Data-driven distribution network operation characteristic simulation method

Abstract

The invention relates to the technical field of power distribution networks, in particular to a data-driven power distribution network operation characteristic simulation method. Comprising the following steps: S1, collecting and preprocessing multi-source data; s2, performing adaptive fusion on the preprocessed data to form a data set; s3, constructing an initial simulation model, and training based on the data set to obtain a main simulation model; outputting a prediction result of the distribution network characteristics through the main simulation model; and S4, based on the prediction result, carrying out dynamic check and simulation optimization so as to output a simulation result. In the working process, technical modules of multi-source data fusion, deep learning model construction, dynamic checking, self-adaptive optimization and the like are covered. By fusing multi-source information related to real-time load data, equipment operation data and environment data and combining dynamic adjustment and an incremental learning mechanism, high-precision prediction and optimization of the operation state of the distribution network are realized.

Description

Technical Field

[0001] The present invention relates to the technical field of distribution network, and in particular to a data-driven distribution network operation characteristic simulation method. Background Art

[0002] With the continuous expansion and complexity of distribution networks, traditional distribution network optimization methods have shown obvious limitations in processing large-scale data, real-time scheduling and precise control.

[0003] Traditional optimization methods are mostly based on linear models and classical algorithms. Although they are suitable for small-scale distribution networks, they are often unable to achieve timely response and efficient regulation for complex and dynamically changing modern distribution networks, especially when dealing with large amounts of changing real-time data.

[0004] If you have the following problems:

[0005] 1) Inaccurate modeling: Static data models cannot reflect the real-time dynamic changes of load, equipment status and environmental conditions;

[0006] 2) Low scheduling efficiency: It relies on manual labor, and the manual decision-making process is long and complicated, making it difficult to deal with emergencies such as high load and equipment failure;

[0007] 3) High operating cost: Single-objective optimization leads to line loss and low utilization of backup power supply. Summary of the invention

[0008] The present invention aims to solve at least one of the technical problems existing in the prior art.

[0009] The technical solution of the present invention is: a data-driven distribution network operation characteristic simulation method, comprising the following steps:

[0010] S1. Collection and preprocessing of multi-source data;

[0011] S2, adaptively fusing the preprocessed data to form a data set;

[0012] S3, constructing an initial simulation model, training it based on the data set, and obtaining a main simulation model; outputting the prediction results of the distribution network characteristics through the main simulation model;

[0013] S4. Based on the prediction results, dynamic verification and simulation optimization are performed to output simulation results.

[0014] In step S1,

[0015] Multi-source data includes real-time load data, equipment operation data, and environmental data;

[0016] The collected multi-source data is cleaned and standardized, including: missing value filling, data denoising and time alignment.

[0017] Real-time load data includes the power change trend of each node;

[0018] Equipment operation data includes transformer status, voltage, and current;

[0019] Environmental data includes meteorological information and market fluctuations, wherein meteorological information includes temperature and humidity; market fluctuations include electricity price fluctuation data, electricity demand fluctuation data and electricity supply fluctuation data.

[0020] In step S2, the weight of each data dimension is dynamically adjusted according to the importance of the data, and a unified distribution network operation data set is generated by fusion;

[0021] The importance of data includes historical relevance, real-time volatility, and operational status sensitivity.

[0022] Historical relevance i , using the following formula:

[0023]

[0024] Where, X i : the value of the i-th data dimension; Y: key indicator of operating status; Cov(X i , Y): X i Covariance with Y; σ Xi and σ Y :X i and the standard deviation of Y;

[0025] Real-time volatility var i , using the following formula:

[0026]

[0027] Where, X i,j : The value of the i-th data dimension in the j-th time window; The average value of the i-th data dimension in the current window; N: sliding window size;

[0028] Operating state sensitivity sens i , using the following formula:

[0029]

[0030] In the formula, Derivative function.

[0031] Dynamically adjust the weight of each data dimension, including: when the weight of a data source changes, the weights of other data sources need to be redistributed proportionally to keep the total equal to 1;

[0032] Assume that w1, w2, and w3 correspond to the weights of real-time load data, equipment operation data, and environmental data, respectively. When w1 is adjusted, w2 and w3 are adjusted according to the following formula:

[0033] w1′=w1·(1+Δ1)

[0034]

[0035] In the formula,

[0036] w1′: adjusted load weight;

[0037] w2′ and w3′: equipment operation weight and environment weight after redistribution;

[0038] Δ1: The incremental ratio of load weight change due to volatility.

[0039] Each data dimension is fused according to the weight to generate a unified distribution network operation data set as follows:

[0040]

[0041] Where D fused : The integrated dataset after fusion; X i : The value of the i-th data dimension; w i : The weight of the i-th data dimension; n: The number of data.

[0042] In step S3,

[0043] Construct an initial simulation model with a deep neural network as the core, and obtain the main simulation model after training. The formula of the main simulation model is:

[0044] F=W (L) ·f(W (L-1) ·f(···f(W (L) ·X+b (L) )···)

[0045] +b (L-1) )b (L)

[0046] Where, F: main simulation model function; X: input feature matrix; W (l) : The weight matrix of the lth layer; W (L-1) : The weight matrix of the l-1th layer; b (l) : bias vector of the lth layer; b (L-1) : bias vector of the l-1th layer; f: activation function; L: total number of layers.

[0047] It also includes model optimization, using adaptive L2 regularization to constrain the output weight matrix of the deep neural network, using the following formula:

[0048] L total =L loss +γ·R(W)

[0049] Where, L total : total objective function value; L loss : prediction error; R(W): regularization term, used to constrain the size of the weight matrix W; γ: regularization strength coefficient.

[0050] In step S4, the dynamic verification includes:

[0051] The prediction results are compared with the real-time monitoring data, and the following relative error is used:

[0052]

[0053] In the formula, E rel : Relative error; Y pred : Model predicted value; Y real : Real-time monitoring data;

[0054] Error checks include:

[0055] Small error: E rel <5%, the prediction results are considered acceptable, and the model directly outputs the simulation results;

[0056] Medium error: 5% ≤ E rel <10%, indicating the need to calibrate the model;

[0057] Large error: E rel ≥10%, triggering model parameter adjustment or data input optimization.

[0058] In the case of large errors, model parameters are dynamically adjusted through incremental learning;

[0059] Including: adding new data samples and updating model weights;

[0060] Weight update formula: For each layer's weight matrix W (l) , the update formula is:

[0061]

[0062] Where V: loss function; μ: learning rate; ΔW (l) : incremental adjustment of weights by new samples; β: incremental learning intensity;

[0063] or redistribute the weight of data sources;

[0064] The redistribution formula is: Data source weight is based on historical relevance rel i, volatility var i and sensitivity i Recalculate:

[0065]

[0066] In its work, the present invention covers technical modules such as multi-source data fusion, deep learning model construction, dynamic verification and adaptive optimization. By integrating multi-source information related to real-time load data, equipment operation data and environmental data, combined with dynamic adjustment and incremental learning mechanisms, high-precision prediction and optimization of the distribution network operation status are achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] In order to more clearly illustrate the specific implementation of the present invention or the technical solution in the prior art, the following briefly introduces the drawings required for the specific implementation or the prior art description. In the drawings, each part is not necessarily drawn according to the actual scale.

[0068] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0069] The present invention Figure 1 As shown, a data-driven distribution network operation characteristic simulation method includes the following steps:

[0070] S1. Collection and preprocessing of multi-source data;

[0071] S2, adaptively fusing the preprocessed data to form a data set;

[0072] The present invention generates a unified high-quality input data set through multi-source data preprocessing and fusion, providing support for model construction and verification.

[0073] S3. Build an initial simulation model and train it based on the data set to obtain the main simulation model; output the prediction results of the distribution network characteristics through the main simulation model; build a deep learning model based on the fused data to output the key simulation results of the distribution network operation characteristics.

[0074] S4. Based on the prediction results, dynamic verification and simulation optimization are performed to output simulation results. Through verification and simulation optimization, the real-time and adaptability of the model are enhanced to ensure the accuracy and practicality of the simulation results.

[0075] In its work, the present invention covers technical modules such as multi-source data fusion, deep learning model construction, dynamic verification and adaptive optimization. By integrating multi-source information related to real-time load data, equipment operation data and environmental data, combined with dynamic adjustment and incremental learning mechanisms, high-precision prediction and optimization of the distribution network operation status are achieved.

[0076] In step S1,

[0077] Multi-source data includes real-time load data, equipment operation data, and environmental data;

[0078] The collected multi-source data is cleaned and standardized, including: missing value filling, data denoising and time alignment.

[0079] Missing value filling: fill in the negative missing values ​​through time series interpolation algorithm;

[0080] Data denoising: Use sliding window filtering algorithm to remove noise;

[0081] Time alignment: unify the timestamps of data from different sources to ensure data synchronization.

[0082] Real-time load data includes the power change trend of each node;

[0083] Equipment operation data includes transformer status, voltage, and current;

[0084] Environmental data includes meteorological information and market fluctuations, wherein meteorological information includes temperature and humidity; market fluctuations include electricity price fluctuation data, electricity demand fluctuation data and electricity supply fluctuation data.

[0085] In step S2, the weight of each data dimension is dynamically adjusted according to the importance of the data, and a unified distribution network operation data set is generated by fusion;

[0086] The importance of data is a measure of the impact of a certain data dimension on the operating status of the distribution network, including historical correlation, real-time volatility, and operating status sensitivity.

[0087] Historical relevance i The contribution of a certain data dimension to the key indicators of distribution network operation status (such as voltage stability, load distribution, etc.) in historical data is calculated through correlation analysis (such as Pearson correlation coefficient). The following formula is used:

[0088]

[0089] Where, X i : the value of the i-th data dimension; Y: key indicator of operating status; Cov(X i , Y): X i Covariance with Y; σ Xi and σ Y :X i and the standard deviation of Y;

[0090] Real-time volatility var i, which indicates the degree of change of a data dimension at the current moment. The change amplitude of the data is calculated by sliding the window, using the following formula:

[0091]

[0092] Where, X i,j : The value of the i-th data dimension in the j-th time window; The average value of the i-th data dimension in the current window; N: sliding window size;

[0093] Operating state sensitivity sens i , reflects the direct impact of the fluctuation of a certain data dimension on the key indicators of the operating status, and is defined as the partial derivative of the operating status indicator with respect to the data dimension; the following formula is used:

[0094]

[0095] In the formula, Derivative function.

[0096] When the load fluctuates violently at a certain moment (such as during peak hours), var i Improve, the real-time load data weight w1 increases;

[0097] When a device is in an abnormal state, the operating state sensitivity sens i Improve, the weight w2 of equipment operation data increases;

[0098] The current temperature rises sharply, causing var i Improved, the real-time load data weight w1 increases.

[0099] Dynamically adjust the weights of each data dimension, including: when the weight of a data source changes due to changes in characteristics (such as increased volatility or sensitivity), the weights of other data sources need to be redistributed proportionally to keep the total equal to 1.

[0100] Assume w1, w2, and w3 (corresponding to real-time load data, equipment operation data, and environmental data, respectively). When w1 (the weight of real-time load data) increases due to increased volatility, the other weights w2 and w3 are adjusted according to the following formula:

[0101] w1′=w1·(1+Δ1)

[0102]

[0103] in:

[0104] w1′: adjusted load weight;

[0105] w2′ and w3′: equipment operation weight and environment weight after redistribution;

[0106] Δ1: The incremental ratio of load weight change due to volatility.

[0107] Each data dimension is fused according to the weight to generate a unified distribution network operation data set as follows:

[0108]

[0109] Where D fused : The integrated dataset after fusion; X i : The value of the i-th data dimension; w i : The weight of the i-th data dimension; n: The number of data.

[0110] The present invention generates a distribution network operation data set with strong consistency and complete characterization through multi-source data fusion, improves the accuracy and timeliness of data input, and provides a reliable foundation for subsequent simulation models.

[0111] In step S3,

[0112] Construct an initial simulation model with a deep neural network as the core, and obtain the main simulation model after training. The formula of the main simulation model is:

[0113] F=W (L) ·f(W (L-1) ·f(···f(W (L) ·X+b (L) )···)

[0114] +b (L-1) )b (L)

[0115] Where, F: main simulation model function; X: input feature matrix; W (l) : The weight matrix of the lth layer; W (L-1) : The weight matrix of the l-1th layer; b (l) : bias vector of the lth layer; b (L-1) : bias vector of the l-1th layer; f: activation function; L: total number of layers;

[0116] The main simulation model is used to capture the nonlinear relationship of the distribution network operation characteristics:

[0117] Input features: feature matrix from fused data, including multi-dimensional features of distribution network operation (such as load, voltage, current, equipment status, environmental conditions, etc.);

[0118] Output target: node voltage, current, power flow and other characteristics.

[0119] It also includes model optimization, using adaptive L2 regularization to constrain the output weight matrix of the deep neural network to improve the generalization ability of the model and adapt to different scenarios and operating conditions. The following formula is used:

[0120] L total =L loss +γ·R(W)

[0121] Where, L total : total objective function value; L loss : prediction error; R(W): regularization term, used to constrain the size of the weight matrix W; γ: regularization strength coefficient.

[0122] The main simulation model is the core of the present invention, which is used to accurately capture the complex nonlinear relationship of the distribution network operation characteristics. It is a core component that can process multi-dimensional data input and generate high-precision simulation results of the distribution network operation status after data training output.

[0123] The initial simulation model relies on fixed parameters and is suitable for predictions in small-scale, simple scenarios. It is difficult to meet the modeling needs of dynamic, multi-source, and nonlinear relationships.

[0124] Main simulation model: A dynamically optimized deep learning model that can handle multi-dimensional complex inputs and perform real-time prediction and optimization; supports dynamic verification and adaptive adjustment to adapt to a variety of scenarios.

[0125] In step S4, the dynamic verification includes:

[0126] The prediction results are compared with the real-time monitoring data, and the following relative error is used:

[0127]

[0128] In the formula, E rel : Relative error; Y pred : Model predicted value; Y real : Real-time monitoring data;

[0129] The error is the deviation between the model prediction value and the real-time monitoring data;

[0130] Error checks include:

[0131] Small error: E rel <5%, the prediction results are considered acceptable, and the model directly outputs the simulation results;

[0132] Medium error: 5% ≤ E rel <10%, indicating the need to calibrate the model;

[0133] Large error: E rel≥10%, triggering model parameter adjustment or data input optimization.

[0134] In the case of moderate errors, calibrate the model, including:

[0135] 1) Recalculate the weights using the weight formula:

[0136] If the error source is mainly concentrated in the load data, increase w1;

[0137] If the device status is highly sensitive, adjust w2.

[0138] 2) Regularization parameter calibration

[0139] Check whether the model is overfitting or underfitting:

[0140] Overfitting: The weight parameter is too large, adjust the regularization strength γ;

[0141] Underfitting: The model complexity is insufficient, and the number of hidden layers or neurons should be appropriately increased.

[0142] By dynamically adjusting γ, the problem of weight being too large or too small can be suppressed.

[0143] In the case of large errors, model parameters are dynamically adjusted through incremental learning;

[0144] Including: adding new data samples, updating model weights, or reallocating data source weights;

[0145] The details are as follows:

[0146] 1) Add data samples of high volatility or abnormal scenarios and update model weights;

[0147] (1) Incremental data sample update:

[0148] Add new samples from highly volatile or abnormal scenarios to the training set;

[0149] High volatility refers to a data source that changes dramatically in a short period of time, which is usually quantified by calculating the rate of change using a sliding window.

[0150] Abnormal scenarios refer to situations in which the operation of the distribution network deviates significantly from historical trends or normal conditions.

[0151] Increase the training weight of a sample according to its importance.

[0152] (2) Weight update formula: For each layer’s weight matrix W (l) , the update formula is:

[0153]

[0154] Where V: loss function; μ: learning rate; ΔW (l) : Incremental adjustment of weights by new samples; β: incremental learning strength.

[0155] The model weights are adjusted based on real-time monitoring data, and the model parameters are updated through an incremental learning mechanism, which significantly improves the adaptability of the model in high volatility and abnormal scenarios.

[0156] Through real-time verification and dynamic optimization, optimization solutions can be quickly generated in load fluctuations or abnormal scenarios, significantly shortening response time and reducing power outages and energy losses.

[0157] 2) Reallocate the weight of data sources to increase the proportion of key data.

[0158] Key data refers to the data dimensions that have a greater impact on the operation status of the distribution network, usually including:

[0159] Load data during peak load periods;

[0160] Abnormal equipment operation data;

[0161] Meteorological data for extreme weather conditions.

[0162] Redistribution formula: Data source weights are based on historical relevance rel i , volatility var i and sensitivity i Recalculate:

[0163]

[0164] Finally, the simulation results (such as node voltage change curve, line power distribution diagram, power flow optimization suggestions, etc.) are output based on the verified model.

[0165] The present invention improves the real-time performance and accuracy of simulation results through dynamic verification; and the adaptive adjustment enhances the stability of the model in complex environments.

[0166] The present invention has the following advantages:

[0167] Through multi-source data fusion and adaptive weight adjustment, the impact of load fluctuations, equipment anomalies and environmental changes can be dynamically captured.

[0168] Based on nonlinear modeling capabilities, it captures the complex relationships between multi-dimensional features and improves prediction accuracy.

[0169] The dynamic verification mechanism is combined with incremental learning to enhance the model's adaptability to high volatility and abnormal scenarios.

[0170] Dynamically adjust model weights and parameters to generate optimization results in real time. Integrate simulation and decision support to quickly respond to load fluctuations, equipment anomalies and environmental changes.

[0171] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A data-driven distribution network operation characteristics simulation method, characterized in that: The following steps are involved: S1. Collection and preprocessing of multi-source data; S2, adaptively fusing the preprocessed data to form a data set; S3, constructing an initial simulation model, training based on the data set, and obtaining a main simulation model; Output the prediction results of distribution network characteristics through the main simulation model; S4. Based on the prediction results, dynamic verification and simulation optimization are performed to output simulation results.

2. A data-driven distribution network operation characteristic simulation method according to claim 1, characterized in that: In step S1, Multi-source data includes real-time load data, equipment operation data, and environmental data; The collected multi-source data is cleaned and standardized, including: missing value filling, data denoising and time alignment.

3. A data-driven distribution network operation characteristic simulation method according to claim 2, characterized in that: Real-time load data includes the power change trend of each node; Equipment operation data includes transformer status, voltage, and current; Environmental data includes meteorological information and market fluctuations, wherein meteorological information includes temperature and humidity; market fluctuations include electricity price fluctuation data, electricity demand fluctuation data and electricity supply fluctuation data.

4. A data-driven distribution network operation characteristic simulation method according to claim 1, characterized in that: In step S2, the weight of each data dimension is dynamically adjusted according to the importance of the data, and a unified distribution network operation data set is generated by fusion; The importance of data includes historical relevance, real-time volatility, and operational status sensitivity. Historical relevance i , using the following formula: Where, X i : the value of the i-th data dimension; Y: key indicator of operating status; Cov(X i , Y): X i The covariance of and Y; and σ Y :X i and the standard deviation of Y; Real-time volatility var i , using the following formula: Where, X i,j : The value of the i-th data dimension in the j-th time window; The average value of the i-th data dimension in the current window; N: sliding window size; Operating state sensitivity sens i , using the following formula: In the formula, Derivative function.

5. A data-driven distribution network operation characteristic simulation method according to claim 4, characterized in that: Dynamically adjust the weight of each data dimension, including: when the weight of a data source changes, the weights of other data sources need to be redistributed proportionally to keep the total equal to 1; Assume that w1, w2, and w3 correspond to the weights of real-time load data, equipment operation data, and environmental data, respectively. When w1 is adjusted, w2 and w3 are adjusted according to the following formula: w1′=w1·(1+Δ1) In the formula, w1′: adjusted load weight; w2′ and w3′: equipment operation weight and environment weight after redistribution; Δ1: The incremental ratio of load weight change due to volatility.

6. A data-driven distribution network operation characteristic simulation method according to claim 1, characterized in that: Each data dimension is fused according to the weight to generate a unified distribution network operation data set as follows: Where D fused : The integrated dataset after fusion; X i : The value of the i-th data dimension; w i : The weight of the i-th data dimension; n: The number of data.

7. A data-driven distribution network operation characteristic simulation method according to claim 1, characterized in that: In step S3, Construct an initial simulation model with a deep neural network as the core, and obtain the main simulation model after training. The formula of the main simulation model is: F=W (L) ·f(W (L-1) ·f(···f(W (L) ·X+b (L) )···)+b (L-1) )b (L) Where, F: main simulation model function; X : Input feature matrix; W (l) : The weight matrix of the lth layer; W (L-1) : The weight matrix of the l-1th layer; b (l) : bias vector of the lth layer; b (L-1) : Bias vector of the l-1th layer; f: activation function; L: total number of layers.

8. A data-driven distribution network operation characteristic simulation method according to claim 7, characterized in that: It also includes model optimization, using adaptive L2 regularization to constrain the output weight matrix of the deep neural network, using the following formula: L total =L loss +γ·R(W) Where, L total : total objective function value; L loss : prediction error; R(W): regularization term, used to constrain the size of the weight matrix W; γ: Regularization strength coefficient.

9. A data-driven distribution network operation characteristic simulation method according to claim 1, characterized in that: In step S4, the dynamic verification includes: The prediction results are compared with the real-time monitoring data, and the following relative error is used: In the formula, E rel : Relative error; Y pred : Model predicted value; Y real : Real-time monitoring data; Error checks include: Small error: E rel <5%, the prediction results are considered acceptable, and the model directly outputs the simulation results; Medium error: 5% ≤ E rel <10%, indicating the need to calibrate the model; Large error: E rel ≥10%, triggering model parameter adjustment or data input optimization.

10. A data-driven distribution network operation characteristic simulation method according to claim 9, characterized in that: In the case of large errors, model parameters are dynamically adjusted through incremental learning; Including: adding new data samples and updating model weights; Weight update formula: For each layer's weight matrix W (l) , the update formula is: Where V: loss function; μ: learning rate; ΔW (l) : incremental adjustment of weights by new samples; β: incremental learning intensity; or redistribute the weight of data sources; The redistribution formula is: Data source weight is based on historical relevance rel i , volatility var i and sensitivity i Recalculate:

Images