Medium-voltage distribution network capacity-load ratio calculation method based on deep neural network

Abstract

The invention provides a medium-voltage power distribution network capacity-load ratio calculation method based on a deep neural network. The method comprises the following steps: collecting multi-source data of a medium-voltage power distribution network; designing an adaptive hierarchical clustering algorithm to dynamically identify the load type according to the high-dimensional feature matrix; constructing time sequence multi-stage recurrent neural network load prediction; calculating the current capacity-load ratio of the power grid according to the load prediction, and carrying out the preliminary optimization of the current capacity-load ratio; constructing a deep learning model according to the preliminarily optimized load prediction to deeply optimize an energy management strategy; and monitoring the operation condition of the medium-voltage power distribution network in real time, and optimizing the deep learning model and the energy management strategy by using the reinforcement learning model. According to the method, the accuracy and real-time performance of calculation of the capacity-load ratio of the medium-voltage distribution network can be remarkably improved, the defect that dynamic adjustment and optimization are difficult to carry out in a complex load environment in the prior art is overcome, and stable operation and power supply safety of the power grid are effectively guaranteed.

Description

Technical Field

[0001] The present invention belongs to the technical field of deep neural networks, and in particular to a method for calculating the capacity-load ratio of a medium-voltage distribution network based on a deep neural network. Background Art

[0002] The medium-voltage distribution network is an important part of the power system, responsible for transmitting electricity from the substation to each user terminal. In the power supply system of modern cities, the load changes of the medium-voltage distribution network are complex and diverse, and are affected by various factors such as seasonality, weather, and user behavior, resulting in a high degree of uncertainty in its operating status. In order to ensure power supply safety and power quality, the operators of the distribution network need to accurately grasp the capacity utilization ratio of the power grid, that is, the ratio between the load capacity of the transformer and the line in actual operation and its rated capacity. The accurate calculation and dynamic adjustment of the capacity utilization ratio not only contributes to the rational allocation of power resources, but also effectively prevents equipment overload and power outages, and improves the reliability and economy of the power grid.

[0003] At present, the traditional methods for calculating the capacity-load ratio of medium-voltage distribution networks mainly rely on static models and fixed parameter settings. These methods are usually based on historical data and empirical rules and cannot respond to dynamic changes in load in real time. With the popularization of distributed energy, smart meters and various new types of power equipment in distribution networks, the load form of power grids has become more diverse and complex, making traditional static models unable to effectively adapt to rapidly changing power demand. In addition, traditional methods lack comprehensive consideration of factors such as power grid topology, load type and time series changes during the calculation process, resulting in low accuracy in capacity-load ratio prediction and difficulty in coping with the complex nonlinear dynamic behavior of power systems.

[0004] Another shortcoming of the existing technology is the lack of responsiveness to real-time load fluctuations and intelligent adjustment mechanisms. Traditional capacity-to-load ratio calculation methods cannot fully utilize the increasingly rich real-time data and environmental information in the distribution network, and it is difficult to accurately predict and optimize scheduling for different load conditions. Due to changes in the power grid operating environment and the volatility of power demand, these methods are slow to respond to sudden load changes and equipment failures, and adjustments are not timely, which may cause local overload of the power grid or uneven load distribution, which in turn triggers a chain reaction and affects the stability and security of the entire power grid. Summary of the invention

[0005] The purpose of this invention is to design a method for calculating the capacity-to-load ratio of a medium-voltage distribution network based on a deep neural network. By combining the latest time series deep learning technology and load type identification technology, the present invention can predict the load changes of the power grid in real time and dynamically optimize the load through an intelligent energy management system. This method breaks through the limitations of the traditional capacity-to-load ratio calculation model, can provide higher prediction accuracy and optimization capabilities in a complex and changeable power environment, and significantly improve the operating efficiency and safety of the power grid.

[0006] In order to achieve the above object, the present invention provides a method for calculating the capacity-load ratio of a medium voltage distribution network based on a deep neural network, and the method comprises the following steps:

[0007] S1. Collect multi-source data of the medium voltage distribution network, and construct a first high-dimensional feature matrix based on the multi-source data;

[0008] S2. Design an adaptive hierarchical clustering algorithm to dynamically identify the load type according to the high-dimensional feature matrix, then extract the time series characteristics of the load and the autocorrelation characteristics in the time series characteristics from the load type, and build the time series characteristics and autocorrelation characteristics into a second high-dimensional feature matrix;

[0009] S3. Construct a time-series multi-level recursive neural network, take the second highest dimensional feature matrix as input, and use the recursive structure and regularization method to combine short-term and long-term features for load forecasting.

[0010] Wherein, the temporal multi-level recursive neural network includes a first layer recursive network and a second layer recursive network;

[0011] The first layer of recursive network processes the timing characteristics of each load type, which is expressed as follows:

[0012] h1(t)=RNN1(T C (t))+λ1·reg(T C (t))

[0013] Where h1(t) is the hidden state of the first layer of the recurrent network, λ1 is the regularization coefficient, and reg(T C (t)) is a specially designed local smoothing regularization term, RNN1 is the first layer of recurrent network, T C (t) is a time series feature; wherein the local smoothing regularization term reg(T C (t)) is expressed as follows:

[0014]

[0015] Where n is the total number of windows, i is the window size, and t is the current time;

[0016] The second layer recursive network combines the hidden state and autocorrelation features of the first layer to process the long-term trend of the load, which is expressed as follows:

[0017] h2(t)=RNN2(h1(t),A C (t))+λ2·reg(A C (t))

[0018] Among them, h2(t) is the hidden state of the second layer recurrent network, A C (t) is the autocorrelation feature, λ2 is the regularization coefficient; reg(A C (t)) is the long-term periodic regularization term, expressed as follows:

[0019]

[0020] Where m is the total number of time steps and k is the number of time steps;

[0021] After the output of the time-series multi-level recurrent neural network, the final hidden state h2(t) is passed to the fully connected layer for mapping to generate a load prediction value P(t), which is expressed as follows:

[0022] P(t)=W p ·h2(t)+b p

[0023] Among them, W p is the weight matrix of the fully connected layer, b p is the bias term, P(t) represents the load forecast value at the future time step t;

[0024] S4. Calculate the current capacity-load ratio of the power grid according to the load forecast, and preliminarily optimize the current capacity-load ratio to obtain a preliminarily optimized load forecast;

[0025] The preliminary optimization of the current capacity-to-load ratio to obtain a preliminary optimized load forecast specifically includes:

[0026] Construct an adaptive load balancing optimization model based on the capacity-load ratio R of each node i (t), the load is transferred to the nodes whose load is higher than the set threshold, which is expressed as follows:

[0027]

[0028] Among them, ΔP i(t) is the node load adjustment item, which is dynamically adjusted according to the current capacity-load ratio of each node and the load of its adjacent nodes; α is the load balancing adjustment coefficient, which is used to control the load distribution speed; β is the adjacent node adjustment coefficient, which is used to adjust the load difference with adjacent nodes; N(i) is the set of nodes adjacent to node i, |N(i)| is the number of adjacent nodes; P h (t) is the load prediction value of adjacent node j; Capacity i is the capacity of the grid node corresponding to the load type; P i (t) is the predicted value of the i-th type of load;

[0029] At the same time, the grid dynamic adjustment function D(t) is introduced to describe the dynamic change rate of the capacity-load ratio in the current grid, which is expressed as follows:

[0030]

[0031] According to the value of the current power grid dynamic adjustment function D(t), the area of ​​load change in the power grid is identified, and load scheduling is performed in advance according to the degree of change;

[0032] S5. Build a deep learning model based on the initially optimized load forecast to further optimize the energy management strategy, and adjust the model parameters of the deep learning model to adapt to the actual operating conditions;

[0033] S6. Monitor the operation of the medium-voltage distribution network in real time and use reinforcement learning models to optimize deep learning models and energy management strategies.

[0034] Further, the multi-source data includes electrical data, environmental data and user behavior data;

[0035] The electrical data includes voltage, current and power;

[0036] The environmental data is environmental changes;

[0037] The user behavior data is load fluctuation caused by power usage patterns of different users, and the load fluctuation includes load power factor and power usage characteristics;

[0038] The step of constructing a first high-dimensional feature matrix according to multi-source data specifically includes:

[0039] The multi-source data is preprocessed, and the preprocessing includes data synchronization and normalization processing, noise filtering and outlier processing, and feature expansion and time series feature generation processing. Finally, time series features are generated, and the generated time series features are integrated into a first high-dimensional feature matrix.

[0040] Furthermore, the design of an adaptive hierarchical clustering algorithm to dynamically identify load types according to a high-dimensional feature matrix specifically includes:

[0041] The first high-dimensional feature matrix is ​​segmented according to a fixed window to form a segmentation matrix S(t) containing the past n time steps. Then, principal component analysis is used to reduce the dimension of the segmentation matrix S(t) to generate a feature vector R(t). Finally, an adaptive hierarchical clustering algorithm is introduced to perform preliminary clustering of the feature vector R(t) through distance measurement and adjust the clustering level according to the time series characteristics, and finally output the load type C(t).

[0042] The step of extracting the time series features of the load and the autocorrelation features in the time series features from the load type and constructing the time series features and the autocorrelation features into a second high-dimensional feature matrix specifically includes:

[0043] For each load type C(t), the sliding window method is used to expand its features to form the time series features T of each load type. C (t), and calculate the autocorrelation function A under load type C(t) C (t), identify the periodic characteristics of long-term loads, and finally convert the autocorrelation function A C (t) and the timing characteristics T of each load type C (t) Reconstruct the second high-dimensional feature matrix E(t).

[0044] Furthermore, in the S3, it also includes: adding a load distribution constraint item to the prediction output layer of the time series multi-level recurrent neural network to constrain the load to meet the grid capacity, which is expressed as follows:

[0045]

[0046] Among them, C(t) is the constraint term, P i (t) is the predicted value of the i-th type of load, Capacity i is the corresponding distribution network capacity.

[0047] Furthermore, in the training process of the time series multi-level recursive neural network, historical load data is used for supervised learning, and a weighted mean square error loss function is adopted to assign different weights to errors of different load types, which is expressed as follows:

[0048]

[0049] Among them, Y(t) is the actual load, P(t) is the predicted load, and α is the weight factor dynamically adjusted according to the load type.

[0050] Furthermore, the capacity ratio R of each node i (t) is the ratio between the load of each node in the power grid and its available capacity, which is used to measure whether the power grid is under normal load and is expressed as follows:

[0051]

[0052] Among them, P i (t) is the predicted value of the load of the i-th type; Capacity i is the capacity of the grid node corresponding to the load type; R(t) is the total capacity-to-load ratio at time t.

[0053] Furthermore, the S5 specifically includes:

[0054] S5.1. Design an adaptive load optimization mechanism to perform real-time correction on the capacity-load ratio of each node to ensure that the capacity-load ratio of each node is within its capacity range; wherein, the adaptive load optimization mechanism ensures the balance of the load by minimizing the adaptive optimization function O(t) based on the capacity-load ratio, and dynamically adjusts the load prediction value so that the actual operation state of the power grid is closest to the load prediction result; the adaptive optimization function O(t) is expressed as follows:

[0055]

[0056] Among them, R i (t) is the capacity ratio of the i-th node; P i (t) is the load prediction value of the node; Capacity i is the capacity of the node; ΔP i (t) is the node load adjustment amount; γ is the weight coefficient of the load balance adjustment term, which is used to control the relative importance of capacity-to-load ratio optimization and node load adjustment;

[0057] S5.2. Design an adaptive regression update mechanism to adaptively update the deep learning model according to load changes;

[0058] S5.3. Perform closed-loop feedback on the deep learning model, feed back the load prediction error and capacity-load ratio to the scheduling system, and dynamically adjust the load prediction model for the next round. At the same time, feed back the node load adjustment amount of each load distribution to the deep learning model, and dynamically adjust the model parameters of the deep learning model.

[0059] Furthermore, the S5.2 specifically includes:

[0060] First, the current load prediction error ∈(t) is calculated, which is expressed as follows:

[0061]

[0062] Among them, Y i (t) is the actual load value;

[0063] Designing an update mechanism with weight decay allows the deep learning model to have the best adaptability when the load fluctuates drastically, as shown below:

[0064]

[0065] Among them, ΔW(t) is the adjustment amount of the model weight, η is the learning rate, which is used to control the speed of weight update. It is the load-to-capacity ratio term, used to slow down weight adjustment when the load is close to the node capacity.

[0066] Furthermore, the state of the reinforcement learning model includes the capacity-load ratio of each node in the current power grid, the prediction error, and the load adjustment between adjacent nodes; the action includes dynamically adjusting the node load distribution and updating the model weight of the deep learning model; the reward function R(t) is expressed as follows:

[0067]

[0068] Among them, R i (t) is the load-to-capacity ratio of node i, and the goal is to be close to 1, indicating that the load of the node is within its capacity; α is a coefficient used to regulate the smoothness of load adjustment to prevent instability caused by over-adjustment; β is used to control the impact of prediction error on rewards to ensure that the prediction accuracy of the model continues to improve.

[0069] Furthermore, the reinforcement learning model is optimized using a deep Q learning algorithm, which calculates the Q value at the current moment through the state S(t) and action A(t), reflecting the expected cumulative reward of taking a certain action in a specific state. The Q value update formula is as follows:

[0070]

[0071] Among them, η is the learning rate, which is used to control the update amplitude of the Q value, γ is the discount factor, which weighs the impact of current rewards and future benefits, and R(t) is the current reward function, which indicates the operating effect under the current load scheduling.

[0072] The beneficial technical effects of the present invention are at least as follows:

[0073] (1) By introducing a time series deep learning model (such as LSTM, GRU or Transformer), the present invention can accurately predict future changes in different load types based on historical data and real-time input. This high-precision prediction capability effectively solves the problem that traditional methods are difficult to cope with complex dynamic changes in the power system, allowing the power grid to respond to load changes in advance and prevent overload and equipment damage.

[0074] (2) The present invention uses load type identification technology to subdivide the loads in the power grid into multiple types (such as residential, commercial, industrial loads, etc.), and extracts features and models each type separately. This multi-dimensional identification and optimization method breaks through the limitation of traditional methods that only predict based on a single load feature, making the calculation of the capacity-to-load ratio more precise and accurate, and better able to adapt to different load conditions and user behavior patterns.

[0075] (3) The present invention introduces an intelligent energy management system (EMS) that can monitor the operating status and load conditions of the distribution network in real time, and perform dynamic load scheduling and optimal allocation based on the prediction results. Through the load balancing optimization algorithm, the system can quickly adjust the load distribution of each node in the event of load changes and failures, avoid grid overload and imbalance, and improve the grid's operational stability and ability to respond to emergencies.

[0076] (4) The present invention introduces an adaptive learning mechanism that can continuously optimize the deep learning model and energy management algorithm based on real-time feedback data, thereby improving the robustness and adaptability of the model. Compared with the traditional fixed parameter method, this adaptive optimization capability enables the system to continuously learn and adjust, adapt to the long-term change trend of the power grid and new load patterns, and improve the real-time and reliability of the capacity-to-load ratio calculation.

[0077] (5) The present invention makes full use of the rich real-time data in the distribution network, including electrical data, environmental data, and user behavior data, and integrates this information into the model through multi-source data fusion technology, thereby improving the accuracy of prediction and optimization. This data-driven approach solves the problem that traditional methods cannot use real-time information for precise adjustments, making power grid management more intelligent and efficient. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] The present invention is further described using the accompanying drawings, but the embodiments in the accompanying drawings do not constitute any limitation to the present invention. A person skilled in the art can obtain other drawings based on the following drawings without creative work.

[0079] Figure 1 This is a flow chart of a method for calculating the capacity-load ratio of a medium-voltage distribution network based on a deep neural network according to an embodiment of the present invention. DETAILED DESCRIPTION

[0080] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0081] In one or more embodiments, Figure 1 As shown, a method for calculating the capacity-load ratio of a medium voltage distribution network based on a deep neural network is disclosed. The method includes steps 1 to 6, including:

[0082] S1. Collect multi-source data of the medium voltage distribution network, and construct a first high-dimensional feature matrix based on the multi-source data.

[0083] Specifically, in a medium voltage distribution network, the load is affected by multiple factors, so the present invention needs to collect data from multiple dimensions. The data here include:

[0084] Electrical data: Collect real-time voltage V(t), current I(t), and power P(t). These data directly reflect the operating status of the power grid load.

[0085] Environmental data: Environmental changes, such as temperature T(t) and humidity H(t), affect the power load, especially the energy consumption patterns of industrial and commercial users.

[0086] User behavior data: The power consumption patterns of different users will cause load fluctuations, so the load power factor cosφ(t) and power consumption characteristics L(t) are taken into consideration.

[0087] These data are collected in the form of time series with timestamp t, which ensures the temporal consistency of the information. Integrating multi-source data helps to fully reflect the actual operating status of the power grid, rather than being limited to the analysis of a single data source.

[0088] Furthermore, data from different sources often have different acquisition frequencies and time steps. In order to ensure their synchronization, a linear interpolation synchronization algorithm is used. This algorithm unifies data with different acquisition frequencies to the same time step Δt through interpolation. The interpolation process ensures the time alignment of the data and avoids model deviations caused by inconsistent data.

[0089] After data synchronization, normalization is performed. For each data type, the present invention scales it to the interval [0,1] to reduce the impact of the dimension on the model. The normalization formula is as follows:

[0090]

[0091] Where X(t) is the original data, X min and X max Respectively represent the minimum and maximum values ​​of the data source. After normalization, each data feature will change on the same scale, avoiding errors in model training due to different dimensions. This synchronization and normalization step ensures that multi-source data can be effectively integrated to provide consistent and usable features for subsequent load forecasting.

[0092] Furthermore, power grid data is often subject to various interferences, especially electromagnetic interference, equipment aging, and signal jitter. To ensure data quality, an adaptive median filter is used to process noise and outliers. This filter dynamically adjusts the filter window size to adapt to the influence of different noise levels. The core formula of the filter is:

[0093] Y(t)=median{X(ti),X(t-i+1),…,X(t+i)}

[0094] Where Y(t) is the filtered data, X(t) is the original data, and the window size i is dynamically adjusted according to the noise level. Compared with the traditional fixed window filter, this method can better preserve the main characteristics of the signal, especially suitable for the fluctuation characteristics of load data. This adaptive noise filtering can not only effectively remove noise, but also preserve the key fluctuation characteristics of the power grid load, ensuring that the model training in the subsequent steps is based on high-quality data.

[0095] Furthermore, the load changes in the power grid have obvious time dependence, so it is not possible to use only the data at the current time point for analysis. In order to capture the past load behavior, a sliding window method is used to generate time series features. For each time step t, the feature set Z(t) is expanded, which contains the data of the past n time steps:

[0096] Z(t)=[X(t),X(t-1),…,X(tn)]

[0097] Here, Z(t) is the extended feature of time t. The window size n can be dynamically adjusted according to the characteristics of the power grid. The optimal window size is usually selected through cross-validation. In this way, the model can not only rely on the data of the current time, but also learn the load change trend in the past period of time. This step ensures that the subsequent model can effectively capture the short-term and long-term fluctuation characteristics of the power grid load by expanding the time series features, thereby enhancing the accuracy of load prediction and the ability to capture time series dependencies.

[0098] Furthermore, through the different data sources (voltage, current, power, environmental data, user behavior data) processed in the previous steps, these expanded time series features are now integrated into a first high-dimensional feature matrix F(t). The specific construction method is as follows:

[0099] F(t)=[Z V (t),Z I (t),Z P (t),Z T (t),Z H (t),Z cosφ (t),Z L (t)]

[0100] Among them, Z V (t) represents the time series characteristics of voltage, Z I (t) represents the current, Z P (t) represents power, Z T (t) and Z H (t) represent the temporal characteristics of temperature and humidity, respectively, Z C0sφ (t) represents the time series characteristics of the load power factor, Z L (t) represents the timing characteristics of the power consumption pattern.

[0101] The final output is the first high-dimensional feature matrix F(t), which combines multi-source data in time series to form an input matrix containing rich time series and multi-dimensional information for subsequent deep learning models. The construction of this feature matrix is ​​specially designed for the complexity of the medium-voltage distribution network, ensuring that it not only combines electrical parameters, but also combines the multi-dimensional influence of the environment and user behavior, greatly enhancing the accuracy of load prediction.

[0102] This step provides a structured and rich input data matrix for subsequent model training and load forecasting through five sub-steps (data collection, synchronization and normalization, noise filtering, feature expansion and data fusion). In particular, through adaptive noise filtering and sliding window time series feature generation, the problem that traditional methods cannot accurately capture the complex time series behavior of power grid loads is solved. The high-dimensional feature matrix F(t) outputted in the end is the input of the subsequent steps. It integrates the multi-dimensional dynamic information of the power grid and effectively improves the training quality and prediction performance of the model.

[0103] S2. Design an adaptive hierarchical clustering algorithm to dynamically identify the load type according to the high-dimensional feature matrix, then extract the load time series characteristics and the autocorrelation characteristics in the time series characteristics from the load type, and build the time series characteristics and autocorrelation characteristics into a second high-dimensional feature matrix.

[0104] The input is the high-dimensional feature matrix F(t) generated from step 1, which is in the form of:

[0105] F(t)=[Z V (t),Z I (t),Z P (t),Z T (t),Z H (t),Z cosφ (t),Z L (t)]

[0106] These data include electrical characteristics (such as voltage Z V (t), current Z I (t), power Z P (t)), and environmental parameters (such as temperature ZT (t), humidity Z H (t)) and user behavior data (such as load power factor Z cosφ (t), user power consumption pattern Z L (t)). These features have been preprocessed in step 1 and can accurately reflect the load changes of the power grid at different time points.

[0107] Specifically, in a medium voltage distribution network, different types of loads (such as industrial loads, commercial loads, and residential loads) exhibit significantly different characteristics, and these characteristics change over time. Therefore, traditional static classification methods cannot accurately identify these load types. The present invention designs an adaptive hierarchical clustering algorithm that can dynamically identify load types by combining electrical characteristics, environmental characteristics, and power consumption behavior characteristics.

[0108] Furthermore, due to the time-varying characteristics of the load, the present invention segments the feature matrix F(t) according to fixed windows to form a segmented matrix S(t) containing the past n time steps, so as to capture the short-term fluctuations and long-term trends of the load.

[0109] S(t)=[F(t),F(t-1),…,F(tn)]

[0110] This time-series segmentation processing method can combine the historical behavior of the load and is particularly suitable for the fluctuating characteristics of the load in the medium-voltage distribution network, such as the peak of industrial load and the periodic changes of commercial load.

[0111] Furthermore, the feature matrix S(t) is a high-dimensional matrix, and direct clustering will lead to interference of redundant information. Therefore, the present invention uses principal component analysis (PCA) to reduce the dimension of S(t) and retain 99% of the variance information. After dimensionality reduction, the feature vector R(t) is generated:

[0112] R(t)=PCA(S(t))

[0113] By reducing the dimension, the present invention effectively reduces the computational complexity of high-dimensional matrices while retaining the most critical information of the load, which provides a concise and core feature representation for processing multi-dimensional data (such as electrical and environmental data) in load characteristics.

[0114] Furthermore, based on the reduced eigenvector R(t), the present invention introduces an adaptive hierarchical clustering algorithm, which ensures accurate classification of load types by adjusting the clustering level. The clustering process performs preliminary clustering on R(t) through distance metrics (such as Euclidean distance), and adjusts the clustering level according to the time series characteristics, and finally outputs the load classification label C(t):

[0115] C(t)=AdaptiveCluster(R(t))

[0116] Here C(t) represents the load type at time t (such as industrial, commercial, residential). Adaptive hierarchical clustering can dynamically adapt to the fluctuation of load types by continuously optimizing clustering results, especially in time periods with drastic load changes.

[0117] Furthermore, after identifying the load type, the present invention further extracts the time series characteristics of the load so that the subsequent load prediction model can understand the short-term and long-term changes of different load types. The present invention uses a sliding window and an autocorrelation function to extract the time series characteristics of the load type.

[0118] Furthermore, for each load type C(t), the sliding window method is used to expand its features to form a short-term time series feature vector T C (t):

[0119] T C (t)=[F C (t),F C (t-1),…,F C (tn)]

[0120] By expanding the sliding window, the characteristics of short-term fluctuations under the load type C(t) are captured, especially the load with high-frequency fluctuations such as industrial load. This ensures that the subsequent prediction model can make accurate judgments on short-term load changes (such as the instantaneous peak of industrial equipment).

[0121] Furthermore, in order to further capture the periodic changes of the load, the present invention calculates the autocorrelation function A under the load type C(t): C (t), used to identify the cyclic characteristics of long-term loads:

[0122]

[0123] Autocorrelation feature A C (t) can help the present invention capture the periodic behavior of the load, especially the low-frequency fluctuations of residential loads and the high-frequency fluctuations of industrial loads. This feature extraction method enhances the performance of the subsequent load prediction model under periodic load conditions.

[0124] Furthermore, after load type identification and time series feature extraction, the output second feature matrix E(t) contains the time series features and autocorrelation features of different load types, providing rich input for the subsequent load forecasting model:

[0125] E(t)=[T C (t),A C (t)]

[0126] Here, T C(t) is the sliding window timing characteristics of each load type, A C (t) is the autocorrelation feature. These data can help the model understand the short-term fluctuations and long-term periodic changes of different loads, thereby enhancing the accuracy and robustness of the forecast.

[0127] S3. Construct a time-series multi-level recursive neural network, take the second highest dimensional feature matrix as input, and use the recursive structure and regularization method to combine short-term and long-term features for load forecasting.

[0128] Specifically, the load changes in the medium-voltage distribution network are extremely complex, and different types of loads show significant differences on different time scales. Existing load forecasting methods often find it difficult to simultaneously cope with the short-term fluctuations of industrial loads and the periodic characteristics of residential loads. Therefore, the present invention designs a time-series multi-level recursive neural network (T-MRNN) to address this complexity, and uses an innovative recursive structure and regularization method to combine short-term and long-term features for load forecasting.

[0129] The first layer of recurrent network: processing the short-term temporal features T of the input C (t). This layer of network is designed to capture short-term fluctuations in load, and is particularly suitable for handling frequent fluctuations such as industrial loads. The model formula is:

[0130] h1(t)=RNN1(T C (t))+λ1·reg(T C (t))

[0131] Where h1(t) is the hidden state of the first layer of the recurrent network, λ1 is the regularization coefficient, and reg(T C (t)) is a specially designed regularization term. The regularization term targets common sudden load changes in the power grid, such as load spikes caused by the start and stop of industrial equipment, and adds a local smoothing regularization term to avoid overfitting the model to abnormal fluctuations.

[0132] The formula for the local smoothing regularization term is as follows:

[0133]

[0134] This regularization term prevents the model from overfitting local fluctuations when industrial load fluctuates frequently by limiting the sharp changes in short-term load characteristics, ensuring that the model can capture the overall change trend of the load.

[0135] The second layer of recurrent network: combines the hidden state h1(t) of the first layer and the autocorrelation feature A C (t) Processes the long-term trend of the load. This layer of the network is specifically used to process periodic features to ensure that the model can capture the long-term dependence of the load. The model formula is:

[0136] h2(t)=RNN2(h1(t),A C (t))+λ2·reg(A C (t))

[0137] Where h2(t) is the hidden state of the second layer recurrent network, λ2 is the regularization coefficient, and reg(A C (t)) is the long-term periodic regularization term, which is used to suppress over-fitting of periodic oscillations and ensure that load cycle fluctuations are correctly captured rather than amplified. The periodic regularization term is designed as:

[0138]

[0139] This regularization term suppresses the model's fitting of transient periodic changes by comparing the autocorrelation features at the current time step t with the previous m time steps. This regularization term helps the model better capture long-term fluctuations in periodic loads, such as periodic electricity consumption behavior in residential loads.

[0140] This hierarchical recursive network design can simultaneously adapt to both short-term sudden changes and long-term stable changes in the load in the power grid by combining short-term fluctuation processing and long-term trend capture. In particular, by introducing regularization terms to suppress the fitting of abnormal load fluctuations and transition period changes, the model can provide more stable load forecasting.

[0141] Furthermore, after being processed by the two-layer recursive network, the final hidden state h2(t) is passed to the fully connected layer for mapping to generate the load prediction value P(t), as shown in the following formula:

[0142] P(t)=W p ·h2(t)+b p

[0143] Among them, W p is the weight matrix of the fully connected layer, b p is the bias term, and P(t) represents the load forecast value at the future time step t. Through this forecasting structure, the model can effectively convert multi-time scale features into specific load values. In particular, for the short-term fluctuations of industrial loads and the long-term periodicity of residential loads, the model can generate stable and accurate forecast results.

[0144] At the same time, the present invention adds a load distribution constraint item in the prediction output layer to ensure that the predicted load meets the constraints of the power grid capacity. The constraint item C(t) is added to the output layer in the form of a matrix:

[0145]

[0146] Here, P i (t) is the predicted value of the i-th type of load, Capacityi is the corresponding distribution network capacity. This constraint is used to control the total load forecast to not exceed the maximum capacity-load ratio of the distribution network, thereby ensuring that the predicted load can be reasonably distributed under the power grid operation constraints.

[0147] Furthermore, during the training process, the present invention uses historical load data for supervised learning. In order to optimize the model, the present invention uses a weighted mean square error (WMSE) loss function, that is, different weights are assigned to errors of different load types to ensure that both high-frequency fluctuations and periodic loads can be accurately optimized. The formula is as follows:

[0148]

[0149] Among them, Y(t) is the actual load, P(t) is the predicted load, and α is the weight factor that is dynamically adjusted according to the load type. Through the weighted loss function, the model can better adapt to the error adjustment of different load types, especially for the different needs of industrial loads and residential loads.

[0150] Furthermore, the trained T-MRNN model finally generates a load prediction value P(t), which not only reflects the short-term fluctuations of the power grid, but also combines long-term periodic changes, providing accurate data support for subsequent energy management and load scheduling. The prediction results are also adjusted by load distribution constraints to ensure that they meet the actual capacity limitations of the power grid and are practical.

[0151] S4. Calculate the current capacity-load ratio of the power grid according to the load forecast, and preliminarily optimize the current capacity-load ratio to obtain a preliminarily optimized load forecast.

[0152] Specifically, the capacity-to-load ratio (R(t)) of the medium voltage distribution network is the ratio between the load of each node in the power grid and its available capacity, which is used to measure whether the power grid is under normal load. In order to calculate the capacity-to-load ratio, the present invention combines the predicted values ​​P of different load types i (t) and the available capacity of each node i , design a formula specifically for real-time capacity-load ratio calculation. The calculation formula for capacity-load ratio is:

[0153]

[0154] Among them, P i (t) is the predicted value of the i-th type of load, Capacity iis the capacity of the grid node corresponding to the load type. R(t) is the total capacity-to-load ratio at time t. This calculation ensures that the grid load does not exceed the node capacity and can reflect the operating status of the grid in real time. If R(t) ≥ 1, it means that the grid load is overloaded and needs to be dispatched through the energy management system; if R(t) < 1, it means that the grid load is at a normal level.

[0155] Furthermore, since the load in the medium voltage distribution network has significant volatility and periodicity, energy allocation directly based on load prediction results may cause local node load overload. In order to cope with this situation, the present invention designs an adaptive load balancing optimization model to ensure the overall stability of the power grid by adjusting the load distribution of each node:

[0156] First, according to the capacity ratio R of each node i (t), and transfer the load to the nodes whose load is higher than the set threshold. To this end, the present invention introduces a node load adjustment term ΔP i (t), which is dynamically adjusted according to the current capacity-load ratio of each node and the load of its adjacent nodes.

[0157] The node load adjustment formula is:

[0158]

[0159] Where α is the load balancing adjustment coefficient, which is used to control the load distribution speed. β is the adjacent node adjustment coefficient, which is used to adjust the load difference with adjacent nodes. N(i) is the set of nodes adjacent to node i, and |N(i)| is the number of adjacent nodes. j (t) is the load prediction value of the adjacent node j. Through this load adjustment formula, the overloaded node load can be reasonably distributed to the adjacent nodes with lighter loads to ensure the balanced distribution of the overall load of the power grid. Especially when the industrial load is too high, this solution can dynamically adjust the load to avoid overload in local areas of the power grid and cause power supply failure.

[0160] Furthermore, through the above capacity-to-load ratio calculation and energy allocation adjustment, the real-time optimization solution can ensure that the power grid remains stable under various complex load conditions. The entire optimization process is not only based on the load and capacity calculation of each node, but also further dynamic adjustments are made through the load information of adjacent nodes to ensure that local loads can be quickly balanced when overloaded.

[0161] In order to more accurately reflect the actual operation of the power grid, the present invention introduces a power grid dynamic adjustment function D(t) to describe the dynamic change rate of the capacity-load ratio in the current power grid:

[0162]

[0163] This formula represents the impact of load change on the capacity-to-load ratio at each time step. By observing D(t), the present invention can identify the area in the power grid where the load changes faster and perform load scheduling in advance according to this change trend. If the value of D(t) increases positively, it indicates that the load is growing faster and further optimization and adjustment are required; if the value decreases, it indicates that the power grid is in a load-balanced state.

[0164] Furthermore, in the actual energy management system, the present invention performs real-time optimization based on capacity-load ratio calculation and node adjustment to ensure the stability of the entire power grid. In order to improve the adaptive ability of the system, the present invention designs a feedback mechanism to feed back the power grid state after each optimization to the next prediction model. The specific steps are as follows:

[0165] The capacity-to-load ratio R(t) of each node is fed back to the load prediction model to adjust the prediction weight of future load.

[0166] Gradually adjust the overloaded nodes to avoid local overload in the next time period.

[0167] Through this feedback mechanism, the entire optimization process forms a closed loop, allowing the calculation of the capacity-to-load ratio to be linked with the energy management system in real time, ensuring the system's sensitivity and adaptability to load fluctuations.

[0168] S5. Build a deep learning model based on the preliminary optimized load forecast to deeply optimize the energy management strategy, and adjust the model parameters of the deep learning model to adapt to the actual operating conditions.

[0169] Specifically, in order to achieve load balancing of different nodes in the power grid, the present invention designs an adaptive load optimization mechanism, which corrects the prediction results of the model in real time by combining the load prediction and adjustment information in step 4, thereby improving the overall scheduling efficiency. The optimization goal is to minimize the imbalance of the load in the power grid while ensuring that the load of each node is within its capacity. To this end, the present invention introduces an adaptive optimization function based on the capacity-to-load ratio, and the formula is as follows:

[0170]

[0171] Among them, R i (t) is the capacity ratio of the i-th node. i (t) is the load forecast value of the node. i is the capacity of the node. i(t) is the node load adjustment amount. γ is the weight coefficient of the load balance adjustment term, which is used to control the relative importance of capacity-load ratio optimization and node load adjustment. The adaptive optimization mechanism ensures the balance of the load by minimizing O(t), and dynamically adjusts the load prediction value to make the actual operation state of the power grid closer to the load prediction result and reduce the imbalance of load between nodes. This optimization process can handle the load changes between nodes in real time, especially when the industrial load fluctuates violently, and has good adaptability to load scheduling.

[0172] Furthermore, based on the above optimization results, the present invention adaptively updates the deep learning model to improve its accuracy and stability in predicting future loads. In order to capture the nonlinear changes in loads, especially to cope with the dynamic adjustment of unbalanced loads, the present invention introduces an adaptive regression update mechanism for load changes:

[0173] First, calculate the current load prediction error ∈(t), the formula is as follows:

[0174]

[0175] Among them, Y i (t) is the actual load value. By calculating the prediction error of each node, the present invention can adjust the weight of the model.

[0176] In order to make the model more adaptable when the load fluctuates violently, the present invention designs an update mechanism with weight decay, the formula is as follows:

[0177]

[0178] Where ΔW(t) is the adjustment amount of the model weight. η is the learning rate, which is used to control the speed of weight update. It is the load-to-capacity ratio term, which is used to appropriately slow down the weight adjustment when the load is close to the node capacity.

[0179] Through this adaptive update mechanism, the model can dynamically adjust the weights according to the real-time load prediction error. Especially when the load is too large, the model can gradually reduce the sensitivity to the current load to avoid overfitting. At the same time, when the load is low, the model can more sensitively capture load changes to ensure prediction accuracy.

[0180] Furthermore, in order to further enhance the adaptability of power grid load scheduling, the present invention designs a feedback closed-loop control mechanism to feed back the model's prediction output and optimization results to the load scheduling system. The specific steps are as follows:

[0181] Load forecast feedback: The load forecast error ∈(t) and capacity-load ratio R(t) are fed back to the scheduling system, and the next round of load forecast model is dynamically adjusted to make it more in line with the actual load changes.

[0182] Load distribution optimization feedback: The adjustment result of each load distribution ΔP i (t) Feedback to the deep learning model to dynamically adjust the model parameters so that the model can more accurately reflect the actual situation of load scheduling.

[0183] Through this feedback loop, the power grid system can self-adjust according to the actual operating status in each cycle, improve the accuracy of load forecasting and the efficiency of load distribution, and ensure that the power grid remains stable under various load fluctuations.

[0184] This step can achieve real-time optimization and scheduling in complex load scenarios through the combination of adaptive optimization and model update, especially in industrial and commercial load scenarios with drastic load changes. The solution optimizes the capacity-to-load ratio of each node in the power grid and combines the adaptive update of the deep learning model to achieve dynamic load distribution and high-precision load prediction, further improving the operation efficiency of the power grid.

[0185] S6. Monitor the operation of the medium-voltage distribution network in real time and use reinforcement learning models to optimize deep learning models and energy management strategies.

[0186] Specifically, in response to the uncertainty and volatility of the power grid load in the patent scenario, the present invention designs a deep reinforcement learning mechanism that is specifically adapted to the load fluctuations of the medium-voltage distribution network. This mechanism can intelligently adjust the load distribution of the power grid to ensure continuous optimization of the power grid operation when the load changes drastically. The design of the present invention not only takes into account the prediction error of the load, but also optimizes the dispatching decision of the power grid through real-time adjustment of the load distribution.

[0187] At each time step, the system state S(t) includes the capacity load ratio R of each node in the current power grid. i (t), prediction error ∈(t), and load adjustment ΔP between adjacent nodes i (t). These state variables comprehensively reflect the real-time operating status of the power grid and are the basis for reinforcement learning to make optimal scheduling. The state is defined as:

[0188] S(t)={R1(t),R2(t),…,R n (t),∈(t),ΔP1(t),…,ΔP n (t)}

[0189] Different from traditional power grid dispatching, the load changes in the scenario of the present invention are more random, so the state S(t) is specially designed to adjust the load of adjacent nodes ΔPi The sensitivity of (t) is determined so that the model can monitor the changes in local node loads in real time to ensure the balance of the overall power grid load.

[0190] For each node state, the actions that the reinforcement learning model can select include load distribution adjustment and model parameter update. Different from common power grid dispatching, this invention introduces a load distribution mechanism based on regional optimization, that is, the action A(t) is not only for the load of a single node, but also considers the interaction with its neighboring nodes to ensure regional load balance. The action vector is defined as:

[0191] A(t)={ΔP i (t),ΔW(t)}

[0192] By dynamically adjusting the node load distribution ΔP i (t) and model weight update ΔW(t), the reinforcement learning model can flexibly adapt to the changes in loads in different areas and intelligently optimize load scheduling.

[0193] In view of the load fluctuation in the patent scenario, the present invention designs a reward function that is closely integrated with grid energy efficiency, load balance and prediction accuracy. i (t), prediction error ∈(t) and load adjustment ΔP i (t) Perform a comprehensive evaluation to ensure that the power grid maintains load balance while operating efficiently. The reward function is defined as:

[0194]

[0195] Among them, R i (t) is the capacity-to-load ratio of node i. The target is close to 1, indicating that the load of the node is within its capacity. α is a coefficient used to regulate the smoothness of load adjustment to prevent instability caused by excessive adjustment. β is used to control the impact of prediction error on rewards to ensure that the prediction accuracy of the model continues to improve. This reward function ensures load balance and efficiency optimization during power grid dispatching, especially for the frequent load fluctuations in medium-voltage distribution networks. By penalizing excessive load adjustment and prediction errors, the adaptive dispatching capability of the model is strengthened.

[0196] Furthermore, the present invention uses a deep Q learning algorithm (DQN) for optimization. Through the reinforcement learning mechanism, the system can continuously optimize the load scheduling strategy in real-time operation. The specific process is as follows:

[0197] Through the state S(t) and action A(t), the Q value at the current moment is calculated, reflecting the expected cumulative reward for taking a certain action in a specific state. The Q value update formula is as follows:

[0198]

[0199] Among them, η is the learning rate, which is used to control the update amplitude of the Q value. γ is the discount factor, which weighs the impact of current rewards and future benefits. R(t) is the current reward signal, which indicates the operating effect of the system under the current load scheduling. This Q value update process allows the system to continuously optimize its scheduling strategy to ensure that the optimal resource allocation solution is found in complex load fluctuations.

[0200] Through the balance of exploration and exploitation in reinforcement learning, the model selects the action A(t) that maximizes the Q value at each moment, and then performs load adjustment ΔP i (t) and the model weight update ΔW(t). This adaptive decision-making method ensures that the system can adjust its strategy according to the real-time fluctuation of the load, so that the power grid can still operate efficiently under complex conditions.

[0201] Furthermore, in response to sudden load changes and long-term fluctuations in the power grid, the present invention designs a continuous optimization feedback mechanism based on reinforcement learning. After each decision cycle, the system will calculate the load prediction error ∈(t) and the load adjustment amount ΔP i (t) Feedback to the reinforcement learning model, so that it can dynamically adjust the next round of scheduling decisions based on the latest status. The specific feedback process is as follows:

[0202] The results of each round of load scheduling are fed back to the reinforcement learning model to adjust the state evaluation and reward function of the next round, ensuring that the system can continuously improve the scheduling strategy based on historical data.

[0203] After each round of decision-making, the reinforcement learning model adjusts the node load feedback to make the load tend to be balanced in the future time step. The system can make adaptive adjustments based on the real-time data changes of the power grid to avoid excessive fluctuations and sudden load impacts on the power grid.

[0204] For the medium-voltage distribution network in the patent scenario, load fluctuations and regionalized load imbalance are the key issues that the present invention needs to address. By combining the reinforcement learning mechanism with adaptive load adjustment, the present invention can achieve high-frequency load scheduling and optimization, enabling the system to cope with complex load environments of multiple nodes, ensuring that the power grid remains stable when the load fluctuates violently. The present invention specifically designs a regionalized load adjustment mechanism for the problem of load imbalance, ensuring that the load distribution between adjacent nodes can be dynamically adjusted, thereby enhancing the flexibility of the system.

[0205] These are just some preferred embodiments of the present invention, which certainly cannot be used to limit the scope of rights of the present invention. Ordinary technicians in this field can understand that all or part of the processes of implementing the above embodiments and making equivalent changes according to the claims of the present invention still fall within the scope of the invention.

Claims

1. A method for calculating the capacity-load ratio of a medium voltage distribution network based on a deep neural network, characterized in that: The method comprises the following steps: S1. Collect multi-source data of the medium voltage distribution network, and construct a first high-dimensional feature matrix based on the multi-source data; S2. Design an adaptive hierarchical clustering algorithm to dynamically identify the load type according to the high-dimensional feature matrix, then extract the time series characteristics of the load and the autocorrelation characteristics in the time series characteristics from the load type, and build the time series characteristics and autocorrelation characteristics into a second high-dimensional feature matrix; S3. Construct a time-series multi-level recursive neural network, take the second highest dimensional feature matrix as input, and use the recursive structure and regularization method to combine short-term and long-term features for load forecasting. S4. Calculate the current capacity-load ratio of the power grid according to the load forecast, and preliminarily optimize the current capacity-load ratio to obtain a preliminarily optimized load forecast; S5. Build a deep learning model based on the initially optimized load forecast to further optimize the energy management strategy, and adjust the model parameters of the deep learning model to adapt to the actual operating conditions; S6. Monitor the operation of the medium-voltage distribution network in real time and use reinforcement learning models to optimize deep learning models and energy management strategies.

2. A method for calculating the capacity-load ratio of a medium voltage distribution network based on a deep neural network according to claim 1, characterized in that: The multi-source data includes electrical data, environmental data and user behavior data; The electrical data includes voltage, current and power; The environmental data is environmental changes; The user behavior data is load fluctuation caused by power usage patterns of different users, and the load fluctuation includes load power factor and power usage characteristics; The step of constructing a first high-dimensional feature matrix according to multi-source data specifically includes: The multi-source data is preprocessed, and the preprocessing includes data synchronization and normalization processing, noise filtering and outlier processing, and feature expansion and time series feature generation processing. Finally, time series features are generated, and the generated time series features are integrated into a first high-dimensional feature matrix.

3. The method for calculating the capacity-load ratio of a medium voltage distribution network based on a deep neural network according to claim 1, characterized in that: The design of the adaptive hierarchical clustering algorithm dynamically identifies the load type according to the high-dimensional feature matrix, specifically including: The first high-dimensional feature matrix is ​​segmented according to a fixed window to form a segmentation matrix S(t) containing the past n time steps. Then, principal component analysis is used to reduce the dimension of the segmentation matrix S(t) to generate a feature vector R(t). Finally, an adaptive hierarchical clustering algorithm is introduced to perform preliminary clustering of the feature vector R(t) through distance measurement and adjust the clustering level according to the time series characteristics, and finally output the load type C(t). The step of extracting the time series features of the load and the autocorrelation features in the time series features from the load type and constructing the time series features and the autocorrelation features into a second high-dimensional feature matrix specifically includes: For each load type C(t), the sliding window method is used to expand its features to form the time series features T of each load type. C (t), and calculate the autocorrelation function A under load type C(t) C (t), identify the periodic characteristics of long-term loads, and finally convert the autocorrelation function A C (t) and the timing characteristics T of each load type C (t) Reconstruct the second high-dimensional feature matrix E(t).

4. The method for calculating the capacity-load ratio of a medium voltage distribution network based on a deep neural network according to claim 1, characterized in that: The temporal multi-level recursive neural network includes a first layer recursive network and a second layer recursive network; The first layer of recursive network processes the timing characteristics of each load type, which is expressed as follows: h1(t)=RNN1(T C (t))+λ1·reg(T C (t)) Where h1(t) is the hidden state of the first layer of the recurrent network, λ1 is the regularization coefficient, and reg(T C (t)) is a specially designed local smoothing regularization term, RNN1 is the first layer of recurrent network, T C (t) is a time series feature; wherein the local smoothing regularization term reg(T C (t)) is expressed as follows: Where n is the total number of windows, i is the window size, and t is the current time; The second layer recursive network combines the hidden state and autocorrelation features of the first layer to process the long-term trend of the load, which is expressed as follows: h2(t)=RNN2(h1(t),A C (t))+λ2·reg(A C (t)) Among them, h2(t) is the hidden state of the second layer recurrent network, A C (t) is the autocorrelation feature, λ2 is the regularization coefficient; reg(A C (t)) is the long-term periodic regularization term, expressed as follows: Where m is the total number of time steps and k is the number of time steps; After the output of the time-series multi-level recurrent neural network, the final hidden state h2(t) is passed to the fully connected layer for mapping to generate a load prediction value P(t), which is expressed as follows: P(t)=W p ·h2(t)+b p Among them, W p is the weight matrix of the fully connected layer, b p is the bias term, P(t) represents the load forecast value at the future time step t; In the S3, it also includes: adding a load distribution constraint item to the prediction output layer of the time series multi-level recurrent neural network to constrain the load to meet the power grid capacity, which is expressed as follows: Among them, C(t) is the constraint term, P i (t) is the predicted value of the i-th type of load, Capacity i is the corresponding distribution network capacity.

5. A method for calculating the capacity-load ratio of a medium voltage distribution network based on a deep neural network according to claim 4, characterized in that: In the training process of the time series multi-level recursive neural network, historical load data is used for supervised learning, and a weighted mean square error loss function is adopted to assign different weights to errors of different load types, which is expressed as follows: Among them, Y(t) is the actual load, P(t) is the predicted load, and α is the weight factor dynamically adjusted according to the load type.

6. A method for calculating capacity-load ratio of a medium voltage distribution network based on a deep neural network according to claim 1, characterized in that: The preliminary optimization of the current capacity-to-load ratio to obtain a preliminary optimized load forecast specifically includes: Construct an adaptive load balancing optimization model based on the capacity-load ratio R of each node i (t), the load is transferred to the nodes whose load is higher than the set threshold, which is expressed as follows: Among them, ΔP i (t) is the node load adjustment item, which is dynamically adjusted according to the current capacity-load ratio of each node and the load of its adjacent nodes; α is the load balancing adjustment coefficient, which is used to control the load distribution speed; β is the adjacent node adjustment coefficient, which is used to adjust the load difference with adjacent nodes; N(i) is the set of nodes adjacent to node i, |N(i)| is the number of adjacent nodes; P j (t) is the load prediction value of adjacent node j; Capacity i is the capacity of the grid node corresponding to the load type; P i (t) is the predicted value of the i-th type of load; At the same time, the grid dynamic adjustment function D(t) is introduced to describe the dynamic change rate of the capacity-load ratio in the current grid, which is expressed as follows: According to the value of the current power grid dynamic adjustment function D(t), the area of ​​load change in the power grid is identified, and load scheduling is performed in advance according to the degree of change; Among them, the capacity ratio of each node is R i (t) is the ratio between the load of each node in the power grid and its available capacity, which is used to measure whether the power grid is under normal load and is expressed as follows: Among them, P i (t) is the predicted value of the load of the i-th type; Capacity i is the capacity of the grid node corresponding to the load type; R(t) is the total capacity-to-load ratio at time t.

7. A method for calculating capacity-load ratio of a medium voltage distribution network based on a deep neural network according to claim 1, characterized in that: The S5 specifically includes: S5.

1. Design an adaptive load optimization mechanism to perform real-time correction on the capacity-load ratio of each node to ensure that the capacity-load ratio of each node is within its capacity range; wherein, the adaptive load optimization mechanism ensures the balance of the load by minimizing the adaptive optimization function O(t) based on the capacity-load ratio, and dynamically adjusts the load prediction value so that the actual operation state of the power grid is closest to the load prediction result; the adaptive optimization function O(t) is expressed as follows: Among them, R i (t) is the capacity ratio of the i-th node; P i (t) is the load prediction value of the node; Capacity i is the capacity of the node; ΔP i (t) is the node load adjustment amount; γ is the weight coefficient of the load balance adjustment term, which is used to control the relative importance of capacity-to-load ratio optimization and node load adjustment; S5.

2. Design an adaptive regression update mechanism to adaptively update the deep learning model according to load changes; S5.

3. Perform closed-loop feedback on the deep learning model, feed back the load prediction error and capacity-load ratio to the scheduling system, and dynamically adjust the load prediction model for the next round. At the same time, feed back the node load adjustment amount of each load distribution to the deep learning model, and dynamically adjust the model parameters of the deep learning model.

8. A method for calculating the capacity-load ratio of a medium voltage distribution network based on a deep neural network according to claim 7, characterized in that: S5.2 specifically includes: First, the current load prediction error ∈(t) is calculated, which is expressed as follows: Among them, Y i (t) is the actual load value; Designing an update mechanism with weight decay allows the deep learning model to have the best adaptability when the load fluctuates drastically, as shown below: Among them, ΔW(t) is the adjustment amount of the model weight, η is the learning rate, which is used to control the speed of weight update. It is the load-to-capacity ratio term, used to slow down weight adjustment when the load is close to the node capacity.

9. The method for calculating the capacity-load ratio of a medium voltage distribution network based on a deep neural network according to claim 1, characterized in that: The state of the reinforcement learning model includes the capacity-load ratio of each node in the current power grid, the prediction error, and the load adjustment between adjacent nodes; the action includes dynamically adjusting the node load distribution and updating the model weight of the deep learning model; the reward function R(t) is expressed as follows: Among them, R i (t) is the load-to-capacity ratio of node i, and the goal is to be close to 1, indicating that the load of the node is within its capacity; α is a coefficient used to regulate the smoothness of load adjustment to prevent instability caused by over-adjustment; β is used to control the impact of prediction error on rewards to ensure that the prediction accuracy of the model continues to improve.

10. A method for calculating capacity-load ratio of a medium voltage distribution network based on a deep neural network according to claim 9, characterized in that: The reinforcement learning model is optimized using a deep Q learning algorithm. Through the state S(t) and action A(t), the Q value at the current moment is calculated, reflecting the expected cumulative reward for taking an action in a specific state. The Q value update formula is as follows: Among them, η is the learning rate, which is used to control the update amplitude of the Q value, γ is the discount factor, which weighs the impact of current rewards and future benefits, and R(t) is the current reward function, which indicates the operating effect under the current load scheduling.

Images