Substation flood prevention risk probability prediction method

Abstract

The invention discloses a transformer substation flood prevention risk probability prediction method, which comprises the following steps: firstly, collecting historical multi-dimensional flood prevention data related to a transformer substation, carrying out data preprocessing and flood prevention feature optimization on the historical data, and constructing a flood prevention risk probability prediction model by utilizing an improved DeepAR algorithm based on the optimized historical flood prevention features; during the period, an attention mechanism, a dynamic L2 regularization strategy and a multi-modal data fusion optimization prediction model are introduced; then collecting real-time data of the multi-dimensional flood prevention data of the transformer substation, performing data preprocessing and flood prevention feature optimization on the real-time data, performing flood prevention risk probability prediction on the transformer substation by using the optimized flood prevention risk probability prediction model based on the optimized real-time flood prevention features, and outputting a flood prevention risk probability value and a prediction result; according to the method, data in multiple aspects can be more efficiently utilized, and the flood prevention risk prediction probability of the transformer substation can be accurately predicted through the improved DeepAR algorithm and the optimized prediction model.

Description

Technical Field

[0001] The present invention belongs to the technical field of flood control risk prediction for power systems, and in particular relates to a method for predicting flood control risk probability for a substation. Background Art

[0002] In recent years, extreme weather such as heavy rainfall, typhoons, and severe convection have occurred frequently. Power safety accidents caused by natural disasters such as heavy rainfall have brought great threats and challenges to the normal operation of substation equipment. In order to improve the flood control capabilities of substations, avoid untimely flood control response measures that bring unbearable strong impacts to substation equipment, effectively prevent and curb various types of power production safety accidents, and effectively prevent the occurrence of large-scale power outages in the power grid, it is currently an important means to ensure the safe and stable operation of the power grid to conduct risk forecasting and assessment of various natural disasters in advance, and to promptly investigate and eliminate hidden dangers based on the forecast and assessment results.

[0003] At present, flood prevention work for substations is mostly based on extreme weather meteorological warnings issued by the Meteorological Bureau, and then based on the empirical judgment of historical data and the preliminary application of simple models, it is completed by relying on manual empirical judgment and manual fixed-point and squatting. There are problems such as manual empirical judgment is prone to loopholes, low efficiency and high work intensity of manual fixed-point and squatting, insufficient use of big data information, shallow historical data mining, limited deviations caused by simple models, and lack of dynamic adaptability. In order to solve the above problems, it is necessary to develop a substation flood prevention risk probability prediction method. Summary of the invention

[0004] The purpose of the present invention is to overcome the shortcomings of the prior art and to provide a method for predicting the probability of flood control risk in a substation, which can more efficiently utilize data from multiple aspects, accurately predict the probability of flood control risk prediction in a substation through an improved DeepAR algorithm and an optimized prediction model, improve the safety and reliability of the substation in flood control, provide decision-making support for the flood control work of the substation, thereby improving the comprehensive processing capability of the flood control work of the substation, and further ensuring the ability of the substation to maintain normal and safe operation as much as possible under extreme weather conditions.

[0005] The object of the present invention is achieved in this way: a method for predicting the probability of flood control risk in a substation comprises the following steps: Step a, collecting historical multi-dimensional flood control data related to the substation; Step b, preprocessing the historical multi-dimensional flood control data and optimizing the flood control features; Step c: Based on the selected historical flood control characteristics, a flood control risk probability prediction model is constructed using the improved DeepAR algorithm; Step d, introduce the attention mechanism to optimize the prediction model; Step e, introducing a dynamic L2 regularization strategy to optimize the prediction model; Step f, introducing a multimodal data fusion optimization prediction model; Step g, collecting real-time data of multi-dimensional flood control data of the substation; Step h, preprocessing the real-time multi-dimensional flood control data and optimizing flood control features; Step i: Based on the selected real-time flood control features, the optimized flood control risk probability prediction model is used to predict the flood control risk probability of the substation, and the flood control risk probability value and prediction result are output.

[0006] In step a and step g, the multidimensional flood control data includes the equipment waterproof information, geographical environment information, hydrological characteristic information, micro-meteorological information and historical flood information of the substation, wherein the equipment waterproof information includes the equipment waterproof level and the equipment installation height data, wherein the geographical environment information includes the collection well volume, the collection well water pump discharge, the surrounding terrain data, the terrain height data and the drainage path data, wherein the hydrological characteristic information includes the average water level, the highest water level, the lowest water level, the runoff and the flow rate data, wherein the micro-meteorological information includes the rainfall, rainfall duration and rainfall intensity data in the area where the substation is located, and wherein the historical flood information includes the flood occurrence time period and flood level data.

[0007] In the step i, the prediction results include the flood inundation range, flood depth distribution, and flood retention time.

[0008] In the step b, data preprocessing is one or more of multi-source data integration, data time synchronization, data cleaning, data standardization and normalization, and data reconstruction.

[0009] In step b, the flood control features preferably include automated feature extraction and selection, and new feature construction.

[0010] In step d, introducing the attention mechanism includes two sub-steps: selecting the type of attention mechanism and embedding the attention mechanism.

[0011] In the step e, the introduction of the dynamic L2 regularization strategy includes three sub-steps: selecting the key features of L2 regularization, customizing the function, and adding the regularization term in the custom function.

[0012] In the step f, three sub-steps of multimodal data fusion are introduced, including selecting multimodal data type, data synchronization and alignment, and fusion at the model level.

[0013] It also includes step j. During use, as time accumulates, the prediction model is continuously iterated and optimized through historical multi-dimensional flood control data, attention mechanism, dynamic L2 regularization strategy and multimodal data fusion.

[0014] Due to the adoption of the above technical solution, the beneficial effects of the present invention are: The present invention is based on historical multi-dimensional flood control data and real-time multi-dimensional flood control data, adopts data preprocessing and flood control feature optimization, and uses the improved DeepAR algorithm to construct a flood control risk probability prediction model from multiple dimensional data. At the same time, it introduces the attention mechanism, dynamic L2 regularization strategy and multi-mode data fusion technology, which can more efficiently utilize data from multiple aspects, accurately predict the substation flood control risk prediction probability through the improved DeepAR algorithm and optimized prediction model, improve the safety, reliability and real-time performance of the substation in flood control, and provide more powerful decision-making support for the flood control work of the substation, thereby improving the comprehensive processing capability of the substation flood control work, and then ensuring the substation's ability to maintain normal and safe operation as much as possible under extreme weather conditions; The present invention adopts continuous iteration and optimization of the prediction model. With the accumulation of big data and the development of power grid technology, the prediction model can have stronger prediction ability and adaptability. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a schematic flow chart of the working steps of the method of the present invention. DETAILED DESCRIPTION

[0016] The technical solution of the present invention is further specifically described below through embodiments and in conjunction with the accompanying drawings.

[0017] like Figure 1 As shown, the present invention provides a method for predicting the probability of flood control risk of a substation, which is characterized by comprising the following steps: Step a, collecting historical multi-dimensional flood control data related to the substation; Step b, preprocessing the historical multi-dimensional flood control data and optimizing the flood control features; Step c: Based on the selected historical flood control characteristics, a flood control risk probability prediction model is constructed using the improved DeepAR algorithm; Step d, introduce the attention mechanism to optimize the prediction model and enhance the prediction model's ability to mine the time series characteristics of multi-dimensional flood control dynamic meteorological data; Step e: Introduce dynamic L2 regularization strategy to optimize the prediction model, enhance the adaptability of the prediction model to multi-dimensional flood control data, and improve the generalization ability of the prediction model; Step f, introducing a multimodal data fusion optimization prediction model; Step g, collecting real-time data of multi-dimensional flood control data of the substation; Step h, preprocessing the real-time multi-dimensional flood control data and optimizing flood control features; Step i, based on the selected real-time flood control features, using the optimized flood control risk probability prediction model to predict the flood control risk probability of the substation, and outputting the flood control risk probability value and prediction results; Step j: During use, as time accumulates, the prediction model is continuously iterated and optimized through historical multi-dimensional flood control data, attention mechanism, dynamic L2 regularization strategy and multimodal data fusion, so that it can better adapt to new data characteristics and new flood control needs.

[0018] Among them, the multi-dimensional flood control data include the substation's equipment waterproof information, geographical environment information, hydrological characteristic information, micro-meteorological information and historical flood information, among which the equipment waterproof information includes the equipment waterproof level and equipment installation height data, among which the geographical environment information includes the collection well volume, the collection well water pump discharge, the surrounding terrain data, the terrain height data and the drainage path data, among which the hydrological characteristic information includes the average water level, the highest water level, the lowest water level, the runoff and the flow rate data, among which the micro-meteorological information includes the rainfall, rainfall duration and rainfall intensity data in the area where the substation is located, and among which the historical flood information includes the flood occurrence time period and flood level data.

[0019] Among them, the prediction results include flood inundation range, flood depth distribution, and flood retention time.

[0020] Among them, data preprocessing is one or more of multi-source data integration, data time synchronization, data cleaning, missing value processing, outlier detection, data standardization and normalization, and data reconstruction. Multi-source data integration is used to effectively integrate multi-source data related to the substation area from different institutions and platforms. Data time synchronization is to ensure that the time tags of different data sources may be inconsistent, and time synchronization processing is required to ensure that all data are aligned in time for subsequent processing. Data reconstruction takes into account the seasonal and cyclical characteristics of flood season data. It is necessary to adjust the frequency and form of the data through time series reconstruction to make it more suitable for time series analysis, such as reconstructing hourly data into daily data to meet the needs of daily flood control decisions.

[0021] Among them, flood control feature optimization includes automated feature extraction and selection, as well as new feature construction. Specifically, it uses automated feature extraction technology to identify important features, and combines domain knowledge to construct new feature variables, such as rainfall intensity, surrounding river water levels, and terrain influences, to enhance the predictive ability of the prediction model.

[0022] Among them, the introduction of the attention mechanism includes two sub-steps: selecting the type of attention mechanism and embedding the attention mechanism.

[0023] The attention mechanism enables the prediction model to pay more attention to data features that are closely related to flood control risks, thereby improving its ability to handle complex patterns. Attention mechanisms include self-attention mechanisms and sequence attention mechanisms. The self-attention mechanism calculates the internal association weights of the data to allow the prediction model to learn the connection between asynchronous time or features. The sequence attention mechanism can highlight the impact of recent data on the prediction results, which helps the prediction model capture data changes during the critical period of flood control. One or more attention modules can be designed in the prediction model, and then the attention modules can be embedded in the key parts of the prediction model, such as before each time step prediction of the DeepAR model.

[0024] Among them, the introduction of the dynamic L2 regularization strategy includes three sub-steps: selecting the key features of L2 regularization, customizing the function, and adding regularization terms in the custom function.

[0025] The dynamic L2 regularization strategy can effectively control the complexity of the prediction model, prevent overfitting, and enhance the generalization ability of the prediction model. L2 regularization is also called Ridge regularization and weight decay. Specifically, it limits the complexity of the prediction model by penalizing the L2 norm of the weight vector to avoid excessive model weights, thereby reducing the risk of overfitting. In response to the flood prevention needs of substations, stronger regularization is selectively applied to certain key features, such as rainfall and water level, while weaker regularization is applied to other secondary features. Specifically, different regularization strengths are assigned through the weights of different features.

[0026] Define a function such as l2_penalty(weights, lambda), the input is the weight parameter weights of the prediction model and the regularization coefficient lambda, and the output is the calculated L2 regularization term value l2_penalty. The function uses PyTorch tensor operations for calculation.

[0027] Among them, three sub-steps of multimodal data fusion are introduced, including selection of multimodal data type, data synchronization and alignment, and fusion at the model level.

[0028] Multimodal data fusion can significantly improve the prediction progress and reliability of the model, make full use of information from different data sources, and enhance the prediction model's understanding and prediction capabilities of complex substation flood control scenarios, thereby greatly improving the accuracy and robustness of the substation flood control risk probability prediction model.

[0029] Multimodal data types require identifying and classifying data types of various modalities, such as text, images, time series, sounds, etc. Each type of data provides different aspects of the substation environment. The application scenarios and functions of each data modality in flood risk prediction are analyzed to determine how they contribute to the final prediction results.

[0030] Since the data collection frequency and timestamp of different modalities may be different, time synchronization processing is needed to ensure that all data are aligned at the same time point. At the same time, for images or geographic information, it is necessary to align data from different sources to the same coordinate system through geo-registration technology to ensure the consistency of spatial information.

[0031] Fusion can be performed directly at the model level through specific fusion models, such as the Multi-layers LSTM (Multi-layers LSTM) used to handle sequence data fusion problems. This can dig deeper into the intrinsic connections between different modal components, thereby improving the prediction performance of the model.

[0032] In practical applications, the collected data is first cleaned and normalized. Then, during the training process, the key weight parameters of the attention mechanism, the regularization parameters of the dynamic L2 regularization strategy, and the fusion method of multimodal data are continuously adjusted according to the performance on the validation set to accurately predict the size and changing trend of the substation flood risk probability value and take flood prevention measures in advance.

[0033] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the specific implementation modes of the present invention can still be modified or replaced by equivalents, and any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be included in the scope of the claims of the present invention.

Claims

1. A method for predicting the probability of flood risk in a substation, characterized in that: The steps include: Step a, collecting historical multi-dimensional flood control data related to the substation; Step b, preprocessing the historical multi-dimensional flood control data and optimizing the flood control features; Step c: Based on the selected historical flood control characteristics, a flood control risk probability prediction model is constructed using the improved DeepAR algorithm; Step d, introduce the attention mechanism to optimize the prediction model; Step e, introducing a dynamic L2 regularization strategy to optimize the prediction model; Step f, introducing a multimodal data fusion optimization prediction model; Step g, collecting real-time data of multi-dimensional flood control data of the substation; Step h, preprocessing the real-time multi-dimensional flood control data and optimizing flood control features; Step i: Based on the selected real-time flood control features, the optimized flood control risk probability prediction model is used to predict the flood control risk probability of the substation, and the flood control risk probability value and prediction result are output.

2. The method for predicting substation flood risk probability according to claim 1, characterized in that: In step a and step g, the multidimensional flood control data includes the equipment waterproof information, geographical environment information, hydrological characteristic information, micro-meteorological information and historical flood information of the substation, wherein the equipment waterproof information includes the equipment waterproof level and the equipment installation height data, wherein the geographical environment information includes the collection well volume, the collection well water pump discharge, the surrounding terrain data, the terrain height data and the drainage path data, wherein the hydrological characteristic information includes the average water level, the highest water level, the lowest water level, the runoff and the flow rate data, wherein the micro-meteorological information includes the rainfall, rainfall duration and rainfall intensity data in the area where the substation is located, and wherein the historical flood information includes the flood occurrence time period and flood level data.

3. The method for predicting substation flood risk probability according to claim 1, characterized in that: In the step i, the prediction results include the flood inundation range, flood depth distribution, and flood retention time.

4. The method for predicting substation flood risk probability according to claim 1, characterized in that: In the step b, data preprocessing is one or more of multi-source data integration, data time synchronization, data cleaning, data standardization and normalization, and data reconstruction.

5. The method for predicting substation flood risk probability according to claim 1, characterized in that: In step b, the flood control features preferably include automated feature extraction and selection, and new feature construction.

6. The method for predicting substation flood risk probability according to claim 1, characterized in that: In step d, introducing the attention mechanism includes two sub-steps: selecting the type of attention mechanism and embedding the attention mechanism.

7. The method for predicting substation flood risk probability according to claim 1, characterized in that: In the step e, the introduction of the dynamic L2 regularization strategy includes three sub-steps: selecting the key features of L2 regularization, customizing the function, and adding the regularization term in the custom function.

8. The method for predicting substation flood risk probability according to claim 1, characterized in that: In the step f, three sub-steps of multimodal data fusion are introduced, including selecting multimodal data type, data synchronization and alignment, and fusion at the model level.

9. The method for predicting substation flood risk probability according to claim 1, characterized in that: It also includes step j. During use, as time accumulates, the prediction model is continuously iterated and optimized through historical multi-dimensional flood control data, attention mechanism, dynamic L2 regularization strategy and multimodal data fusion.

Images