Power distribution network reliability prediction method based on influence of weather on power distribution network fault

Abstract

The invention discloses a power distribution network reliability prediction method based on the influence of weather on power distribution network faults, and the method comprises the steps: defining an acquisition prediction window on a time axis, and obtaining a fault type and a fault level of a to-be-predicted time point; updating the weather vector at the current moment in real time in the acquisition prediction window; updating the time point weight of the current moment in real time; according to the weather vector and the time point weight which are updated in real time, the fault type and the fault level of the to-be-predicted time point are updated; determining the reliability of the power distribution network at the to-be-predicted time point according to the fault type and the fault level of the to-be-predicted time point; matching a predefined prevention strategy according to the reliability of the power distribution network at the to-be-predicted time point; according to the method, the combination of the time weight and the weather vector is introduced, and the influence of the time proximity and the meteorological change on the power distribution network fault is comprehensively considered, so that the fault type and grade prediction can be dynamically adjusted on the basis of the real-time weather condition and the time factor, and the prediction precision and timeliness are improved.

Description

Technical Field

[0001] This invention relates to distribution network reliability prediction, specifically a distribution network reliability prediction method based on the impact of weather on distribution network faults. Background Technology

[0002] With the development of smart grid and distribution network technologies, the reliability assessment of distribution networks has received increasing attention, especially under extreme weather conditions, where weather factors have a significant impact on the stability and failure rate of distribution networks. Traditional distribution network reliability prediction methods are usually based on historical fault data and load conditions, failing to fully consider the impact of meteorological factors on fault types and levels. Moreover, existing technologies suffer from numerous problems with single-dimensional predictions, lacking multi-dimensional prediction models that comprehensively consider fault types and fault labels. Patent publication number CN108596449A discloses a distribution network reliability prediction method that considers the impact of weather on the probability of distribution network failures, capable of predicting distribution network reliability for different weather variables.

[0003] However, despite incorporating weather influence factors, the lack of dynamic time considerations and the simplistic output of fault prediction result in low accuracy in calculating the reliability of the distribution network. Therefore, this embodiment proposes a reliability prediction method based on the impact of weather on distribution network faults, combining time weights and weather factors to achieve more accurate distribution network fault prediction and reliability assessment. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method for predicting the reliability of distribution networks based on the impact of weather on distribution network faults. This method solves the technical problems mentioned in the background by combining time weights and weather factors.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A method for predicting the reliability of distribution networks based on the impact of weather on distribution network faults includes the following steps: S1. Define the acquisition and prediction window on the time axis to obtain the fault type and fault level at the time point to be predicted; S2. Within the acquisition and prediction window, update the current time in real time. Weather vector; S3, Real-time update of the current time Time point weighting; S4. Based on the real-time updated weather vector and time point weights, update the fault type and fault level of the time point to be predicted; S5. Determine the reliability of the distribution network at the time point to be predicted based on the fault type and fault level. S6. Match predefined prevention strategies based on the distribution network reliability at the time point to be predicted.

[0006] In some embodiments, obtaining the fault type and fault level at the time point to be predicted includes: S1-1, Define the left and right endpoints of the acquisition prediction window; Wherein, the left endpoint is the acquisition time point that is always slidably aligned with the current moment, and the right endpoint is the time point to be predicted that is fixed on the time axis; S1-2. In the acquisition and prediction window, obtain the initial time. Weather vector; S1-3, the initial time The weather vector is input into the fine-tuned fault prediction model, which outputs the fault type and fault level at the time point to be predicted.

[0007] In some of these embodiments, the current time is updated in real time. The weather vectors include: S2-1. Obtain real-time updated meteorological characteristics; S2-2. Perform data cleaning and preprocessing on the real-time updated meteorological features to obtain standardized meteorological features; S2-3. Perform feature stitching on all standardized meteorological features to generate the current time. Weather vector.

[0008] In some of these embodiments, the current time is updated in real time. The time point weights include: S3-1, According to the initial time Calculate the total prediction duration based on the time difference with the time point to be predicted; S3-2. Calculate the time difference between the current time and the time point to be predicted, based on the real-time updated current time. S3-3. Define the weighted ratio of the time difference at the current moment to the total prediction duration as the time weight at the current moment; The definition expression for the time weight at the current moment is: in, The time weight represents the current moment and is used to quantify the impact of the current moment on the prediction result. This represents the time difference between the current moment and the point in time to be predicted. This represents the total prediction duration, which is the time from the initial moment to the point in time to be predicted.

[0009] In some of these embodiments, updating the fault type and fault level at the predicted time point includes: S4-1, Based on the current time Time point weights and the current time The weather vector is used to generate the current time. The weighted weather vector; S4-2, For all current moments The weighted weather vectors are averaged to generate the input vector for the time to be predicted; S4-3. Input the input vector of the time to be predicted into the fine-tuned fault prediction model to update the fault type and fault level of the time point to be predicted.

[0010] In some of these embodiments, determining the distribution network reliability at the point in time to be predicted includes: S5-1. Determine the fault influencing factors based on the fault type and fault level at the time point to be predicted; The expression for the fault impact factor is: ; in, Indicates the failure impact factor. This represents the influence factor corresponding to the combination of fault type i and fault level j.

[0011] S5-2. Define the meteorological weight of each meteorological feature in the weather vector; S5-3. Based on the meteorological weight, sum all meteorological features by weight to generate the weather influence factor of the weather vector; The expression for the weather influencing factor is: ; in, Indicates weather influencing factors, Represents the i-th meteorological feature. This represents the meteorological weight of the i-th meteorological feature.

[0012] S5-4. Based on the fault impact factor, weather impact factor, and time weight, the reliability of the distribution network is obtained; The expression for the reliability of the distribution network is: ; in, Indicates the reliability of the power distribution network. The weights of the failure impact factors, The weights of weather-related factors.

[0013] In some embodiments, the modeling steps of the fault prediction model include: A1. Obtain a weather fault training set; wherein, the weather fault training set includes M*N weather fault training samples; A2. Initialize the general neural network model as a dual-channel model; the initialization parameters include: minimum first objective loss and minimum second objective loss; A3. Extract the first batch of weather fault training samples from the weather fault training set as the input to the dual-channel model; A4. The fault prediction model is obtained by supervising the training of the first batch of weather fault training samples through iterative training.

[0014] In some embodiments, obtaining a weather fault training set includes: A1-1. In the historical fault database, find M different distribution network fault types; where each fault type has at least K fault levels, and the fault levels of different fault types are the same; A1-2. For each distribution cabinet fault of the same fault type and the same fault level, select and mark its most recent occurrence time. A1-3. Based on the most recent occurrence time point, match the weather conditions at the most recent time point and standardize them into a weather input vector; A1-4. Based on the weather input vector at the most recent time point, slide the time length of the acquisition and prediction window forward to match the corresponding fault type and its fault level. A1-5. Define the fault type and its fault level as dual-target labels; A1-6. Collect M*N bi-target labels and combine them with the corresponding weather input vectors to construct a weather fault training set.

[0015] In some of these embodiments, the general neural network model is initialized as a two-channel model, including: A2-1. Define a dual-channel output node in the output layer of a general neural network model; including: a first channel node and a second node; wherein, the first channel node outputs the fault prediction type; and the second channel node outputs the fault level; A2-2. Define the target loss of the first channel node as the multi-class cross-entropy loss and the target loss of the second channel node as the mean squared error loss. A2-3. Preset the minimum values ​​of multi-class cross-entropy loss and mean squared error respectively, and define them as the minimum first target loss of the first channel node and the minimum second target loss of the second channel node respectively.

[0016] In some embodiments, the fault prediction model is obtained by supervising the first batch of weather fault training samples through iterative training, including: A4-1. Based on the first batch of weather fault training samples, output the fault prediction type at the first channel node and the fault prediction level at the second channel node. A4-2. Calculate the multi-class cross-entropy loss of the first channel node and the mean square error loss of the second channel node in the first batch, respectively. A4-3. Calculate the gradients of the first channel node relative to the current model parameters based on the multi-class cross-entropy loss and the mean squared error loss of the second channel node. A4-4. Update the model parameters according to the calculated gradient using the gradient descent method; A4-5. Extract the next batch of weather fault training samples and use them as input to the dual-channel model after parameter update. Repeat A4-1 to A4-4 until the minimum first target loss and minimum second target loss are output.

[0017] A4-6. When the minimum first target loss and the minimum second target loss are output, the corresponding parameters are updated and the dual-channel model is defined as the fault prediction model.

[0018] This invention provides a method for predicting the reliability of distribution networks based on the impact of weather on distribution network faults, which has the following beneficial effects: This invention combines time weights with weather vectors to comprehensively consider the impact of time proximity and meteorological changes on power distribution network faults. This allows fault type and level predictions to be dynamically adjusted based on real-time weather conditions and time factors. Especially under severe weather conditions, the time weights make the predictions more sensitive to real-time weather changes close to the time to be predicted, thereby effectively reducing prediction errors and improving the accuracy and timeliness of predictions.

[0019] Furthermore, this invention uses a dual-channel neural network model to assign the prediction tasks of fault types and fault levels in the distribution network to different channels for processing, and optimizes the prediction results of each task respectively, thereby achieving dual prediction of fault types and fault levels in the distribution network and ensuring comprehensive prediction of distribution network faults. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating the distribution network reliability prediction method based on the impact of weather on distribution network faults according to the present invention. Figure 2 This is a schematic diagram illustrating the definition process of the time weight described in this invention; Figure 3 This is a schematic diagram of the modeling process for the fault prediction model described in this invention; Figure 4 This is a schematic diagram of the initialization of the dual-channel model described in this invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Example 1: See Figure 1 This invention provides a method for predicting the reliability of a distribution network based on the impact of weather on distribution network faults, comprising the following steps: Step S1: Define the acquisition and prediction window on the time axis to obtain the fault type and fault level at the time point to be predicted; Step S1 specifically includes: S1-1, Define the left and right endpoints of the acquisition prediction window; Wherein, the left endpoint is the acquisition time point that is always slidably aligned with the current moment, and the right endpoint is the time point to be predicted that is fixed on the time axis; S1-2. In the acquisition and prediction window, obtain the initial time. Weather vector; S1-3, the initial time The weather vector is input into the fine-tuned fault prediction model, which outputs the fault type and fault level at the time point to be predicted.

[0023] This embodiment uses a defined acquisition and prediction window to obtain the fault type and fault level at the predicted time point. First, the left endpoint of the acquisition and prediction window is always aligned with the current time and updated in real time via a sliding mechanism. The right endpoint is a fixed time point, i.e., the time point to be predicted. Within this window, the weather vector at the initial time is first acquired, representing the meteorological data at the current time. Then, this weather vector is input into a fine-tuned fault prediction model. The model uses learned historical data to output the fault type and fault level at the predicted time point, thus achieving the prediction of distribution network fault types and levels based on real-time weather data.

[0024] See Figure 1 The method further includes: Step S2: Update the current time in real time within the acquisition and prediction window. The weather vector; step S2 specifically includes: S2-1. Obtain real-time updated meteorological characteristics; The meteorological characteristics include basic meteorological features such as temperature, humidity, wind speed, air pressure, and precipitation. S2-2. Perform data cleaning and preprocessing on the real-time updated meteorological features to obtain standardized meteorological features; The standardization features include interpolating missing basic meteorological features, identifying and removing abnormal basic meteorological features, and standardizing weather data from different sources.

[0025] S2-3. Perform feature stitching on all standardized meteorological features to generate the current time. Weather vector.

[0026] Feature concatenation refers to combining standardized meteorological features (such as temperature, humidity, and wind speed) into a complete vector in sequence or through a pre-defined algorithm. Feature concatenation can connect feature values ​​into a long vector according to feature type order; or it can use methods such as weighted averaging or PCA dimensionality reduction for combination.

[0027] This embodiment achieves power distribution network fault prediction by updating the weather vector at the current moment in real time. First, real-time updated meteorological features are acquired, including basic meteorological characteristics such as temperature, humidity, wind speed, air pressure, and precipitation. These meteorological features undergo data cleaning and preprocessing to ensure data integrity and accuracy. Specific processing methods include imputing missing meteorological features, identifying and removing outliers, and standardizing weather data from different sources to ensure consistency. Next, the standardized meteorological features are concatenated to generate the weather vector at the current moment. Feature concatenation can be achieved by sequentially connecting feature values ​​or by combining them using methods such as weighted averaging and PCA dimensionality reduction, ensuring effective integration of real-time weather data and providing accurate meteorological data support for subsequent power distribution network fault prediction.

[0028] See Figures 1 to 2 The method further includes: Step S3: Update the current time in real time. Time point weights; step S3 specifically includes: S3-1, According to the initial time Calculate the total prediction duration based on the time difference with the time point to be predicted; S3-2. Calculate the time difference between the current time and the time point to be predicted, based on the real-time updated current time. S3-3. Define the weighted ratio of the time difference at the current moment to the total prediction duration as the time weight at the current moment; The definition expression for the time weight at the current moment is: in, The time weight represents the current moment and is used to quantify the impact of the current moment on the prediction result. This represents the time difference between the current moment and the point in time to be predicted. The total prediction duration represents the time interval from the initial time to the predicted time point. The time weights are calculated using a logarithmic function to ensure that the weight of the current time point does not approach zero as the distance to the predicted time point increases, while also preventing excessively high weights for closer time points. The use of the logarithmic function makes the increase in weight values ​​smoother, and the rate of increase slows down as the time difference increases.

[0029] The weight of the current moment is calculated based on the ratio of the time difference between the current moment and the time point to be predicted to the total prediction duration. The time weight reflects the relative importance of the current moment from the time point to be predicted; generally, the larger the weight value, the greater the influence of the current moment.

[0030] This embodiment achieves dynamic adjustment of distribution network fault prediction results by updating the time point weight in real time. The calculation method of the time weight ensures that the influence of the current moment on the predicted time point can be reasonably quantified according to the time proximity. By calculating the time difference between the current moment and the predicted time point, and combining it with the total prediction duration, the time weight can effectively reflect the relative importance of the current moment to the prediction result.

[0031] Using a logarithmic function to define time weights ensures that the rate of increase in time weights gradually slows down as the time difference increases. This prevents the influence of distant time points on the prediction results from being excessively weakened, while also preventing excessively large weights for very recent time points, which could lead to unreasonable impacts. This method can smoothly handle the influence of time on distribution network fault prediction, making distribution network reliability prediction more accurate and stable. Especially when dealing with real-time data close to the prediction time, it can reasonably amplify the impact of the current moment on the prediction results.

[0032] See Figure 1 The method further includes: Step S4: Update the fault type and fault level of the time point to be predicted based on the real-time updated weather vector and time point weight; Step S4 specifically includes: S4-1, Based on the current time Time point weights and the current time The weather vector is used to generate the current time. The weighted weather vector; S4-2, For all current moments The weighted weather vectors are averaged to generate the input vector for the time to be predicted; S4-3. Input the input vector of the time to be predicted into the fine-tuned fault prediction model to update the fault type and fault level of the time point to be predicted.

[0033] This embodiment achieves accurate updates to the fault type and fault level at the predicted time point by combining the weight of the current moment with the weather vector. First, the generated weighted weather vector reasonably reflects the impact of current weather conditions on fault prediction. Since the time point weight can adjust the importance of time proximity to the prediction result, the current weather conditions are weighted according to their proximity to the predicted time, thus giving weather conditions closer to the prediction time a greater predictive impact.

[0034] By averaging all weighted weather vectors to generate the input vector for the time to be predicted, the system ensures that all real-time changing weather conditions are fully integrated from data collected at multiple times, thereby improving the stability and accuracy of the prediction. By inputting this input vector into the fine-tuned fault prediction model, the fault type and severity at the predicted time point can be updated in real time.

[0035] See Figure 1 The method further includes: step S5, determining the reliability of the distribution network at the time point to be predicted based on the fault type and fault level; step S5 specifically includes: S5-1. Determine the fault influencing factors based on the fault type and fault level at the time point to be predicted; The expression for the fault impact factor is: ; in, Indicates the failure impact factor. This represents the influence factor corresponding to the combination of fault type i and fault level j.

[0036] Specifically, the fault impact factor is calculated based on the fault type and fault level. In this embodiment, it is assumed that the fault type and level are determined using a lookup table or mapping matrix A to determine the degree of fault impact. Therefore, each combination of fault type and fault level has a corresponding impact factor.

[0037] S5-2. Define the meteorological weight of each meteorological feature in the weather vector; wherein the meteorological weight of each meteorological feature has a different impact on the reliability of the distribution network.

[0038] S5-3. Based on the meteorological weights, sum all meteorological characteristics by weight to generate the weather influence factor of the weather vector; The expression for the weather influencing factor is: ; in, Indicates weather influencing factors, Represents the i-th meteorological feature. This represents the meteorological weight of the i-th meteorological feature. By multiplying each meteorological feature by its corresponding meteorological weight and then summing the weighted sums of all meteorological features, we can obtain the weather impact factor, which is used to represent the impact of weather conditions on the reliability of the power distribution network.

[0039] S5-4. Based on the fault impact factor, weather impact factor, and time weight, the reliability of the distribution network is obtained; The expression for the reliability of the distribution network is: ; in, Indicates the reliability of the power distribution network. The weights of the failure impact factors, The weights of weather-related factors.

[0040] This embodiment combines fault impact factors, weather impact factors, and the current time relative to the time point to be predicted, i.e. the time point when the fault may occur, to finally generate the distribution network reliability corresponding to the possible occurrence of the fault.

[0041] Specifically, the predicted fault type and fault level are based on the predicted time point. Distribution network faults may or may not occur. The probability of a fault occurring gradually increases as the predicted time point approaches, especially under the influence of meteorological disasters.

[0042] Severe weather events (such as strong winds, blizzards, and freezing temperatures) typically occur continuously rather than suddenly. For example, wind speeds may gradually increase, and temperatures may gradually decrease, leading to snow and ice accumulation; these are all examples of the gradual accumulation process of meteorological disasters. The duration and cumulative effect of meteorological disasters gradually increase their impact on the power distribution network, leading to a continuous decline in the network's reliability.

[0043] The reliability of a distribution network changes over time and with the accumulation of weather conditions. Because weather conditions are constantly changing, their impact on the distribution network gradually increases. When weather conditions worsen, the reliability of the distribution network continuously declines, especially under the long-term cumulative effects of meteorological disasters. By weighting and combining fault impact factors, weather impact factors, and time weights, the reliability of the distribution network can be dynamically adjusted and predicted.

[0044] In summary, the reliability of the distribution network gradually decreases with the accumulation of meteorological disasters, especially as the risk of failure increases as the predicted time approaches. The gradual impact of each meteorological feature and the cumulative effect over time work together to ultimately determine the reliability of the distribution network.

[0045] Step S6: Match the distribution network reliability at the predicted time point with a predefined prevention strategy. The predefined prevention strategy means the pre-prepared fault contingency plan based on different distribution network reliability levels.

[0046] In summary, this embodiment combines real-time weather conditions, time weights, and a fault prediction model to realize a reliability prediction method based on the impact of weather on distribution network faults. In this embodiment, a data acquisition and prediction window is first defined, within which the weather vector and time point weights are updated in real time, allowing the prediction results to be dynamically adjusted based on current meteorological data and time factors. The real-time updated weather vector and time weights work together to update the fault type and fault level at the predicted time point, thereby ensuring the real-time nature and accuracy of distribution network fault prediction.

[0047] Furthermore, based on the predicted fault type and fault level, the reliability of the distribution network at the predicted time point is calculated, and predefined prevention strategies are automatically matched according to this reliability. This method can timely assess the reliability of the distribution network based on real-time changing weather conditions and time factors, and provide effective preventive measures, thereby reducing the risk of distribution network faults and improving the stability and security of the power system. Through this multi-dimensional prediction that comprehensively considers meteorological conditions, time factors, and fault types, distribution network faults can be predicted more accurately, thus enabling targeted preventive measures to be taken.

[0048] Example 2: See Figures 2 to 4 The technical solution of this embodiment 2 differs from that of embodiment 1 in that it discloses the modeling steps of the fault prediction model described in embodiment 1, the modeling steps including: A1. Obtain a weather fault training set; wherein, the weather fault training set includes M*N weather fault training samples; A2. Initialize the general neural network model as a dual-channel model; the initialization parameters include: minimum first objective loss and minimum second objective loss; A3. Extract the first batch of weather fault training samples from the weather fault training set as the input to the dual-channel model; A4. The fault prediction model is obtained by supervising the training of the first batch of weather fault training samples through iterative training.

[0049] This embodiment constructs a fault prediction model based on weather data, utilizing a weather fault training set and a neural network model to predict the fault type and level of a power distribution network. First, a weather fault training set containing M*N weather fault training samples is obtained, covering different weather conditions and corresponding fault data. Next, the general neural network model is initialized with a dual-channel structure to simultaneously handle classification and linear regression problems, ensuring the model can learn the complex relationship between weather conditions and fault types. During initialization, minimum first objective loss and minimum second objective loss are set to optimize multiple prediction objectives simultaneously. Then, the first batch of weather fault training samples is extracted as model input for training. Within the framework of supervised learning, the parameters of the neural network are gradually optimized through multiple iterations of training, enabling it to accurately predict the fault type and level at the predicted time point.

[0050] Specifically, step A1 includes: A1-1. In the historical fault database, find M different distribution network fault types; where each fault type has at least K fault levels, and the fault levels of different fault types are the same; A1-2. For each distribution cabinet fault of the same fault type and the same fault level, select and mark its most recent occurrence time. A1-3. Based on the most recent occurrence time point, match the weather conditions at the most recent time point and standardize them into a weather input vector; A1-4. Based on the weather input vector at the most recent time point, slide the time length of the acquisition and prediction window forward to match the corresponding fault type and its fault level. A1-5. Define the fault type and its fault level as dual-target labels; A1-6. Collect M*N bi-target labels and combine them with the corresponding weather input vectors to construct a weather fault training set.

[0051] This embodiment constructs a training dataset for predicting distribution network faults based on weather conditions by acquiring a weather-based fault training set. First, M different distribution network fault types are selected from the historical fault database. Each fault type contains at least K fault levels, and all fault types have the same fault level. This ensures that the training set data covers different types and levels of faults, thereby enhancing the model's generalization ability.

[0052] Next, for each distribution network fault of the same fault type and fault level, the most recent time point of occurrence is selected and marked so that the generated weather data can reflect the actual meteorological conditions at the time of the fault as much as possible.

[0053] Based on these time points, match the most recent weather conditions and standardize them into a weather input vector.

[0054] Subsequently, by sliding the acquisition prediction window time length, the corresponding fault type and level are matched with the weather input vector, and training data is generated according to the time series.

[0055] Ultimately, the fault type and its fault level were defined as bi-target labels. By collecting M×N bi-target labels and corresponding weather input vectors, a complete weather fault training set was constructed, providing high-quality input data for the subsequent training of the fault prediction model and ensuring that the model can make predictions when facing power distribution network faults under different weather conditions.

[0056] Step A2 includes: A2-1. Define a dual-channel output node in the output layer of a general neural network model; including: a first channel node and a second node; wherein, the first channel node outputs the fault prediction type; and the second channel node outputs the fault level; A2-2. Define the target loss of the first channel node as the multi-class cross-entropy loss and the target loss of the second channel node as the mean squared error loss. A2-3. Preset the minimum values ​​of multi-class cross-entropy loss and mean squared error respectively, and define them as the minimum first target loss of the first channel node and the minimum second target loss of the second channel node respectively.

[0057] This embodiment initializes the general neural network model as a dual-channel structure, enabling it to simultaneously process and output the type and level of distribution network faults. The first channel focuses on the classification of fault types, while the second channel predicts the level of faults. The combination of the two output nodes ensures the multi-dimensionality of fault prediction.

[0058] In terms of optimizing training results, two different loss functions, multi-class cross-entropy loss and mean squared error loss, were used, respectively targeting the classification problem of fault types and the regression problem of fault levels. This allowed each task to be specifically optimized, avoiding mutual interference between loss functions, thereby improving training efficiency and model accuracy.

[0059] In summary, this embodiment, through its dual-channel structure, differentiated loss function, and precise setting of optimization objectives, enables more accurate performance in distribution network fault prediction in terms of fault type and level, effectively improving the performance of distribution network reliability prediction.

[0060] Specifically, step A4 includes: A4-1. Based on the first batch of weather fault training samples, output the fault prediction type at the first channel node and the fault prediction level at the second channel node. A4-2. Calculate the multi-class cross-entropy loss of the first channel node and the mean square error loss of the second channel node in the first batch, respectively. A4-3. Calculate the gradients of the first channel node relative to the current model parameters based on the multi-class cross-entropy loss and the mean squared error loss of the second channel node. A4-4. Update the model parameters according to the calculated gradient using the gradient descent method; A4-5. Extract the next batch of weather fault training samples and use them as input to the dual-channel model after parameter update. Repeat A4-1 to A4-4 until the minimum first target loss and minimum second target loss are output.

[0061] A4-6. When the minimum first target loss and the minimum second target loss are output, the corresponding parameters are updated and the dual-channel model is defined as the fault prediction model.

[0062] This embodiment optimizes and trains the fault prediction model through supervised iterative training. First, during the training process, the fault prediction type and fault level are output at the first and second nodes of the dual-channel model using the first batch of weather fault training samples, allowing it to independently optimize the classification task of fault type and the regression task of fault level.

[0063] Next, the multi-class cross-entropy loss (for fault type) and mean squared error loss (for fault level) are calculated and optimized to ensure that the task for each channel is specifically optimized. After calculating the loss, gradient descent is used to update the model parameters, ensuring that model performance is gradually optimized through iterative training. This process is repeated continuously, with the next batch of training samples being drawn, and the model parameters are gradually optimized until the minimum first objective loss and the minimum second objective loss are reached. Finally, the trained dual-channel model accurately reflects the prediction of fault type and fault level, thereby improving the performance and accuracy of distribution network reliability prediction.

[0064] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer program are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means.

[0065] The computer-readable storage medium can be any available medium that a computer can access, or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g.,...), etc. DVD ( ), or semiconductor media. Semiconductor media can be solid-state drives (SSDs).

[0066] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division of a waterway underwater topography change analysis system and method. In actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, indirect coupling or communication connection between devices or units, and may be electrical, mechanical, or other forms.

[0067] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A power distribution network reliability prediction method based on weather influence on power distribution network failure, characterized by, The application relates to a power distribution network reliability prediction method, which comprises the following steps: S1, defining a collection prediction window on a time axis to obtain a fault type and a fault level of a to-be-predicted time point; S2, within the collection prediction window, real-time update of the current time weather vector; S3, updating the current time point in real time weight of the time point; S4, updating the fault type and the fault level of the to-be-predicted time point according to a real-time updated weather vector and a time point weight; S5, determining power distribution network reliability of the to-be-predicted time point according to the fault type and the fault level of the to-be-predicted time point; S6, matching a predefined prevention strategy according to the power distribution network reliability of the to-be-predicted time point.

2. The method for reliability forecasting of power distribution network based on weather influence on power distribution network faults according to claim 1, characterized in that, The fault type and the fault level of the to-be-predicted time point are obtained, and the method comprises the following steps: S1-1, defining a left end point and a right end point of the collection prediction window; The left end point is a collection time point that is always slidingly aligned with a current time, and the right end point is a to-be-predicted time point fixed on a time axis; S1-2, in the collection prediction window, obtaining a weather vector at an initial time ; S1-3, inputting the weather vector at the initial moment to the fine-tuned failure prediction model, outputting the failure type and its failure level at the time point to be predicted.

3. The method for reliability forecasting of distribution network based on weather influence on distribution network fault according to claim 1, characterized in that, updating the weather vector in real time of the current time, comprising: S2-1, obtaining real-time updated meteorological features; S2-2, performing data cleaning and preprocessing on the real-time updated meteorological features to obtain standardized meteorological features; S2-3, splice all normalized weather features to generate the weather vector of the current time point. of the current time point.

4. The method for reliability forecasting of power distribution network based on weather impact on power distribution network faults according to claim 2, characterized in that, Real-time updates of the current time The time point weights include: S3-1、according to the initial time point a time difference from the time point to be predicted, and calculate the total prediction duration. S3-2, calculating a time difference between a current time and the to-be-predicted time point according to the real-time updated current time; S3-3, defining a time weight of the current time as a weighted ratio of the time difference of the current time and a total prediction duration; The definition expression of the time weight of the current time is as follows: wherein, denotes a time weight of a current time, used to quantify the influence of the current time on the prediction result; denotes a time difference between the current time and the time point to be predicted, denotes a total prediction duration, and denotes a time from an initial time to the time point to be predicted.

5. The method for reliability forecasting of distribution network based on weather influence on distribution network fault according to claim 1, characterized in that, The fault type and the fault level of the to-be-predicted time point are updated, and the method comprises the following steps: S4-1, Based on the current time Time point weights and the current time The weather vector is used to generate the current time. The weighted weather vector; S4-2, For all current moments The weighted weather vectors are averaged to generate the input vector for the time to be predicted; S4-3, inputting an input vector of the to-be-predicted time into a fine-tuned fault prediction model to update the fault type and the fault level of the to-be-predicted time point.

6. The method for reliability forecasting of distribution network based on weather influence on distribution network fault according to claim 1, characterized in that, The power distribution network reliability of the to-be-predicted time point is determined, and the method comprises the following steps: S5-1, determining a fault influence factor according to the fault type and the fault level of the to-be-predicted time point; The expression of the fault influence factor is as follows: ； wherein, represents a failure impact factor, represents an impact factor corresponding to the combination of failure type i and failure level j; S5-2, defining a meteorological weight of each meteorological feature in the weather vector; S5-3, generating a weather influence factor of the weather vector by weighted summation of all meteorological features according to the meteorological weight; The expression of the weather influence factor is as follows: ； wherein, represents a weather impact factor, represents the i-th weather feature, represents a weather weight for the i-th weather feature; S5-4, obtaining the power distribution network reliability according to the fault influence factor, the weather influence factor and the time weight; The expression of the power distribution network reliability is as follows: ； wherein, represents the reliability of the power distribution network, is the weight of the fault impact factor, is the weight of the weather impact factor.

7. The method for power distribution network reliability forecasting based on weather impact on power distribution network faults according to claim 5, characterized in that, The modeling steps of the fault prediction model comprise the following steps: A1, obtaining a weather fault training set; wherein the weather fault training set comprises M*N weather fault training samples; A2, initializing a general neural network model into a double-channel model; wherein the initialization parameters comprise a minimum first target loss and a minimum second target loss; A3, extracting a first batch of weather fault training samples in the weather fault training set as inputs of the double-channel model; A4, obtaining the fault prediction model after the first batch of weather fault training samples are trained through supervised iteration.

8. The method for reliability forecasting of distribution network based on weather influence on distribution network fault according to claim 5, characterized in that, The weather fault training set is obtained, and the method comprises the following steps: A1-1, searching for M different power distribution network fault types in a historical fault database; wherein each fault type has at least K or more fault levels, and the fault levels of different fault types are the same; A1-2, for each power distribution cabinet fault of the same fault type and the same fault level, selecting and marking a latest occurrence time point thereof; A1-3, according to the latest time point, match the weather conditions of the latest time point, and standardize it into a weather input vector; A1-4, according to the weather input vector of the latest time point, slide the time length of the collection prediction window forward, match the corresponding fault type and its fault level; A1-5, define the fault type and its fault level as a double-target label; A1-6, collect M*N double-target labels, combine the corresponding weather input vector, and construct a weather fault training set.

9. The method for reliability forecasting of distribution network based on weather influence on distribution network faults according to claim 5, characterized in that, Initialize the general neural network model as a double-channel model, including: A2-1, define double-channel output nodes in the output layer of the general neural network model; including: first channel node and second channel node; wherein the first channel node outputs the fault prediction type; the second channel node outputs the fault level; A2-2, define the target loss of the first channel node as the multi-classification cross-entropy loss, and the target loss of the second channel node as the mean square error loss; A2-3, respectively preset the minimum values of the multi-classification cross-entropy loss and the mean square error, and respectively define them as the minimum first target loss of the first channel node and the minimum second target loss of the second channel node.

10. The method for reliability forecasting of distribution network based on weather influence on distribution network faults according to claim 5, characterized in that, After the first batch of weather fault training samples are trained by supervised iteration, the fault prediction model is obtained, including: A4-1, according to the first batch of weather fault training samples, output the fault prediction type in the first channel output node, and output the fault prediction level in the second channel node; A4-2, calculate the multi-classification cross-entropy loss of the first channel node and the mean square error loss of the second channel node in the first batch respectively; A4-3, according to the multi-classification cross-entropy loss of the first channel node and the mean square error loss of the second channel node, calculate the gradient of the current model parameters respectively; A4-4, according to the calculated gradient, update the model parameters according to the gradient descent method; A4-5, extract the next batch of weather fault training samples, take them as the input of the parameter updated double-channel model, and repeat A4-1 to A4-4 until the minimum first target loss and the minimum second target loss are outputted; A4-6, when the minimum first target loss and the minimum second target loss are outputted, define the corresponding parameter updated double-channel model as the fault prediction model.

Images