Article Preview
Top1. Introduction
With the continuous development of computer technology and communication technology, State Grid Corporation of China proposed the development strategy of “International Leading Energy Internet Enterprise” to promote the intelligent upgrading of power grid business and realize the digital transformation of marketing measurement system (Ersheng et al., 2020). Currently in traditional laboratories, the information between various devices is isolated, lacking effective connections and with a high workload of data maintenance; the experimental equipment lacks necessary identity verification information, and the usage and location information of the equipment cannot be mastered; there is also a lack of linkage between experimental personnel and equipment (Fernandes et al., 2018; Sardjono, 2021). The traditional laboratory has reached the stage where it must be upgraded and transformed. It is needed to build an interconnected intelligent laboratory integrated with the management platform using advanced technologies such as big data and microservices in the background of ubiquitous power from Internet of Things (IoT) and centering on communication management, data storage, computing analysis, business applications, data sharing, etc. (Giacomo et al., 2023).
Although laboratories in China have developed rapidly, various types of power grid equipment are also constantly increasing, and different equipment manufacturers are not the same. The interface types and data transmission protocols of the equipment are not the same, rendering it difficult to automate equipment inspections. At the same time, the laboratory testing process is very cumbersome, and the obtained experiments and test results cannot be automatically processed, requiring a large amount of manual collection, analysis, and uploading, resulting in very low efficiency. Therefore, there is an urgent need for intelligent and information-based management methods to achieve efficient analysis and to process equipment testing information.
The image recognition and error correction for testing reports of power grid equipment is a key business capability of the laboratory. At present, traditional image recognition methods include feature detection method, support vector machine (SVM) machine learning recognition method, BP neural network method, etc. However, these methods have a cumbersome processing process, low accuracy, and are not suitable for the recognition of test report images for power grid equipment (Lu et al., 2022; Masayuki et al., 2022). Recently, scholars are gradually increasing research on the direct use of deep reinforcement learning (DRL) for classification, but the classification effect is greatly influenced by the selection of agents. Meanwhile, present research on image error correction is mostly based on error correction output codes; however, redundant information is added in this approach, causing the waste of computing resources (Najib et al., 2022). As a result, drawing on the increasingly mature deep learning network and IoT technology, this paper proposes an image identification and error correction method for test report based on DRL and IoT platform in the smart laboratory. Compared with traditional methods, the innovation of the proposed method lies in:
- 1)
In order to improve the accuracy of image recognition, the proposed method adopts deep separable convolution to simplify the structure of convolutional neural networks (CNN) and combines it with bidirectional recurrent neural networks to ensure the network’s fast feature extraction.
- 2)
Due to the possibility of dimensionality disasters in traditional Q-learning algorithms, the proposed method uses the ICNN-BiLSTM model as an intelligent agent for reinforcement learning to make action decisions, further ensuring the reliability of error correction results.