A method and system for predicting substation flooding based on neural networks

Abstract

The present invention discloses a method and system for predicting waterlogging in a substation based on a neural network, which belongs to the field of data processing technology, and includes: step S1: establishing a prediction model, outputting the prediction accuracy, predicted precipitation data within the prediction time and precipitation offset time based on the first sequence data and the second sequence data; step S2: establishing a neural network model, inputting each waterlogging distribution image into the neural network model, generating a waterlogging image sequence, and obtaining the waterlogging level of each precipitation intensity sequence; step S3: obtaining the future precipitation data and future precipitation time of the target area within the prediction time; step S4: generating a change sequence based on the future precipitation data, obtaining a real-time waterlogging image set and the waterlogging time of the target area, and generating the waterlogging level of the target area within the prediction time. The present invention can predict the waterlogging image and waterlogging level of the substation, thereby reducing the phenomenon that the substation facilities are damaged due to waterlogging.

Description

Technical Field

[0001] The invention belongs to the field of data processing, and in particular relates to a method and system for predicting waterlogging in a substation based on a neural network. Background Art

[0002] Waterlogging refers to the phenomenon that heavy rainfall or continuous rainfall exceeds the drainage capacity of the current area, resulting in waterlogging disasters in the current area. Waterlogging in substations will cause damage to power facilities. Therefore, it is necessary to build a substation waterlogging prediction method based on technologies such as the Internet of Things, big data, neural networks, and image recognition, and generate a waterlogging monitoring plan to conduct real-time waterlogging monitoring and early warning for substations.

[0003] For example, the Chinese patent application "CN109544865B" discloses a method and system for early warning of waterlogging in a substation. By calculating the vacant area not occupied by fixed objects at different spatial heights above the ground in the substation's internal space, a function is established to determine the relationship between the amount of water accumulated in the substation and the vacant area and spatial height; the maximum drainage capacity of the substation's sewer network is evaluated based on the physical parameters of the drainage pipes in the substation; the accumulated water volume and water level per unit time in the substation are calculated based on the predicted rainfall, the maximum drainage capacity of the substation's sewer network, and the drainage capacity of the emergency drainage facilities; and an early warning is issued based on the calculated amount of water accumulated in the substation and the water level to determine whether waterlogging has occurred in the substation. The method of this invention can achieve the prediction and early warning of waterlogging in substations, effectively guide the flood prevention and drainage work of substations, and reduce the impact of heavy rainfall disasters on power grid companies and the entire society. Another example is the Chinese patent application "CN115640899A" which discloses a substation flooding warning method and system based on real-time monitoring and short-term precipitation prediction. The method includes: calculating the catchment area of ​​the substation: taking the substation to be predicted as the center, calculating the area of ​​the catchment area enclosed by the slope around the substation as the catchment area of ​​the substation; calculating the catchment coefficient of the substation; calculating the drainage capacity of the substation: obtaining the rate of decline of the water level in the catchment area as the drainage capacity of the substation; calculating the water depth of the substation: according to the precipitation, catchment area, catchment coefficient, drainage capacity and real-time water depth of the substation, the water depth of the substation in the next few hours is calculated. This invention can be used for short-term warning of rainstorm flooding of substations.

[0004] However, the above-mentioned prior art only calculates the depth of water accumulation in the early warning method of waterlogging in the substation. In actual situations, changes in the waterlogging area of ​​the substation will also affect the safety of the power facilities, so it is necessary to predict the water accumulation. Summary of the invention

[0005] In order to solve the above problems, the present invention provides a method, system and storage medium for predicting waterlogging in a substation based on a neural network to solve the problems in the prior art.

[0006] In order to achieve the above-mentioned invention object, the present invention proposes a method for predicting waterlogging in a substation based on a neural network, comprising:

[0007] Based on a preset first time interval, a first sequence of historical forecast precipitation data and a second sequence of historical actual precipitation data of a target area in a preset historical time period are acquired, and a prediction model is established, wherein the prediction model outputs a prediction accuracy, predicted precipitation data within a prediction time, and a precipitation offset time based on the first sequence of data and the second sequence of data;

[0008] Based on each of the second sequence data and a preset second time interval, a waterlogging distribution image corresponding to the target area is obtained, a plurality of first numerical intervals are set, a plurality of precipitation intensities are generated based on the first numerical intervals, each precipitation intensity sequence of the second sequence data is obtained based on the precipitation intensities, a neural network model is established, each of the waterlogging distribution images is input into the neural network model based on the precipitation intensity sequence, a waterlogging image sequence corresponding to each of the precipitation intensity sequences is generated, and a waterlogging level of each of the precipitation intensity sequences is obtained based on the waterlogging image sequence;

[0009] Acquire future forecast data of the target area within the forecast time, and acquire future precipitation data and future precipitation time of the target area within the forecast time based on the future forecast data, the forecast accuracy, the precipitation offset time and the forecast precipitation data;

[0010] A change sequence of each precipitation intensity is generated based on the future precipitation data to obtain the corresponding waterlogging distribution image in the waterlogging image sequence, and a real-time waterlogging image set is generated. The waterlogging time of the target area is obtained based on the change sequence and the real-time waterlogging image set, and the waterlogging level of the target area within the predicted time is generated based on the real-time waterlogging image set.

[0011] Further, the waterlogging image sequences and the waterlogging levels corresponding to the respective precipitation intensities are generated based on the following steps:

[0012] At the start time of each of the second sequence data, the camera unit of the target area is started, the camera unit obtains the water distribution image of the target area at every preset second time interval, and the time difference corresponding to any two adjacent historical actual precipitation data is calculated in the second sequence data. If the time difference is greater than or equal to a preset third time interval, the two historical actual precipitation data are split to generate a plurality of subsequence data from the second sequence data, the precipitation intensity of each of the subsequence data is obtained to generate a corresponding precipitation intensity sequence, each of the subsequence data of the same precipitation intensity sequence is grouped into a group to generate a plurality of small group sequences, the water distribution image corresponding to each of the subsequence data contained in the small group sequence is obtained, and the image is input into the neural network model, the neural network model is trained to learn and recognize the edge features of each of the water distribution images with the same arrangement number, and each standard water distribution image corresponding to the precipitation intensity sequence is output, and each standard water distribution image is set as the water image sequence;

[0013] The edge features of any two adjacent standard waterlogging distribution images are obtained, a coordinate system is established based on the standard waterlogging distribution images, and the pixel coordinates corresponding to each edge feature are located; if the pixel coordinates in any two adjacent standard waterlogging distribution images are different, the pixel coordinates are connected to generate an edge graph, and the pixel area of ​​the edge graph is calculated; the standard waterlogging distribution image with the largest pixel area is defined as a maximum waterlogging image; and the waterlogging level of the target area is generated based on the maximum waterlogging image and the precipitation intensity sequence.

[0014] Furthermore, the real-time water accumulation image set and the water accumulation time of the target area are generated based on the following steps:

[0015] Selecting similar sequences from each of the precipitation intensity sequences based on the change sequence, and setting the waterlogging image sequences corresponding to the similar sequences as the real-time waterlogging image set;

[0016] Obtain the time length of the change sequence, obtain the number of pixel points of the edge features of each image in the real-time waterlogging image set in turn, generate a first curve based on the time length and the number of pixel points, obtain the first inflection point and the last inflection point of the first curve, set the time point corresponding to the first inflection point in the time length as the start time of waterlogging in the target area, set the time point corresponding to the last inflection point in the time length as the end time of waterlogging in the target area, and the waterlogging time is the end time of waterlogging minus the start time of waterlogging.

[0017] Further, obtaining the similar sequence comprises the following steps:

[0018] If each of the precipitation intensity sequences contains a numerical sequence identical to the change sequence, the corresponding precipitation intensity sequence is set as the similar sequence; otherwise, a sequence combination of the same time length is extracted from each precipitation intensity sequence to generate a first screening group, a first overlap between the change sequence and each of the precipitation intensity sequences in the first screening group is obtained, each of the precipitation intensity sequences whose first overlap is greater than or equal to a preset first threshold is obtained and set as a second screening group, the precipitation intensity of each of the precipitation intensity sequences at the waterlogging start time, the maximum waterlogging image and the waterlogging end time is obtained in the second screening group and defined as each comparison point, the precipitation intensity sequence is split into a plurality of sub-intensity sequences based on each of the comparison points and numbered in sequence, the sub-intensity sequences including the precipitation intensity and the number, a second overlap between each of the sub-intensity sequences having the same number and the change sequence is obtained respectively, the sub-intensity sequences corresponding to the maximum second overlap among each of the numbers are combined to generate the similar sequence, and the standard waterlogging distribution image combination is extracted from the precipitation intensity sequences corresponding to each of the sub-intensity sequences to generate the real-time waterlogging image set.

[0019] Furthermore, outputting the forecast accuracy of the first sequence data and the second sequence data, the forecast precipitation data within the forecast time, and the precipitation offset time based on the forecast model includes the following steps:

[0020] The first sequence data and the second sequence data include the same number of time series, and the absolute value of the difference between the historical forecast precipitation data and the historical actual precipitation data at the same position in the time series is obtained. If the absolute value is less than or equal to a preset second threshold, it is determined that the historical forecast precipitation data is accurate, and the forecast accuracy rate R is calculated based on a first formula, and the first formula is: , wherein r1 is the number of accurate historical forecast precipitation data in the first sequence data, r2 is the number of the first sequence data that is not 0 and corresponds to the second sequence data that is 0, and r3 is the number of the second sequence data that is not 0 and corresponds to the first sequence data that is 0;

[0021] A preset number of sequence data are extracted in reverse order from the first sequence data and the second sequence data, respectively defined as first comparison data and second comparison data, and the i-th predicted precipitation data Q within the prediction time is generated based on the prediction model and the second formula. i , the second formula is: , where n is the preset number, P i is the i-th prediction result output by the prediction model, a iis the first comparison data of the ith one, b i is the i-th second comparison data;

[0022] Obtain discontinuous points of time characteristics in the time series, split the first sequence data and the second sequence data into multiple first data groups based on the discontinuity points, and calculate the precipitation offset time M based on a third formula, where the third formula is: , where J is the number of the first data set, T wj is the time when the second sequence data in the jth first data group is first not zero, T pj It is the time when the first sequence data in the j-th first data group is not 0.

[0023] Further, the future precipitation data and future precipitation time of the target area within the prediction time are obtained based on the following steps:

[0024] Generate the i-th future precipitation data H of the target area within the prediction time based on the fourth formula i , the fourth formula is: , where F i For the i-th future forecast data, the precipitation offset time is set to the future precipitation time. If the precipitation offset time is greater than 0, the prediction time is the start time of the future forecast data plus the precipitation offset time. Otherwise, the prediction time is the start time of the future forecast data minus the precipitation offset time.

[0025] The present invention also provides a substation waterlogging prediction system based on a neural network, which is used to implement the above-mentioned substation waterlogging prediction method based on a neural network. The system mainly includes:

[0026] A data prediction module, which obtains a first sequence of historical forecast precipitation data and a second sequence of historical actual precipitation data of a target area in a preset historical time period based on a preset first time interval, and establishes a prediction model, wherein the prediction model outputs a prediction accuracy, predicted precipitation data within a prediction time, and a precipitation offset time based on the first sequence of data and the second sequence of data;

[0027] an image processing module, which obtains a waterlogging distribution image corresponding to the target area based on each of the second sequence data and a preset second time interval, sets a plurality of first numerical intervals, generates a plurality of precipitation intensities based on the first numerical intervals, obtains each precipitation intensity sequence of the second sequence data based on the precipitation intensities, establishes a neural network model, inputs each of the waterlogging distribution images into the neural network model based on the precipitation intensity sequence, generates a waterlogging image sequence corresponding to each of the precipitation intensity sequences, and obtains the waterlogging level of each of the precipitation intensity sequences based on the waterlogging image sequence;

[0028] A data output module, used to obtain future forecast data of the target area within the forecast time, and obtain future precipitation data and future precipitation time of the target area within the forecast time based on the future forecast data, the forecast accuracy, the precipitation offset time and the forecast precipitation data;

[0029] An image generation module generates a change sequence of each precipitation intensity based on the future precipitation data to obtain the corresponding waterlogging distribution image in the waterlogging image sequence and generate a real-time waterlogging image set, obtains the waterlogging time of the target area based on the change sequence and the real-time waterlogging image set, and generates the waterlogging level of the target area within the predicted time based on the real-time waterlogging image set.

[0030] Compared with the prior art, the beneficial effects of the present invention are at least as follows:

[0031] The present invention first collects historical forecast precipitation data and historical actual precipitation data of the target area through the above scheme, and inputs the corresponding sequence data into the prediction model constructed by the neural network. The prediction model outputs the predicted precipitation data within the prediction time, and further obtains the forecast accuracy and precipitation offset time of the forecast through statistical analysis. Then, each waterlogging distribution image corresponding to the second sequence data in the historical actual precipitation data is collected, and the precipitation intensity of the second sequence data is obtained to generate a plurality of precipitation intensity sequences. The waterlogging distribution images corresponding to each precipitation intensity sequence can be statistically analyzed, and the neural network model performs image feature recognition on each waterlogging distribution image of the same precipitation intensity sequence to generate a standard waterlogging distribution image corresponding to each precipitation intensity sequence. The waterlogging image sequence is generated in combination, which can reduce the error of image feature extraction and recognition in the waterlogging image of the target area. Finally, the future forecast data, the predicted precipitation data and the forecast accuracy are combined to generate the future precipitation data and the future precipitation time, which can improve the accuracy of the future actual precipitation, and further accurately predict the waterlogging in the target area.

[0032] The present invention also extracts similar sequences from various precipitation intensity sequences through the precipitation intensity corresponding to future precipitation data, and generates a real-time waterlogging image set from each waterlogging distribution sequence corresponding to the similar sequences. The waterlogging time and waterlogging level in the target area within the predicted time can be identified, which is convenient for manual formulation of drainage plans and avoids damage to various target facilities in the target area caused by waterlogging. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art are briefly introduced below. Obviously, the drawings in the following description are only embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on the provided drawings without creative work.

[0034] Figure 1 A flowchart of the steps of a method for predicting waterlogging in a substation based on a neural network according to the present invention;

[0035] Figure 2 This is a structural diagram of a substation waterlogging prediction system based on a neural network according to the present invention. DETAILED DESCRIPTION

[0036] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0037] It is understood that the terms "first", "second", etc. used in this application may be used herein to describe various elements, but unless otherwise specified, these elements are not limited by these terms. These terms are only used to distinguish a first element from another element. For example, without departing from the scope of this application, a first xx script may be referred to as a second xx script, and similarly, a second xx script may be referred to as a first xx script.

[0038] like Figure 1 As shown, a method for predicting waterlogging in a substation based on a neural network includes:

[0039] Step S1: Based on a preset first time interval, a first sequence of historical forecast precipitation data and a second sequence of historical actual precipitation data of the target area in a preset historical time period are obtained, and a prediction model is established. The prediction model outputs the prediction accuracy, the predicted precipitation data within the prediction time, and the precipitation offset time based on the first sequence of data and the second sequence of data.

[0040] Specifically, the future precipitation data of the target area is predicted through analysis, and the waterlogging phenomenon in the target area is further predicted; in this embodiment, the target area refers to the area that needs to be monitored, including but not limited to each substation, the first time interval refers to the time interval for collecting precipitation data, for example, the first time interval is set to every hour, the historical time period refers to a period of time in the past, for example, the historical time period is set to the time period of the past year corresponding to the current date, the historical forecast precipitation data refers to the precipitation forecast data within the target area within the historical time period, the first sequence data is time series data composed of hourly forecast precipitation, similarly, the historical actual precipitation data refers to the actual precipitation data within the target area within the historical time period, and the second sequence data is time series data composed of actual precipitation per hour, that is, the historical actual precipitation data corresponds to each precipitation data in the second sequence data, all of which are obtained through the meteorological station or network platform.

[0041] The prediction model refers to the use of the LSTM neural network model, the first sequence data and the second sequence data are input into the model for learning and training, and then the first sequence data of the current future time forecast is input into the prediction model to obtain the actual second sequence data in the future. The prediction model outputs the forecast accuracy, predicted precipitation data and precipitation offset time. The forecast accuracy refers to the accuracy of the weather forecast data. The credibility of the forecast data is judged based on the forecast accuracy. Since the start time and end time of each time period in the first sequence data may be offset from the second sequence data, it is necessary to record and analyze the offset time. Time refers to the time offset between the start time of the historical forecast precipitation data and the historical actual precipitation data in the same time period, which can be used as one of the ways to measure the accuracy of the forecast data. For example, the forecast data for 2023 / 01 / 01 / 12:00 in the first sequence of data is 10 mm / h, and the forecast data for 2023 / 01 / 01 / 13:00 is 15 mm / h. The corresponding precipitation data for 2023 / 01 / 01 / 12:00 in the second sequence of data is 0 mm / h, and the precipitation data for 2023 / 01 / 01 / 13:00 is 12 mm / h. The corresponding precipitation offset time is 1 hour in advance.

[0042] Step S2: Based on each second sequence data and a preset second time interval, obtain a waterlogging distribution image corresponding to the target area, set multiple first numerical intervals, generate multiple precipitation intensities based on the first numerical intervals, obtain each precipitation intensity sequence of the second sequence data based on the precipitation intensity, establish a neural network model, input each waterlogging distribution image into the neural network model based on the precipitation intensity sequence, generate a waterlogging image sequence corresponding to each precipitation intensity sequence, and obtain the waterlogging level of each precipitation intensity sequence based on the waterlogging image sequence.

[0043] Specifically, in this embodiment, the water accumulation distribution image is an image of the target area obtained by starting the camera unit during precipitation, which can display the image changes of water accumulation in the target area. Each second sequence data has a corresponding water accumulation distribution image. The second time interval is set to 10 minutes, that is, a corresponding water accumulation distribution image is generated every 10 minutes in the time feature included in the second sequence data. The first numerical interval refers to grading the precipitation by setting the numerical interval to generate various precipitation intensities. For example, the first numerical intervals are (0, 5], (5, 10], (10, 15], (15, 25], (25, 35], (35, 45], (45, 55], (55, 65], (65, 75], (75, 85], (85, 95] and greater than 95, in millimeters per hour, and the corresponding The precipitation intensities are 5, 10, 15, 25, 35, 45, 55, 65, 75, 85, 95 and 100 respectively; the precipitation intensity sequence refers to obtaining the first numerical intervals where the historical actual precipitation data in the second sequence data are located, and further obtaining each precipitation intensities, thereby generating a precipitation intensity sequence from the second sequence data, which is convenient for statistics and analysis of precipitation data; the neural network model refers to a convolutional neural network model, which can extract the underlying features of the input image, and input all the waterlogging distribution images contained in each precipitation intensity sequence into the neural network model, and the images can be classified and identified through feature detection to generate a waterlogging image sequence corresponding to the precipitation intensity sequence. The specific generation method will be described later; the waterlogging level refers to setting the level of each precipitation intensity sequence, which is convenient for predicting the waterlogging level in the target area.

[0044] Step S3: Obtain future forecast data for the target area within the prediction time, and obtain future precipitation data and future precipitation time for the target area within the prediction time based on the future forecast data, forecast accuracy, precipitation offset time and predicted precipitation data.

[0045] Specifically, in the present embodiment, the future forecast data is the forecast precipitation data of the target area within the forecast time obtained through the meteorological station or the network platform, the future precipitation data refers to the precipitation data of the target area within the forecast time, and the future precipitation time refers to the start time of the future precipitation data. The future precipitation data, the forecast accuracy and the future forecast data output by the prediction model can be combined to output the future precipitation data, and the future precipitation time is obtained by the precipitation offset time. The accuracy of the urban waterlogging waterlogging prediction in the present invention can be further improved through the future precipitation data and the future precipitation time.

[0046] Step S4: Generate a change sequence of each precipitation intensity based on future precipitation data to obtain the corresponding waterlogging distribution image in the waterlogging image sequence, and generate a real-time waterlogging image set. Based on the change sequence and the real-time waterlogging image set, obtain the waterlogging time of the target area, and generate the waterlogging level of the target area within the predicted time based on the real-time waterlogging image set.

[0047] Specifically, in the present embodiment, the change sequence is a time sequence of changes in each precipitation intensity. A precipitation intensity sequence is extracted from each precipitation intensity sequence through the change sequence, and each waterlogging distribution image in the corresponding waterlogging image sequence is set as a real-time waterlogging image set of future precipitation data. The waterlogging time refers to the length of time that waterlogging exists in the target area. The waterlogging level and waterlogging time of the target area within the predicted time can be calculated through the real-time waterlogging image set. The waterlogging level refers to a numerical value that measures the degree of waterlogging. The higher the waterlogging level, the greater the corresponding degree of waterlogging. The pixel positions of each waterlogging point and target facility in the target area can be further obtained, an image coordinate system can be established, and the pixel coordinates corresponding to each waterlogging point and target facility are marked in the real-time waterlogging image set. The pixel distance between the target facility and each waterlogging point is obtained based on the pixel coordinates. The flow direction diagram of each waterlogging point is generated by obtaining the pixel extension direction of two adjacent waterlogging distribution images. The water flow rate of each waterlogging point is obtained based on the flow direction diagram. Therefore, a drainage plan can be manually formulated in advance to avoid losses caused by waterlogging to the target facilities in the target area.

[0048] Generate waterlogging image sequences and waterlogging levels corresponding to each precipitation intensity based on the following steps:

[0049] At the start time of each second sequence data, the camera unit of the target area is started, and the camera unit obtains the water distribution image of the target area at every preset second time interval, and the time difference corresponding to any two adjacent historical actual precipitation data is calculated in the second sequence data. If the time difference is greater than or equal to the preset third time interval, the two historical actual precipitation data are split to generate a plurality of subsequence data from the second sequence data, and the precipitation intensity of each subsequence data is obtained to generate a corresponding precipitation intensity sequence, and each subsequence data of the same precipitation intensity sequence is divided into a group to generate a plurality of small group sequences, and the water distribution image corresponding to each subsequence data contained in the small group sequence is obtained, and the image is input into the neural network model. The neural network model is trained to learn and recognize the edge features of each water distribution image with the same arrangement number, and each standard water distribution image corresponding to the precipitation intensity sequence is output, and each standard water distribution image is set as a water image sequence;

[0050] The edge features of any two adjacent standard waterlogging distribution images are obtained, a coordinate system is established based on the standard waterlogging distribution images, and the pixel coordinates corresponding to each edge feature are located. If the pixel coordinates in any two adjacent standard waterlogging distribution images are different, the pixel coordinates are connected to generate an edge graph, and the pixel area of ​​the edge graph is calculated. The standard waterlogging distribution image with the largest pixel area is defined as the maximum waterlogging image. The waterlogging level of the target area is generated based on the maximum waterlogging image and the precipitation intensity sequence.

[0051] Specifically, in this embodiment, the precipitation state is monitored by a sensor to start the camera unit of the target area in real time to generate various water distribution images with the same time series as the second sequence data. This can reduce the image data processing load. The water distribution image includes a bird's-eye view of the target area. The time difference refers to the time interval length of the second sequence data. For example, the third time interval is set to 5 hours, and the second sequence data has precipitation time points of 01:00, 02:00, 03:00, 04:00, 11:00, 13:00 and 14:00. The time difference between 04:00 and 11:00 is 6 hours, which is greater than the third time interval. interval, the second sequence data is split between 04:00 and 11:00, the time difference between 11:00 and 13:00 is 1 hour, which is less than the third time interval, and it is not split, that is, the time points corresponding to the two split subsequence data are subsequence data 01:00, 02:00, 03:00 and 04:00, and subsequence data 11:00, 13:00 and 14:00. Repeat this step to split the second sequence data into multiple subsequence data. By splitting the second sequence data, the water accumulation image of the target area corresponding to each precipitation time can be analyzed in detail. If the precipitation time is not continuous, the water accumulation image of the target area will change greatly.

[0052] A precipitation intensity sequence is generated according to each precipitation intensity corresponding to the subsequence data, so as to facilitate statistics and classification of each subsequence data. A small group sequence refers to classifying each subsequence data, that is, the small group sequence contains multiple subsequence data of the same precipitation intensity sequence, and each water distribution image of each subsequence data is obtained. For example, it can be calculated from the first time interval and the second time interval, that is, a water distribution image is generated every 10 minutes in 1 hour, and a historical actual precipitation data in the subsequence data contains 6 water distribution images; the neural network model is used for image processing and feature recognition. Through learning a certain amount of image data, feature extraction and classification selection can be performed on each image, and the set image result can be output. Each water distribution image of the subsequence data has an arrangement order, and the same water distribution image in the small group sequence is The waterlogging distribution images in the arrangement order are input into the neural network model, and the neural network model extracts the edge features of each waterlogging distribution image and outputs a standard image containing edge features, namely, the standard waterlogging distribution image. The edge feature refers to the feature point generated after edge detection of the image, which is the junction of the image area and another attribute area. By extracting the edge features of each image, the adjacent edge features are connected. The contour shape of the waterlogging area can be distinguished through the contour shape composed of the total edge features of the continuous waterlogging distribution images, and the changing trend of the waterlogging area in the target area can be further effectively described. The waterlogging distribution images corresponding to each arrangement order are input into the neural network model to output the standard waterlogging distribution images of the corresponding precipitation intensity sequence in the group sequence, which are combined in sequence to generate a waterlogging image sequence.

[0053] The waterlogging image sequence is an image description of the real-time waterlogging degree of the precipitation intensity sequence. The edge graph refers to connecting the two edge features with the smallest distance in sequence until all edge features are connected into an edge graph. The ratio value can be obtained by (the number of pixels contained in the edge graph / the total number of pixels of the standard waterlogging distribution image). The maximum waterlogging image refers to the standard waterlogging distribution image corresponding to the maximum ratio value. The time when the maximum waterlogging occurs in the intensity precipitation sequence data is calculated by the arrangement sequence number of the maximum waterlogging image in the waterlogging image sequence. The waterlogging level can be set by setting the waterlogging interval. For example, the waterlogging interval is (0, 2], (2, 5], (5, 10], (10, 15], (15, 20], (20, 30], (30, 40], (40, 50] and greater than 50, and the corresponding waterlogging levels are 1, 2, 3, 4, 5, 6, 7, 8 and 9. The waterlogging degree calculation formula is: , where t1 is the arrangement number in the precipitation intensity sequence corresponding to the minimum pixel area, t2 is the arrangement number in the precipitation intensity sequence corresponding to the maximum waterlogging image, and G is the precipitation intensity corresponding to the arrangement number. For example, the precipitation intensity sequence is 5, 10, 15, 25, 10, 5, where waterlogging occurs at the second precipitation intensity of 10, that is, the edge graph is the minimum, and the maximum waterlogging image appears at the fifth precipitation intensity of 10. The waterlogging degree of this precipitation intensity sequence is (10+15+25+10) / (5-2+1)=15, and the corresponding waterlogging level is level 4.

[0054] Generate a real-time waterlogging image set and waterlogging time of the target area based on the following steps:

[0055] Based on the change sequence, similar sequences are selected from each precipitation intensity sequence, and the waterlogging image sequence corresponding to the similar sequence is set as the real-time waterlogging image set;

[0056] The time length of the change sequence is obtained, and the number of pixel points of the edge features of each image in the real-time waterlogging image set is obtained in turn. A first curve is generated by fitting based on the time length and the number of pixel points. The first inflection point and the last inflection point of the first curve are obtained, and the time point corresponding to the first inflection point in the time length is set as the start time of waterlogging in the target area, and the time point corresponding to the last inflection point in the time length is set as the end time of waterlogging in the target area. The waterlogging time is the end time of waterlogging minus the start time of waterlogging.

[0057] Specifically, in this embodiment, the change sequence is generated by each precipitation intensity included in the future precipitation data, and a precipitation intensity sequence that is the same as the change sequence is selected from each precipitation intensity sequence as a similar sequence, and the water accumulation image sequence included in the similar sequence is further set as a real-time water accumulation image set of the future precipitation data, so that a real-time water accumulation image of the target area within the prediction time can be directly generated;

[0058] The time length refers to the time feature sequence contained in the change sequence, that is, the time point corresponding to the prediction time. The waterlogging time refers to the difference between the start time and the end time of waterlogging. The first curve is generated by fitting the number of pixels containing edge features in each image in the real-time waterlogging image set, which can describe the edge feature change curve of each image. The inflection point mathematically refers to the point where the curve changes upward or downward. The first inflection point refers to the time point when the number of pixels in the target area of ​​the image first changes significantly, which means the waterlogging point generated by waterlogging. Therefore, the time point of the first inflection point on the time length is set as the start time of waterlogging. Similarly, the last inflection point is the number of pixels contained in the edge feature generated by the stable and unchanged waterlogging. The time point of the last inflection point on the time length is set as the end time of waterlogging. Note that the end time of waterlogging does not mean that there is no waterlogging in the target area, but only indicates the time when the waterlogging range in the target area no longer changes. The waterlogging time is the end time of waterlogging minus the start time of waterlogging.

[0059] Obtaining similar sequences includes the following steps:

[0060] If each precipitation intensity sequence contains a numerical sequence identical to the change sequence, the corresponding precipitation intensity sequence is set as a similar sequence; otherwise, a sequence combination of the same time length is extracted from each precipitation intensity sequence to generate a first screening group, a first overlap between the change sequence and each precipitation intensity sequence in the first screening group is obtained, each precipitation intensity sequence whose first overlap is greater than or equal to a preset first threshold is obtained and set as a second screening group, the precipitation intensity of each precipitation intensity sequence at the start time of waterlogging, the maximum waterlogging image and the end time of waterlogging is obtained in the second screening group and defined as each comparison point, the precipitation intensity sequence is split into multiple sub-intensity sequences based on each comparison point and numbered in sequence, the sub-intensity sequence includes precipitation intensity and number, the second overlap between each sub-intensity sequence with the same number and the change sequence is obtained respectively, the sub-intensity sequences corresponding to the maximum second overlap in each number are combined to generate a similar sequence, and a standard waterlogging distribution image combination is extracted from the precipitation intensity sequence corresponding to each sub-intensity sequence to generate a real-time waterlogging image set.

[0061] Specifically, in this embodiment, similar sequences are obtained in each precipitation intensity sequence to facilitate further effective prediction of water accumulation distribution in the target area. For example, the precipitation intensities in the change sequence G1 are 5, 10, 10, 5 and 5. If there is a sequence L1 in each precipitation intensity sequence that is 5, 10, 10, 5 and 5, then the precipitation intensity sequence is set as a similar sequence. Otherwise, it is necessary to further extract and generate similar sequences in each precipitation intensity sequence. The extraction can be performed through time length, first overlap and second overlap. Each precipitation intensity sequence with the same time length is set as the first screening group. The formula for calculating the first overlap Z is: Z=Z1 / Z2, wherein Z1 is the number of precipitation intensities in the precipitation intensity sequence that are the same as the change sequence and have the same arrangement number, and Z2 is the number of precipitation intensities in the change sequence. For example, the precipitation intensity sequence L2 is 5, 5, 10, 5 and 5, then the first overlap between the precipitation intensity sequence L2 and the change sequence G1 is 4 / 5=0.8. If the first overlap is 0.5, then the precipitation intensity sequence L2 is 1 / 2=0.6. If the degree is greater than or equal to the preset first threshold, it means that the corresponding precipitation intensity sequence has a high degree of fit with the change sequence, and a second screening group is formed; for example, each precipitation intensity sequence in the second screening group can be split into 4 sub-intensity sequences by setting each comparison point, and the corresponding numbers are number 1, number 2, number 3 and number 4, and the precipitation intensity corresponding to each sub-intensity sequence with the same number is obtained, and the second coincidence degree is calculated with the precipitation intensity of the change sequence respectively. The calculation formula of the second coincidence degree U is U=u1 / u2, wherein u1 is the number of consecutive identical precipitation intensities in the sub-intensity sequence and the change sequence, and u2 is the number of precipitation intensities in the sub-intensity sequence. The sub-intensity sequences corresponding to the maximum second coincidence degree of each number are sequentially combined into similar sequences. For example, the sub-intensity sequence K1 of number 1 is 5 and 10, and the second coincidence degree between the sub-intensity sequence K1 of number 1 and the first two precipitation intensities of the change sequence G1 is 2 / 2=100%, and each precipitation intensity of number 1 is selected as the sub-intensity sequence K1.

[0062] Outputting the forecast accuracy of the first sequence data and the second sequence data based on the forecast model, the forecast precipitation data within the forecast time, and the precipitation offset time includes the following steps:

[0063] The first sequence data and the second sequence data contain the same number of time series. The absolute value of the difference between the historical forecast precipitation data and the historical actual precipitation data at the same position in the time series is obtained. If the absolute value is less than or equal to the preset second threshold, the historical forecast precipitation data is judged to be accurate. The forecast accuracy R is calculated based on the first formula. The first formula is: , where r1 is the number of accurate historical forecast precipitation data in the first sequence data, r2 is the number of first sequence data that is not 0 and corresponds to the second sequence data that is 0, and r3 is the number of second sequence data that is not 0 and corresponds to the first sequence data that is 0;

[0064] A preset number of sequence data are extracted in reverse order from the first sequence data and the second sequence data, respectively defined as the first comparison data and the second comparison data, and the i-th predicted precipitation data Q within the prediction time is generated based on the prediction model and the second formula. i , the second formula is: , where n is the preset number, P i is the i-th prediction result output by the prediction model, a i is the first comparison data of the ith i is the i-th second comparison data;

[0065] Obtain discontinuous points of time characteristics in the time series, split the first series data and the second series data into multiple first data groups based on the discontinuity points, and calculate the precipitation offset time M based on the third formula, which is: , where J is the number of the first data set, T wj is the time when the second sequence data in the jth first data group is first not zero, T pj It is the time when the first sequence data in the j-th first data group is not 0.

[0066] Specifically, in this embodiment, the first formula is to calculate the forecast accuracy by using a statistical analysis method to further evaluate the accuracy of the forecast data. For example, the second threshold is set to 3 mm / h, the forecast data at 2023 / 01 / 01 / 13:00 in the first sequence data is 15 mm / h, and the corresponding precipitation data at 2023 / 01 / 01 / 13:00 in the second sequence data is 12 mm / h, where 15-12=3, which is equal to the second threshold. It is then determined that the historical forecast precipitation data of the first sequence data at 2023 / 01 / 01 / 13:00 is accurate, and all the first sequence data and the second sequence data are further counted to generate a forecast accuracy R. In the first formula, r1 refers to the number of empty reports in the forecast, and r2 refers to the number of missed reports in the forecast.

[0067] The prediction model can output the prediction result within the prediction time by training and learning the first sequence data and the second sequence data. In this embodiment, the prediction result of the prediction model is corrected by the second formula to generate the predicted precipitation data, wherein the first comparison data and the second comparison data refer to the sequence data with the same preset number extracted from the last position of the first sequence data and the second sequence data, respectively, because they are closest to the prediction time and have higher usability. For example, if the preset number of prediction time is 10, the last 10 data in the first sequence data are set as the first comparison data, and the last 10 data in the second sequence data are set as the second comparison data;

[0068] The third formula is used to count the offset of the forecast time, which can further obtain the future precipitation time. The first data group is generated by splitting the time series in the first sequence data and the second sequence data, wherein the first sequence data and the second sequence data contained in the first data group are one-to-one corresponding, that is, the first sequence data and the second sequence data exist at the same time point. The discontinuity point refers to the time point where the time series is discontinuous. For example, the time series is 2023 / 01 / 01 / 12:00, 2023 / 01 / 01 / 13:00 and 2023 / 01 / 02 / 11:00, wherein there is a time discontinuity between 2023 / 01 / 01 / 13:00 and 2023 / 01 / 02 / 11:00, and the corresponding discontinuity point is between 2023 / 01 / 01 / 13:00 and 2023 / 01 / 02 / 11:00, which is split into 2023 / 01 / 01 / 12:00 and 2023 / 01 / 01 / 13:00, 2023 / 01 / 02 / 11:00. Simply understood, the time features in the first data group are all continuous time points.

[0069] Get the future precipitation data and future precipitation time of the target area within the forecast time based on the following steps:

[0070] Generate the i-th future precipitation data H of the target area within the prediction time based on the fourth formula i , the fourth formula is: , where F i For the i-th future forecast data, the precipitation offset time is set to the future precipitation time. If the precipitation offset time is greater than 0, the prediction time is the start time of the future forecast data plus the precipitation offset time. Otherwise, the prediction time is the start time of the future forecast data minus the precipitation offset time.

[0071] Specifically, in this embodiment, future precipitation data is obtained by combining future forecast data, forecast accuracy and predicted precipitation data, which can improve the precipitation accuracy within the forecast time. For example, the forecast time is set to 5 hours, the future forecast data are 2, 4, 6, 8 and 5, the predicted precipitation data are 1, 3, 6, 7 and 4, and the forecast accuracy R is 0.8. The corresponding future precipitation data are 1.8, 3.8, 6, 7.8 and 4.8, in units of mm / h.

[0072] The present invention first collects historical forecast precipitation data and historical actual precipitation data of the target area through the above scheme, and inputs the corresponding sequence data into the prediction model constructed by the neural network. The prediction model outputs the predicted precipitation data within the prediction time, and further obtains the forecast accuracy and precipitation offset time of the forecast through statistical analysis. Then, each waterlogging distribution image corresponding to the second sequence data in the historical actual precipitation data is collected, and the precipitation intensity of the second sequence data is obtained to generate a plurality of precipitation intensity sequences. The waterlogging distribution images corresponding to each precipitation intensity sequence can be statistically analyzed, and the neural network model performs image feature recognition on each waterlogging distribution image of the same precipitation intensity sequence to generate a standard waterlogging distribution image corresponding to each precipitation intensity sequence. The waterlogging image sequence is generated in combination, which can reduce the error of image feature extraction and recognition in the waterlogging image of the target area. Finally, the future forecast data, the predicted precipitation data and the forecast accuracy are combined to generate the future precipitation data and the future precipitation time, which can improve the accuracy of the future actual precipitation, and further accurately predict the waterlogging in the target area.

[0073] The present invention also extracts similar sequences from various precipitation intensity sequences through the precipitation intensity corresponding to future precipitation data, and generates a real-time waterlogging image set from each waterlogging distribution sequence corresponding to the similar sequences. The waterlogging time and waterlogging level in the target area within the predicted time can be identified, which is convenient for manual formulation of drainage plans and avoids damage to various target facilities in the target area caused by waterlogging.

[0074] It is particularly noteworthy that the present invention can predict future precipitation data of the target area, and further predict waterlogging in the target area, thereby reducing the problem of substation facilities being damaged due to waterlogging.

[0075] like Figure 2 As shown, the present invention also provides a substation waterlogging prediction system based on a neural network, which is used to implement the above-mentioned substation waterlogging prediction method based on a neural network. The system mainly includes:

[0076] A data prediction module, which obtains a first sequence of historical forecast precipitation data and a second sequence of historical actual precipitation data of a target area in a preset historical time period based on a preset first time interval, and establishes a prediction model. The prediction model outputs a prediction accuracy, predicted precipitation data within a prediction time, and a precipitation offset time based on the first sequence of data and the second sequence of data;

[0077] An image processing module, which obtains a waterlogging distribution image corresponding to a target area based on each second sequence data and a preset second time interval, sets multiple first numerical intervals, generates multiple precipitation intensities based on the first numerical intervals, obtains each precipitation intensity sequence of the second sequence data based on the precipitation intensity, establishes a neural network model, inputs each waterlogging distribution image into the neural network model based on the precipitation intensity sequence, generates a waterlogging image sequence corresponding to each precipitation intensity sequence, and obtains the waterlogging level of each precipitation intensity sequence based on the waterlogging image sequence;

[0078] A data output module is used to obtain future forecast data of a target area within a forecast time, and obtain future precipitation data and future precipitation time of the target area within the forecast time based on the future forecast data, forecast accuracy, precipitation offset time and forecast precipitation data;

[0079] The image generation module generates a change sequence of each precipitation intensity based on future precipitation data to obtain the corresponding waterlogging distribution image in the waterlogging image sequence and generates a real-time waterlogging image set. The waterlogging time of the target area is obtained based on the change sequence and the real-time waterlogging image set, and the waterlogging level of the target area within the predicted time is generated based on the real-time waterlogging image set.

[0080] It should be understood that, although each step in the flow chart of each embodiment of the present invention is shown in sequence according to the indication of the arrow, these steps are not necessarily performed in sequence according to the order indicated by the arrow. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be performed in other orders. Moreover, at least a portion of the steps in each embodiment may include a plurality of sub-steps or a plurality of stages, and these sub-steps or stages are not necessarily performed at the same time, but can be performed at different times, and the execution order of these sub-steps or stages is not necessarily performed in sequence, but can be performed in turn or alternately with at least a portion of other steps or sub-steps or stages of other steps.

[0081] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the above-mentioned program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0082] The technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above-mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0083] The above embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the patent of the present invention. It should be pointed out that, for those of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

[0084] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for predicting waterlogging in a substation based on a neural network, characterized in that: The method comprises the following steps: Based on a preset first time interval, a first sequence of historical forecast precipitation data and a second sequence of historical actual precipitation data of a target area in a preset historical time period are acquired, and a prediction model is established, wherein the prediction model outputs a prediction accuracy, predicted precipitation data within a prediction time, and a precipitation offset time based on the first sequence of data and the second sequence of data; Based on each of the second sequence data and a preset second time interval, a waterlogging distribution image corresponding to the target area is obtained, a plurality of first numerical intervals are set, a plurality of precipitation intensities are generated based on the first numerical intervals, each precipitation intensity sequence of the second sequence data is obtained based on the precipitation intensities, a neural network model is established, each of the waterlogging distribution images is input into the neural network model based on the precipitation intensity sequence, a waterlogging image sequence corresponding to each of the precipitation intensity sequences is generated, and a waterlogging level of each of the precipitation intensity sequences is obtained based on the waterlogging image sequence; The waterlogging image sequences and the waterlogging levels corresponding to the respective precipitation intensities are generated based on the following steps: At the start time of each of the second sequence data, the camera unit of the target area is started, the camera unit obtains the water distribution image of the target area at every preset second time interval, and the time difference corresponding to any two adjacent historical actual precipitation data is calculated in the second sequence data. If the time difference is greater than or equal to a preset third time interval, the two historical actual precipitation data are split to generate a plurality of subsequence data from the second sequence data, the precipitation intensity of each of the subsequence data is obtained to generate a corresponding precipitation intensity sequence, each of the subsequence data of the same precipitation intensity sequence is grouped into a group to generate a plurality of small group sequences, the water distribution image corresponding to each of the subsequence data contained in the small group sequence is obtained, and the image is input into the neural network model, the neural network model is trained to learn and recognize the edge features of each of the water distribution images with the same arrangement number, and each standard water distribution image corresponding to the precipitation intensity sequence is output, and each standard water distribution image is set as the water image sequence; Acquire the edge features of any two adjacent standard waterlogging distribution images, establish a coordinate system based on the standard waterlogging distribution images, locate the pixel coordinates corresponding to each edge feature, if the pixel coordinates in any two adjacent standard waterlogging distribution images are different, connect the pixel coordinates to generate an edge graph, calculate the pixel area of ​​the edge graph, define the standard waterlogging distribution image with the largest pixel area as a maximum waterlogging image, and generate the waterlogging level of the target area based on the maximum waterlogging image and the precipitation intensity sequence; Acquire future forecast data of the target area within the forecast time, and acquire future precipitation data and future precipitation time of the target area within the forecast time based on the future forecast data, the forecast accuracy, the precipitation offset time and the forecast precipitation data; Generate a variation sequence of each precipitation intensity based on the future precipitation data, so as to obtain the corresponding waterlogging distribution image in the waterlogging image sequence, and generate a real-time waterlogging image set, obtain the waterlogging time of the target area based on the variation sequence and the real-time waterlogging image set, and generate the waterlogging level of the target area within the predicted time based on the real-time waterlogging image set; Wherein, the real-time waterlogging image set and the waterlogging time of the target area are generated based on the following steps: Selecting similar sequences from each of the precipitation intensity sequences based on the change sequence, and setting the waterlogging image sequences corresponding to the similar sequences as the real-time waterlogging image set; Obtain the time length of the change sequence, sequentially obtain the number of pixel points of the edge feature of each image in the real-time waterlogging image set, generate a first curve based on the time length and the number of each pixel point, obtain the first inflection point and the last inflection point of the first curve, set the time point corresponding to the first inflection point in the time length as the waterlogging start time of the target area, set the time point corresponding to the last inflection point in the time length as the waterlogging end time of the target area, and the waterlogging time is the waterlogging end time minus the waterlogging start time; Wherein, obtaining the similar sequence comprises the following steps: If each of the precipitation intensity sequences contains a numerical sequence identical to the change sequence, the corresponding precipitation intensity sequence is set as the similar sequence; otherwise, a sequence combination of the same time length is extracted from each precipitation intensity sequence to generate a first screening group, a first overlap between the change sequence and each of the precipitation intensity sequences in the first screening group is obtained, each of the precipitation intensity sequences whose first overlap is greater than or equal to a preset first threshold is obtained and set as a second screening group, the precipitation intensity of each of the precipitation intensity sequences at the waterlogging start time, the maximum waterlogging image and the waterlogging end time is obtained in the second screening group and defined as each comparison point, the precipitation intensity sequence is split into a plurality of sub-intensity sequences based on each of the comparison points and numbered in sequence, the sub-intensity sequences including the precipitation intensity and the number, a second overlap between each of the sub-intensity sequences having the same number and the change sequence is obtained respectively, the sub-intensity sequences corresponding to the maximum second overlap among each of the numbers are combined to generate the similar sequence, and the standard waterlogging distribution image combination is extracted from the precipitation intensity sequences corresponding to each of the sub-intensity sequences to generate the real-time waterlogging image set.

2. A method for predicting waterlogging in a substation based on a neural network according to claim 1, characterized in that: Outputting the forecast accuracy of the first sequence data and the second sequence data, the forecast precipitation data within the forecast time, and the precipitation offset time based on the forecast model comprises the following steps: The first sequence data and the second sequence data include the same number of time series, and the absolute value of the difference between the historical forecast precipitation data and the historical actual precipitation data at the same position in the time series is obtained. If the absolute value is less than or equal to a preset second threshold, it is determined that the historical forecast precipitation data is accurate, and the forecast accuracy rate R is calculated based on a first formula, and the first formula is: ,in, is the accurate number of the historical forecast precipitation data in the first sequence of data, is the number of the first sequence data that is not 0 and the corresponding second sequence data that is 0, is the number of the second sequence data which is not 0 and corresponds to the first sequence data which is 0; A preset number of sequence data are extracted in reverse order from the first sequence data and the second sequence data, respectively defined as first comparison data and second comparison data, and the i-th predicted precipitation data within the prediction time is generated based on the prediction model and the second formula. , the second formula is: , where n is the preset number, is the i-th prediction result output by the prediction model, is the i-th first comparison data, is the i-th second comparison data; Obtain discontinuous points of time characteristics in the time series, split the first sequence data and the second sequence data into multiple first data groups based on the discontinuity points, and calculate the precipitation offset time M based on a third formula, where the third formula is: , where J is the number of the first data groups, is the first time when the second sequence data in the jth first data group is not 0, It is the time when the first sequence data in the j-th first data group is not 0.

3. A method for predicting waterlogging in a substation based on a neural network according to claim 2, characterized in that: The future precipitation data and future precipitation time of the target area within the prediction time are obtained based on the following steps: Generate the i-th future precipitation data of the target area within the prediction time based on the fourth formula , the fourth formula is: ,in, For the i-th future forecast data, the precipitation offset time is set to the future precipitation time. If the precipitation offset time is greater than 0, the predicted precipitation time is the start time of the future forecast data plus the precipitation offset time. Otherwise, the predicted precipitation time is the start time of the future forecast data minus the precipitation offset time.

4. A substation waterlogging prediction system based on a neural network, used to implement a substation waterlogging prediction method based on a neural network as claimed in any one of claims 1 to 3, characterized in that: The system includes the following modules: A data prediction module, which obtains a first sequence of historical forecast precipitation data and a second sequence of historical actual precipitation data of a target area in a preset historical time period based on a preset first time interval, and establishes a prediction model, wherein the prediction model outputs a prediction accuracy, predicted precipitation data within a prediction time, and a precipitation offset time based on the first sequence of data and the second sequence of data; The image processing module acquires the water accumulation distribution image corresponding to the target area based on each of the second sequence data and a preset second time interval, sets multiple first numerical intervals, generates multiple precipitation intensities based on the first numerical intervals, acquires each precipitation intensity sequence of the second sequence data based on the precipitation intensity, establishes a neural network model, inputs each of the water accumulation distribution images into the neural network model based on the precipitation intensity sequence, generates a water accumulation image sequence corresponding to each of the precipitation intensity sequences, and acquires the water accumulation level of each of the precipitation intensity sequences based on the water accumulation image sequence; starts the camera unit of the target area at the start time of each of the second sequence data, the camera unit acquires the water accumulation distribution image of the target area at every preset second time interval, calculates any The time difference corresponding to two adjacent historical actual precipitation data is determined. If the time difference is greater than or equal to a preset third time interval, the two historical actual precipitation data are split to generate a plurality of subsequence data from the second sequence data, the precipitation intensity of each subsequence data is obtained to generate a corresponding precipitation intensity sequence, each subsequence data of the same precipitation intensity sequence is grouped into a group to generate a plurality of small group sequences, the water accumulation distribution image corresponding to each subsequence data contained in the small group sequence is obtained, and the images are input into the neural network model, the neural network model is trained to learn and recognize the edge features of each water accumulation distribution image of the same arrangement number, and each standard water accumulation distribution image corresponding to the precipitation intensity sequence is output, and each standard water accumulation distribution image is set as the water accumulation image sequence; Acquire the edge features of any two adjacent standard waterlogging distribution images, establish a coordinate system based on the standard waterlogging distribution images, locate the pixel coordinates corresponding to each edge feature, if the pixel coordinates in any two adjacent standard waterlogging distribution images are different, connect the pixel coordinates to generate an edge graph, calculate the pixel area of ​​the edge graph, define the standard waterlogging distribution image with the largest pixel area as a maximum waterlogging image, and generate the waterlogging level of the target area based on the maximum waterlogging image and the precipitation intensity sequence; A data output module, used to obtain future forecast data of the target area within the forecast time, and obtain future precipitation data and future precipitation time of the target area within the forecast time based on the future forecast data, the forecast accuracy, the precipitation offset time and the forecast precipitation data; An image generation module generates a variation sequence of each precipitation intensity based on the future precipitation data, so as to obtain the corresponding waterlogging distribution image in the waterlogging image sequence, and generates a real-time waterlogging image set, obtains the waterlogging time of the target area based on the variation sequence and the real-time waterlogging image set, and generates the waterlogging level of the target area within the predicted time based on the real-time waterlogging image set; Based on the change sequence, similar sequences are selected from each of the precipitation intensity sequences, and the waterlogging image sequence corresponding to the similar sequences is set as the real-time waterlogging image set; the time length of the change sequence is obtained, and the number of pixel points of the edge features of each image in the real-time waterlogging image set is obtained in turn, and a first curve is generated by fitting based on the time length and the number of each pixel point, and the first inflection point and the last inflection point of the first curve are obtained, and the time point corresponding to the first inflection point in the time length is set as the waterlogging start time of the target area, and the time point corresponding to the last inflection point in the time length is set as the waterlogging end time of the target area, and the waterlogging time is the waterlogging end time minus the waterlogging start time; if each of the precipitation intensity sequences contains the same numerical sequence as the change sequence, the corresponding precipitation intensity sequence is set as the similar sequence, otherwise, the same numerical sequence is extracted from each precipitation intensity sequence. A first screening group is generated by combining the sequences of the change sequence and each precipitation intensity sequence in the first screening group, obtaining a first overlap degree between the change sequence and each precipitation intensity sequence in the first screening group, obtaining each precipitation intensity sequence whose first overlap degree is greater than or equal to a preset first threshold value, and setting them as a second screening group, obtaining the precipitation intensity of each precipitation intensity sequence at the waterlogging start time, the maximum waterlogging image, and the waterlogging end time in the second screening group and defining them as each comparison point, splitting the precipitation intensity sequence into a plurality of sub-intensity sequences based on each of the comparison points and numbering them in sequence, wherein the sub-intensity sequence includes the precipitation intensity and the number, obtaining a second overlap degree between each of the sub-intensity sequences with the same number and the change sequence, combining the sub-intensity sequences corresponding to the maximum second overlap degree in each of the numbers to generate the similarity sequence, and extracting the standard waterlogging distribution image from the precipitation intensity sequences corresponding to each of the sub-intensity sequences to combine and generate the real-time waterlogging image set.

Images