Power distribution network risk prediction system and method based on big data

Abstract

The invention relates to a big data-based power distribution network risk prediction system and method. The system comprises a comprehensive data acquisition module, an image recognition processing module, a risk prediction model construction module, a judgment and decision module and a risk prediction module. The prediction result is more accurate and reliable.

Description

Technical Field

[0001] The present invention belongs to the technical field of early warning of electric power engineering, and relates to a distribution network risk prediction system and method, in particular to a distribution network risk prediction system and method based on big data. Background Art

[0002] A distribution network refers to an electric power grid that receives electrical energy from a transmission network or a regional power plant and distributes it locally or step by step according to voltage to various types of users through distribution facilities.

[0003] There are many power grid devices in the distribution network, and these power grid devices are in the wild environment. Generally speaking, power grid equipment failures only occur in extreme weather conditions, such as strong winds, heavy rains, etc. In the existing technology, the distribution network is combined with the meteorological system to judge the environment in which the power grid equipment is located. Based on this information, it is determined whether the power grid equipment will fail. It is crucial to determine whether the established risk prediction model is accurate.

[0004] Although the current meteorological system can predict the weather very accurately, all meteorological data represent weather conditions in a large range, while power grid equipment is specific to a certain location. Therefore, the actual surrounding environment of power grid equipment may be different. This leads to a lack of accuracy in the existing risk prediction models, at least in terms of specific location and power grid equipment.

[0005] Therefore, how to develop a distribution network risk prediction system and method based on big data with accurate prediction results and strong reliability is a technical problem that needs to be urgently solved by technical personnel in this field.

[0006] After searching, no prior art documents identical or similar to the present invention were found. Summary of the invention

[0007] The purpose of the present invention is to overcome the deficiencies of the prior art and to propose a distribution network risk prediction system and method based on big data with accurate prediction results and strong reliability.

[0008] The present invention solves the practical problem by adopting the following technical solutions:

[0009] A distribution network risk prediction system based on big data, including a comprehensive data acquisition module, an image recognition processing module, a risk prediction model building module, a judgment and decision module and a risk prediction module;

[0010] The comprehensive data acquisition module is used to obtain fault data and corresponding meteorological data of power grid equipment, and obtain corresponding image data through monitoring equipment;

[0011] The image recognition processing module obtains corresponding image data based on the monitoring device and uses a convolutional neural network to perform recognition to obtain feature data;

[0012] The risk prediction model building module is used to establish a risk prediction model based on the fault data of the power grid equipment and the corresponding meteorological data and characteristic data obtained by the comprehensive data acquisition module, and the risk prediction model includes a simple prediction model and a deep prediction model;

[0013] The judgment and decision module is used to obtain real-time power grid data, real-time meteorological data and real-time image data, and establish input decisions of the risk prediction model based on the three;

[0014] The risk prediction module is used to select a risk prediction model according to an input decision and retrieve corresponding data to input into it for risk prediction.

[0015] Moreover, the output end of the comprehensive data acquisition module is connected to the image recognition processing module, and the comprehensive data acquisition module transmits the collected image data to the image recognition processing module, and the image recognition processing module needs to recognize these images;

[0016] The output end of the image recognition processing module and the output end of the comprehensive data acquisition module are both connected to the risk prediction model construction module. The risk prediction model construction module will receive the fault data and meteorological data sent by the comprehensive data acquisition module, and will also receive the feature data sent by the image recognition processing module. After all the data are available, the risk prediction model is constructed;

[0017] The judgment and decision module first needs to rely on the monitoring equipment in the comprehensive data acquisition module and the related equipment to obtain the data of the power grid and the weather. According to different data acquisition strategies, the comprehensive data acquisition module will transmit these data to the judgment and decision module, and then the judgment and decision module will process these data and determine which risk prediction model these data can be input into;

[0018] The output end of the judgment and decision module and the output end of the risk prediction model construction module are both connected to the risk prediction module. The risk prediction module will store the risk prediction model output by the risk prediction model construction module. Then, after the judgment and decision module obtains the input decision based on the collected data, it will transmit the data to the risk prediction module, select the corresponding risk prediction model based on the input decision, and then complete the risk prediction of the power grid failure.

[0019] A distribution network risk prediction method based on big data, such as Figure 2 As shown, the following steps are included:

[0020] Step 1: Obtain fault data of power grid equipment and corresponding meteorological data, and obtain corresponding image data through monitoring equipment;

[0021] Step 2: Using a convolutional neural network to identify image data to obtain feature data;

[0022] Step 3: Establish a risk prediction model based on the obtained fault data, meteorological data and characteristic data, wherein the risk prediction model includes a simple prediction model and a deep prediction model;

[0023] Step 4: Obtain real-time power grid data, real-time meteorological data, and real-time image data, and establish input decisions for the risk prediction model based on the three;

[0024] Step 5: Select a risk prediction model based on the input decision and retrieve the corresponding data to input into it for risk prediction.

[0025] Moreover, the specific steps of step 1 include:

[0026] (1) Using the SCADA system to collect status data when equipment fails, i.e., the failure data and corresponding meteorological data;

[0027] (2) Determine the specific monitoring device to acquire image data through location information.

[0028] Moreover, the specific steps of step 2 include:

[0029] (1) Determine the time point when the fault occurs based on the fault data, thereby determining a frame in the image data;

[0030] (2) Input the three frames into the trained convolutional neural network model for recognition;

[0031] (3) performing dot matrix processing on the features in the feature image output by the model to obtain a whitened image;

[0032] (4) Analyzing the whitened image with many points, the density data of the points can be determined by comparing the number of points, and the slope data of the points can be determined according to the distribution of the points in the whitened image;

[0033] (5) The accuracy of the image data is determined by comparing the whitened image;

[0034] (6) When the whitened image is correctly compared, the density data and slope data of the specified whitened image are selected and integrated to obtain feature data.

[0035] Moreover, the specific steps of step 3 include:

[0036] (1) Establish a simple prediction model using fault data and meteorological data;

[0037] (2) Use the above fault data, meteorological data and feature data to establish a deep prediction model.

[0038] Moreover, the specific steps of step 4 include:

[0039] (1) Obtain real-time power grid data and corresponding real-time meteorological data and real-time image data;

[0040] (2) dividing and processing the real-time meteorological data according to the set value; the meteorological data is divided into large, medium and small levels, and when the meteorological data is of a small level, generating an input decision for inputting the real-time power grid data, real-time meteorological data and real-time image data into a simple prediction model;

[0041] (3) When the meteorological data is greater than or equal to the medium level, the accuracy of the meteorological data classification is judged based on the feature data in the real-time image data;

[0042] (4) When the accuracy judgment is passed, an input decision is generated to input the real-time power grid data, real-time meteorological data and real-time image data into the deep prediction model. When the accuracy judgment is not passed, an input decision is generated to input the real-time power grid data, real-time meteorological data and real-time image data into the simple prediction model.

[0043] Moreover, the specific steps of step 5 include:

[0044] (1) Selecting the corresponding risk prediction model based on the input decision;

[0045] (2) According to the input decision, the corresponding data is retrieved and input into the selected risk prediction model to perform risk prediction.

[0046] Advantages and beneficial effects of the present invention:

[0047] 1. The present invention proposes a distribution network risk prediction system and method based on big data. By utilizing the fact that corresponding monitoring equipment is installed on power grid equipment, meteorological data is associated with image data around the power grid equipment. A large amount of image data is used to train a convolutional neural network to obtain a risk prediction model. Meteorological data is used as a standard definition for identification, thereby discovering the correlation between the real environment and the failure of power grid equipment. The risk prediction model established by this method is more accurate.

[0048] 2. The present invention utilizes the fact that existing power grid equipment is generally equipped with monitoring equipment, collects image data from the power grid equipment site, and then associates it with the acquired meteorological data. The image data is used to train a convolutional neural network and establish a risk prediction model. The meteorological data is used as a reference for convolutional neural network recognition, focusing on the actual environment of the power grid equipment reflected by the image data. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 It is a system configuration diagram of the present invention;

[0050] Figure 2 It is a process flow chart of the present invention. DETAILED DESCRIPTION

[0051] The embodiments of the present invention are further described in detail below with reference to the accompanying drawings:

[0052] A distribution network risk prediction system based on big data, including a comprehensive data acquisition module, an image recognition processing module, a risk prediction model building module, a judgment and decision module and a risk prediction module;

[0053] The comprehensive data acquisition module is used to obtain fault data and corresponding meteorological data of power grid equipment, and obtain corresponding image data through monitoring equipment;

[0054] The image recognition processing module obtains corresponding image data based on the monitoring device and uses a convolutional neural network to perform recognition to obtain feature data;

[0055] The risk prediction model building module is used to establish a risk prediction model based on the fault data of the power grid equipment and the corresponding meteorological data and characteristic data obtained by the comprehensive data acquisition module, and the risk prediction model includes a simple prediction model and a deep prediction model;

[0056] The judgment and decision module is used to obtain real-time power grid data, real-time meteorological data and real-time image data, and establish input decisions of the risk prediction model based on the three;

[0057] The risk prediction module is used to select a risk prediction model according to an input decision and retrieve corresponding data to input into it for risk prediction.

[0058] In this embodiment, the output end of the comprehensive data acquisition module is connected to the image recognition processing module, and the comprehensive data acquisition module transmits the collected image data to the image recognition processing module, and the image recognition processing module needs to recognize these images;

[0059] The output end of the image recognition processing module and the output end of the comprehensive data acquisition module are both connected to the risk prediction model construction module. The risk prediction model construction module will receive the fault data and meteorological data sent by the comprehensive data acquisition module, and will also receive the feature data sent by the image recognition processing module. After all the data are available, the risk prediction model is constructed;

[0060] The judgment and decision module first needs to rely on the monitoring equipment in the comprehensive data acquisition module and the related equipment to obtain the data of the power grid and the weather. According to different data acquisition strategies, the comprehensive data acquisition module will transmit these data to the judgment and decision module, and then the judgment and decision module will process these data and determine which risk prediction model these data can be input into;

[0061] The output end of the judgment and decision module and the output end of the risk prediction model construction module are both connected to the risk prediction module. The risk prediction module will store the risk prediction model output by the risk prediction model construction module. Then, after the judgment and decision module obtains the input decision based on the collected data, it will transmit the data to the risk prediction module, select the corresponding risk prediction model based on the input decision, and then complete the risk prediction of the power grid failure.

[0062] The functions of each module in the system are further explained below:

[0063] 1. Comprehensive data collection module

[0064] Obtain fault data and corresponding meteorological data of power grid equipment, and obtain corresponding image data through monitoring equipment.

[0065] The SCADA system is used to collect status data of power grid equipment, such as switch status, current and voltage values, and combined with some set sensors, the temperature, sound and other characteristics of the power grid equipment can be collected. These data become the fault data.

[0066] When a power grid device fails, the fault data and corresponding meteorological data can be obtained. The specific monitoring device can be determined through the location information. The image data, fault data and meteorological data also need to determine the corresponding time nodes.

[0067] 2. Image recognition processing module

[0068] Based on the image data, convolutional neural network is used to identify and obtain feature data;

[0069] The steps to achieve this mainly include:

[0070] 1) Image data preprocessing

[0071] The time node when the fault occurred is determined according to the fault data, thereby determining a frame in the image data, and at the same time obtaining the two frames before and after according to the set time threshold.

[0072] 2) Input the three frames into the trained convolutional neural network model for recognition

[0073] The convolutional neural network model is trained through images obtained by a large number of monitoring devices. Of course, these images contain some meteorological features. After being processed by filters and pooling layers, feature images are finally output from the output layer.

[0074] 3) Feature analysis

[0075] The features in the feature image are dotted, that is, the individual features in the feature image are replaced and covered in the form of dots, and other contents are cleared after removing these features.

[0076] 4) Scalability calculation analysis

[0077] By using the above-mentioned whitened image with many points for analysis, the density data of the points can be determined by comparing the number of points, and the movement trend of the points can be determined according to the distribution of the points in the whitened image, so as to discover the linear movement law of the points, and thereby determine the slope data about the point movement trend.

[0078] 4) Frame Screening

[0079] Based on the whitened image obtained above, three frames are compared to determine the accuracy of the image when the fault occurs. The density data and slope data of the designated whitened image can be used as a reference, and the two are characteristic data.

[0080] 3. Establish a risk prediction model

[0081] A risk prediction model is established based on the obtained fault data, meteorological data and characteristic data, and the risk prediction model includes a simple prediction model and a deep prediction model.

[0082] The risk prediction model mainly includes two parts: one is a simple prediction model based on fault data and meteorological data, and the other is a deep prediction model based on feature data, fault data and meteorological data.

[0083] Through time and space matching, data preprocessing, feature extraction and correlation analysis, the connection between power grid data and meteorological data can be established. The steps to achieve this include:

[0084] 1) Fusion and preprocessing of power grid data and meteorological data

[0085] First, ensure the temporal and spatial consistency of power grid data and meteorological data to facilitate analysis on the same dimension. Specifically, if a power grid device fails in a certain time period, find the meteorological data of a certain meteorological station in this time period to discover the correlation; then remove the outliers in the two data to ensure the quality of the data.

[0086] 2) Feature extraction

[0087] Extract key features from meteorological and power grid data.

[0088] 3) Correlation analysis

[0089] Statistical methods were used to evaluate the associations between the extracted features.

[0090] 4) Model construction

[0091] Select a suitable model for training. The training method is to input various data into the model, update the model parameters through back propagation, and repeat the process continuously until the model converges. There are many types of models, such as linear models, decision tree models, neural network models, etc.

[0092] The above method is also used to establish a correlation model among feature data, fault data and meteorological data. The feature data is mainly used to verify the accuracy of meteorological data. The density data in the feature data can be directly related to the amount of rain and snow, and the slope data can be directly related to the current wind speed.

[0093] 4. Judgment and decision-making module

[0094] Obtain real-time power grid data, real-time meteorological data, and real-time image data, and use the three to establish the input decision of the risk prediction model

[0095] 1) Obtain real-time power grid data and corresponding real-time meteorological data and real-time image data

[0096] 2) Divide and process real-time meteorological data according to set values

[0097] 3) Verify the result of the segmentation process using the feature data in the real-time image data

[0098] 4) Generate input decisions based on the results of the division and verification

[0099] Real-time meteorological data indicates that the environment in which the power grid equipment is located contains some meteorological characteristics. Therefore, real-time meteorological data can be used as a reference for content recognition of real-time image data, and the accuracy of real-time meteorological data can be judged by the characteristic data of real-time image data.

[0100] 4) Determine the data input decision based on the above processing process

[0101] 5. Risk prediction module

[0102] A risk prediction model is selected based on the input decision and the corresponding data is retrieved and input into it for risk prediction.

[0103] A distribution network risk prediction method based on big data, such as Figure 2 As shown, the following steps are included:

[0104] Step 1: Obtain fault data of power grid equipment and corresponding meteorological data, and obtain corresponding image data through monitoring equipment;

[0105] The specific steps of step 1 include:

[0106] (1) Using the SCADA system to collect status data when equipment fails, i.e., the failure data and corresponding meteorological data;

[0107] (2) Determine the specific monitoring device to acquire image data through location information.

[0108] Step 2: Using a convolutional neural network to identify image data to obtain feature data;

[0109] The specific steps of step 2 include:

[0110] (1) Determine the time point when the fault occurs based on the fault data, thereby determining a frame in the image data;

[0111] (2) Input the three frames into the trained convolutional neural network model for recognition;

[0112] (3) performing dot matrix processing on the features in the feature image output by the model to obtain a whitened image;

[0113] (5) Using the above-mentioned whitened image with many points for analysis, the density data of the points can be determined by comparing the number of points, and the slope data of the points can be determined according to the distribution of the points in the whitened image;

[0114] (5) The accuracy of the image data is determined by comparing the whitened image;

[0115] (6) When the whitened image is correctly compared, the density data and slope data of the specified whitened image are selected and integrated to obtain feature data.

[0116] Step 3: Establish a risk prediction model based on the obtained fault data, meteorological data and characteristic data, wherein the risk prediction model includes a simple prediction model and a deep prediction model;

[0117] The specific steps of step 3 include:

[0118] (1) Establish a simple prediction model using fault data and meteorological data;

[0119] (2) Use the above fault data, meteorological data and feature data to establish a deep prediction model.

[0120] Step 4: Obtain real-time power grid data, real-time meteorological data, and real-time image data, and establish input decisions for the risk prediction model based on the three;

[0121] The specific steps of step 4 include:

[0122] (5) Obtain real-time power grid data and corresponding real-time meteorological data and real-time image data;

[0123] (6) dividing and processing the real-time meteorological data according to the set value; the meteorological data is divided into large, medium and small levels, and when the meteorological data is of a small level, generating an input decision for inputting the real-time power grid data, real-time meteorological data and real-time image data into a simple prediction model;

[0124] (7) When the meteorological data is greater than or equal to the medium level, the accuracy of the meteorological data classification is determined based on the feature data in the real-time image data;

[0125] (8) When the accuracy judgment is passed, an input decision is generated to input the real-time power grid data, real-time meteorological data and real-time image data into the deep prediction model. When the accuracy judgment is not passed, an input decision is generated to input the real-time power grid data, real-time meteorological data and real-time image data into the simple prediction model.

[0126] Step 5: Select a risk prediction model based on the input decision and retrieve the corresponding data to input into it for risk prediction.

[0127] The specific steps of step 5 include:

[0128] (3) Selecting the corresponding risk prediction model based on the input decision;

[0129] (4) According to the input decision, the corresponding data is retrieved and input into the selected risk prediction model to perform risk prediction.

[0130] like Figure 1 As shown, the embodiment of the present invention provides a distribution network risk prediction system based on big data, which mainly includes five modules: comprehensive data collection, image recognition processing, risk prediction model establishment, judgment and decision-making, and risk prediction;

[0131] The comprehensive data acquisition module is mainly used to obtain the fault data of the power grid equipment and the corresponding meteorological data, and obtain the corresponding image data through the monitoring equipment. Here is also the corresponding implementation method.

[0132] Currently, the SCADA (Supervisory Control and Data Acquisition) system is used in power grids to monitor and control power grid equipment, and collect equipment status data, such as switch status, current and voltage values. In combination with some installed sensors, the temperature, sound and other characteristics of power grid equipment can be collected, and changes in these data can be determined when power grid equipment fails.

[0133] When a power grid device fails, the fault data and corresponding meteorological data can be obtained (the association between meteorology and power grid is already an existing technology, so the acquisition of the two is easy to achieve). The monitoring equipment is set up around the power grid equipment to monitor the environment. The specific monitoring equipment can be determined by location information, so the image data of the monitoring equipment and the fault data can also be corresponded. Of course, the image data, fault data and meteorological data also need to determine the corresponding time nodes. The power grid equipment can specifically be transformers, distribution boxes and other equipment. These power grid equipment are all key nodes on the distribution network, and the monitoring equipment can be existing monitoring cameras, which are mainly used to monitor the environment around the power grid equipment and also have a warning function.

[0134] The image recognition processing module is mainly used to obtain feature data based on image data using a convolutional neural network;

[0135] Image data is specifically a video of a period of time. It is difficult to use the complete video to identify the meteorological information contained in it. Therefore, a specific image can be obtained from the video for identification. By analyzing the features in the image, meteorological feature data, such as the number and density of raindrops, wind speed, etc., can be obtained. The implementation steps mainly include:

[0136] 1) Image data preprocessing

[0137] The time node when the fault occurred is determined according to the fault data, so as to determine a frame in the image data, and the two frames before and after are obtained according to the set time threshold. Assuming that the time threshold is 10s, the three frames obtained are the image when the fault occurred, the image 10s before the fault occurred, and the image 10s after the fault occurred.

[0138] 2) Input the three frames into the trained convolutional neural network model for recognition

[0139] The convolutional neural network model is trained through images obtained by a large number of monitoring devices. Of course, these images contain some meteorological features, such as rain and snow. After the frame is input into the convolutional neural network model, the filter slides on the image to extract local features. The pooling layer reduces the dimension of the image features so that only important features are retained. The output layer outputs a feature image with meteorological features. The meteorological features here are mainly about rain and snow.

[0140] 3) Feature analysis

[0141] The features in the feature image are dotted, that is, the individual features in the feature image are replaced and covered in the form of dots, and other contents are cleared except these features. For example, on a rainy day, the feature image contains many falling raindrops, and on a snowy day, the feature image contains many falling snowflakes. The image after clearing is a whitened image with many dots.

[0142] 4) Scalability calculation analysis

[0143] By using the above-mentioned whitened image with many points for analysis, the density data of the points can be determined by comparing the number of points, and the movement trend of the points can be determined according to the distribution of the points in the whitened image, so as to discover the linear movement law of the points, and use this to determine the slope data about the movement trend of the points. For example, if the characteristic image is an image of a snowstorm, then the image must contain many snowflakes, and the wind will drive these snowflakes to move in an upward regular direction. Correspondingly, in the whitened image, there are multiple points with linear arrangement and distribution. Connecting the arranged and distributed points will obtain multiple parallel lines, and the slope data can be determined by the parallel lines. Of course, the appearance of slope data requires certain conditions, and it must be when the wind force reaches a certain level to appear. Of course, this wind force level is not enough to affect the power grid equipment.

[0144] 4) Frame Screening

[0145] Based on the whitened image obtained above, three frames are compared to determine the accuracy of the image when the fault occurs, so that the density data and slope data of which whitened image can be used as a reference can be selected, which are the characteristic data mentioned above.

[0146] The risk prediction model is established to establish the risk prediction model according to the obtained fault data, meteorological data and characteristic data, and the risk prediction model includes a simple prediction model and a deep prediction model.

[0147] The risk prediction model mainly includes two parts: one is a simple prediction model based on fault data and meteorological data, and the other is a deep prediction model based on feature data, fault data and meteorological data.

[0148] Through time and space matching, data preprocessing, feature extraction and correlation analysis, the connection between power grid data and meteorological data can be established. The steps to achieve this include:

[0149] 1) Fusion and preprocessing of power grid data and meteorological data

[0150] First, ensure the temporal and spatial consistency of power grid data and meteorological data to facilitate analysis in the same dimension. Specifically, if a power grid device fails in a certain time period, find the meteorological data of a certain meteorological station in this time period to explore the correlation; then remove the outliers in the two data to ensure the quality of the data, for example, correct some abnormal temperature and wind speed values ​​in the meteorological data; convert data from different sources into the same format and unit, for example, convert wind speed into m / s and convert current in power grid data into amperes.

[0151] 2) Feature extraction

[0152] Extract key features from meteorological data and power grid data, such as temperature, humidity, wind speed, current, voltage, power, equipment temperature, etc.

[0153] 3) Correlation analysis

[0154] Statistical methods are used to evaluate the association between the extracted features, which can be achieved through methods such as Pearson correlation coefficient and Spearman rank correlation coefficient, and then a regression model is used to establish the relationship between the two.

[0155] 4) Model construction

[0156] Select a suitable model for training. The training method is to input various data into the model, update the model parameters through back propagation, and repeat the process continuously until the model converges. There are many types of models, such as linear models, decision tree models, neural network models, etc.

[0157] The above method is also used to establish a correlation model among feature data, fault data and meteorological data. The main function of feature data is to verify the accuracy of meteorological data (especially in rainy and snowy weather). The density data in the feature data can be directly related to the amount of rain and snow, and the slope data can be directly related to the current wind speed. In summary, the simple prediction model based on fault data and meteorological data can respond to more conventional situations, while the deep prediction model of feature data, fault data and meteorological data can respond to situations in extreme weather.

[0158] The judgment and decision module is used to obtain real-time power grid data, real-time meteorological data and real-time image data, and establish the input decision of the risk prediction model based on the three.

[0159] 1) Obtain real-time power grid data and corresponding real-time meteorological data and real-time image data

[0160] These data acquisition methods have been mentioned in the previous article and will not be repeated here.

[0161] 2) Divide and process real-time meteorological data according to set values

[0162] Specifically, the meteorological conditions can be divided into large, medium and small levels, and the result of the division process can be used to determine which risk prediction model the three data should be input into.

[0163] 3) Determine the accuracy of meteorological data classification based on the characteristic data in real-time image data

[0164] 4) Generate input decisions based on the above division results and accuracy judgment structure

[0165] Real-time meteorological data indicates that the environment in which the power grid equipment is located contains some meteorological characteristics. Therefore, real-time meteorological data can be used as a reference for real-time image data content recognition. For example, when the real-time meteorological data shows that the area is experiencing heavy rain, the real-time image data will be retrieved first for verification. When the weather displayed by the real-time meteorological data is verified to be accurate through the feature data in the real-time image data, the above three data can be input into the deep prediction model. Therefore, the feature data obtained based on the real-time meteorological data is very important. It can be used as a basis for judgment. The feature data is used to judge the accuracy of the weather. If the weather is normal, such as light rain, only the simple prediction model is enabled to predict risk faults.

[0166] 4) Determine the data input decision based on the above processing process

[0167] The risk prediction module is used to select a risk prediction model based on the input decision and retrieve the corresponding data input into it for risk prediction.

[0168] For the convenience of subsequent elaboration, an example is given for illustration. For example, the acquired real-time power grid data and the corresponding real-time meteorological data and real-time image data are displayed in a rainy day. Then the level of the rainy weather can be judged by the real-time data judgment module. When the rainy weather is below the intermediate level, the real-time power grid data and the real-time meteorological data will be input into the prediction model established by the fault data and the meteorological data. Because the natural weather at this time does not meet the requirements for more accurate prediction, it can be simplified to reduce the amount of data processing. When it is concluded from the real-time meteorological data that the rainy weather is intermediate or above, it means that the power grid equipment is in a more severe environment. First, the correspondence between the real-time image data and the real-time meteorological data is determined according to the real-time data module. If the real-time meteorological data can be verified in the real-time image data, then the real-time power grid data, the real-time meteorological data and the image data will be input into the prediction model established by the feature data, the fault data and the meteorological data. If the real-time image data cannot verify the real-time meteorological data, then the real-time power grid data and the real-time meteorological data will be input into the prediction model established by the fault data and the meteorological data. Of course, the rain level used in the embodiment is just a reference, which can be rainfall, snowfall, etc.

[0169] like Figure 2 As shown, the embodiment of the present invention also provides a distribution network risk prediction method based on big data, the steps of which include:

[0170] Step 1: Obtain fault data and corresponding meteorological data of power grid equipment, and obtain corresponding image data through monitoring equipment;

[0171] Step 2: Using a convolutional neural network to identify image data to obtain feature data;

[0172] Step 3: establishing a risk prediction model based on the obtained fault data, meteorological data and characteristic data, wherein the risk prediction model includes a simple prediction model and a deep prediction model;

[0173] Step 4: Obtain real-time power grid data, real-time meteorological data, and real-time image data, and establish input decisions for the risk prediction model based on the three;

[0174] Step 5: Select a risk prediction model based on the input decision and retrieve the corresponding data to input into it for risk prediction.

[0175] The purpose is to take advantage of the fact that power grid equipment will be equipped with corresponding monitoring equipment, associate meteorological data with image data around the power grid equipment, and use a large amount of image data to train a convolutional neural network to obtain a risk prediction model. Meteorological data is used as the standard definition for identification, so as to discover the correlation between the real environment and the failure of power grid equipment. The risk prediction model established by this method is more accurate.

[0176] For example, the specific methods for implementing step 2 include:

[0177] Determine the time point when the fault occurs based on the fault data, thereby determining a frame in the image data, and at the same time obtain the two frames before and after according to the set time threshold;

[0178] Input the three frames into the trained convolutional neural network model for recognition;

[0179] The features in the feature image are subjected to dot matrix processing to obtain a whitened image;

[0180] By using the above-mentioned whitened image with many points for analysis, the density data of the points can be determined by comparing the number of points, and the movement trend of the points can be determined according to the distribution of the points in the whitened image, so as to discover the linear movement law of the points, and thereby determine the slope data about the point movement trend.

[0181] The three frames are compared based on the whitened image obtained above.

[0182] It should be emphasized that the embodiments of the present invention are illustrative rather than restrictive. Therefore, the present invention includes but is not limited to the embodiments described in the specific implementation modes. Any other implementation modes derived by those skilled in the art based on the technical solutions of the present invention also fall within the scope of protection of the present invention.

Claims

1. A distribution network risk prediction system based on big data, characterized by: It includes comprehensive data acquisition module, image recognition processing module, risk prediction model building module, judgment and decision module and risk prediction module; The comprehensive data acquisition module is used to obtain fault data and corresponding meteorological data of power grid equipment, and obtain corresponding image data through monitoring equipment; The image recognition processing module obtains corresponding image data based on the monitoring device and uses a convolutional neural network to perform recognition to obtain feature data; The risk prediction model building module is used to establish a risk prediction model based on the fault data of the power grid equipment and the corresponding meteorological data and characteristic data obtained by the comprehensive data acquisition module, and the risk prediction model includes a simple prediction model and a deep prediction model; The judgment and decision module is used to obtain real-time power grid data, real-time meteorological data and real-time image data, and establish input decisions of the risk prediction model based on the three; The risk prediction module is used to select a risk prediction model according to an input decision and retrieve corresponding data to input into it for risk prediction.

2. A distribution network risk prediction system based on big data according to claim 1, characterized in that: The output end of the comprehensive data acquisition module is connected to the image recognition processing module. The comprehensive data acquisition module transmits the collected image data to the image recognition processing module, and the image recognition processing module needs to recognize these images; The output end of the image recognition processing module and the output end of the comprehensive data acquisition module are both connected to the risk prediction model construction module. The risk prediction model construction module will receive the fault data and meteorological data sent by the comprehensive data acquisition module, and will also receive the feature data sent by the image recognition processing module. After all the data are available, the risk prediction model is constructed; The judgment and decision module first needs to rely on the monitoring equipment in the comprehensive data acquisition module and the related equipment to obtain the data of the power grid and the weather. According to different data acquisition strategies, the comprehensive data acquisition module will transmit these data to the judgment and decision module, and then the judgment and decision module will process these data and determine which risk prediction model these data can be input into; The output end of the judgment and decision module and the output end of the risk prediction model construction module are both connected to the risk prediction module. The risk prediction module will store the risk prediction model output by the risk prediction model construction module. Then, after the judgment and decision module obtains the input decision based on the collected data, it will transmit the data to the risk prediction module, select the corresponding risk prediction model based on the input decision, and then complete the risk prediction of the power grid failure.

3. A distribution network risk prediction method based on big data, characterized by: The following steps are involved: Step 1: Obtain fault data of power grid equipment and corresponding meteorological data, and obtain corresponding image data through monitoring equipment; Step 2: Using a convolutional neural network to identify image data to obtain feature data; Step 3: Establish a risk prediction model based on the obtained fault data, meteorological data and characteristic data, wherein the risk prediction model includes a simple prediction model and a deep prediction model; Step 4: Obtain real-time power grid data, real-time meteorological data, and real-time image data, and establish input decisions for the risk prediction model based on the three; Step 5: Select a risk prediction model based on the input decision and retrieve the corresponding data to input into it for risk prediction.

4. The method for predicting distribution network risk based on big data according to claim 3 is characterized in that: The specific steps of step 1 include: (1) Using the SCADA system to collect status data when equipment fails, i.e., the failure data and corresponding meteorological data; (2) Determine the specific monitoring device to acquire image data through location information.

5. The method for predicting distribution network risk based on big data according to claim 3 is characterized in that: The specific steps of step 2 include: (1) Determine the time point when the fault occurs based on the fault data, thereby determining a frame in the image data; (2) Input the three frames into the trained convolutional neural network model for recognition; (3) performing dot matrix processing on the features in the feature image output by the model to obtain a whitened image; (4) Analyzing the whitened image with many points, the density data of the points can be determined by comparing the number of points, and the slope data of the points can be determined according to the distribution of the points in the whitened image; (5) The accuracy of the image data is determined by comparing the whitened image; (6) When the whitened image is correctly compared, the density data and slope data of the specified whitened image are selected and integrated to obtain feature data.

6. The method for predicting distribution network risk based on big data according to claim 3 is characterized by: The specific steps of step 3 include: (1) Establish a simple prediction model using fault data and meteorological data; (2) Use the above fault data, meteorological data and feature data to establish a deep prediction model.

7. The method for predicting distribution network risk based on big data according to claim 3 is characterized by: The specific steps of step 4 include: (1) Obtain real-time power grid data and corresponding real-time meteorological data and real-time image data; (2) dividing and processing the real-time meteorological data according to the set value; the meteorological data is divided into large, medium and small levels, and when the meteorological data is of a small level, generating an input decision for inputting the real-time power grid data, real-time meteorological data and real-time image data into a simple prediction model; (3) When the meteorological data is greater than or equal to the medium level, the accuracy of the meteorological data classification is judged based on the feature data in the real-time image data; (4) When the accuracy judgment is passed, an input decision is generated to input the real-time power grid data, real-time meteorological data and real-time image data into the deep prediction model. When the accuracy judgment is not passed, an input decision is generated to input the real-time power grid data, real-time meteorological data and real-time image data into the simple prediction model.

8. The method for predicting distribution network risk based on big data according to claim 3 is characterized by: The specific steps of step 5 include: (1) Selecting the corresponding risk prediction model based on the input decision; (2) According to the input decision, the corresponding data is retrieved and input into the selected risk prediction model to perform risk prediction.

Images