Line loss optimization method and system for new energy access power distribution network

Abstract

The invention provides a line loss optimization method and system for a new energy access power distribution network. The method comprises the steps of obtaining associated line loss data of the new energy access power distribution network; according to the associated line loss data, using a time sequence prediction model to output fluctuation characteristic data of the new energy access power distribution network; based on the fluctuation characteristic data, performing line loss calculation on the new energy access power distribution network through Monte Carlo simulation to obtain line loss distribution data; performing optimization analysis on the line loss distribution data by using a deep reinforcement learning algorithm according to the line loss distribution data to obtain an optimal access point and an optimal configuration capacity of the new energy access power distribution network as a line loss optimization scheme of the new energy access power distribution network; according to the invention, through the time sequence prediction model constructed based on the transformation model and the generalized autoregression condition heterovariance model, the fluctuation characteristic data of the power distribution network can be accurately captured, and the deep reinforcement learning algorithm is combined to dynamically adapt to the fluctuation characteristics brought by the access of new energy to the power distribution network, so that the line loss optimization efficiency of the access power distribution network is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power distribution network line loss optimization, and in particular to a method and system for optimizing line loss when a new energy source is connected to a power distribution network. Background Art

[0002] At present, with the rapid development and widespread application of renewable energy technology, more and more new energy sources (such as wind energy, solar energy, etc.) are connected to the distribution network. However, the access of new energy sources has brought many new challenges, one of which is the line loss problem. Line loss refers to the energy loss caused by resistance, inductance and other factors during the transmission of electric energy, which is one of the key factors affecting the efficiency and economy of the power system.

[0003] Existing line loss optimization methods mainly rely on traditional static analysis and optimization techniques, which perform well when dealing with fixed loads and power sources. However, the access of new energy makes the load and power supply of the distribution network highly dynamic and unpredictable, and its volatility characteristics are significant. Specifically, traditional line loss optimization methods often assume that the load and power supply are relatively stable, and fail to effectively consider the volatility characteristics brought about by the access of new energy. This assumption will lead to large errors in practical applications, which in turn affects the accuracy of the optimization results. Due to the failure to accurately capture the volatility characteristics of new energy, this will lead to low accuracy of the line loss calculation results, which cannot truly reflect the actual operation status of the system. It is difficult to find the optimal new energy access point and configuration capacity based on inaccurate line loss calculation results, which will lead to reduced energy utilization efficiency of new energy access to the distribution network and increase the operating cost of the system. Summary of the invention

[0004] In order to solve the problem that the existing line loss optimization method cannot effectively consider the volatility characteristics of the access of renewable energy to the distribution network, thereby reducing the energy utilization efficiency of the access of renewable energy to the distribution network, the present invention proposes a line loss optimization method, system, device and medium for the access of renewable energy to the distribution network, including:

[0005] Obtain the line loss data associated with the access of new energy to the distribution network;

[0006] Outputting the fluctuation characteristic data of the new energy access to the distribution network using a pre-built time series prediction model according to the associated line loss data;

[0007] Based on the fluctuation characteristic data, line loss calculation is performed on the renewable energy access distribution network through Monte Carlo simulation to obtain line loss distribution data of the renewable energy access distribution network;

[0008] According to the line loss distribution data, the line loss distribution data is optimized and analyzed using a deep reinforcement learning algorithm to obtain an optimal access point and an optimal configuration capacity for the new energy to access the distribution network, and the optimal access point and the optimal configuration capacity are used as a line loss optimization solution for the new energy to access the distribution network;

[0009] Wherein, the time series prediction model is constructed based on the transformation model and the generalized autoregressive conditional heteroskedasticity model.

[0010] Optionally, the time series prediction model includes the following construction process:

[0011] Use historical correlation line loss data as training input data;

[0012] Using the fluctuation characteristics corresponding to the historical associated line loss data as training output data;

[0013] Calculating the time series correlation corresponding to the training input data by transforming the multi-head self-attention mechanism in the model;

[0014] Using the time series correlation as an input for training data of a generalized autoregressive conditional heteroskedasticity model;

[0015] Using the training output data as an output item of training data of the generalized autoregressive conditional heteroskedasticity model;

[0016] The generalized autoregressive conditional variance model is trained according to the input items and output items of the training data to obtain a time series prediction model.

[0017] Optionally, the generalized autoregressive conditional variance model is trained according to the input items and output items of the training data to obtain a time series prediction model, including:

[0018] Initialize the model's parameters;

[0019] Inputting the input items of the training data into the generalized autoregressive conditional variance model, and outputting the corresponding volatility prediction value;

[0020] Calculating a negative log-likelihood value between the fluctuation prediction value and the output item of the training data according to the fluctuation prediction value and the output item of the training data;

[0021] According to the negative log-likelihood value, the parameters of the model are adjusted using a gradient descent algorithm until a maximum number of iterations is reached to obtain a time series prediction model.

[0022] Optionally, the expression corresponding to the time series prediction model is as follows:

[0023]

[0024] Among them, h t represents the fluctuation characteristic data at time t; α0 represents a constant value; i represents the lag index of the error square term in the generalized autoregressive conditional heteroskedasticity model; i=1…q; q represents the lag order of the error square term; α i represents the weight coefficient of the i-th error square term; ρ t-i represents the volatility residual value at time ti; β j represents the lag weight coefficient of the jth conditional variance; j = 1…p; p represents the lag order of the conditional variance; h t-j represents the conditional variance at time tj; Attention() represents the multi-head self-attention mechanism in the transformation model; Q represents the features at the current time point; K represents the time series corresponding to the associated line loss data; V represents the weight of the feature value in the time series.

[0025] Optionally, the line loss calculation of the renewable energy access distribution network is performed through Monte Carlo simulation based on the fluctuation characteristic data to obtain the line loss distribution data of the renewable energy access distribution network, including:

[0026] According to the fluctuation characteristic data, a density peak clustering algorithm is used to perform cluster analysis on the fluctuation characteristic data, and output load characteristic data;

[0027] Based on the load characteristic data, line loss calculation is performed on the renewable energy access distribution network through Monte Carlo simulation to obtain line loss distribution data of the renewable energy access distribution network.

[0028] Optionally, the line loss calculation of the renewable energy access distribution network is performed through Monte Carlo simulation based on the load characteristic data to obtain line loss distribution data of the renewable energy access distribution network, including:

[0029] Based on the load characteristic data, a fast decoupling power flow method is used to perform power flow calculation on the new energy access distribution network to obtain power flow distribution data of the new energy access distribution network;

[0030] According to the power flow distribution data, line loss calculation is performed on the power distribution network for accessing the new energy source through Monte Carlo simulation to obtain line loss distribution data for accessing the power distribution network for accessing the new energy source.

[0031] Optionally, based on the load characteristic data, using a fast decoupling power flow method to perform power flow calculation on the new energy access distribution network to obtain power flow distribution data of the new energy access distribution network includes:

[0032] According to the load characteristic data, a multi-scale graph convolutional network is used to extract key nodes from the load characteristic data to obtain node loss data of the new energy access distribution network;

[0033] According to the node loss data, a fast decoupling power flow method is used to perform power flow calculation on the node loss data to obtain power flow distribution data of the new energy access to the power distribution network.

[0034] Optionally, the calculating line loss of the renewable energy access distribution network through Monte Carlo simulation according to the power flow distribution data to obtain the line loss distribution data of the renewable energy access distribution network includes:

[0035] Randomly perturbing the power flow distribution data to generate random power flow data under different power scenarios;

[0036] Performing power flow calculations on the random power flow data under the different power scenarios respectively, and outputting power flow distribution data corresponding to each power scenario;

[0037] According to the power flow distribution data corresponding to each power scenario, the line loss of the new energy access distribution network is calculated to obtain the line loss distribution data of the new energy access distribution network.

[0038] Optionally, the optimizing and analyzing the line loss distribution data using a deep reinforcement learning algorithm according to the line loss distribution data to obtain an optimal access point and an optimal configuration capacity for the new energy to access the distribution network includes:

[0039] Extracting characteristic states of the line loss distribution data to obtain corresponding state characteristic data;

[0040] According to the state feature data, a strategy network for accessing the new energy to the distribution network is established using a deep reinforcement learning algorithm, and topological features of the accessing the new energy to the distribution network are extracted from the strategy network;

[0041] According to the topological characteristics, an optimal access point and an optimal configuration capacity for the new energy to access the distribution network are obtained;

[0042] The state characteristic data includes one or more of the following: node distribution characteristics, line loss distribution characteristics, power output characteristics and load characteristics.

[0043] Optionally, the associated line loss data includes one or more of the following: power generation data of new energy power stations, operation data of new energy access to distribution networks, and user load data;

[0044] The new energy power station power generation data includes one or more of the following: power output data, fluctuation curve data and weather forecast data;

[0045] The operation data of the new energy access distribution network includes one or more of the following: node voltage data, node current data, topology structure and equipment parameter data;

[0046] The user load data includes one or more of the following: electricity consumption classification data, time period load curve data and demand response behavior data.

[0047] Based on the same inventive concept, the present invention also provides a line loss optimization system for accessing a new energy source to a distribution network, comprising:

[0048] A data acquisition module is used to obtain the line loss data associated with the access of new energy to the distribution network;

[0049] A time series prediction module, used to output the fluctuation characteristic data of the new energy access to the distribution network based on the associated line loss data and using a pre-built time series prediction model;

[0050] A line loss calculation module, used to calculate the line loss of the new energy access distribution network through Monte Carlo simulation based on the fluctuation characteristic data, and obtain the line loss distribution data of the new energy access distribution network;

[0051] An optimization analysis module is used to optimize and analyze the line loss distribution data using a deep reinforcement learning algorithm according to the line loss distribution data, obtain an optimal access point and an optimal configuration capacity for the new energy to access the distribution network, and use the optimal access point and the optimal configuration capacity as a line loss optimization solution for the new energy to access the distribution network;

[0052] Wherein, the time series prediction model is constructed based on the transformation model and the generalized autoregressive conditional heteroskedasticity model.

[0053] Optionally, the line loss optimization system further includes: a model building module, which is used to:

[0054] Use historical correlation line loss data as training input data;

[0055] Using the fluctuation characteristics corresponding to the historical associated line loss data as training output data;

[0056] Calculating the time series correlation corresponding to the training input data by transforming the multi-head self-attention mechanism in the model;

[0057] Using the time series correlation as an input for training data of a generalized autoregressive conditional heteroskedasticity model;

[0058] Using the training output data as an output item of training data of the generalized autoregressive conditional heteroskedasticity model;

[0059] The generalized autoregressive conditional variance model is trained according to the input items and output items of the training data to obtain a time series prediction model.

[0060] Optionally, the model building module includes:

[0061] Initialization submodule, used to initialize the parameters of the model;

[0062] A fluctuation prediction submodule, used for inputting the input items of the training data into the generalized autoregressive conditional variance model, and outputting corresponding fluctuation prediction values;

[0063] a parameter optimization submodule, for calculating a negative log-likelihood value between the fluctuation prediction value and the output item of the training data according to the fluctuation prediction value and the output item of the training data;

[0064] The parameter adjustment submodule is used to adjust the parameters of the model according to the negative log-likelihood value using a gradient descent algorithm until a maximum number of iterations is reached to obtain a time series prediction model.

[0065] Optionally, the expression corresponding to the time series prediction model is as follows:

[0066]

[0067] Among them, h t represents the fluctuation characteristic data at time t; α0 represents a constant value; i represents the lag index of the error square term in the generalized autoregressive conditional heteroskedasticity model; i=1…q; q represents the lag order of the error square term; α i represents the weight coefficient of the i-th error square term; ρ t-i represents the volatility residual value at time ti; β j represents the lag weight coefficient of the jth conditional variance; j = 1…p; p represents the lag order of the conditional variance; h t-j represents the conditional variance at time tj; Attention() represents the multi-head self-attention mechanism in the transformation model; Q represents the features at the current time point; K represents the time series corresponding to the associated line loss data; V represents the weight of the feature value in the time series.

[0068] Optionally, the line loss calculation module includes:

[0069] A cluster analysis submodule, used to perform cluster analysis on the fluctuation characteristic data using a density peak clustering algorithm according to the fluctuation characteristic data, and output load characteristic data;

[0070] The line loss distribution submodule is used to calculate the line loss of the renewable energy access distribution network through Monte Carlo simulation based on the load characteristic data to obtain the line loss distribution data of the renewable energy access distribution network.

[0071] Optionally, the line loss distribution submodule includes:

[0072] A power flow calculation unit, configured to perform power flow calculation on the new energy access distribution network using a fast decoupling power flow method based on the load characteristic data, and obtain power flow distribution data of the new energy access distribution network;

[0073] The line loss simulation unit is used to calculate the line loss of the renewable energy access distribution network through Monte Carlo simulation according to the power flow distribution data, so as to obtain the line loss distribution data of the renewable energy access distribution network.

[0074] Optionally, the power flow calculation unit includes:

[0075] A graph analysis subunit is used to extract key nodes from the load characteristic data using a multi-scale graph convolutional network to obtain node loss data of the new energy access distribution network;

[0076] The power flow distribution subunit is used to perform power flow calculation on the node loss data through a fast decoupling power flow method according to the node loss data, so as to obtain the power flow distribution data of the new energy access to the distribution network.

[0077] Optionally, the line loss simulation unit includes:

[0078] A random perturbation subunit, used for randomly perturbing the power flow distribution data to generate random power flow data under different power scenarios;

[0079] A random power flow calculation subunit, used to perform power flow calculation on the random power flow data under the different power scenarios respectively, and output power flow distribution data corresponding to each power scenario;

[0080] The distribution network line loss calculation subunit is used to calculate the line loss of the renewable energy access distribution network according to the power flow distribution data corresponding to each power scenario, and obtain the line loss distribution data of the renewable energy access distribution network.

[0081] Optionally, the optimization analysis module includes:

[0082] A feature extraction submodule is used to extract feature states of the line loss distribution data to obtain corresponding state feature data;

[0083] A strategy network establishment submodule is used to establish a strategy network for accessing the new energy to the distribution network using a deep reinforcement learning algorithm according to the state feature data, and to extract topological features of the accessing the new energy to the distribution network from the strategy network;

[0084] An optimal extraction submodule, used to obtain the optimal access point and optimal configuration capacity of the new energy access to the distribution network according to the topological characteristics;

[0085] The state characteristic data includes one or more of the following: node distribution characteristics, line loss distribution characteristics, power output characteristics and load characteristics.

[0086] Optionally, the associated line loss data includes one or more of the following: power generation data of new energy power stations, operation data of new energy access to distribution networks, and user load data;

[0087] The new energy power station power generation data includes one or more of the following: power output data, fluctuation curve data and weather forecast data;

[0088] The operation data of the new energy access distribution network includes one or more of the following: node voltage data, node current data, topology structure and equipment parameter data;

[0089] The user load data includes one or more of the following: electricity consumption classification data, time period load curve data and demand response behavior data.

[0090] In yet another aspect, the present invention further provides a computer device, comprising: one or more processors;

[0091] A memory for storing one or more programs;

[0092] When the one or more programs are executed by the one or more processors, a line loss optimization method for connecting new energy to a distribution network as described above is implemented.

[0093] On the other hand, the present invention further provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed, the line loss optimization method for connecting a new energy source to a distribution network as described above is implemented.

[0094] Compared with the prior art, the present invention has the following beneficial effects:

[0095] The present invention provides a line loss optimization method, system, device and medium for accessing a new energy distribution network, comprising: obtaining associated line loss data of accessing a new energy distribution network; outputting fluctuation characteristic data of accessing a new energy distribution network by using a pre-constructed time series prediction model according to the associated line loss data; performing line loss calculation on the accessing new energy distribution network by Monte Carlo simulation based on the fluctuation characteristic data, and obtaining line loss distribution data of accessing a new energy distribution network; optimizing and analyzing the line loss distribution data by using a deep reinforcement learning algorithm according to the line loss distribution data, obtaining an optimal access point and an optimal configuration capacity of accessing a new energy distribution network, and using the optimal access point and the optimal configuration capacity as a line loss optimization scheme for accessing a new energy distribution network; wherein the time series prediction model is constructed based on a transformation model and a generalized autoregressive conditional heteroscedasticity model; the present application uses a time series prediction model constructed based on a transformation model and a generalized autoregressive conditional heteroscedasticity model, which is conducive to accurately capturing the fluctuation characteristic data of accessing a new energy distribution network, and can dynamically adapt to the fluctuation characteristics brought by accessing a new energy distribution network in combination with a deep reinforcement learning algorithm, thereby improving the line loss optimization efficiency and increasing the utilization efficiency of new energy. BRIEF DESCRIPTION OF THE DRAWINGS

[0096] Figure 1 A schematic flow chart of a line loss optimization method for connecting a new energy source to a distribution network provided by the present invention;

[0097] Figure 2 A schematic diagram of data flow using a time series prediction model to output fluctuation characteristic data in a line loss optimization method for connecting new energy to a distribution network provided by the present invention;

[0098] Figure 3 A schematic diagram of the structure of a line loss optimization system for connecting new energy to a distribution network provided by the present invention;

[0099] Figure 4 The present invention provides a schematic structural diagram of an electronic device. DETAILED DESCRIPTION

[0100] The present invention proposes a line loss optimization method, system, device and medium for connecting new energy to a distribution network. The specific implementation methods of the present invention are further described in detail below with reference to the accompanying drawings.

[0101] Embodiment 1:

[0102] The present invention provides a method for optimizing line loss when a new energy source is connected to a distribution network. The flow chart is as follows: Figure 1 As shown, including:

[0103] Step 1: Obtain the associated line loss data of renewable energy access to the distribution network;

[0104] Step 2: Based on the associated line loss data, use the pre-built time series prediction model to output the fluctuation characteristic data of the new energy access to the distribution network;

[0105] Step 3: Based on the fluctuation characteristic data, the line loss of the renewable energy access distribution network is calculated through Monte Carlo simulation to obtain the line loss distribution data of the renewable energy access distribution network;

[0106] Step 4: Based on the line loss distribution data, the deep reinforcement learning algorithm is used to optimize and analyze the line loss distribution data to obtain the optimal access point and optimal configuration capacity for the new energy access to the distribution network, and the optimal access point and optimal configuration capacity are used as the line loss optimization solution for the new energy access to the distribution network;

[0107] Among them, the time series forecasting model is constructed based on the transformation model (also called the Transformer model) and the generalized autoregressive conditional heteroskedasticity model (also called the GARCH model).

[0108] Generally, when optimizing the line loss analysis of the new energy access to the distribution network, it is necessary to first obtain the associated line loss data of the distribution network. The acquisition of associated line loss data is directly related to the actual impact of the new energy access on the operation of the distribution network, and provides key input for the construction of the time series prediction model and the line loss optimization analysis. In practical applications, the sources of associated line loss data are wide and diverse. In order to fully reflect the dynamic impact of the new energy access on the operation of the distribution network, it is necessary to combine the power generation characteristics of the new energy power station, the operating status of the distribution network, and the user load behavior and other multi-dimensional data;

[0109] For example, the above-mentioned associated line loss data may include one or more of the following: power generation data of new energy power stations, operation data of new energy access to distribution networks, and user load data;

[0110] For example, the above-mentioned new energy power station power generation data may include one or more of the following: power output data, fluctuation curve data and weather forecast data (such as wind speed, solar radiation intensity, etc.);

[0111] For example, the above-mentioned operation data of the new energy access to the distribution network may include one or more of the following: node voltage data, node current data, topology structure and equipment parameter data;

[0112] For example, the above-mentioned user load data may include one or more of the following: electricity classification data, time period load curve data, and demand response behavior data;

[0113] After collecting the above-mentioned associated line loss data, the associated line loss data is preprocessed (such as de-redundancy and time alignment) through the edge computing node to reduce the burden on the central node and improve the data processing efficiency and stability of the entire system. As a computing unit close to the data source, the edge node can first filter the redundant data after the data collection is completed, such as detecting and deleting duplicate data or invalid data, thereby reducing the proportion of meaningless data in the transmission process. At the same time, the edge node can perform time alignment operations on the associated line loss data from different sources, and ensure the consistency and comparability of the data in the time dimension by standardizing timestamps, filling missing time points and resampling. This preprocessing method not only optimizes data quality, but also reduces the computing load of the central node, allowing it to focus on high-value data analysis tasks; in addition, by performing preprocessing at the edge node, the amount of data transmission can be significantly reduced, thereby reducing bandwidth requirements and communication delays, and improving the real-time and stability of the system.

[0114] In order to further utilize the multi-dimensional information in the above-mentioned associated line loss data to achieve accurate prediction of the fluctuation characteristics of renewable energy access to the distribution network, this can be accomplished by constructing a time series prediction model that adapts to complex time series. Specifically:

[0115] In one implementation, the above time series prediction model may include the following construction process:

[0116] Use historical correlation line loss data as training input data;

[0117] The fluctuation characteristics corresponding to the historical correlation line loss data are used as training output data;

[0118] By transforming the multi-head self-attention mechanism in the model, the time series correlation corresponding to the training input data is calculated;

[0119] Use time series correlation as input for training data of generalized autoregressive conditional heteroskedasticity model;

[0120] The training output data is used as the output item of the training data of the generalized autoregressive conditional heteroskedasticity model;

[0121] According to the input and output items of the training data, the generalized autoregressive conditional variance model is trained to obtain a time series prediction model;

[0122] In this implementation, by combining the multi-head self-attention mechanism and the generalized autoregressive conditional heteroskedasticity model in the transformation model, the Transform model effectively captures the time series characteristics of the data, while the GARCH model is specifically used to analyze and predict data series with strong volatility (especially suitable for the output prediction of new energy such as wind power and solar energy). The correlation modeling of complex time series is integrated with the prediction of volatility characteristics, which significantly improves the adaptability and accuracy of the time series prediction model in the scenario of new energy access to the distribution network. The multi-head self-attention mechanism can capture multi-dimensional and multi-time scale correlations from historical correlated line loss data, especially for the correlation characteristics of long-term and short-term dynamic changes. The generalized autoregressive conditional heteroskedasticity model further focuses on the modeling of volatility characteristics, which can accurately describe the uncertainty and conditional variance characteristics caused by the access of new energy. The synergy of the two can not only improve the model's ability to process complex time series data, but also provide more accurate prediction results for the dynamic analysis of the access of new energy to the distribution network. This implementation method can solve the bottleneck that traditional models are difficult to simultaneously capture nonlinear characteristics and model conditional fluctuations. It provides a reliable technical foundation for optimizing new energy access points, improving grid operation efficiency, and promoting intelligent energy management. It has significant application value and promotion prospects.

[0123] For example, the calculation formula for calculating the time series correlation corresponding to the training input data using the multi-head self-attention mechanism can be as follows:

[0124]

[0125] Among them, Q is the query vector obtained by linear transformation (in this technical scenario, it can be expressed as the feature of the current time point); K is the key vector obtained by linear transformation (in this technical scenario, it can be expressed as the time series corresponding to the associated line loss data); V is the value vector obtained by linear transformation (in this technical scenario, it can be expressed as the weight of the feature value in the time series); d kis the dimension of the key vector (in this technical scenario, it can represent the feature representation of the output time series after position encoding and multi-layer stacking); Softmax represents the activation function; T represents transposition; in this example, the time series correlation of the training input data is calculated through the multi-head self-attention mechanism, which can accurately capture the correlation between the complex dynamic characteristics and multi-dimensional data in the new energy access distribution network, thereby improving the accuracy and robustness of the time series prediction; and the query vector, key vector and value vector are used to model the dependency relationship between the current time point and the historical data respectively, focusing on the reference value of the historical high volatility period to the current power flow distribution, and fully integrating the historical power generation power and load demand characteristics of each node. This mechanism can dynamically weigh the long-term and short-term time dependencies, and further enhance the feature extraction capability of the time series through position encoding and multi-layer stacking, which provides higher modeling flexibility and adaptability for the time series analysis of complex power systems. Finally, accurate time series modeling of power grid power distribution, load demand fluctuations and new energy power generation characteristics is achieved, which not only improves the prediction accuracy, but also provides a scientific basis for optimizing the new energy access strategy and power grid operation efficiency, and lays a solid foundation for the intelligent development of the power system.

[0126] The above time series prediction model fully exploits the time series correlation and volatility characteristics in the historical correlation line loss data by combining the multi-head self-attention mechanism and the generalized autoregressive conditional heteroskedasticity model in the transformation model, laying the foundation for high-precision time series prediction. In order to further improve the model construction process, it is necessary to describe in detail how to train the generalized autoregressive conditional heteroskedasticity model so that it can effectively capture the volatility characteristics and optimize the model parameters. Specifically:

[0127] In one implementation, the process of training the generalized autoregressive conditional variance model based on the input items and output items of the training data to obtain the time series prediction model may include:

[0128] Initialize the model's parameters;

[0129] Input the input items of the training data into the generalized autoregressive conditional variance model and output the corresponding volatility forecast value;

[0130] According to the fluctuation prediction value and the output item of the training data, a negative log likelihood value between the fluctuation prediction value and the output item of the training data is calculated;

[0131] According to the negative log-likelihood value, the gradient descent algorithm is used to adjust the parameters of the model until the maximum number of iterations is reached to obtain the time series prediction model; in this implementation method, by accurately describing the training process of the generalized autoregressive conditional variance model, combined with the negative log-likelihood value as the loss function, the gradient descent algorithm is used to iteratively optimize the model parameters, thereby achieving high-precision modeling of complex fluctuation characteristics. In the process of model training, the initialization parameters, step-by-step optimization and maximum iteration strategies are used to ensure the convergence and stability of the model, providing strong technical support for capturing the fluctuation characteristics of new energy access to the distribution network; and the probability distribution difference between the predicted value and the true value is measured by the negative log-likelihood value, which can not only quantify the model's fitting effect on the fluctuation characteristics, but also gradually improve the model's adaptability to complex data patterns through iterative optimization, which is significantly innovative. Compared with traditional methods, this implementation method shows obvious advantages in the accuracy, robustness and stability of dynamic fluctuation characteristics prediction, providing a scientific prediction basis for the optimization of new energy access points, the improvement of power grid operation efficiency and load management, thereby greatly promoting the development of theoretical research and application practice of smart grids;

[0132] For example, the expression corresponding to the above time series prediction model can be as follows:

[0133]

[0134] Among them, h t represents the fluctuation characteristic data at time t; α0 represents a constant value; i represents the lag index of the error square term in the generalized autoregressive conditional heteroskedasticity model; i = 1…q; q represents the lag order of the error square term; α i represents the weight coefficient of the i-th error square term; ρ t-i represents the volatility residual value at time ti; β j represents the lag weight coefficient of the jth conditional variance; j = 1…p; p represents the lag order of the conditional variance; h t-jrepresents the conditional variance at time tj; Attention() represents the multi-head self-attention mechanism in the transformation model; Q represents the characteristics of the current time point; K represents the time series corresponding to the associated line loss data; V represents the weight of the eigenvalue in the time series; In this example, by combining the generalized autoregressive conditional heteroskedasticity model with the multi-head self-attention mechanism, an improved time series prediction model expression is proposed, which can accurately capture the complex fluctuation characteristics and dynamic correlation characteristics of the new energy access to the distribution network. The conditional variance part in the expression uses the lagged information of the squared error term and the historical characteristics of the conditional variance to provide a deep characterization of the volatility of the time series, and the multi-head self-attention mechanism further enhances the model's time series correlation mining ability, and dynamically matches and weights the characteristics of the current time point with the historical associated line loss data. The introduction of the attention mechanism not only expands the scope of application of the traditional GARCH model, but also significantly improves the model's sensitivity to nonlinearity and long-term and short-term dependencies, and can show higher prediction accuracy and robustness in complex power system scenarios. This fusion expression breaks the linear limitations of the traditional conditional variance model, dynamically adjusts the influence weight of historical data through the attention mechanism, solves the traditional modeling problems caused by the volatility of renewable energy power generation and the randomness of load demand, and can provide highly reliable prediction support for the optimization of renewable energy access points, the improvement of grid operation efficiency, and load dispatch management. It not only injects new theoretical innovations into the intelligent development of power systems, but also opens up new solutions to time series prediction and optimization problems in practical application scenarios.

[0135] For example, the calculation formula corresponding to the negative log-likelihood value between the above-mentioned fluctuation prediction value and the output item of the training data is as follows:

[0136]

[0137] Where L represents the negative log-likelihood value; Y t is the true value at time t; y t is the predicted value at time t; h t represents the fluctuation characteristic data at time t; t = 1…N; N represents the total number of time points; in this example, the deviation between the predicted value and the true value and the dynamic change of the fluctuation characteristic data are unified into the evaluation framework through the calculation formula of the negative logarithmic likelihood value, creatively realizing the quantification and optimization of the fitting effect of the time series prediction model. By constructing L in the formula, it is possible to capture not only the prediction error (Y t -y t ) 2 The influence of the fluctuation characteristics h tThe logarithmic constraint dynamically balances the contribution of the error to the overall loss. This design combines the dynamic fluctuation characteristics of the data and the accuracy requirements of the prediction, so that the time series prediction model can more effectively adapt to the complex time series data characteristics in the access of renewable energy to the distribution network, especially in scenarios with high volatility and randomness. It significantly improves the robustness and prediction accuracy of the model, and provides a clear direction and theoretical basis for model parameter optimization.

[0138] Through the calculation of the negative log-likelihood value and the optimization of model parameters, the generalized autoregressive conditional variance model can accurately extract the fluctuation characteristics in the training data, laying a solid foundation for further analysis of the dynamic characteristics and line loss distribution of renewable energy access to the distribution network. On this basis, in order to more comprehensively quantify the specific impact of fluctuation characteristics on the line loss of the distribution network, cluster analysis and Monte Carlo simulation can be used to combine the fluctuation characteristic data output by the model with the actual operation of the distribution network, so as to deeply explore the line loss distribution characteristics under different load characteristics. Specifically:

[0139] In one implementation, the above-mentioned process of calculating the line loss of the renewable energy access distribution network through Monte Carlo simulation based on the fluctuation characteristic data to obtain the line loss distribution data of the renewable energy access distribution network may include:

[0140] According to the fluctuation characteristic data, the density peaks clustering algorithm (DPC) is used to cluster the fluctuation characteristic data and output the load characteristic data;

[0141] Based on the load characteristic data, the line loss of the renewable energy access distribution network is calculated through Monte Carlo simulation to obtain the line loss distribution data of the renewable energy access distribution network;

[0142] In this implementation, by combining the density peak clustering algorithm and Monte Carlo simulation, the distribution calculation of distribution network line loss based on fluctuation characteristic data is realized, providing an efficient and accurate solution for the dynamic characteristic analysis of new energy access to the distribution network; the density peak clustering algorithm can automatically identify representative load characteristic groupings according to the distribution of fluctuation characteristic data, thereby effectively capturing the characteristic differences of different load types in the new energy access scenario. This unsupervised clustering analysis not only avoids the deviation of artificially set categories, but also can more flexibly adapt to the complex power grid operation environment. By combining Monte Carlo simulation with comprehensive simulation and analysis of random scenarios, it can accurately evaluate the line loss distribution under different load characteristics, and provide reliable quantitative results for new energy access scenarios with strong dynamic fluctuations and randomness. This implementation combines clustering with probabilistic simulation, deepens the understanding of line loss distribution from both data structure and random scenarios, extracts regular features of load characteristics through clustering process, and uses Monte Carlo simulation to perform multi-scenario iterative calculations, which can fully reflect the distribution characteristics and fluctuation range of line losses, and provides a high-value scientific basis for distribution network operation optimization, energy dispatch and policy formulation. This method not only improves the accuracy and applicability of line loss distribution calculations, but also opens up a new direction for complex power system operation analysis, and provides technical support for uncertainty management brought about by new energy access.

[0143] For example, the above-mentioned load characteristic data may include: user classification labels and load center curves of each type of user;

[0144] For example, the expression for clustering analysis of fluctuation feature data using the density peak clustering algorithm can be as follows:

[0145]

[0146] in, represents the i0th user cluster center; is the local density of the i0th user (in this technical scenario, it can represent the user load aggregation density); is the minimum distance from the ith user to a higher density point (in this technical scenario, it can represent the difference between the user's load and other users); in this example, by applying the core expression of the density peak clustering algorithm to the technical scenario of new energy access to the distribution network, accurate clustering of user load characteristics is achieved. This calculation formula combines the local aggregation density of user loads with the differences between user load characteristics, so that the clustering process can simultaneously consider the concentration and discrete characteristics of user loads, thereby automatically identifying user load centers with significant representativeness. This clustering strategy that combines density and difference not only improves the algorithm's adaptability to complex data distribution, but also effectively avoids the problem of sensitivity to the initial clustering center in traditional clustering methods.

[0147] By using the density peak clustering algorithm to cluster the fluctuation characteristic data, we can obtain load characteristic data that can accurately reflect the user load characteristic law. These load characteristic data not only provide high-quality input for the calculation of line loss distribution of new energy access to the distribution network, but also lay a data foundation for the subsequent analysis of the distribution network operation status and power flow distribution. In order to calculate the line loss distribution characteristics more precisely, the Monte Carlo simulation process based on load characteristic data needs to be combined with a specific power flow calculation method to accurately describe the dynamic operation characteristics of the distribution network. Specifically:

[0148] In one implementation, the above process of calculating the line loss of the renewable energy access distribution network through Monte Carlo simulation based on the load characteristic data to obtain the line loss distribution data of the renewable energy access distribution network may include:

[0149] Based on the load characteristic data, the fast decoupling power flow method is used to calculate the power flow of the new energy access distribution network, and the power flow distribution data of the new energy access distribution network is obtained;

[0150] According to the power flow distribution data, the line loss of the new energy access distribution network is calculated through Monte Carlo simulation to obtain the line loss distribution data of the new energy access distribution network;

[0151] In this implementation, the fast decoupled power flow method can use its efficient computing characteristics to quickly obtain power flow distribution data in large-scale power grids and accurately reflect the operating status of distribution networks under different load characteristics; Monte Carlo simulation simulates the operation of the power grid multiple times through random sampling methods, which can fully consider the various possibilities of new energy fluctuations and provide more accurate line loss distribution information. Through efficient line loss distribution simulation, it can evaluate the impact of new energy access to the distribution network in real time and dynamically, and quantitatively analyze the line loss under different access points and configuration capacities, providing a basis for optimization decision-making. This implementation integrates physical modeling with probabilistic analysis, so that line loss calculation is not limited to a single static scene, but can dynamically capture the global impact of load characteristic changes on line loss distribution. This method breaks through the limitations of traditional single-scenario line loss calculation, and can provide comprehensive line loss analysis results under complex conditions caused by new energy output volatility and load demand diversity, providing high-value decision-making basis for power grid operation optimization, energy scheduling and system planning. By accurately analyzing the relationship between different load characteristics and power flow distribution, the accuracy and adaptability of line loss calculation are significantly improved, providing theoretical support and practical feasibility for building an intelligent and dynamic modern distribution network.

[0152] By combining the fast decoupling power flow method based on load characteristic data with Monte Carlo simulation, the line loss distribution data of the new energy access distribution network can be effectively obtained, providing accurate quantitative results for the dynamic analysis of the power grid operation status. However, in order to further improve the accuracy and efficiency of power flow calculation, especially when dealing with complex distribution network topology, a key node extraction mechanism can be introduced to perform a more detailed analysis of the load characteristic data. By extracting key node data through a multi-scale graph convolutional network, it can not only efficiently capture the node losses with significant characteristics in the distribution network, but also provide more accurate input for the power flow calculation process. Specifically:

[0153] In one implementation, the process of calculating the power flow of the new energy access distribution network using the fast decoupling power flow method based on the load characteristic data to obtain the power flow distribution data of the new energy access distribution network may include:

[0154] A topological graph is constructed based on the load characteristic data, and a multi-scale graph convolutional network is used to extract key nodes from the load characteristic data to obtain node loss data for the new energy access distribution network.

[0155] According to the node loss data, the power flow calculation is performed on the node loss data through the fast decoupling power flow method to obtain the power flow distribution data of the new energy access distribution network;

[0156] By way of example, the node loss data mentioned above may include: key nodes and high-loss branches;

[0157] In this implementation, by combining the multi-scale graph convolutional network and the fast decoupling power flow method, the criticality and loss characteristics of nodes in the new energy access distribution network are accurately extracted and efficient power flow calculation is achieved, thereby achieving the dual goals of improving computing efficiency and analysis accuracy under complex power grid operation environments; the topology map is constructed using load characteristic data, and the key nodes are extracted through the multi-scale graph convolutional network, and the active and reactive power balance equations are decoupled by the fast decoupling power flow method to perform power flow calculation. This can not only capture the nodes and high-loss branches with significant influence in the distribution network, but also dynamically adapt to the changes in network topology under different operating conditions. The data extraction of key nodes and high-loss branches provides a more focused input for power flow calculation, effectively reduces the calculation scope, reduces redundant calculations, and significantly improves the computational efficiency of the fast decoupling power flow method.

[0158] By using load characteristic data to construct a topological map and combining a multi-scale graph convolutional network to extract key nodes and high-loss branches, we can efficiently obtain node loss data for the access of new energy to the distribution network and provide accurate input for the rapid decoupling power flow method. Based on these node loss data, the power flow distribution data calculated by the rapid decoupling power flow method can not only reflect the power flow law in the operation of the power grid, but also provide basic support for further analysis of line loss distribution under different operating scenarios. In order to quantify the impact of power flow fluctuations caused by the access of new energy on the line loss of the distribution network, based on the power flow distribution data, a Monte Carlo simulation method can be introduced to generate power flow data under various power scenarios through random disturbances, so as to comprehensively evaluate the line loss distribution characteristics of the distribution network in multi-scenario analysis. Specifically:

[0159] In one implementation, the above process of calculating the line loss of the new energy access distribution network through Monte Carlo simulation based on the power flow distribution data to obtain the line loss distribution data of the new energy access distribution network may include:

[0160] Randomly perturb the power flow distribution data to generate random power flow data under different power scenarios;

[0161] Perform power flow calculations on random power flow data under different power scenarios, and output power flow distribution data corresponding to each power scenario;

[0162] According to the power flow distribution data corresponding to each power scenario, the line loss of the new energy access distribution network is calculated to obtain the line loss distribution data of the new energy access distribution network;

[0163] In this implementation, the Monte Carlo simulation method is used to randomly perturb the power flow distribution data, and multiple power scenarios are generated from the predicted fluctuation data, which can effectively simulate the uncertainty and randomness caused by the fluctuation of new energy output and the change of load demand. This scenario-based dynamic analysis method makes the line loss calculation no longer limited to a single or static operating condition, but can fully reflect the line loss change characteristics of the distribution network under different operating conditions. The introduction of Monte Carlo simulation further improves the accuracy and robustness of line loss analysis through multiple iterative calculations and result statistics; the combination of multi-scenario power flow calculation and line loss assessment can capture the global laws of line loss distribution and potential risks under extreme conditions, providing a scientific basis and high-value decision support for the optimization of power grid operation under the scenario of new energy access.

[0164] The line loss distribution data of renewable energy access to the distribution network obtained through Monte Carlo simulation not only fully reflects the line loss characteristics under various power scenarios, but also provides high-quality basic data for further optimization analysis. These line loss distribution data cover the fluctuation patterns and extreme scenarios of the distribution network under different operating conditions, laying a solid foundation for identifying potential optimization space and decision variables in the system. In order to further improve the operating efficiency and rationality of resource allocation of renewable energy access to the distribution network, we can use deep reinforcement learning algorithms to optimize and analyze these line loss distribution data and explore the key characteristics of the distribution network operation. Specifically:

[0165] In one implementation, the above process of optimizing and analyzing the line loss distribution data using a deep reinforcement learning algorithm based on the line loss distribution data to obtain the optimal access point and optimal configuration capacity for the new energy to access the distribution network may include:

[0166] Extract characteristic states of line loss distribution data to obtain corresponding state characteristic data;

[0167] Based on the state feature data, a deep reinforcement learning algorithm is used to establish a strategy network for the access of new energy to the distribution network, and the topological features of the access of new energy to the distribution network are extracted from the strategy network;

[0168] According to the topological characteristics, the optimal access point and optimal configuration capacity of the new energy access distribution network are obtained;

[0169] For example, the above-mentioned state characteristic data may include one or more of the following: node distribution characteristics, line loss distribution characteristics, power output characteristics and load characteristics.

[0170] In this implementation, the policy network automatically selects the node with the smallest line loss and the highest reliability as the optimal access point by analyzing the topological characteristics, and the new energy capacity allocated to the access point is the optimal configuration capacity; it performs optimization analysis based on the line loss distribution data and automatically learns the optimal access point and configuration capacity; and introduces deep reinforcement learning to continuously optimize the strategy through the interaction between the policy network and the environment. Ultimately, it can determine the new energy access point and the optimal configuration capacity that are most suitable for the current state of the distribution network, and can dynamically adjust the specific location and capacity of new energy access, thereby minimizing the negative impact of new energy access on grid operation (such as increased line loss and energy waste) and achieving higher energy utilization efficiency.

[0171] In summary, the present invention aims at the problem that the existing line loss optimization method cannot effectively consider the volatility characteristics of the access of renewable energy to the distribution network, thereby reducing the energy utilization efficiency of the access of renewable energy to the distribution network. A line loss optimization method for the access of renewable energy to the distribution network is proposed, such as Figure 2As shown, by collecting associated line loss data (including power generation data of local power sources such as new energy, operation data of distribution network and user load data), the time series prediction model is used to predict the fluctuation characteristic data of new energy access to distribution network. The time series prediction model can accurately capture the fluctuation data of new energy. Taking into account the historical volatility, fluctuation law and the influence of external factors, through accurate fluctuation prediction, the changing trend of new energy output can be predicted in advance, providing reliable input data for subsequent optimization and reducing the impact of volatility on power grid operation; using the Monte Carlo simulation method, through multi-scenario random analysis, the line loss distribution characteristics under different operating conditions are quantified, providing a scientific basis for evaluating the impact of new energy access; and combining deep reinforcement learning with the interaction of the environment to optimize the access point and capacity configuration, the access point and capacity are accurately optimized, and the method of the present invention comprehensively considers the influence of various factors on the line loss of the distribution network, can dynamically adapt to the changing distribution network environment, effectively reduce the line loss of the power grid, and reduce the operating cost caused by invalid access, and improve the efficiency and economy of power grid operation, so as to realize the optimization analysis of the line loss of the distribution network.

[0172] Embodiment 2:

[0173] The present invention based on the same inventive concept also provides a line loss optimization system for connecting new energy to a distribution network, the structural composition diagram of which is shown in FIG. Figure 3 As shown, including:

[0174] A data acquisition module is used to obtain the line loss data associated with the access of new energy to the distribution network;

[0175] The time series prediction module is used to output the fluctuation characteristic data of the new energy access to the distribution network based on the associated line loss data and the pre-built time series prediction model;

[0176] The line loss calculation module is used to calculate the line loss of the new energy access distribution network through Monte Carlo simulation based on the fluctuation characteristic data, and obtain the line loss distribution data of the new energy access distribution network;

[0177] The optimization analysis module is used to optimize and analyze the line loss distribution data using a deep reinforcement learning algorithm, obtain the optimal access point and optimal configuration capacity for the new energy access to the distribution network, and use the optimal access point and optimal configuration capacity as the line loss optimization solution for the new energy access to the distribution network;

[0178] Among them, the time series forecasting model is constructed based on the transformation model and the generalized autoregressive conditional heteroskedasticity model;

[0179] For example, the above-mentioned associated line loss data may include one or more of the following: power generation data of new energy power stations, operation data of new energy access to distribution networks, and user load data;

[0180] For example, the above-mentioned new energy power station power generation data may include one or more of the following: power output data, fluctuation curve data and weather forecast data;

[0181] For example, the above-mentioned operation data of the new energy access to the distribution network may include one or more of the following: node voltage data, node current data, topology structure and equipment parameter data;

[0182] For example, the above-mentioned user load data may include one or more of the following: electricity consumption classification data, time period load curve data, and demand response behavior data.

[0183] In one implementation, the above-mentioned line loss optimization system may further include: a model building module, specifically used for:

[0184] Use historical correlation line loss data as training input data;

[0185] The fluctuation characteristics corresponding to the historical correlation line loss data are used as training output data;

[0186] By transforming the multi-head self-attention mechanism in the model, the time series correlation corresponding to the training input data is calculated;

[0187] Use time series correlation as input for training data of generalized autoregressive conditional heteroskedasticity model;

[0188] The training output data is used as the output item of the training data of the generalized autoregressive conditional heteroskedasticity model;

[0189] According to the input and output items of the training data, the generalized autoregressive conditional variance model is trained to obtain the time series prediction model.

[0190] In one implementation, the above-mentioned model building module may include:

[0191] Initialization submodule, used to initialize the parameters of the model;

[0192] The volatility prediction submodule is used to input the input items of the training data into the generalized autoregressive conditional variance model and output the corresponding volatility prediction value;

[0193] A parameter optimization submodule, for calculating a negative log-likelihood value between the fluctuation prediction value and the output item of the training data according to the fluctuation prediction value and the output item of the training data;

[0194] The parameter adjustment submodule is used to adjust the parameters of the model according to the negative log-likelihood value using the gradient descent algorithm until the maximum number of iterations is reached to obtain a time series prediction model.

[0195] For example, the expression corresponding to the above time series prediction model can be as follows:

[0196]

[0197] Among them, h t represents the fluctuation characteristic data at time t; α0 represents a constant value; i represents the lag index of the error square term in the generalized autoregressive conditional heteroskedasticity model; i = 1…q; q represents the lag order of the error square term; α i represents the weight coefficient of the i-th error square term; ρ t-i represents the volatility residual value at time ti; β j represents the lag weight coefficient of the jth conditional variance; j = 1…p; p represents the lag order of the conditional variance; h t-j represents the conditional variance at time tj; Attention() represents the multi-head self-attention mechanism in the transformation model; Q represents the features at the current time point; K represents the time series corresponding to the associated line loss data; V represents the weight of the feature value in the time series.

[0198] In one implementation, the line loss calculation module may include:

[0199] The cluster analysis submodule is used to perform cluster analysis on the fluctuation characteristic data by using the density peak clustering algorithm according to the fluctuation characteristic data, and output the load characteristic data;

[0200] The line loss distribution submodule is used to calculate the line loss of the renewable energy access distribution network based on the load characteristic data through Monte Carlo simulation to obtain the line loss distribution data of the renewable energy access distribution network.

[0201] In this implementation, the line loss distribution submodule may include:

[0202] A power flow calculation unit is used to calculate the power flow of the new energy access distribution network based on the load characteristic data by using the fast decoupling power flow method to obtain the power flow distribution data of the new energy access distribution network;

[0203] The line loss simulation unit is used to calculate the line loss of the renewable energy access distribution network through Monte Carlo simulation according to the power flow distribution data, and obtain the line loss distribution data of the renewable energy access distribution network.

[0204] In this implementation, the power flow calculation unit may include:

[0205] The graph analysis subunit is used to extract key nodes from the load characteristic data using a multi-scale graph convolutional network to obtain node loss data for the new energy access distribution network;

[0206] The power flow distribution subunit is used to perform power flow calculation on the node loss data through the fast decoupling power flow method according to the node loss data, and obtain the power flow distribution data of the new energy access to the distribution network.

[0207] In this implementation, the line loss simulation unit may include:

[0208] The random perturbation subunit is used to randomly perturb the power flow distribution data to generate random power flow data under different power scenarios;

[0209] The random power flow calculation subunit is used to perform power flow calculation on the random power flow data under different power scenarios and output the power flow distribution data corresponding to each power scenario;

[0210] The distribution network line loss calculation subunit is used to calculate the line loss of the new energy access distribution network according to the power flow distribution data corresponding to each power scenario, and obtain the line loss distribution data of the new energy access distribution network.

[0211] In one implementation, the above-mentioned optimization analysis module may include:

[0212] The feature extraction submodule is used to extract the feature state of the line loss distribution data to obtain the corresponding state feature data;

[0213] The strategy network establishment submodule is used to establish a strategy network for connecting new energy to the distribution network based on state feature data using a deep reinforcement learning algorithm, and to extract topological features of connecting new energy to the distribution network from the strategy network;

[0214] The optimal extraction submodule is used to obtain the optimal access point and optimal configuration capacity of the new energy access distribution network according to the topological characteristics;

[0215] For example, the above-mentioned state characteristic data may include one or more of the following: node distribution characteristics, line loss distribution characteristics, power output characteristics and load characteristics.

[0216] Embodiment 3:

[0217] like Figure 4 As shown, the present invention also provides an electronic device, which may be a computer device, a single-chip device, an intelligent mobile device, etc. The electronic device in this embodiment may include a processor, a memory, a transceiver component, etc. The memory, the processor, and the transceiver component are connected via a bus; the memory may be used to store an execution program, and an exemplary execution program may include instructions; the processor is used to execute the instructions stored in the memory. The memory may also be used to store data, which may be called and / or modified when the instructions are executed.

[0218] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions in a storage medium to implement the corresponding method flow or corresponding functions, so as to realize the steps of a line loss optimization method for accessing a new energy distribution network in the above-mentioned embodiment.

[0219] Embodiment 4:

[0220] Based on the same inventive concept, the present invention also provides a readable storage medium, specifically an electronic device readable storage medium (Memory), which is a memory device in an electronic device for storing programs and data. It can be understood that the storage medium here can include both built-in storage media in electronic devices and, of course, extended storage media supported by electronic devices. The storage medium provides a storage space, which stores the operating system of the terminal. In addition, one or more instructions suitable for being loaded and executed by a processor are also stored in the storage space, and these instructions can be one or more execution programs (including program codes). It should be noted that the storage medium here can be a high-speed RAM memory or a non-volatile memory, such as at least one disk memory. The processor loads and executes one or more instructions stored in the storage medium, which can implement the steps of a line loss optimization method for connecting new energy to a distribution network in the above embodiment.

[0221] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0222] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0223] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0224] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0225] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit its protection scope. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that after reading the present invention, those skilled in the art can still make various changes, modifications or equivalent substitutions to the specific implementation methods of the application, but these changes, modifications or equivalent substitutions are all within the protection scope of the claims to be approved.

Claims

1. A line loss optimization method for connecting new energy to a distribution network, characterized in that: include: Obtain the line loss data associated with the access of new energy to the distribution network; Outputting the fluctuation characteristic data of the new energy access to the distribution network using a pre-built time series prediction model according to the associated line loss data; Based on the fluctuation characteristic data, line loss calculation is performed on the renewable energy access distribution network through Monte Carlo simulation to obtain line loss distribution data of the renewable energy access distribution network; According to the line loss distribution data, the line loss distribution data is optimized and analyzed using a deep reinforcement learning algorithm to obtain an optimal access point and an optimal configuration capacity for the new energy to access the distribution network, and the optimal access point and the optimal configuration capacity are used as a line loss optimization solution for the new energy to access the distribution network; Wherein, the time series prediction model is constructed based on the transformation model and the generalized autoregressive conditional heteroskedasticity model.

2. The method according to claim 1, characterized in that The time series prediction model includes the following construction process: Use historical correlation line loss data as training input data; Using the fluctuation characteristics corresponding to the historical associated line loss data as training output data; Calculating the time series correlation corresponding to the training input data by transforming the multi-head self-attention mechanism in the model; Using the time series correlation as an input for training data of a generalized autoregressive conditional heteroskedasticity model; Using the training output data as an output item of training data of the generalized autoregressive conditional heteroskedasticity model; Training the generalized autoregressive conditional variance model according to the input items and output items of the training data to obtain a time series prediction model; The associated line loss data includes one or more of the following: power generation data of new energy power stations, operation data of new energy access to distribution networks, and user load data; The new energy power station power generation data includes one or more of the following: power output data, fluctuation curve data and weather forecast data; The operation data of the new energy access distribution network includes one or more of the following: node voltage data, node current data, topology structure and equipment parameter data; The user load data includes one or more of the following: electricity consumption classification data, time period load curve data and demand response behavior data.

3. The method according to claim 2, characterized in that The generalized autoregressive conditional variance model is trained according to the input items and output items of the training data to obtain a time series prediction model, including: Initialize the model parameters; Inputting the input items of the training data into the generalized autoregressive conditional variance model, and outputting the corresponding volatility prediction value; Calculating a negative log-likelihood value between the fluctuation prediction value and the output item of the training data according to the fluctuation prediction value and the output item of the training data; According to the negative log-likelihood value, the parameters of the model are adjusted using a gradient descent algorithm until a maximum number of iterations is reached to obtain a time series prediction model.

4. The method according to any one of claims 1 to 3, characterized in that: The expression corresponding to the time series prediction model is as follows: Among them, h t represents the fluctuation characteristic data at time t; α0 represents a constant value; i represents the lag index of the error square term in the generalized autoregressive conditional heteroskedasticity model; i=1…q; q represents the lag order of the error square term; α i represents the weight coefficient of the i-th error square term; ρ t-i represents the volatility residual value at time ti; β j represents the lag weight coefficient of the jth conditional variance; j = 1…p; p represents the lag order of the conditional variance; h t-j represents the conditional variance at time tj; Attention() represents the multi-head self-attention mechanism in the transformation model; Q represents the features at the current time point; K represents the time series corresponding to the associated line loss data; V represents the weight of the feature value in the time series.

5. The method according to claim 1, characterized in that The line loss calculation of the renewable energy access distribution network is performed based on the fluctuation characteristic data through Monte Carlo simulation to obtain the line loss distribution data of the renewable energy access distribution network, including: According to the fluctuation characteristic data, a density peak clustering algorithm is used to perform cluster analysis on the fluctuation characteristic data, and output load characteristic data; Based on the load characteristic data, line loss calculation is performed on the renewable energy access distribution network through Monte Carlo simulation to obtain line loss distribution data of the renewable energy access distribution network.

6. The method according to claim 5, characterized in that The method of calculating the line loss of the new energy access distribution network through Monte Carlo simulation based on the load characteristic data to obtain the line loss distribution data of the new energy access distribution network includes: Based on the load characteristic data, a fast decoupling power flow method is used to perform power flow calculation on the new energy access distribution network to obtain power flow distribution data of the new energy access distribution network; According to the power flow distribution data, line loss calculation is performed on the power distribution network for accessing the new energy source through Monte Carlo simulation to obtain line loss distribution data for accessing the power distribution network for accessing the new energy source.

7. The method according to claim 6, characterized in that The method of performing power flow calculation on the new energy access distribution network based on the load characteristic data using a fast decoupling power flow method to obtain power flow distribution data of the new energy access distribution network includes: According to the load characteristic data, a multi-scale graph convolutional network is used to extract key nodes from the load characteristic data to obtain node loss data of the new energy access distribution network; According to the node loss data, a fast decoupling power flow method is used to perform power flow calculation on the node loss data to obtain power flow distribution data of the new energy access to the power distribution network.

8. The method according to claim 6, characterized in that The method of calculating the line loss of the new energy access distribution network through Monte Carlo simulation according to the power flow distribution data to obtain the line loss distribution data of the new energy access distribution network includes: Randomly perturbing the power flow distribution data to generate random power flow data under different power scenarios; Performing power flow calculations on the random power flow data under the different power scenarios respectively, and outputting power flow distribution data corresponding to each power scenario; According to the power flow distribution data corresponding to each power scenario, the line loss of the new energy access distribution network is calculated to obtain the line loss distribution data of the new energy access distribution network.

9. The method according to claim 1, characterized in that The optimizing and analyzing the line loss distribution data by using a deep reinforcement learning algorithm according to the line loss distribution data to obtain the optimal access point and optimal configuration capacity of the new energy access to the distribution network includes: Extracting characteristic states of the line loss distribution data to obtain corresponding state characteristic data; According to the state feature data, a strategy network for accessing the new energy to the distribution network is established using a deep reinforcement learning algorithm, and topological features of the accessing the new energy to the distribution network are extracted from the strategy network; According to the topological characteristics, an optimal access point and an optimal configuration capacity for the new energy to access the distribution network are obtained; The state characteristic data includes one or more of the following: node distribution characteristics, line loss distribution characteristics, power output characteristics and load characteristics.

10. A line loss optimization system for connecting new energy to a distribution network, characterized in that: include: A data acquisition module is used to obtain the line loss data associated with the access of new energy to the distribution network; A time series prediction module, used to output the fluctuation characteristic data of the new energy access to the distribution network based on the associated line loss data and using a pre-built time series prediction model; A line loss calculation module, used to calculate the line loss of the new energy access distribution network through Monte Carlo simulation based on the fluctuation characteristic data, and obtain the line loss distribution data of the new energy access distribution network; An optimization analysis module is used to optimize and analyze the line loss distribution data using a deep reinforcement learning algorithm according to the line loss distribution data, obtain an optimal access point and an optimal configuration capacity for the new energy to access the distribution network, and use the optimal access point and the optimal configuration capacity as a line loss optimization solution for the new energy to access the distribution network; Wherein, the time series prediction model is constructed based on the transformation model and the generalized autoregressive conditional heteroskedasticity model.

Images