A dual-channel CNN-based RPA automated archiving risk warning method

Abstract

The present invention discloses an RPA automated archiving risk sheet early warning method based on dual-path CNN, comprising: automatically logging into the OMS scheduling support system through multi-threaded network technology, and obtaining archived risk sheets in the system in real time; using dual-path CNN technology, artificial intelligence automatically compares the various information of the previously "archived" list to obtain a communication list of relevant feedback units; using relevant data transmission technology, the obtained communication list is transmitted to the back-end database server and an Excel file is generated; and the SMS voice interface is called to notify each unit. The present invention considers the intelligent reading and comparison technology of the archived list using a dual-path convolutional neural network, extracts features through the convolution layer and pooling layer, then performs feature mapping and combination, fusion and normalization, and classifies and outputs relevant information of the risk sheet, automatically compares and calculates to determine whether the risk sheet has been completed, thereby realizing intelligent archiving of risk sheets and improving service quality and efficiency.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent power data processing, and specifically relates to an RPA automatic archiving risk single early warning method based on dual-channel CNN. Background Art

[0002] With the rapid development of internet, artificial intelligence, and the Internet of Things (IoT) technologies in recent years, traditional power grids are rapidly transforming into modern, intelligent, and specialized ones. As power grid systems continue to evolve, the resulting data volume is also growing exponentially. This includes basic data related to the properties of power facilities like transformers and generators, real-time data during power grid operation and maintenance, and management data generated by daily power system management. Automated data processing is crucial for power systems. The power early warning risk system is a crucial component of ensuring the normal operation of the power system. It ensures that power departments can efficiently develop appropriate emergency response plans, promptly initiate relevant emergency response policies, and maintain the monitoring and management of power grid equipment. Early warning risk information is key data within this system, and the timely and accurate archiving of early warning risk information is a crucial task for power personnel today. Robotic Process Automation (RPA) is intelligent software that non-invasively connects various business systems and replaces human employees in executing tasks with clear rules. Based on a dual-channel CNN, RPA can intelligently compare the specific information between current warning risk tickets and archived warning risk tickets, automatically and accurately archive relevant unarchived risk tickets, generate communication methods for corresponding departments, and use SMS language interfaces to promptly notify relevant units. Recent developments in computer science have shown that artificial intelligence has played an increasingly important role in power grid systems. Leveraging dual-channel CNN networks to intelligently and dynamically process data and improve the real-time and adaptability of applications is a key research direction for the future. Summary of the Invention

[0003] In view of this, the purpose of the present invention is to provide an RPA automatic archiving risk sheet warning method based on dual-channel CNN, which can achieve high-efficiency and high-precision accurate automatic archiving of warning risk sheets in power systems.

[0004] The purpose of the present invention is achieved through the following technical solutions:

[0005] A dual-channel CNN-based RPA automated archiving risk order early warning method, characterized by comprising the following steps:

[0006] Step 1: Use multi-threaded network technology to automatically log in to the OMS scheduling support system, obtain archived risk orders in the system in real time, and obtain relevant completed archived risk order data and uncompleted archived risk order data;

[0007] Through multi-threaded network technology, the OMS scheduling support system website is automatically opened, and the team leader's account is used to log in and pass the verification. All risk warning orders in the system are automatically obtained, including the start and end time of the "archived" risk order, the change start time, the change end time, and the archived time. The update frequency is once an hour, and the corresponding files are generated and placed in the corresponding folders.

[0008] Step 2: Collect a large number of risk warning samples as a training set, manually annotate them, and input them into the dual-channel CNN network for training to obtain the corresponding network model parameter file;

[0009] Step 3: Input the risk sheet to be judged into the network, compare it with the relevant information of the "archived" risk sheet, output the completion status of this risk sheet, and generate a list of feedback units;

[0010] Step 4: Call the SMS and voice interface to notify relevant personnel of the completion of the risk order.

[0011] The step 2 includes:

[0012] Step 2.1: Collect datasets containing targets through the Internet. For the collected image datasets, use LabelImg to label the datasets with targets. For the non-image datasets, use manual labeling to label the targets and obtain the labeling files. Place the datasets and labeling files in corresponding folders respectively and wait for network training.

[0013] Step 2.2: Before putting the dataset into network training, set the categories to be recognized, select the network model to be used, and set the path for reading training data; select an appropriate training batch size to improve training efficiency and achieve the optimal final training convergence effect;

[0014] In step 2-3, the data set is input into the network model for classification training, and the network model parameter file is output after iterative training.

[0015] The step 3 comprises:

[0016] Step 3.1: For the two risk tickets, divide them into multiple fixed-size patches. A patch is an image unit between a pixel and a graph. These patches are fed into the network in batches for recognition.

[0017] Step 3.2: Pass the input data into the convolution layer, and each patch is convolved with the convolution kernel in turn;

[0018] Step 3.3: Based on the above steps, each risk list is subjected to feature extraction through the convolutional layer and pooling layer, and the features are combined by feature mapping;

[0019] Step 3.4: Vectorize the merged features to obtain feature vectors, and then fuse and normalize the feature vectors to improve the performance of the network;

[0020] Step 3.5: Input the normalized vector into the fully connected layer and use the softmax function to calculate and classify the specific filing information of the risk order;

[0021] Step 3.6: Generate a feedback unit list based on the acquired risk order archive information, transfer it to the traceability layer to modify the MySQL database data, and transfer it to the application layer to generate the corresponding Excel table.

[0022] The step 4 includes: calling the SMS platform and the voice outbound call interface, sending SMS messages, making phone calls to notify relevant personnel of the automatic completion of the early warning risk order.

[0023] In the dual-channel CNN RPA automatic archiving risk sheet early warning system, the present invention inputs the early warning risk sheet obtained in real time into the network, and outputs the archiving information of the unarchived risk sheet through dual-channel CNN comparative calculation and classification, thereby realizing high-efficiency and high-precision accurate automatic archiving of early warning risk sheets in the power system.

[0024] Compared with the existing technology, the present invention has the following significant advantages: (1) The present invention considers the intelligent calculation and comparison of unarchived risk orders with archived risk orders, outputs the start and end time of unarchived risk orders, changes the start and end time and other specific information, and automatically and accurately archives the risk orders. (2) The present invention considers multi-threaded automatic scheduling technology, obtains risk order data in the OMS system in real time, realizes all-weather data update scheduling, and effectively improves service quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the technical solution of the present invention, the following is a brief introduction to the drawings required for use in the embodiments. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0026] Figure 1 This is a flow chart of the RPA automated archiving risk order warning method based on dual-channel CNN of the present invention.

[0027] Figure 2This is a flow chart of the use of a dual-channel CNN network for intelligent calculation and comparison of risk order data in the RPA automated archiving risk order warning method based on the dual-channel CNN of the present invention. DETAILED DESCRIPTION

[0028] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0029] like Figure 1 As shown in the figure, a dual-channel CNN-based RPA automated archiving risk order warning method specifically includes the following steps:

[0030] Step 1: Use multi-threaded network technology to automatically log in to the OMS scheduling support system, obtain archived risk orders in the system in real time, and obtain relevant completed archived risk order data and uncompleted archived risk order data;

[0031] Step 2: Collect a large number of risk warning samples as a training set, manually annotate them, and input them into the dual-channel CNN network for training to obtain the corresponding network model parameter file;

[0032] Step 3: Input the risk sheet to be judged into the network, compare it with the relevant information of the "archived" risk sheet, output the completion status of this risk sheet, and generate a list of feedback units;

[0033] Step 4: Call the SMS and voice interface to notify relevant personnel of the completion of the risk order.

[0034] Step 1 involves obtaining the relevant completed and uncompleted archived risk order data, as follows:

[0035] Through multi-threaded network technology and other means, the OMS scheduling support system website is automatically opened, and the team leader's account is used to log in and pass the verification, and all risk warning orders in the system are automatically obtained, including the start and end time of the "archived" risk order, the change start time, the change end time, and the archived time. The update frequency is once an hour, and the corresponding files are generated and placed in the corresponding folders.

[0036] Step 2 involves collecting and labeling the training set, inputting it into the dual-channel CNN network for training, and obtaining the corresponding network model parameter file. The specific steps are as follows:

[0037] Step 2.1: Collect datasets containing targets through the Internet or other channels. For the collected image datasets, use LabelImg to label the datasets with targets. For the non-image datasets, use manual labeling to label the targets and obtain the labeling files. Place the datasets and labeling files in corresponding folders respectively and wait for network training.

[0038] Step 2.2: Before putting the dataset into network training, set the categories to be recognized, select the network model to be used, and set the path for reading training data. Choose an appropriate training batch size. An appropriate training batch size can improve training efficiency and ultimately achieve optimal training convergence.

[0039] Step 2.3: Input the data set into the network model for classification training, and output the network model parameter file after iterative training.

[0040] Step 3 involves inputting the risk sheet to be assessed into the network, comparing it with the relevant information of the "archived" risk sheet, outputting the completion status of the risk sheet, and generating a list of feedback units. The specific steps are as follows:

[0041] Step 3.1: For the two risk orders, divide them into multiple fixed-size patches. A patch is an image unit between a pixel and a graph. These patches are fed into the network in batches for recognition:

[0042] Step 3.2: Pass the input data into the convolution layer, and each patch is convolved with the convolution kernel in turn. Here is the propagation method between feature maps:

[0043]

[0044] in, and They represent the i-th input feature and j-th output feature of the r-th convolutional layer, respectively. and They represent the jth convolution kernel and the bias of the convolution layer respectively, and the f(·) function is the activation function ReLU;

[0045] Step 3.3: According to the above steps, each risk list is subjected to feature extraction through the convolution layer and pooling layer, and the features are combined by feature mapping:

[0046]

[0047] Among them, f(x1) and f(x2) refer to the output of the two risk orders after passing through the convolution layer, P1 and P2 refer to the maximum pooling feature operation, which maps and combines the two features;

[0048] Step 3.4: Vectorize the merged features to obtain feature vectors, and then fuse and normalize the feature vectors to improve the performance of the network:

[0049]

[0050]

[0051] Here, v refers to the feature vector corresponding to the fused feature, to which the signed square root operation is applied and L2 norm normalization is performed to improve performance.

[0052] Step 3.5: Input the normalized vector into the fully connected layer and use the softmax function to calculate and classify the specific filing information of the risk order:

[0053]

[0054] where α k Refers to the score of risk order information k, K refers to the total number of risk order information types, p z，k Refers to the possibility that the final information prediction data of the risk single sample z is k.

[0055] Step 3.6: Generate a feedback unit list based on the acquired risk order archive information, transfer it to the traceability layer, modify the MySQL database data through SQL statements, and transfer it to the application layer to generate the corresponding Excel table.

[0056] Step 4 involves RPA calling the provincial communication platform to notify relevant personnel after the archiving task is completed. Specifically, it calls the provincial company's SMS platform and voice outbound call interface to send SMS messages and make phone calls to notify relevant personnel of the automatic completion of the warning risk order.

[0057] Those skilled in the art will clearly understand that the techniques in the embodiments of the present invention can be implemented using software plus a necessary general-purpose hardware platform. Based on this understanding, the technical solutions in the embodiments of the present invention, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods described in various embodiments of the present invention or certain portions of the embodiments.

[0058] The same or similar parts between the various embodiments in this specification can be referred to each other. The above-described embodiments of the present invention do not constitute a limitation on the scope of protection of the present invention.

Claims

1. A dual-channel CNN-based RPA automated archiving risk order warning method, characterized by: The following steps are involved: Step 1: Automatically log in to the OMS scheduling support system through a multi-threaded network, obtain archived risk orders in the system in real time, and obtain relevant completed archived risk order data and uncompleted archived risk order data; Step 2: Collect a large number of risk warning samples as a training set, manually annotate them, and input them into the dual-channel CNN network for training to obtain the corresponding network model parameter file; Step 3: Enter the risk order to be judged into the network, compare it with the relevant information of the archived risk order, output the completion status of the risk order, and generate a list of feedback units; Step 4: Call SMS or voice interface to notify relevant personnel of the completion of the risk order; The step 2 includes: Step 2.1: Collect datasets containing targets through the Internet. For the collected image datasets, use LabelImg to label the datasets with targets. For the non-image datasets, use manual labeling to label the targets and obtain the labeling files. Place the datasets and labeling files in corresponding folders respectively and wait for network training. Step 2.2: Before putting the dataset into network training, set the categories to be recognized, select the network model to be used, and set the path for reading training data; Step 2-3: Input the data set into the network model for classification training, and output the network model parameter file after iterative training; The step 3 comprises: Step 3.1: For the two risk orders, divide them into multiple fixed-size patches. A patch is an image unit between a pixel and a graph. These patches are fed into the network in batches for recognition. Step 3.2: Pass the input data into the convolution layer, and each patch is convolved with the convolution kernel in turn. Here is the propagation method between feature maps: in, and They represent the i-th input feature and j-th output feature of the r-th convolutional layer, respectively. and They represent the jth convolution kernel and the bias of the convolution layer respectively, and the f() function is the activation function ReLU; Step 3.3: According to the above steps, each risk list is subjected to feature extraction through the convolution layer and pooling layer, and the features are combined by feature mapping: Among them, f(x1) and f(x2) refer to the output of the two risk orders after passing through the convolution layer, P1 and P2 refer to the maximum pooling feature operation, which maps and combines the two features; Step 3.4: Vectorize the merged features to obtain feature vectors, and then fuse and normalize the feature vectors to improve the performance of the network: Here, v refers to the feature vector corresponding to the fused feature, and the signed square root operation is applied to it and the L2 norm is normalized to improve performance; Step 3.5: Input the normalized vector into the fully connected layer and use the softmax function to calculate and classify the specific filing information of the risk order: where α k Refers to the score of risk order information k, K refers to the total number of risk order information types, p z,k Refers to the possibility that the final information prediction data of the risk single sample z is k; Step 3.6: Generate a feedback unit list based on the acquired risk order archive information, transfer it to the traceability layer to modify the MySQL database data, and transfer it to the application layer to generate the corresponding Excel table.

2. The RPA automated archiving risk order warning method based on dual-channel CNN according to claim 1 is characterized in that: The step 1 comprises: Through the multi-threaded network, the OMS scheduling support system website is automatically opened. After logging in and passing the verification using the team leader's account, all risk warning orders in the system are automatically obtained, including the start and end time of the "archived" risk order, the change start time, the change end time, and the archived time. The update frequency is once an hour, and the corresponding files are generated and placed in the corresponding folders.

3. The RPA automated archiving risk order warning method based on dual-channel CNN according to claim 1 is characterized in that: The step 4 includes: calling the SMS platform and the voice outbound call interface, sending SMS messages, making phone calls to notify relevant personnel of the automatic completion of the early warning risk order.

Images