Heavy overload risk prediction method and device of power distribution network, terminal equipment and storage medium

Abstract

The invention discloses a weight and overload risk prediction method and device for a power distribution network, terminal equipment and a storage medium, and belongs to the technical field of electric power. The method comprises the steps of obtaining a first target characteristic parameter of each transformer substation, a second target characteristic parameter of each feeder line and a third target characteristic parameter of each transformer area; the method comprises the following steps: providing three-layer characteristic parameters of a transformer substation, a feeder line and a transformer area transformer for a load prediction model, constructing a global view, facilitating the load prediction model in subsequent steps to capture an association relationship between upper and lower level equipment loads, improving the accuracy of the predicted load rate of each equipment, and finally, according to the predicted load rate, calculating the load rate of each equipment. And generating a heavy overload risk early warning when the heavy overload risk of the transformer substation, the feeder line and the transformer area transformer is identified respectively. Therefore, by implementing the method and the device, the problem of low accuracy of heavy overload risk early warning in the prior art can be solved.

Description

Technical Field

[0001] This invention relates to the field of power technology, and in particular to a method, device, terminal equipment, and storage medium for predicting heavy overload risks in power distribution networks. Background Technology

[0002] With the large-scale integration of new energy sources and loads with high randomness and intermittency, such as distributed photovoltaic, wind power, and electric vehicle charging stations, into distribution networks, the traditional load characteristics of the power grid are undergoing a profound and dramatic transformation. In the past, electricity load was mainly influenced by relatively regular social production and residential lifestyles, resulting in a relatively stable load curve with strong predictability. However, the widespread adoption of these new elements has led to unprecedented levels of randomness, intermittency, and volatility in load curves. This change has significantly increased the risk of severe overload on key equipment such as transformers and lines in the distribution network, becoming a core hidden danger directly threatening the safety, stability, and high-quality operation of the regional power grid. Against this backdrop, proactive and accurate prediction of power grid equipment load is undoubtedly a crucial technological foundation for shifting from passive response to proactive operation and maintenance, thereby accurately identifying and effectively preventing severe overload accidents.

[0003] However, the traditional early warning methods widely used in distribution network operation and maintenance are proving inadequate in the face of this new challenge. Their fundamental flaw lies in the limited analytical perspective and lack of data dimensions. Traditional methods often focus only on data from a single level or isolated nodes; for example, monitoring only the total load at the substation level or only the total current of a particular important feeder. This approach, attempting to judge the overall situation based on limited information, inevitably leads to cognitive distortion and misjudgment. A typical dilemma is that the overall load rate of a 110kV substation may show a healthy 70%, but a 10kV feeder under its jurisdiction may be severely overloaded, with an actual load rate as high as 130% due to distributed photovoltaic backfeeding or concentrated power consumption at local charging stations. Similarly, the total current of a feeder may be normal, but a transformer in a residential area at its power supply end may already be overloaded and overheating due to a surge in afternoon air conditioning load. This phenomenon of normal load at higher levels masking actual overload at lower levels is precisely due to the lack of systematic, cross-level collaborative analysis. The direct consequence is that the current overload risk warning in the power distribution network is not accurate enough, making it difficult to capture the real risk points in a timely and accurate manner, and failing to meet the requirements of the new power distribution system for the foresight and accuracy of safety warnings. Summary of the Invention

[0004] This invention provides a method, device, terminal equipment, and storage medium for predicting heavy overload risks in power distribution networks. The method can solve the problem of low accuracy in the early warning of heavy overload risks in the prior art.

[0005] An embodiment of the present invention provides a method for predicting the heavy overload risk of a power distribution network, comprising: The topology of the distribution network is obtained, along with first target characteristic parameters for several first target characteristics that characterize the operating trends and system stability of each substation in the distribution network, second target characteristic parameters for several second target characteristics that characterize the power flow distribution characteristics and load trends of each feeder in the distribution network, and third target characteristic parameters for several third target characteristics that characterize the electricity consumption behavior and load fluctuations of each distribution area in the distribution network. The load characteristic matrix of the distribution network is constructed based on the topology, the first target characteristic parameter, the second target characteristic parameter, and the third target characteristic parameter. The load feature matrix is ​​input into a preset load prediction model so that the load prediction model can capture the correlation between the loads of substations, feeders and transformer substations, and output the predicted load rates of substations, feeders and transformer substations. When a heavy overload risk is identified in a substation, feeder, and transformer area based on the predicted load rate, a heavy overload risk warning is generated.

[0006] Furthermore, the construction of the load prediction model includes: The system acquires the first characteristic parameters of several first initial characteristics of each substation, the second characteristic parameters of several second initial characteristics of each feeder, and the third characteristic parameters of the third initial characteristics of each distribution area before several historical heavy overload accidents, as well as the historical load rates of substations, feeders, and distribution area transformers after the occurrence of historical heavy overload accidents. The first initial characteristics include: reactive power voltage support index, main transformer oil temperature change rate, regional load, and regional load prediction deviation. The second initial characteristics include: power flow transfer coefficient, power flow distribution, feeder margin, meteorological data, and feeder temperature. The third initial characteristics include: apparent power of transformers, electricity consumption behavior characteristics, three-phase current imbalance, environmental data, and load fluctuation trends. Based on historical load rates, first characteristic parameters, second characteristic parameters, and third characteristic parameters, the importance scores of each first initial characteristic, each second initial characteristic, and each third initial characteristic in the task of predicting the load rates of substations, feeders, and transformer substations are evaluated. Based on the importance score, the first target feature, the second target feature, and the third target feature are selected from the first initial feature, the second initial feature, and the third initial feature; Based on the first feature parameter of the first target feature, the second feature parameter of the second target feature, the third feature parameter of the third target feature, the topology of the distribution network, and the historical load rate, several training samples are constructed. The training samples are used to iteratively train the preset initial load prediction model. In each training round, a preset loss function is used to calculate the corresponding loss function value based on the load rate prediction result output by the initial load prediction model. The model parameters of the initial load prediction model are updated based on the loss function value. When the loss function value converges, the last updated initial load prediction model is used as the load prediction model.

[0007] Furthermore, the historical load rate includes: the first historical load rate of the substation, the second historical load rate of the feeder, and the third historical load rate of the transformer in the distribution area. The process involves evaluating the importance scores of each initial feature (first, second, and third) in predicting the load rate of substations, feeders, and transformer substations based on historical load rates, first characteristic parameters, second characteristic parameters, and third characteristic parameters. This includes: Based on the first characteristic parameter and the first historical load rate, calculate the first nonlinear dependence coefficient of each first initial feature; based on the second characteristic parameter and the second historical load rate, calculate the second nonlinear dependence coefficient of each second initial feature; and based on the third characteristic parameter and the third historical load rate, calculate the third nonlinear dependence coefficient of each third initial feature. The first, second, and third initial features corresponding to the first, second, and third nonlinear dependency coefficients that are greater than the preset coefficient thresholds are respectively used as the first, second, and third candidate features. Based on the first historical load rate, the second historical load rate, the third historical load rate, the first feature parameter, the second feature parameter, and the third feature parameter, calculate the importance scores of the first candidate feature, the second candidate feature, and the third candidate feature in the task of predicting the load rate of substations, feeders, and transformer substations.

[0008] Further, the step of calculating the first nonlinear dependence coefficient of each first initial feature based on the first feature parameter and the first historical load rate, calculating the second nonlinear dependence coefficient of each second initial feature based on the second feature parameter and the second historical load rate, and calculating the third nonlinear dependence coefficient of each third initial feature based on the third feature parameter and the third historical load rate includes: For each first initial feature, based on the corresponding first parameter features and the first historical load rate, calculate the first marginal probability distribution of each first parameter feature, the second marginal probability distribution of the first historical load rate, and the first joint probability distribution of the first parameter feature and the first historical load rate. Based on the first marginal probability distribution, the first marginal probability distribution and the first joint probability distribution, calculate the first nonlinear dependence coefficient of each first initial feature; For each second initial feature, based on the corresponding second parameter features and the second historical load rate, calculate the third marginal probability distribution of each second parameter feature, the fourth marginal probability distribution of the second historical load rate, and the second joint probability distribution of the second parameter feature and the second historical load rate. Based on the third marginal probability distribution, the fourth marginal probability distribution, and the second joint probability distribution, calculate the second nonlinear dependence coefficient of each second initial feature; For each third initial feature, based on the corresponding third parameter features and the third historical load rate, calculate the fifth marginal probability distribution of each third parameter feature, the sixth marginal probability distribution of the third historical load rate, and the third joint probability distribution of the third parameter feature and the third historical load rate. Based on the fifth marginal probability distribution, the sixth marginal probability distribution, and the third joint probability distribution, calculate the third nonlinear dependence coefficient of each third initial feature.

[0009] Furthermore, the step of calculating the importance scores of the first candidate feature, the second candidate feature, and the third candidate feature in the task of predicting the load rate of substations, feeders, and transformer substations based on the first historical load rate, the second historical load rate, the third historical load rate, the first feature parameter, the second feature parameter, and the third feature parameter includes: Several first candidate features, several second candidate features, and several third candidate features are all used as candidate splitting features. With the goal of maximizing the load prediction accuracy, splitting nodes are selected from the candidate splitting features to construct a target decision tree. Calculate the gain value of each split node in the target decision tree, and sum the gain values ​​of split nodes with the same candidate split feature to determine the importance score of each candidate split feature.

[0010] Furthermore, the step of selecting the first target feature, the second target feature, and the third target feature from the first initial feature, the second initial feature, and the third initial feature based on the importance score includes: The importance scores of the first candidate feature, the second candidate feature, and the third candidate feature are combined based on the importance scores. The cumulative importance of each combination is calculated, and the combination with a cumulative importance greater than a preset threshold is taken as the target combination. The first candidate feature, the second candidate feature, and the third candidate feature in the target combination are respectively used as the first target feature, the second target feature, and the third target feature.

[0011] An embodiment of the present invention also provides a heavy overload risk prediction device for a power distribution network, comprising: The parameter acquisition module is used to acquire the topology of the distribution network, first target feature parameters of several first target features that characterize the operating trend and system stability characteristics of each substation in the distribution network, second target feature parameters of several second target features that characterize the power flow distribution characteristics and load trends of each feeder in the distribution network, and third target feature parameters of several third target features that characterize the electricity consumption behavior and load fluctuations of each distribution area in the distribution network. A feature construction model is used to construct the load feature matrix of the distribution network based on the topology, first target feature parameters, second target feature parameters, and third target feature parameters. The load prediction module is used to input the load feature matrix into a preset load prediction model so that the load prediction model can capture the correlation between the loads of substations, feeders and transformer substations, and output the predicted load rates of substations, feeders and transformer substations. The risk warning module is used to generate a heavy overload risk warning when the substation, feeder and transformer area are identified as having a heavy overload risk based on the predicted load rate.

[0012] Furthermore, the heavy overload risk prediction device for a power distribution network described in the above embodiments also includes a module construction module; The model building module is used to obtain the first characteristic parameters of several first initial characteristics of each substation in the distribution network before several historical heavy overload accidents, the second characteristic parameters of several second initial characteristics of each feeder, the third characteristic parameters of the third initial characteristics of each distribution area, and the historical load rate of substations, feeders, and distribution area transformers after the occurrence of historical heavy overload accidents. The first initial characteristics include: reactive power voltage support index, main transformer oil temperature change rate, regional load, and regional load prediction deviation; the second initial characteristics include: power flow transfer coefficient, power flow distribution, feeder margin, meteorological data, and feeder temperature; the third initial characteristics include: apparent power of transformers, electricity consumption behavior characteristics, three-phase current imbalance, environmental data, and load fluctuation trend. Based on historical load rates, first characteristic parameters, second characteristic parameters, and third characteristic parameters, the importance scores of each first initial characteristic, each second initial characteristic, and each third initial characteristic in the task of predicting the load rates of substations, feeders, and transformer substations are evaluated. Based on the importance score, the first target feature, the second target feature, and the third target feature are selected from the first initial feature, the second initial feature, and the third initial feature; Based on the first feature parameter of the first target feature, the second feature parameter of the second target feature, the third feature parameter of the third target feature, the topology of the distribution network, and the historical load rate, several training samples are constructed. The training samples are used to iteratively train the preset initial load prediction model. In each training round, a preset loss function is used to calculate the corresponding loss function value based on the load rate prediction result output by the initial load prediction model. The model parameters of the initial load prediction model are updated based on the loss function value. When the loss function value converges, the last updated initial load prediction model is used as the load prediction model.

[0013] This application also provides a terminal device, including: One or more processors; A memory, coupled to the processor, for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement a heavy overload risk prediction method for a power distribution network as described in the above embodiments of the invention.

[0014] This application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a method for predicting heavy overload risks in a power distribution network as described in the above embodiments.

[0015] The following benefits can be obtained by implementing the present invention: This invention provides a method, device, terminal equipment, and storage medium for predicting heavy overload risks in a distribution network. The method acquires first target feature parameters representing several first target features that characterize the operating trends and system stability of each substation in the distribution network; second target feature parameters representing several second target features that characterize the power flow distribution characteristics and load trends of each feeder in the distribution network; and third target feature parameters representing several third target features that characterize the electricity consumption behavior and load fluctuations of each transformer area in the distribution network. This provides a three-layer feature parameter set for the load prediction model, including substation, feeder, and transformer area parameters, thus constructing a global view. This facilitates the subsequent steps of the load prediction model in capturing the correlation between the loads of upstream and downstream equipment, improving the accuracy of the predicted load rate of each device, and consequently improving the accuracy of heavy overload risk prediction. Attached Figure Description

[0016] To more clearly illustrate the technical solution of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating a method for predicting heavy overload risks in a power distribution network according to a certain embodiment of this application. Figure 2This is a schematic diagram of the structure of a heavy overload risk prediction device for a power distribution network provided in a certain embodiment of this application; Figure 3 This is a schematic diagram of the structure of a terminal device provided in a certain embodiment of this application; Figure 4 This is a schematic diagram of feeder predicted load rate provided in a certain embodiment of this application; Figure 5 This is a thermal diagram of the three-phase imbalance in a transformer area provided in a certain embodiment of this application; Figure 6 This is a comparison chart of the predicted load rate and the actual heavy overload event provided in a certain embodiment of this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the application; the terms “comprising” and “having”, and any variations thereof, in the specification, claims, and foregoing description of the drawings are intended to cover non-exclusive inclusion.

[0020] In the description of the embodiments of this application, technical terms such as "first" and "second" are used only to distinguish different objects and should not be construed as indicating or implying relative importance or implicitly specifying the number, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly defined.

[0021] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0022] In the description of the embodiments in this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0023] In the description of the embodiments of this application, the term "multiple" refers to two or more (including two), similarly, "multiple sets" refers to two or more (including two sets), and "multiple pieces" refers to two or more (including two pieces).

[0024] In the description of the embodiments of this application, unless otherwise expressly specified and limited, technical terms such as "installation," "connection," "joining," and "fixing" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. For those skilled in the art, the specific meaning of the above terms in the embodiments of this application can be understood according to the specific circumstances.

[0025] See Figure 1 To address the problems in the prior art, an embodiment of the present invention provides a method for predicting the heavy overload risk of a power distribution network, comprising: S1. Obtain the topology of the distribution network, first target characteristic parameters of several first target characteristics used to characterize the operating trend and system stability characteristics of each substation in the distribution network, second target characteristic parameters of several second target characteristics used to characterize the power flow distribution characteristics and load trend of each feeder in the distribution network, and third target characteristic parameters of several third target characteristics used to characterize the electricity consumption behavior and load fluctuation of each distribution area in the distribution network. In a preferred embodiment of the present invention, the first target feature includes: regional load and reactive power voltage support index; the second target feature includes: power flow transfer coefficient, feeder margin, and power flow distribution; the third target feature includes: electricity consumption behavior characteristics and load fluctuation trend. By obtaining the feature parameters of these target features, the following feature matrix is ​​constructed: ; in, It is the first substation The first target feature parameter of the first target feature, It is the feeder layer. The second target feature parameters of the second target feature, It is the first level of the Taiwan area. The third target feature parameter of the third target feature.

[0026] S2. Construct the load characteristic matrix of the distribution network based on the topology, the first target characteristic parameter, the second target characteristic parameter, and the third target characteristic parameter; In a preferred embodiment of the present invention, the first target characteristic parameter, the second target characteristic parameter, and the third target characteristic parameter are integrated according to the connection relationship between each substation, feeder, and transformer area. Specifically, the array of the first target characteristic parameters of each transformer area, the array of the second target characteristic parameters of the feeders connected to each transformer area, and the array of the third target characteristic parameters of the substations connected to the feeders of each transformer area are combined into a row to construct a load characteristic matrix.

[0027] S3. Input the load feature matrix into the preset load prediction model so that the load prediction model can capture the correlation between the loads of substations, feeders and transformer substations, and output the predicted load rates of substations, feeders and transformer substations. In a preferred embodiment of the present invention, the load prediction model can be trained using multilayer perceptrons, convolutional neural networks, and LSTMs. Specifically, in this embodiment, a multilayer perceptron (MLP) is used, leveraging its ability to learn complex nonlinear relationships. Multiple hidden layers are designed to learn high-level feature interactions, capturing the correlation between the loads of substations, feeders, and transformer substations, thereby improving the accuracy of load rate prediction.

[0028] S4. When the substation, feeder and transformer area are identified as having a risk of heavy overload based on the predicted load rate, a heavy overload risk warning is generated.

[0029] In a preferred embodiment of the present invention, when the load rate of any device in the substation, feeder, or transformer exceeds 100%, a heavy overload risk warning is issued for that device. Furthermore, to prevent sudden equipment malfunctions from causing an increase in load rate, a warning is issued for any device when its load rate exceeds 80%, reminding relevant maintenance personnel to pay closer attention to the load on that device.

[0030] To verify this method, this embodiment selects a 10kV distribution network area in a prefecture-level city in China as the case study. This distribution network includes one 110 / 10kV substation with a main transformer capacity of 40 MVA, monitoring the main transformer load rate, oil temperature, and bus voltage deviation. At the feeder level, there are two 10kV feeders with capacities of 12 MVA and 15 MVA respectively, monitoring real-time load rate, N-1 overload risk coefficient, and power interconnection with adjacent feeders. At the transformer substation level, there are six substations, each with a capacity of 1.6–2.0 MVA, covering residential and commercial users, monitoring three-phase imbalance, synchronous load difference, and user electricity consumption patterns. Under multiple scenarios, the daily peak hours are 19:00–21:00, with concentrated residential air conditioning and commercial loads; during the evening peak, the substation load accounts for approximately 95–105% of the transformer capacity. To verify the control effect, the load between feeders A and B can be adjusted by 2–3 MVA through transfer, while users can reduce the load by 0.5–1 MVA during peak hours through demand response, thus forming verification conditions for heavy overload risk under different load scenarios. The example is verified according to the method proposed in this patent. First, multi-source data from substations, feeders, and distribution areas are integrated through a unified data access and preprocessing module, and the time-series data is cleaned, aligned, and standardized to form a high-quality cross-level dataset. Then, features of substations, feeders, and distribution areas are extracted according to a multi-level hierarchical feature construction strategy. Next, key features are extracted using mutual information and the LightGBM embedded feature selection method, and a unified load feature vector is constructed. Finally, load prediction is performed based on this load feature vector, and transfer and demand response measures are triggered within the prediction lead time to verify the method's ability to identify and mitigate heavy overload risks.

[0031] Figure 4 This paper presents the load rate of a feeder over a 24-hour period and the changes in predicted probability based on multi-level feature representation. The graph shows that during the evening peak period from 19:00 to 21:00, the feeder load rate significantly increases to over 100%, corresponding to the highest risk of heavy overload. The predicted probability curve generated using the multi-level, multi-scenario, and multi-scale feature representation construction method proposed in the example accurately reflects the load risk during peak periods. The predicted load probability is highly consistent with the actual peak load, indicating that the selected target features can effectively capture the high-risk period of heavy overload during the evening peak, providing high-value input for the load prediction model and verifying the effectiveness and feasibility of multi-level feature representation in heavy overload prediction.

[0032] Figure 5A heatmap showing the three-phase imbalance characteristics of six transformer substations over a 24-hour period is presented. During the evening peak hours (19:00–21:00), the three-phase imbalance significantly increased in some substations, indicating that uneven local load distribution may lead to transformer overload. Using the feature construction method of this invention, local physical characteristics such as three-phase imbalance are incorporated as key features into the load feature vector, enabling accurate capture of localized heavy overload risks in substations. The heatmap clearly shows the risk differences of each substation at different time periods, verifying the important role of hierarchical feature engineering in predicting localized load risks.

[0033] Figure 6 This study compares the load forecasts and actual heavy overload events at three levels: substations, feeders, and distribution transformers. Actual heavy overload events only occur during the evening peak hours of 19:00–21:00. The forecast curves at each level accurately reflect the heavy overload risk before the peak and demonstrate the differences in contribution between different levels. The substation level provides overall macroscopic status information, the feeder level reflects intermediate transmission risks, and the distribution transformer level captures local detailed features. These results verify that the present invention can systematically integrate macroscopic and local information to achieve high-precision heavy overload early warning, providing reliable data support for smart distribution network safety early warning and operation control.

[0034] In summary, this example fully validates the effectiveness of the proposed method. By constructing hierarchical features at the substation, feeder, and distribution area levels, and through cross-level fusion and multi-scale representation, it can accurately capture heavy overload events during high-risk periods such as evening peak hours and clearly reflect the load characteristics at each level. Experimental results show that this method can not only provide early warning of heavy overload risks but also provide high-value, interpretable input features for load prediction models, offering reliable data support and technical assurance for the safe operation and proactive control of smart distribution networks.

[0035] Preferably, the construction of the load prediction model includes: S21. Obtain the first characteristic parameters of several first initial characteristics of each substation in the distribution network before several historical heavy overload accidents, the second characteristic parameters of several second initial characteristics of each feeder, the third characteristic parameters of the third initial characteristics of each distribution area, and the historical load rate of the substations, feeders, and distribution area transformers after the historical heavy overload accidents; wherein, the first initial characteristics include: reactive power voltage support index, main transformer oil temperature change rate, regional load, and regional load prediction deviation; the second initial characteristics include: power flow transfer coefficient, power flow distribution, feeder margin, meteorological data, and feeder temperature; the third initial characteristics include: transformer apparent power, electricity consumption behavior characteristics, three-phase current imbalance, environmental data, and load fluctuation trend; In a preferred embodiment of the present invention, initial characteristic data from substations, feeders, and distribution areas are respectively accessed through a power distribution automation system, an electricity consumption information acquisition system, and a meteorological data platform. The first initial characteristics include: reactive power voltage support index, main transformer oil temperature change rate, regional load, and regional load prediction deviation; the second initial characteristics include: power flow transfer coefficient, power flow distribution, feeder margin, meteorological data, and feeder temperature; the third initial characteristics include: transformer apparent power, electricity consumption behavior characteristics, three-phase current imbalance, environmental data, and load fluctuation trend. Specifically, the reactive power voltage support index is determined by the bus voltage. With the capacity of reactive power compensation equipment Perform the calculation: ; In the formula, For the total active power load of the busbar For the target power factor angle, This is the rated bus voltage.

[0036] The rate of change of main transformer oil temperature is obtained by collecting main transformer oil temperature data. The calculation involves determining the change in value per unit time. ; in, The duration of a unit of time. This represents the main transformer oil temperature data for the t-th unit of time.

[0037] Regional load forecast deviation characterizes the degree of deviation between operational conditions and expectations. Excessive deviation is a potential risk signal, determined by the current total regional load. Compared with short-term load forecasts Perform calculations. ; in, This represents the regional load forecasting deviation.

[0038] Furthermore, for the second initial feature, the meteorological data is the temperature and weather of the environment where each feeder is located, obtained through a meteorological platform; the power flow distribution is the power transmitted on each feeder, obtained through a distribution automation system; and the feeder margin is the ratio of the difference between the current power transmitted on the feeder and the rated power to the rated power.

[0039] The power flow transfer coefficient represents the power flow transfer of this feeder after the critical equipment of the adjacent feeder is shut down. ; ; in, This represents the current power transmitted on the feeder. To estimate the transfer power, This represents the power increment of adjacent feeders when critical equipment is out of service. The power distribution factor between feeders (which can be determined through power flow calculations or historical power transfer ratios). This represents the maximum allowable transmission power of the line.

[0040] Considering that the line current carrying capacity is affected by ambient temperature The rated current carrying capacity is corrected using formulas provided by international standards, and the real-time feeder temperature is calculated. : ; in, This is the maximum allowable temperature of the conductor. The reference ambient temperature is the temperature corresponding to the rated current carrying capacity.

[0041] Furthermore, for the third initial characteristic, the environmental data consisted of temperature and weather data for each station area obtained from a meteorological platform. The three-phase current imbalance was calculated using the negative sequence imbalance standard from the national standard. or current imbalance To quantify the three-phase imbalance: ; in, This represents the positive sequence component of the current. It represents the negative sequence component of the current.

[0042] Load fluctuation trend, based on the deviation rate between the current load of the transformer in the distribution area and the historical load for the same period. Perform the calculation.

[0043] ; in, For the current load This is the historical load for the same period.

[0044] Electricity consumption behavior characteristics were analyzed using a supervised clustering algorithm to examine the daily load curves of all users within a transformer substation, aiming to uncover electricity consumption behavior patterns. The steps are as follows: 1) Data Preprocessing: Normalize the load curve for each user and smooth noise using a moving average. 2) Cluster Analysis: Input each user's load sequence as a vector into the K-Means algorithm. The optimal number of clusters is determined using the elbow method. 3) Typical Curves and Shape Indicators: Calculate the cluster center of each cluster as a typical curve and extract the following indicators: peak value, peak location, valley value, peak-valley difference, load slope, and fluctuation amplitude. 4) Feature Construction: Combine the user proportions of each cluster and the typical curve indicators to form the transformer substation layer features, refining the overall electricity consumption behavior pattern of the transformer substation.

[0045] S22. Based on the historical load rate, the first characteristic parameter, the second characteristic parameter, and the third characteristic parameter, evaluate the importance score of each first initial characteristic, each second initial characteristic, and each third initial characteristic in the task of predicting the load rate of substations, feeders, and transformer substations. Preferably, the historical load rate includes: the first historical load rate of the substation, the second historical load rate of the feeder, and the third historical load rate of the transformer in the distribution area; The process involves evaluating the importance scores of each initial feature (first, second, and third) in predicting the load rate of substations, feeders, and transformer substations based on historical load rates, first characteristic parameters, second characteristic parameters, and third characteristic parameters. This includes: S221. Calculate the first nonlinear dependence coefficient of each first initial feature based on the first feature parameter and the first historical load rate; calculate the second nonlinear dependence coefficient of each second initial feature based on the second feature parameter and the second historical load rate; and calculate the third nonlinear dependence coefficient of each third initial feature based on the third feature parameter and the third historical load rate. In a preferred embodiment of the present invention, the first historical load rate Apparent power of each winding of the main transformer With rated capacity The ratio: ; Second historical load rate To calculate the feeder current With rated current carrying capacity The ratio: ; Third historical load rate For Taiwan to change the power With rated capacity The ratio: .

[0046] Preferably, the step of calculating the first nonlinear dependence coefficient of each first initial feature based on the first feature parameter and the first historical load rate, calculating the second nonlinear dependence coefficient of each second initial feature based on the second feature parameter and the second historical load rate, and calculating the third nonlinear dependence coefficient of each third initial feature based on the third feature parameter and the third historical load rate includes: For each first initial feature, based on the corresponding plurality of first parameter features and the first historical load rate, calculate the first marginal probability distribution of each first parameter feature, the second marginal probability distribution of the first historical load rate, and the first joint probability distribution of the first parameter feature and the first historical load rate; calculate the first nonlinear dependency coefficient of each first initial feature based on the first marginal probability distribution, the first marginal probability distribution, and the first joint probability distribution; for each second initial feature, based on the corresponding plurality of second parameter features and the second historical load rate, calculate the third marginal probability distribution of each second parameter feature, the fourth marginal probability distribution of the second historical load rate, and the second joint probability distribution of the second parameter feature and the second historical load rate; calculate the second nonlinear dependency coefficient of each second initial feature based on the third marginal probability distribution, the fourth marginal probability distribution, and the second joint probability distribution; for each third initial feature, based on the corresponding plurality of third parameter features and the third historical load rate, calculate the fifth marginal probability distribution of each third parameter feature, the sixth marginal probability distribution of the third historical load rate, and the third joint probability distribution of the third parameter feature and the third historical load rate; calculate the third nonlinear dependency coefficient of each third initial feature based on the fifth marginal probability distribution, the sixth marginal probability distribution, and the third joint probability distribution.

[0047] In a preferred embodiment of the present invention, each initial feature is first calculated. With the predicted target Mutual information (non-linear dependency coefficient) Mutual information measures the non-linear dependency between two variables. That is, to what extent knowing information about feature X can reduce the uncertainty of predicting target Y. Initial features are first screened by calculating the mutual information between the initial features at each level and the predicted load rate of that level.

[0048] The nonlinear dependence coefficient is calculated using the following formula: in, For the first An initial feature, To predict load factor, Features With the goal The joint probability distribution of the initial feature X is P(x) and P(y) is the marginal probability distribution, that is, P(x) is the probability that the initial feature X takes the value of the parameter feature x.

[0049] S222. The first initial feature, the second initial feature, and the third initial feature corresponding to the first nonlinear dependency coefficient, the second nonlinear dependency coefficient, and the third nonlinear dependency coefficient that are greater than the preset coefficient threshold are respectively used as the first candidate feature, the second candidate feature, and the third candidate feature. In a preferred embodiment of the present invention, a threshold is set. Initial features with mutual information values ​​greater than a threshold are retained as candidate features, while initial features with weak relevance to the target are eliminated to reduce feature dimensionality.

[0050] S223. Based on the first historical load rate, the second historical load rate, the third historical load rate, the first feature parameter, the second feature parameter, and the third feature parameter, calculate the importance scores of the first candidate feature, the second candidate feature, and the third candidate feature in the task of predicting the load rate of substations, feeders, and transformer substations.

[0051] Preferably, the step of calculating the importance scores of the first candidate feature, the second candidate feature, and the third candidate feature in the task of predicting the load rate of substations, feeders, and transformer substations based on the first historical load rate, the second historical load rate, the third historical load rate, the first feature parameter, the second feature parameter, and the third feature parameter includes: Several first candidate features, several second candidate features, and several third candidate features are all used as candidate splitting features. With the goal of maximizing the load prediction accuracy, splitting nodes are selected from the candidate splitting features to construct a target decision tree. The gain value of each splitting node in the target decision tree is calculated, and the gain values ​​of splitting nodes with the same candidate splitting feature are accumulated to determine the importance score of each candidate splitting feature.

[0052] S23. Based on the importance score, select the first target feature, the second target feature, and the third target feature from the first initial feature, the second initial feature, and the third initial feature; Preferably, the step of selecting the first target feature, the second target feature, and the third target feature from the first initial feature, the second initial feature, and the third initial feature based on the importance score includes: The importance scores of the first candidate feature, the second candidate feature, and the third candidate feature are combined based on the importance scores. The cumulative importance of each combination is calculated, and the combination with a cumulative importance greater than a preset threshold is taken as the target combination. The first candidate feature, the second candidate feature, and the third candidate feature in the target combination are taken as the first target feature, the second target feature, and the third target feature, respectively.

[0053] In a preferred embodiment of the present invention, the initially selected candidate features are input into the LightGBM classification model for training. After training, the feature importance score of the model is output. Select the feature set whose cumulative importance exceeds 95% of the total importance to form the final key feature subset.

[0054] Specifically, LightGBM is an efficient gradient boosting decision tree algorithm. Its core is to build a series of decision trees sequentially through multiple rounds of iteration. Each tree learns the prediction residuals of all previous tree combinations, and finally the prediction results of all trees are added together to obtain a powerful model.

[0055] Its high efficiency stems primarily from two innovations. First, it discretizes continuous features into multiple buckets, forming a histogram. When finding the optimal split point, only the buckets need to be traversed, rather than all the data, greatly improving speed and saving memory. Second, it employs a leaf-wise growth strategy. Unlike traditional layer-by-layer growth, it selects the leaf with the highest gain for splitting each time. This achieves higher accuracy with the same number of splits, but may require depth control to prevent overfitting.

[0056] The model building process can be simplified into four steps: 1. Initialize using a constant (such as the target mean) as the initial prediction. 2. Calculate the negative gradient between the current model's prediction and the true value (i.e., the direction of the model's error). 3. Build a new decision tree to fit this negative gradient (i.e., learn to correct the error). This process uses histograms to quickly find split points and grows according to a leaf-wise strategy. 4. Multiply the new tree by a learning rate and add it to the existing model to form a stronger prediction. Repeat steps 2-4 until a specified number of trees is reached or the model performance no longer improves.

[0057] S24. Construct several training samples based on the first feature parameter of the first target feature, the second feature parameter of the second target feature, the third feature parameter of the third target feature, the topology of the distribution network, and the historical load rate. In a preferred embodiment of the present invention, the same method as in step S2 is used to construct a training sample matrix based on the first feature parameter of the first target feature, the second feature parameter of the second target feature, the third feature parameter of the third target feature, and the topology of the distribution network, and then construct several training samples by combining the matrix with the corresponding historical load rate.

[0058] S25. The training samples are used to iteratively train the preset initial load prediction model. In each round of training, a preset loss function is used to calculate the corresponding loss function value based on the load rate prediction result output by the initial load prediction model. The model parameters of the initial load prediction model are updated based on the loss function value. When the loss function value converges, the last updated initial load prediction model is used as the load prediction model.

[0059] In a preferred embodiment of the present invention, a multilayer perceptron (MLP) is employed. Taking advantage of its ability to learn complex nonlinear relationships, multiple hidden layers are designed to learn high-level feature interactions, capture the correlation between the loads of substations, feeders, and transformer substations, and improve the accuracy of load rate prediction.

[0060] Furthermore, this embodiment provides a method for predicting heavy overload risks in a distribution network. By acquiring first target feature parameters for several first target features characterizing the operating trends and system stability of each substation in the distribution network, second target feature parameters for several second target features characterizing the power flow distribution characteristics and load trends of each feeder in the distribution network, and third target feature parameters for several third target features characterizing the electricity consumption behavior and load fluctuations of each transformer area in the distribution network, a three-layer feature parameter system for the load prediction model is provided, consisting of substations, feeders, and transformer areas. This constructs a global view, facilitating the subsequent steps of the load prediction model to capture the correlation between the loads of upper and lower level equipment, improving the accuracy of the predicted load rate of each equipment, and thus improving the accuracy of heavy overload risk prediction.

[0061] See Figure 2 This invention provides a heavy overload risk prediction device for a power distribution network, comprising: The parameter acquisition module is used to acquire the topology of the distribution network, first target feature parameters of several first target features that characterize the operating trend and system stability characteristics of each substation in the distribution network, second target feature parameters of several second target features that characterize the power flow distribution characteristics and load trends of each feeder in the distribution network, and third target feature parameters of several third target features that characterize the electricity consumption behavior and load fluctuations of each distribution area in the distribution network. A feature construction model is used to construct the load feature matrix of the distribution network based on the topology, first target feature parameters, second target feature parameters, and third target feature parameters. The load prediction module is used to input the load feature matrix into a preset load prediction model so that the load prediction model can capture the correlation between the loads of substations, feeders and transformer substations, and output the predicted load rates of substations, feeders and transformer substations. The risk warning module is used to generate a heavy overload risk warning when the substation, feeder and transformer area are identified as having a heavy overload risk based on the predicted load rate.

[0062] Furthermore, the heavy overload risk prediction device for a power distribution network described in the above embodiments also includes a module construction module; The model building module is used to obtain the first characteristic parameters of several first initial characteristics of each substation in the distribution network before several historical heavy overload accidents, the second characteristic parameters of several second initial characteristics of each feeder, the third characteristic parameters of the third initial characteristics of each distribution area, and the historical load rate of substations, feeders, and distribution area transformers after the occurrence of historical heavy overload accidents. The first initial characteristics include: reactive power voltage support index, main transformer oil temperature change rate, regional load, and regional load prediction deviation; the second initial characteristics include: power flow transfer coefficient, power flow distribution, feeder margin, meteorological data, and feeder temperature; the third initial characteristics include: apparent power of transformers, electricity consumption behavior characteristics, three-phase current imbalance, environmental data, and load fluctuation trend. Based on historical load rates, first characteristic parameters, second characteristic parameters, and third characteristic parameters, the importance scores of each first initial characteristic, each second initial characteristic, and each third initial characteristic in the task of predicting the load rates of substations, feeders, and transformer substations are evaluated. Based on the importance score, the first target feature, the second target feature, and the third target feature are selected from the first initial feature, the second initial feature, and the third initial feature; Based on the first feature parameter of the first target feature, the second feature parameter of the second target feature, the third feature parameter of the third target feature, the topology of the distribution network, and the historical load rate, several training samples are constructed. The training samples are used to iteratively train the preset initial load prediction model. In each training round, a preset loss function is used to calculate the corresponding loss function value based on the load rate prediction result output by the initial load prediction model. The model parameters of the initial load prediction model are updated based on the loss function value. When the loss function value converges, the last updated initial load prediction model is used as the load prediction model.

[0063] It is understood that the above-described device embodiments correspond to the method embodiments of the present invention, and can realize the method for predicting heavy overload risks in a power distribution network provided by any of the above-described method embodiments of the present invention.

[0064] It should be noted that the device embodiments described above are merely illustrative, and some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can specifically be implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0065] See Figure 3 One embodiment of this application also provides a terminal device, including: One or more processors; A memory, coupled to the processor, for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement a heavy overload risk prediction method for a power distribution network as described above.

[0066] The processor controls the overall operation of the terminal device to complete all or part of the steps of the aforementioned method for predicting heavy overload risks in a power distribution network. The memory stores various types of data to support the operation of the terminal device. This data may include, for example, instructions for any application or method operating on the terminal device, as well as application-related data. The memory can be implemented using any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0067] In an exemplary embodiment, the terminal device may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform a heavy overload risk prediction method for a power distribution network as described in any of the foregoing embodiments, and to achieve the same technical effects as the methods described above.

[0068] In another exemplary embodiment, a computer-readable storage medium including a computer program is also provided. When executed by a processor, the computer program implements the steps of a heavy overload risk prediction method for a distribution network as described in any of the foregoing embodiments. For example, the computer-readable storage medium may be the aforementioned memory including the computer program, which may be executed by a processor of a terminal device to complete the heavy overload risk prediction method for a distribution network as described in any of the foregoing embodiments, and achieve the same technical effects as the aforementioned method.

[0069] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for predicting the risk of overload of an electrical distribution network, characterized in that, The method comprises the following steps: obtaining a topology structure of a power distribution network, first target feature parameters of a plurality of first target features for representing operation trends and system stability characteristics of each substation in the power distribution network, second target feature parameters of a plurality of second target features for representing power flow distribution characteristics and load trends of each feeder in the power distribution network, and third target feature parameters of a plurality of third target features for representing electricity consumption behaviors and load fluctuations of each area in the power distribution network; constructing a load feature matrix of the power distribution network according to the topology structure, the first target feature parameters, the second target feature parameters, and the third target feature parameters; inputting the load feature matrix into a preset load prediction model to enable the load prediction model to capture the correlation between the substations, the feeders, and the area transformers, and output predicted load rates of the substations, the feeders, and the area transformers; generating a heavy overload risk warning when heavy overload risks exist in the substations, the feeders, and the area transformers according to the predicted load rates.

2. The method for predicting the risk of overload of a power distribution network according to claim 1, characterized in that, The construction of the load prediction model comprises the following steps: obtaining first feature parameters of a plurality of first initial features of each substation in the power distribution network, second feature parameters of a plurality of second initial features of each feeder, third feature parameters of a plurality of third initial features of each area before a plurality of historical heavy overload accidents occur, and historical load rates of the substations, the feeders, and the area transformers after the historical heavy overload accidents occur; wherein the first initial features include reactive power voltage support indicators, main transformer oil temperature change rates, regional loads, and regional load prediction deviations; the second initial features include power flow transfer coefficients, power flow distributions, feeder margins, meteorological data, and feeder temperatures; and the third initial features include apparent powers of transformers, electricity consumption behaviors, three-phase current unbalance degrees, environmental data, and load fluctuation trends; evaluating the importance scores of each first initial feature, each second initial feature, and each third initial feature in the task of predicting the load rates of the substations, the feeders, and the area transformers according to the historical load rates, the first feature parameters, the second feature parameters, and the third feature parameters; selecting the first target features, the second target features, and the third target features from the first initial features, the second initial features, and the third initial features according to the importance scores; constructing a plurality of training samples according to the first feature parameters of the first target features, the second feature parameters of the second target features, the third feature parameters of the third target features, the topology structure of the power distribution network, and the historical load rates; iteratively training a preset initial load prediction model using the training samples, and in each training process, calculating a loss function value corresponding to a load rate prediction result output by the initial load prediction model using a preset loss function, and updating model parameters of the initial load prediction model according to the loss function value, and when the loss function value converges, taking the last updated initial load prediction model as the load prediction model.

3. A method of predicting the risk of overload of an electricity distribution network according to claim 2, characterized in that, The historical load rates comprise first historical load rates of the substations, second historical load rates of the feeders, and third historical load rates of the area transformers. The importance score of each first initial feature, each second initial feature and each third initial feature on the substation load rate prediction task, the feeder load rate prediction task and the transformer load rate prediction task is evaluated according to the historical load rate, the first feature parameter, the second feature parameter and the third feature parameter, and the importance score of each first initial feature, each second initial feature and each third initial feature on the substation load rate prediction task, the feeder load rate prediction task and the transformer load rate prediction task is calculated according to the historical load rate, the first feature parameter, the second feature parameter and the third feature parameter. The first nonlinear dependence coefficient of each first initial feature is calculated according to the first feature parameter and the first historical load rate, the second nonlinear dependence coefficient of each second initial feature is calculated according to the second feature parameter and the second historical load rate, and the third nonlinear dependence coefficient of each third initial feature is calculated according to the third feature parameter and the third historical load rate. The first initial feature, the second initial feature and the third initial feature corresponding to the first nonlinear dependence coefficient, the second nonlinear dependence coefficient and the third nonlinear dependence coefficient greater than the preset coefficient threshold are respectively taken as the first candidate feature, the second candidate feature and the third candidate feature. The importance score of each first candidate feature, each second candidate feature and each third candidate feature on the substation load rate prediction task, the feeder load rate prediction task and the transformer load rate prediction task is calculated according to the first historical load rate, the second historical load rate, the third historical load rate, the first feature parameter, the second feature parameter and the third feature parameter.

4. A method of predicting the risk of overload of an electricity distribution network according to claim 3, characterized in that, The first nonlinear dependence coefficient of each first initial feature is calculated according to the first feature parameter and the first historical load rate, the second nonlinear dependence coefficient of each second initial feature is calculated according to the second feature parameter and the second historical load rate, and the third nonlinear dependence coefficient of each third initial feature is calculated according to the third feature parameter and the third historical load rate. For each first initial feature, the first marginal probability distribution of each first parameter feature, the second marginal probability distribution of the first historical load rate and the first joint probability distribution of the first parameter feature and the first historical load rate are respectively calculated according to the corresponding first parameter feature and the first historical load rate. The first nonlinear dependence coefficient of each first initial feature is calculated according to the first marginal probability distribution, the first marginal probability distribution and the first joint probability distribution. For each second initial feature, the third marginal probability distribution of each second parameter feature, the fourth marginal probability distribution of the second historical load rate and the second joint probability distribution of the second parameter feature and the second historical load rate are respectively calculated according to the corresponding second parameter feature and the second historical load rate. The second nonlinear dependence coefficient of each second initial feature is calculated according to the third marginal probability distribution, the fourth marginal probability distribution and the second joint probability distribution. For each third initial feature, the fifth marginal probability distribution of each third parameter feature, the sixth marginal probability distribution of the third historical load rate and the third joint probability distribution of the third parameter feature and the third historical load rate are respectively calculated according to the corresponding third parameter feature and the third historical load rate. The third nonlinear dependence coefficient of each third initial feature is calculated according to the fifth marginal probability distribution, the sixth marginal probability distribution and the third joint probability distribution.

5. A method of predicting the risk of overload of an electricity distribution network according to claim 4, characterized in that, The calculating of the importance scores of the first candidate feature, the second candidate feature and the third candidate feature on the tasks of predicting the load rates of the transformer substation, the feeder line and the transformer substation respectively according to the first historical load rate, the second historical load rate, the third historical load rate, the first characteristic parameter, the second characteristic parameter and the third characteristic parameter comprises: The first candidate feature, the second candidate feature and the third candidate feature are all taken as candidate split features, and a split node is screened from the candidate split features to maximize the load prediction accuracy, so as to construct a target decision tree; The gain value of each split node of the target decision tree is calculated, and the gain values of the split nodes corresponding to the same candidate split feature are accumulated to determine the importance score of each candidate split feature.

6. A method of predicting the risk of overload of an electricity distribution network according to claim 5, characterized in that, The screening of the first target feature, the second target feature and the third target feature from the first initial feature, the second initial feature and the third initial feature according to the importance scores comprises: The importance scores of the first candidate feature, the second candidate feature and the third candidate feature are combined according to the importance scores, the cumulative importance of each combination is calculated, and the combination with the cumulative importance greater than a preset threshold is taken as a target combination; The first candidate feature, the second candidate feature and the third candidate feature in the target combination are taken as the first target feature, the second target feature and the third target feature respectively.

7. An apparatus for predicting a risk of overload of a power distribution network, characterized by, Comprise: The parameter acquisition module is configured to acquire the topological structure of the power distribution network, the first target feature parameters of the first target features representing the operation trend and the system stability characteristics of each transformer substation of the power distribution network, the second target feature parameters of the second target features representing the power flow distribution characteristics and the load trend of each feeder line of the power distribution network, and the third target feature parameters of the third target features representing the power consumption behavior and the load fluctuation of each transformer substation of the power distribution network. The feature construction model is configured to construct a load feature matrix of the power distribution network according to the topological structure, the first target feature parameters, the second target feature parameters and the third target feature parameters. The load prediction module is configured to input the load feature matrix into a preset load prediction model, so that the load prediction model captures the correlation between the loads of the transformer substation, the feeder line and the transformer substation, and outputs the predicted load rates of the transformer substation, the feeder line and the transformer substation. The risk early warning module is configured to generate a heavy overload risk warning when it is identified that the transformer substation, the feeder line and the transformer substation have a heavy overload risk according to the predicted load rates.

8. A device for predicting the risk of overload of an electricity distribution network according to claim 7, characterized in that, Further comprise, the module construction module; The model construction module is configured to obtain first feature parameters of a plurality of first initial features of each substation in the power distribution network, second feature parameters of a plurality of second initial features of each feeder, third feature parameters of a plurality of third initial features of each area before a plurality of historical overload accidents occur, and historical load rates of the substations, the feeders, and the area transformers after the historical overload accidents occur; the first initial features include reactive power voltage support indicators, main transformer oil temperature change rates, regional loads, and regional load prediction deviations; the second initial features include power flow transfer coefficients, power flow distributions, feeder margins, meteorological data, and feeder temperatures; and the third initial features include apparent powers of the transformers, power consumption behavior characteristics, three-phase current unbalance degrees, environmental data, and load fluctuation trends. According to the historical load rates, the first feature parameters, the second feature parameters, and the third feature parameters, importance scores of the first initial features, the second initial features, and the third initial features in predicting load rates of the substations, the feeders, and the area transformers are respectively evaluated. According to the importance scores, first target features, second target features, and third target features are selected from the first initial features, the second initial features, and the third initial features. According to the first feature parameters of the first target features, the second feature parameters of the second target features, the third feature parameters of the third target features, a topology of the power distribution network, and the historical load rates, a plurality of training samples are constructed. The training samples are used to iteratively train a preset initial load prediction model, and in each training process, a preset loss function is used to calculate a corresponding loss function value according to a load rate prediction result output by the initial load prediction model, and model parameters of the initial load prediction model are updated according to the loss function value; when the loss function value converges, the last updated initial load prediction model is taken as a load prediction model.

9. A terminal device, comprising: Comprise: One or more processors; Memory, coupled with the processor, for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement a power distribution network overload risk prediction method according to any one of claims 1-6.

10. A storage medium having stored thereon a computer program, characterized in that The computer program is executed by the processor to implement a power distribution network overload risk prediction method according to any one of claims 1-6.

Images