Transformer substation CT phase sequence measurement system assisted by intelligent image recognition

Abstract

The invention discloses a transformer substation CT phase sequence measurement system assisted by intelligent image recognition. The transformer substation CT phase sequence measurement system comprises a current source on the primary side of a CT, a digital ammeter on the secondary side of the CT and a camera loaded with an intelligent image recognition module. The CT primary side current source outputs stable large current, the reading is displayed on the electric meter on the CT secondary side, the reading on the electric meter is read through the intelligent image recognition module, and data is transmitted back to CT primary side equipment. According to the invention, a very small amount of data communication transmission can be realized, and personnel use of a secondary side is saved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent recognition, and specifically to a system for measuring the CT phase sequence of a substation assisted by intelligent image recognition. Background Technique

[0002] Facing the continuous development of the power grid, the continuous commissioning of large-capacity units, extra-high voltage and ultra-high voltage equipment, the matching relay protection principles are becoming increasingly complex, and the secondary circuit of relay protection is also becoming more and more important. Among them, the secondary circuit of the current transformer (CT) is the top priority for the correct operation of the relay protection device. According to the requirements of the "Eighteen Major Anti-accident Measures for the Power Grid of the State Grid Corporation", for the work such as the commissioning of new substations, the replacement of current transformers (CTs) and the protection transformation, the CT load phase measurement work needs to be carried out.

[0003] The existing debugging and acceptance methods for the secondary circuit of the substation CT include conventional methods such as checking the secondary polarity of the CT with a small direct current, measuring the resistance of the secondary circuit to ensure connectivity, and checking the wiring correctness by secondary line inspection. The inspection process is complex and overly relies on the experience judgment of on-site personnel. Once the acceptance is not in place, it may cause misoperation of the relay protection during the start-up of the renovated equipment, and even endanger personal safety. Summary of the Invention

[0004] The purpose of the present invention is to provide a system for measuring the CT phase sequence of a substation assisted by intelligent image recognition to solve the problems such as endangering personal safety that may occur during the debugging and acceptance of the secondary circuit of the substation CT using the existing methods mentioned in the above background technique.

[0005] To achieve the above purpose, the present invention provides the following technical solution: A system for measuring the CT phase sequence of a substation assisted by intelligent image recognition includes a CT primary side current source, a CT secondary side digital electric meter, and a camera loaded with an intelligent image recognition module on the CT secondary side. The CT primary side current source outputs a stable large current to the primary side equipment, a reading is displayed on the CT secondary side digital electric meter, and then the reading is recognized by the camera loaded with the intelligent image recognition module on the CT secondary side, and the data is wirelessly transmitted back to the equipment on the primary side for comparison with the theoretical value. If the measured value is the same as the theoretical value, the CT phase sequence is correct; otherwise, there is a problem.

[0006] Compared with the existing technology, the beneficial effects of the present invention are:

[0007] 1. The present invention saves the use of secondary side personnel by embedding a camera loaded with an intelligent image recognition module on the secondary side equipment to recognize the electric meter reading and transmitting the data to the equipment on the primary side so that the primary side personnel can observe the change of the reading in real time.

[0008] 2. The present invention uses an intelligent image recognition module to recognize the readings on the electric meter. Compared with the image transmission of ordinary cameras, the present invention can perform data transmission, achieving extremely small amounts of data communication, greatly improving the transmission stability, and making the transmission method more flexible. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 It is a structural block diagram of a substation CT phase sequence measurement system assisted by intelligent image recognition provided by an embodiment of the present invention.

[0010] Figure 2 It is a schematic flowchart of a substation CT phase sequence measurement system assisted by intelligent image recognition provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0011] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be described in detail and completely below with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0012] Referring to Figure 1 as shown, it is a structural block diagram of a substation CT phase sequence measurement system assisted by intelligent image recognition provided by an embodiment of the present invention.

[0013] According to the functions achieved, the substation CT phase sequence measurement system (0) assisted by intelligent image recognition of the present invention includes a current source device (1), a digital electric meter (2), a camera (3) equipped with an intelligent image recognition module, a YOLOv10 detection module (4), an EasyOCR text detection module (5), and an EasyOCR text recognition module (6).

[0014] Next, with reference to specific embodiments, each component of the substation CT phase sequence measurement system assisted by intelligent image recognition and the specific working process will be described.

[0015] First, the current source device (1) passes a large current through the primary side of the CT, and the digital electric meter (2) on the secondary side of the CT displays the measured data, and then the camera (3) equipped with the intelligent image recognition module installed on the secondary side recognizes the readings on the electric meter. During the recognition process:

[0016] The YOLOv10 detection module (4) identifies the position of the digital electric meter and the position of the digital electric meter reading frame. As a one-stage object detection algorithm, the YOLO series is widely used in real-time detection due to its high detection accuracy and fast running speed, and has been verified to be suitable for deployment in industrial detection systems. Although YOLOv1-9 has achieved a delicate balance between performance and efficiency, there are still some deficiencies: the dependence of post-processing on non-maximum suppression (NMS) hinders the end-to-end deployment of YOLO and has an adverse impact on inference latency; the design of each component in YOLO lacks a comprehensive and thorough inspection, resulting in obvious computational redundancy, thus limiting the performance of the model. Therefore, in YOLOv10, a consistent dual assignment strategy for YOLO without NMS is first proposed, with dual label assignment and consistent matching metrics, which solves the redundant prediction problem in post-processing. Enables the model to obtain rich and harmonious supervision during the training process, and at the same time does not require NMS during the inference process, thus obtaining a highly efficient competitive performance. Secondly, an overall efficiency-accuracy-driven model architecture design strategy is proposed, including a lightweight classification head, spatial-channel decoupled downsampling, and rank-guided block design to reduce computational redundancy and achieve a more efficient architecture. To improve accuracy, an effective partial self-attention module is proposed to enhance the model's capabilities. Therefore, the lightweight YOLOv10n is selected as the benchmark model for electric meter detection. The model consists of four parts: Input (input layer), Backbone (backbone network), Neck (neck network), and Head (detection output layer).

[0017] The input layer preprocesses the electric meter image and generates an anchor box generation mechanism, and then inputs it to the Backbone layer, which is responsible for extracting the information features of the electric meter at different development stages. In addition to continuing the convolutional Conv in v8, the C2f module with the characteristics of an efficient aggregation network, and the SPPF module that realizes the fusion of local features and global features in its structure, an SCDown (Spatial-channel decoupled downsampling) module is also introduced. First, the channel dimension is adjusted through pointwise convolution, and then depth convolution is used for spatial downsampling, decoupling the spatial downsampling and channel conversion operations, improving the information retention rate while reducing the computational cost.

[0018] Partial self-attention (PSA) modules are added, which evenly divide cross-channel features into two parts after convolutional operations. Only one part is input into the NPSA block composed of a multi-head self-attention module (MHSA) and a feed-forward network (FFN). Then the two parts are connected and fused through convolution. In addition, the dimensions of queries and keys are assigned to be half of the values in the MHSA, and LayerNorm is replaced with BatchNorm to improve the inference speed. The PSA module is only placed after stage 4 with the lowest resolution to avoid excessive computational complexity overhead, effectively integrating the global representation learning ability to improve the model performance.

[0019] The Neck layer fuses the electricity meter features by enhancing the network's feature learning ability. v10 adds a compact inverted block (CIB) structure in this layer, which uses spatial mixing of cheap depthwise convolutions and channel mixing of cost-effective pointwise convolutions, and can be used as an efficient basic building block to achieve higher efficiency without sacrificing performance. In the Head layer, YOLOv10 introduces a new dual-assignment strategy. The model uses two prediction heads for joint optimization during training: one uses one-to-many assignment and the other uses one-to-one assignment. This can not only allow the Backbone and Neck to enjoy the rich supervision provided by one-to-many assignment, but also utilize the prediction results of one-to-one assignment during inference, thus achieving efficient inference without NMS.

[0020] The EasyOCR character reader reads the readings in the digital electricity meter display box. EasyOCR is a font-related printed character reader based on a template matching algorithm. EasyOCR supports 42 languages, including many dynamic parameters for reading text and detecting colored regions. It is designed to read short texts such as part numbers, serial numbers, expiration dates, manufacturing dates, and batch numbers. Whether printed on labels or directly on parts, EasyOCR provides comprehensive character verification, including automatic training, grayscale analysis, text and character-level checks, contrast, position, and shape defect detection. Images are retrieved from the video as frames and fed into the electricity meter detection system. License plates are extracted from the images and sent to OCR for digital character recognition. The outputs of the module (such as the type, color, and license number of the electricity meter) are saved in the database.

[0021] The EasyOCR text detection module (5) is based on a deep learning model and can accurately locate text regions in images. EasyOCR typically uses the CRAFT algorithm for text detection. CRAFT can accurately detect text regions of arbitrary shapes by predicting the regions of each character and the connection relationships between characters. Through multi-scale feature fusion technology, the algorithm can detect text regions of different sizes and orientations, adapting to complex scenarios. Finally, the detected text regions are output in the form of bounding boxes as the input for subsequent text recognition.

[0022] The text recognition module (6) of EasyOCR recognizes the detected text regions. EasyOCR uses a convolutional neural network (CNN) to extract the features of the text regions. Through multiple layers of convolution and pooling operations, CNN can capture the local features of the text. Next, a recurrent neural network (RNN) or Transformer is used to model the text sequence. RNN captures the context information in the sequence through memory units (such as LSTM, GRU), while Transformer captures long-range dependencies through self-attention mechanisms. Finally, each character is classified through a fully connected layer and a Softmax function, and the recognition result is output. For the digital recognition task, the character classes are 0-9.

[0023] Refer to Figure 2 As shown, it is a schematic flowchart of a substation CT phase sequence measurement system assisted by intelligent image recognition provided by an embodiment of the present invention. In this embodiment, the specific process of the substation CT phase sequence measurement system assisted by intelligent image recognition is as follows:

[0024] A stable large current is input to the primary side equipment of the CT through a current source, and the secondary side ammeter shows the measurement reading;

[0025] The YOLOv10 detection module is used to identify the positions of the ammeter and the reading frame;

[0026] EasyOCR is used to read the reading in the ammeter reading frame.

[0027] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not restrictive. Although the present invention has been described in detail with reference to preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. An intelligent image recognition-assisted substation CT phase sequence measurement system, characterized in that: The system comprises: CT primary side current source, used to output stable large current to primary side equipment; CT secondary side digital meter, used to display measurement data; A camera equipped with an intelligent image recognition module is installed on the secondary side of the CT to intelligently read the readings on the electric meter instead of the secondary side personnel.

2. The intelligent image recognition-assisted substation CT phase sequence measurement system according to claim 1, characterized in that: The intelligent image recognition module includes: a YOLOv10 detection model and an EasyOCR character reader.

3. The intelligent image recognition-assisted substation CT phase sequence measurement system according to claim 2, characterized in that: The YOLOv10 detection model includes: an input layer, a backbone network, a neck network and a detection output layer.

4. The intelligent image recognition-assisted substation CT phase sequence measurement system according to claim 3, characterized in that: The input layer includes a convolution Conv module, a C2f module with efficient aggregation network characteristics, an SPPF module for realizing the fusion of local features and global features, an SCDown module for improving information retention rate, and a self-attention PSA module for improving learning ability.

5. The intelligent image recognition-assisted substation CT phase sequence measurement system according to claim 2, characterized in that: The EasyOCR character reader is based on a template matching algorithm and supports 42 languages. The reader includes a text detection module and a text recognition module.

6. The intelligent image recognition-assisted substation CT phase sequence measurement system according to claim 5, characterized in that: The text detection module performs text detection using the CRAFT algorithm which can accurately detect text regions of any shape.

Images