-dependent characteristics, generates dynamic latent variables as new feature inputs, and at the same time weakens redundant information interference, achieving dimensionality reduction more efficiently than traditional principal component analysis. And provides a high-quality data foundation for improving the accuracy of voltage fitting. It is specifically expressed by the following seventh formula:

[0102]

[0103] Where, X DL,t is the dynamic latent variable matrix at time t; It is the first mapping weight matrix that projects the observed variables into ND dynamic intrinsic latent variables in the feature space.

[0104] Furthermore, this embodiment uses an autoregressive model to characterize the dynamic characteristics of the latent variable, thereby effectively depicting its regular characteristics of evolution over time, as shown in the following eighth formula:

[0105]

[0106] Among them, s is the order of the autoregressive model, β1,…,β s is the autoregressive coefficient; r t is the error term. When s is large enough, r t It is basically white noise and can be ignored.

[0107] In addition, in one embodiment, referring to Figure 5 ,exist Figure 4 After step S42 of the embodiment shown, the following steps are also included but not limited to:

[0108] Step S51, constructing a dynamic time series feature weight matrix based on the autoregressive coefficient, the dynamic latent variable matrix and the input matrix;

[0109] Step S52: Obtain mapping weights according to the dynamic time series feature weight matrix.

[0110] It should be noted that this embodiment introduces an attention mechanism to improve the weight design of width learning. By giving the model the ability to process input data differently, it highlights key features and reduces the interference of secondary information. In dynamic principal component analysis, the mapping weight reflects the importance of the projection of the variable in the principal component space and can characterize the role of dynamic latent variables in time series features. Therefore, as shown in the following tenth formula, this embodiment replaces the random weights of width learning with the mapping weights in dynamic principal component analysis, and constructs a weight matrix that focuses on dynamic time series features, thereby enhancing the model's ability to capture key features and improving overall performance.

[0111]

[0112] In addition, in one embodiment, referring to Figure 6 ,exist Figure 4 Step S15 of the illustrated embodiment also includes but is not limited to the following steps:

[0113] Step S61, obtaining the number of dynamic latent variables and the number of enhanced nodes in the dynamic latent variable data set;

[0114] Step S62, calculating the root mean square error after fitting the width learning model according to the number of dynamic latent variables, the number of groups of feature nodes, and the number of enhanced nodes;

[0115] Step S63 , obtaining the constraint conditions of the width learning model, taking the first parameter as the control variable, the root mean square error as the objective function and combining the constraint conditions to construct an optimization model of the width learning model.

[0116] It should be noted that this embodiment uses parameters as control variables, the root mean square error of voltage fitting as the objective function, and constructs an optimization model in combination with the constraint conditions of the parameters, and adopts an improved particle algorithm for parameter optimization. This method simulates the collaborative movement of a particle swarm in the search space, iteratively updates the speed and position of the particles, and efficiently searches for the optimal parameter combination. In order to further improve the optimization effect and avoid falling into local optimality, the present invention introduces a simulated annealing algorithm to improve the particle swarm algorithm to ensure the global search capability of the algorithm. In the voltage fitting method proposed in the present invention, the selection of dynamic principal component analysis parameters and width learning system hyperparameters is crucial to the model effect, because they directly affect the reliability of feature extraction and the accuracy of voltage fitting. The hyperparameter selection is expressed by the following ninth formula:

[0117]

[0118] Where Z = [ND, NF, NH]; RMSE(Z) is the root mean square error of voltage fitting under the Z parameter.

[0119] like Figure 7 As shown, Figure 7 : is a structural diagram of a voltage fitting device without an electrical model provided by an embodiment of the present invention. The present invention also provides a voltage fitting device without an electrical model, comprising:

[0120] The processor 701 may be implemented as a general-purpose central processing unit (CPU), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is configured to execute relevant programs to implement the technical solutions provided in the embodiments of the present application.

[0121] The memory 702 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 702 can store an operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 702 and is called by the processor 701 to execute the voltage fitting method without an electrical model in the embodiments of this application.

[0122] Input / output interface 703, used to implement information input and output;

[0123] Communication interface 704, used to implement communication interaction between the apparatus and other devices, which can be achieved through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WiFi, Bluetooth, etc.);

[0124] Bus 705 , which transmits information between various components of the device (e.g., processor 701 , memory 702 , input / output interface 703 , and communication interface 704 );

[0125] The processor 701 , the memory 702 , the input / output interface 703 and the communication interface 704 are connected to each other in communication within the device via a bus 705 .

[0126] An embodiment of the present application further provides an electronic device, comprising the voltage fitting device without an electrical model as described above.

[0127] An embodiment of the present application further provides a storage medium, which is a computer-readable storage medium and stores a computer program. When the computer program is executed by a processor, the above-mentioned voltage fitting method without an electrical model is implemented.

[0128] The memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory optionally includes a memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of the above-mentioned networks include but are not limited to the Internet, an intranet, a local area network, a mobile communication network and a combination thereof. The device embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and are located in one place, or may be distributed to multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present embodiment.

[0129] Those skilled in the art will appreciate that all or some of the steps and systems in the method disclosed above can be implemented as software, firmware, hardware, and appropriate combinations thereof. Some physical components or all physical components can be implemented as software executed by a processor, such as a central processing unit, a digital signal processor, or a microprocessor, or implemented as hardware, or implemented as an integrated circuit, such as an application-specific integrated circuit. Such software can be distributed on a computer-readable medium, and the computer-readable medium can include computer storage media (or non-transitory media) and communication media (or temporary media). As known to those skilled in the art, the term computer storage media is included in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data) and is volatile and non-volatile, removable, and non-removable. Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory, or other memory technology, CD-ROM, digital versatile disks (DVD), or other optical disk storage, magnetic cassettes, magnetic tapes, disk storage, or other magnetic storage devices, or any other medium that can be used to store desired information and can be accessed by a computer. Furthermore, as is well known to those skilled in the art, communication media typically includes computer-readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any information delivery media.

[0130] The above is a specific description of the preferred implementation of the present invention, but the present invention is not limited to the above implementation. Those skilled in the art can also make various equivalent modifications or substitutions under the shared conditions that do not violate the spirit of the present invention. These equivalent modifications or substitutions are all included in the scope defined by the claims of the present invention.

Claims

1. A voltage fitting method without an electrical model, characterized in that: include: Obtaining an active power matrix and a reactive power matrix of a low-voltage distribution network, and constructing an input data set according to the active power matrix and the reactive power matrix; Inputting the input data set into a width learning model to obtain the output voltage of the low-voltage distribution network, and using the output voltage as the output data set of the width learning model; Obtaining an original input data set of the low-voltage distribution network, performing dynamic principal component analysis on the original input data set to obtain a dynamic latent variable data set, and inputting the dynamic latent variable data set into the width learning model for training instead of the original input data set; Extracting dynamic feature weights from the dynamic latent variable dataset, and replacing random weights of the width learning model with the dynamic feature weights to obtain a width learning voltage fitting model of the low-voltage distribution network; Parameter optimization is performed on the width learning voltage fitting model according to a preset particle swarm algorithm to obtain optimal parameters of the voltage fitting model.

2. The voltage fitting method without electrical model according to claim 1, characterized in that: The obtaining of the active power matrix and the reactive power matrix of the low-voltage distribution network includes: Obtain a subnode set, an active output matrix, a three-phase current effective value square matrix, a three-phase resistance, and a three-phase active net load matrix of the low-voltage distribution network, and calculate the active power matrix based on the subnode set, the active output matrix, the three-phase current effective value square matrix, the three-phase resistance, and the three-phase active net load matrix; Obtaining a reactive output matrix, a three-phase reactance, and a three-phase reactive net load matrix of the low-voltage distribution network, and calculating the reactive power matrix based on the set of child nodes, the reactive output matrix, the three-phase reactance, the three-phase current, and the reactive net load matrix; Obtain the branch voltage, three-phase impedance matrix and the conjugate matrix of the three-phase impedance matrix of the low-voltage distribution network, and calculate the voltage effective value square matrix according to the branch voltage, the three-phase resistance, the three-phase active power matrix, the three-phase reactive power matrix, the three-phase impedance matrix, the three-phase current and the conjugate matrix.

3. The voltage fitting method without electrical model according to claim 1, characterized in that: The step of obtaining a width learning voltage fitting model for the low-voltage distribution network includes: Acquire power data and voltage data of smart meters in the low-voltage distribution network, use the power data as an input data set of the width learning model, and use the voltage data as an output data set of the width learning model; Inputting the input data set into the width learning model to obtain feature nodes of the width learning model, wherein the width learning model includes an input layer, a feature layer, an enhancement layer, and an output layer; Obtaining an activation function of a feature node of the width learning model, a mapping weight of a feature layer, and a feature layer bias, and constructing a feature node matrix of the width learning model according to the input layer data, the activation function, the mapping weight, and the first feature bias, wherein the feature node represents the feature layer of the width learning model; Obtain the enhancement node deviation of the enhancement layer and the activation function of the enhancement node of the feature layer, and construct an enhancement node matrix of the width learning model according to the activation function and the feature node, wherein the enhancement node matrix represents the enhancement layer of the width learning model.

4. The voltage fitting method without electrical model according to claim 1, characterized in that: The dynamic latent variable dataset is obtained, including: Obtaining the input data matrix of the original input data set, the mapping weight matrix of the dynamic latent variables, the autoregressive coefficients, and the error term information; A dynamic latent variable matrix is ​​obtained according to the input data matrix, the mapping weight matrix, the autoregressive coefficients and the error term information.

5. The voltage fitting method without electrical model according to claim 4, characterized in that: After obtaining the dynamic latent variable matrix, the method further includes: Constructing a dynamic time series feature weight matrix according to the autoregressive coefficient, the dynamic latent variable matrix and the input matrix; The mapping weight is obtained according to the dynamic time series feature weight matrix.

6. The voltage fitting method without electrical model according to claim 4, characterized in that: The dynamic latent variable matrix is ​​obtained and expressed by the following formula: Among them, X DL , t is the dynamic latent variable matrix, (X t ) T is the input data matrix at the first preset moment, is the mapping weight matrix.

7. The voltage fitting method without electrical model according to claim 1, characterized in that: The performing parameter optimization on the voltage fitting model according to a preset particle swarm algorithm includes: Obtaining the number of dynamic latent variables in the dynamic latent variable data set and the number of enhanced nodes of the enhanced node; Calculating a root mean square error after fitting the width learning model according to the number of dynamic latent variables, the number of groups of feature nodes, and the number of enhanced nodes; Obtain the constraints of the width learning model, use the first parameter as a control variable, the root mean square error as an objective function, and construct an optimization model of the width learning model in combination with the constraints.

8. A voltage fitting device without an electrical model, characterized in that: The invention comprises at least one control processor and a memory for communicating with the at least one control processor; the memory stores instructions that can be executed by the at least one control processor, and the instructions are executed by the at least one control processor to enable the at least one control processor to perform the voltage fitting method without an electrical model as described in any one of claims 1 to 7.

9. An electronic device, characterized in that: The invention comprises the voltage fitting device without electrical model as claimed in claim 8.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are used to enable a computer to execute the voltage fitting method without an electrical model according to any one of claims 1 to 7.

Images