AI-based intelligent identification method and system for state of equipment in power distribution area

Abstract

The invention discloses an AI-based intelligent identification method and system for the state of equipment in a power distribution area, and the method comprises the steps: collecting the data of a sensor on the equipment, and obtaining a multi-mode signal; inputting the multi-modal signal into a neural network for state prediction; the method comprises the following steps: acquiring a monitoring image of equipment in real time through image acquisition equipment, and identifying an equipment state in the image; and fusing the recognition result of the image and the prediction result of the neural network to generate a final recognition result. The self-adaptability and learning ability of the system are enhanced, different equipment and environments can be quickly adapted, the operation and maintenance and decision-making process is simplified, and safe and stable operation of the power distribution network is effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent identification of equipment status, and specifically to an AI-based intelligent identification method and system for equipment status in a distribution station area. Background Art

[0002] With the rapid development of smart grid and distribution automation technology, the status monitoring and intelligent identification of distribution substation equipment have become increasingly important. Traditional distribution equipment monitoring methods mainly rely on manual inspections and regular maintenance, which have problems such as low monitoring efficiency and untimely fault warning. In order to solve these problems, intelligent monitoring methods based on artificial intelligence (AI) have emerged. By deploying sensors to collect equipment operation data and combining computer vision technology to monitor equipment status in real time, accurate identification of equipment status and fault warning can be achieved. However, the operating environment of distribution substation equipment is complex and changeable, and the data of a single sensor is often difficult to fully reflect the actual operating status of the equipment. Multimodal data fusion technology has gradually attracted attention. By integrating multiple sensor data such as current, voltage, temperature, vibration, etc., and combining real-time monitoring image information, the operating status of the equipment can be more accurately identified. The fusion and analysis of multimodal data not only improves the accuracy of equipment status identification, but also provides strong support for the safe operation of smart grids.

[0003] In order to further improve the intelligence and automation level of equipment status monitoring, neural network models based on deep learning have been widely used in the field of smart grid monitoring in recent years. Neural networks can effectively process and analyze a large amount of complex nonlinear data, especially in the feature extraction and fusion of multimodal data. For example, convolutional neural networks (CNNs) are good at extracting key features from sensor signals and images, long short-term memory networks (LSTMs) can capture the time series characteristics of equipment operation status, and attention mechanisms (Attention Mechanisms) are helpful for the effective fusion of different modal features. In addition, combining decision tree models for classification and identification of equipment status can further improve the accuracy and stability of identification. Through these advanced AI technical means, real-time and accurate monitoring of distribution substation equipment can be achieved, abnormal conditions can be discovered in time, and the safe and stable operation of the distribution network can be effectively guaranteed. This monitoring method based on multimodal data fusion and AI intelligent identification will be an important direction for future smart grid equipment management. Summary of the invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the technical problem solved by the present invention is that the existing intelligent identification of equipment status in distribution station areas has the problems of high labor costs, inaccurate mechanical identification, etc.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: an AI-based intelligent identification method for equipment status in a distribution station area, comprising:

[0007] Collect data from sensors on the device to obtain multi-modal signals;

[0008] Inputting the multimodal signal into a neural network for state prediction;

[0009] Through the image acquisition device, the monitoring image of the device is collected in real time, and the device status in the image is identified;

[0010] The recognition results of the image and the prediction results of the neural network are combined to generate the final recognition results.

[0011] As a preferred solution of the AI-based intelligent identification method for the status of distribution station equipment in the present invention, the sensors include current sensors, voltage sensors, temperature sensors, vibration sensors, humidity sensors, air pressure sensors, electromagnetic sensors, gas sensors, optical sensors, and position sensors;

[0012] According to the monitorable data of the device, a sensor for acquiring the monitorable data is configured on the device, and a multi-modal signal is obtained through the sensor.

[0013] As a preferred solution of the AI-based intelligent identification method for the status of equipment in a distribution station area described in the present invention, the neural network includes, for different equipment, respectively, constructing a training set and a verification set based on a multimodal signal set obtained by combining sensors matching the equipment;

[0014] Suppose there are N sensors, and the collected multimodal signals are represented by X = {x1, x2, …, x N};

[0015] Among them, each x i Represents the data of the i-th sensor;

[0016] Use convolutional neural network for preliminary feature extraction. The feature extraction process is expressed as:

[0017] F i =CNN i (x i )

[0018] Among them, F i Indicates that from x i Extracted features; CNN i represents the convolutional neural network for the i-th sensor signal; i = 1, 2, ..., N;

[0019] Fusion of features from different modalities through multi-head attention mechanism:

[0020] F fusion =Attention(F1,F2,…,F N )

[0021] Among them, Attention represents a multi-head attention mechanism, which is used for the interaction and fusion of different features;

[0022] The fused features are input into a fully connected layer for state prediction:

[0023] y=σ(WF fusion +b)

[0024] Among them, W represents the weight matrix, b represents the bias vector, and σ represents the activation function ReLU.

[0025] As a preferred solution of the AI-based intelligent identification method for the equipment status of a distribution station area described in the present invention, the monitoring image of the real-time acquisition device includes all the picture information monitored by the image acquisition device, and the equipment status is identified by learning the characteristics of the equipment motion status;

[0026] The device motion state includes the actions of the device when it is working and performing different operations; if multiple device states show the same motion state, then in the process of image recognition, it is identified as a set of all possible device states, and each element in the set represents a possible device state; if one device state only matches one motion state, then the device state set generated by recognition only contains one element.

[0027] As a preferred solution of the method for intelligently identifying the status of equipment in a distribution station area based on AI according to the present invention, the identification of the status of equipment in the image includes: t After standardization and denoising, it is expressed as:

[0028] I pre (t)=Preprocess(I t )

[0029] Among them, Preprocess represents the preprocessing process, I pre (t) represents the processed image;

[0030] Use CNN to extract the spatial features of each frame;

[0031]

[0032] Using a long short-term memory network, the extracted feature sequence is input into a gated recurrent unit, and time series modeling is used to detect whether there is a periodic behavior pattern:

[0033] P cyclic =σ(LSTM(F seq (1),F seq (2),…,F seq (T)))

[0034] Among them, P cyclic It is a binary classification output, indicating whether the behavior is periodic. The output value 1 indicates periodic behavior, and 0 indicates non-periodic behavior. T indicates the length of the image sequence.

[0035] If it is not a periodic behavior, it is identified as a suspected change in the device state; if it is a periodic behavior, a periodic behavior is extracted, a complete action cycle is extracted and its features are constructed:

[0036] Cycle boundary identification: Identify the starting point of the action cycle through time series analysis start and the end point t end ;

[0037] Extracting periodic features: Input the image sequence within the period into LSTM to extract the periodic behavior features F cycle ;

[0038] F cycle =LSTM cycle )F seq (t start ),F seq (t start +1),…,F seq (t end ))

[0039] Among them, LSTM cycle is an LSTM model specifically used for period feature extraction. seq (t start ) indicates t start The spatial characteristics of the moment;

[0040] Use the trained decision tree model DecisionTreeClassifier and input the local feature F cycle , output device state set ε:

[0041] ε=DecisionTreeClassifier(F cycle )

[0042] When the device status is suspected to have changed, the last device status identification result ε is obtained. 上 , obtain the next identification device status result, if the device status is still suspected to have changed, re-obtain the next identification device status result until the device status result ε can be identified 下Until; and change the identification result of suspected device status change during this period to: 上 to ε 下 Status changes.

[0043] As a preferred solution of the AI-based intelligent identification method for distribution substation equipment status of the present invention, the fusion includes: when the equipment status set ε is identified, if the state prediction result through the multimodal signal is in the equipment status set ε, then directly generating a fusion result as the state prediction result of the multimodal signal; if the state prediction result through the multimodal signal is not in the equipment status set ε, then generating an error;

[0044] When the device state is suspected to have changed, the state prediction result of the multimodal signal is output first, and the recognition result of the image is changed to ε 上 to ε 下 When the state of ε changes, the state prediction result of the multimodal signal is verified. If the state prediction result of the priority output is not within ε 上 or ε 下 Then, a multimodal signal with a time length of T before and after the state change and a multimodal signal identified as a period of suspected device state change are obtained and put into the training set of the neural network for training.

[0045] As a preferred solution of the AI-based intelligent identification method for equipment status in a distribution station area described in the present invention, the generated fusion result is used as the final identification result and output to the front end;

[0046] At the same time, a data record of the multimodal signal is generated.

[0047] An AI-based intelligent identification system for equipment status in a distribution station area using the method described in the present invention is characterized in that:

[0048] A collection unit collects data from sensors on the device to obtain multi-modal signals;

[0049] A prediction unit, inputting the multimodal signal into a neural network for state prediction;

[0050] The recognition unit collects the monitoring image of the device in real time through the image acquisition device, and recognizes the device status in the image;

[0051] The generation unit fuses the image recognition result and the prediction result of the neural network to generate the final recognition result.

[0052] A computer device comprises: a memory and a processor; the memory stores a computer program, wherein the processor implements the steps of any one of the methods of the present invention when executing the computer program.

[0053] A computer-readable storage medium stores a computer program, wherein the computer program, when executed by a processor, implements the steps of any one of the methods of the present invention.

[0054] Beneficial effects of the present invention: The AI-based intelligent identification method for the status of distribution substation equipment provided by the present invention realizes efficient and accurate identification of the status of distribution equipment through multimodal data fusion and neural network algorithms. By using a variety of sensors to collect signals such as current, voltage, temperature, etc., and combining them with real-time monitoring images, the operating status of the equipment can be acquired in all directions. By extracting and fusing features through convolutional neural networks and long short-term memory networks, the system can identify the periodic behavior and state changes of the equipment, thereby improving the accuracy and real-time performance of monitoring. In addition, a decision tree model is used to classify the equipment status, effectively handle abnormal situations, and prevent false alarms and missed reports. This method enhances the adaptability and learning ability of the system, can quickly adapt to different equipment and environments, simplify the operation and maintenance and decision-making process, and effectively improve the safe and stable operation of the distribution network. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0056] Figure 1 An overall flow chart of the AI-based intelligent identification method for equipment status in a distribution station area provided for the first embodiment of the present invention. DETAILED DESCRIPTION

[0057] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0058] Example 1, reference Figure 1 , is an embodiment of the present invention, and provides an AI-based intelligent identification method for equipment status in a distribution station area, comprising:

[0059] S1: Collect data from sensors on the device to obtain multi-modal signals.

[0060] Furthermore, the sensor includes a current sensor, a voltage sensor, a temperature sensor, a vibration sensor, a humidity sensor, an air pressure sensor, an electromagnetic sensor, a gas sensor, an optical sensor, and a position sensor. According to the monitorable data of the device, a sensor for acquiring the monitorable data is configured on the device, and a multi-modal signal is obtained through the sensor.

[0061] It should be said that current sensors and voltage sensors provide electrical parameter information, temperature sensors monitor the thermal state of the equipment, vibration sensors detect mechanical vibrations, humidity and air pressure sensors obtain environmental conditions, electromagnetic sensors capture electromagnetic interference, gas sensors monitor chemical reactions inside the equipment, optical sensors are used for optical detection, and position sensors locate the physical state of the equipment. The combination of data from these sensors can fully reflect the operating status of the equipment. Through multimodal signal fusion, it is possible to capture abnormal conditions that are difficult to detect with a single sensor, and improve the sensitivity and recognition accuracy of changes in equipment status. For example, combining temperature, vibration and current data can detect potential equipment failures earlier, thereby achieving early warning. The data redundancy of multiple sensors can improve the reliability of monitoring. Even if a sensor fails, other sensors can still provide necessary data to ensure the continuity and accuracy of monitoring.

[0062] S2: Inputting the multimodal signal into a neural network for state prediction.

[0063] It is important to know that the neural network includes, for different devices, constructing training sets and validation sets based on the multimodal signal sets obtained from the sensor combinations matching the devices.

[0064] Suppose there are N sensors, and the collected multimodal signals are represented by X = {x1, x2, …, x N}.

[0065] Among them, each x i Represents the data of the i-th sensor.

[0066] Use convolutional neural network for preliminary feature extraction. The feature extraction process is expressed as:

[0067] F i =CNN i (x i )

[0068] Among them, F i Indicates that from x i Extracted features; CNN i Represents the convolutional neural network for the i-th sensor signal; i=1,2,…,N.

[0069] Fusion of features from different modalities through multi-head attention mechanism:

[0070] F fusion =Attention(F1,F2,…,F N )

[0071] Among them, Attention represents a multi-head attention mechanism, which is used for the interaction and fusion of different features. The fused features are input into a fully connected layer for state prediction:

[0072] y=σ(WF fusion +b)

[0073] Among them, W represents the weight matrix, b represents the bias vector, and σ represents the activation function ReLU.

[0074] It is also important to know that by using matching sensor combinations for different devices to obtain multimodal signals, various status information of device operation can be fully captured. This information covers electrical parameters, environmental conditions, mechanical vibrations, optical signals and other aspects, which helps to build a more comprehensive description of device status. Using convolutional neural networks (CNN) for preliminary feature extraction can automatically extract meaningful features from multimodal signals, thereby reducing the complexity of manual feature design. At the same time, through the multi-head attention mechanism, the features of different modalities can be effectively interacted and fused. The attention mechanism can dynamically focus on the importance of different modal features and improve the effectiveness and accuracy of feature fusion. The fused features are input into the fully connected layer for state prediction. This deep learning model can process high-dimensional feature spaces and capture complex nonlinear relationships, thereby improving the accuracy of state prediction. The weight matrix and bias vector of the fully connected layer, combined with the ReLU activation function, can achieve effective classification and prediction of device status.

[0075] The design automatically learns and extracts features through neural networks, and can adapt to different equipment and sensor combinations without human intervention. This ability to automatically extract and fuse features allows the system to continuously adjust as equipment and sensors change, improving the versatility and adaptability of the model. Convolutional neural networks have a strong ability in feature extraction, especially when processing time series data and image data, they can capture subtle changes. The multi-head attention mechanism can focus on the relevance and importance of different features to ensure that important information is not ignored. This combined approach helps to improve the accuracy of equipment status recognition and prediction. Through a deep learning framework of feature extraction, fusion, and status prediction, real-time monitoring and identification of equipment status can be achieved. The introduction of convolution operations and attention mechanisms makes the calculation efficient and can quickly process large-scale multimodal signal data to meet the needs of real-time monitoring of equipment in distribution stations.

[0076] S3: The monitoring image of the device is collected in real time through the image acquisition device, and the device status in the image is identified.

[0077] Identifying the device status in the image includes: t After standardization and denoising, it is expressed as:

[0078] I pre (t)=Preprocess(I t )

[0079] Among them, Preprocess represents the preprocessing process, I pre (t) represents the processed image.

[0080] Use CNN to extract the spatial features of each frame.

[0081]

[0082] Using a long short-term memory network, the extracted feature sequence is input into a gated recurrent unit, and time series modeling is used to detect whether there is a periodic behavior pattern:

[0083] P cyclic =σ(LSTM(F seq (1),F seq (2),…,F seq (T)))

[0084] Among them, P cyclic It is a binary classification output, indicating whether the behavior is periodic. The output value 1 indicates periodic behavior, and 0 indicates non-periodic behavior. T represents the length of the image sequence. If it is not a periodic behavior, it is identified as a suspected change in the device state. If it is a periodic behavior, a periodic behavior is extracted, a complete action cycle is extracted, and its features are constructed.

[0085] Cycle boundary identification: Identify the starting point of the action cycle through time series analysis start and the end point t end .

[0086] Extracting periodic features: Input the image sequence within the period into LSTM to extract the periodic behavior features F cycle .

[0087] F cycle =LSTM cycle (F seq (t start ),F seq (t start +1),…,F seq (t end ))

[0088] Among them, LSTM cycle is an LSTM model specifically used for period feature extraction. seq (t start ) indicates t start Use the trained decision tree model DecisionTreeClassifier and input the local feature F cycle , output device state set ε:

[0089] ε=DecisionTreeClassifier(F cycle )

[0090] When the device status is suspected to have changed, the last device status identification result ε is obtained. 上 , obtain the next identification device status result, if the device status is still suspected to have changed, re-obtain the next identification device status result until the device status result ε can be identified 下 Until; and change the identification result of suspected device status change during this period to: 上 to ε 下 Status changes.

[0091] It should be said that by capturing the running image of the equipment in real time through the image acquisition device, it is possible to continuously monitor its status without affecting the normal operation of the equipment. This non-contact monitoring method is not only safe, but also can capture the abnormal status and behavior changes of the equipment in a timely manner. The equipment usually exhibits certain periodic behaviors during operation, such as regular vibration or mechanical action. Using convolutional neural networks (CNN) to extract the spatial features of the image and detecting periodic behaviors through long short-term memory networks (LSTM) can accurately identify the normal operation cycle of the equipment and issue timely warnings when the periodic behavior changes. This helps to discover abnormal conditions in the operation of the equipment and carry out maintenance and prevention in advance. Through LSTM modeling, the behavioral characteristics of the equipment in a complete cycle can be extracted, and the starting and ending points of the periodic behavior can be identified. This step not only helps to understand the normal operating mode of the equipment, but also accurately locates the problem when an abnormality occurs. This feature extraction method provides a basis for further status classification. When periodic behavior is identified or suspected changes in status are detected, the trained decision tree model is used for status classification, which can quickly determine whether the current operating status of the equipment is normal. If a suspected change in the status of the equipment is detected, the system will continuously monitor the subsequent status until the status change is confirmed. This mechanism can effectively avoid false alarms and ensure the authenticity of status changes.

[0092] By combining CNN and LSTM for image feature extraction and time series modeling, the subtle behavioral characteristics of the device can be captured. LSTM can identify periodic patterns in time series and is very sensitive to long-term dependent features. This method improves the accuracy and real-time performance of device state recognition. Using LSTM to detect periodic behavior allows the system to automatically adapt to the operating modes of different devices without pre-setting periodic features. This intelligent detection method can dynamically adapt to different working states of the device, enhancing the flexibility and universality of the system. By introducing a decision tree model to classify local features, the state change of the device can be quickly identified. If the state is detected as a suspected change, the system will continue to track and confirm to ensure that only real state changes are recorded and reported. This strategy reduces false alarms and improves the reliability of monitoring. By processing multiple frames of images and modeling time series, this design has strong robustness in the face of noise and short-term interference. Even if there is interference in a single frame image, as long as the overall behavioral characteristics do not change, the system can maintain its judgment of the normal state of the device. The system can record the change process of the device state and provide detailed time period information from one state to another. This helps operation and maintenance personnel quickly diagnose the cause of equipment failure, take targeted maintenance measures, reduce downtime, and improve the overall operating efficiency of the equipment.

[0093] S4: Fusion the image recognition result and the neural network prediction result to generate the final recognition result.

[0094] When the device state set ε is identified, if the state prediction result through the multimodal signal is in the device state set ε, the fusion result is directly generated as the state prediction result of the multimodal signal; if the state prediction result through the multimodal signal is not in the device state set ε, an error is generated.

[0095] When the device state is suspected to have changed, the state prediction result of the multimodal signal is output first, and the recognition result of the image is changed to ε 上 to ε 下 When the state of ε changes, the state prediction result of the multimodal signal is verified. If the state prediction result of the priority output is not within ε 上 or ε 下 Then, a multimodal signal with a time length of T before and after the state change and a multimodal signal identified as a period of suspected device state change are obtained and put into the training set of the neural network for training.

[0096] It should be said that when the state prediction result of the multimodal signal is in the device state set, the fusion result is directly generated as the prediction result. This design aims to improve the accuracy of state recognition by utilizing the diversity and redundancy of multimodal data. If the prediction result of the multimodal signal matches the state set of image recognition, it means that the system has a high degree of credibility in judging the current state of the device, and thus directly adopts the result. If the prediction result of the multimodal signal is not in the device state set, this may indicate that there is an abnormality in the device state or that the sensor data does not match the actual situation. By generating an error, the system can promptly issue a warning to the operation and maintenance personnel to remind them of potential problems, thereby facilitating rapid response and troubleshooting. When the device state is identified as a suspected change, the state prediction result of the multimodal signal is output first, because the prediction of the multimodal signal has high real-time and flexibility. When the result of image recognition confirms the state change, the prediction result of the multimodal signal is verified to ensure the accuracy of the state change.

[0097] When the multimodal signal prediction results of the priority output are inconsistent with the state set before and after the state change of image recognition, it indicates that the model may need further adjustment. By obtaining multimodal signals with a time length of T before and after the state change and the signal data identified as the suspected state change period, and retraining the neural network training set, the model's ability to recognize specific state changes can be improved. This adaptive learning mechanism enables the model to continuously optimize itself and improve its accuracy and robustness in complex scenarios.

[0098] The design takes advantage of the complementarity of multimodal signals and image recognition. In most cases, the results of multimodal signals and image recognition are consistent, which improves the robustness of the system. By cross-validating the results of two independent signal sources, the system can more reliably identify the device state. By reporting and further verifying the inconsistency between the multimodal signal and image recognition results, the occurrence of false positives and false negatives can be effectively reduced. This strategy ensures that the state change is output only when the state change is confirmed, avoiding unnecessary intervention caused by system false positives. Prioritizing the output of the state prediction results of the multimodal signal enables real-time response, and when it is identified as a suspected change in the device state, high accuracy can be maintained in dynamic adjustment by verifying the image recognition results. This design enables the system to strike a balance between fast response and accuracy. By training the multimodal signals before and after the state change into the training set of the neural network, the system can self-learn and adapt to new state changes. This continuous optimization mechanism helps to improve the performance of the model in complex and changing practical application environments, thereby improving the performance of the system in the long term. This design enables the system to automatically take action when an abnormal state is detected, and adjust the model through retraining, effectively reducing the need for manual intervention. Automated fault diagnosis and learning mechanisms have greatly improved operation and maintenance efficiency and reduced labor costs.

[0099] It should also be noted that during the operation of the equipment, state changes are usually accompanied by some degree of signal mutation. For example, sensor data such as temperature, vibration, or current may have abnormal peaks or sudden changes in a short period of time. If these mutation signals are input into the neural network for state prediction alone, the network may misjudge them as equipment states that do not match the actual situation. This misjudgment will affect the accuracy and reliability of the system. Therefore, by inputting the normal signals and mutation signals in the time period before and after the state change into the neural network for retraining, the network can learn the reasonable interpretation of these mutation signals in different state changes, thereby avoiding misjudgment. The change of equipment state is not an isolated emergency, but a dynamic process. By combining multimodal signals before and after the mutation, the entire process from stable to mutation and then to stable can be captured. This analysis of long-term features helps the neural network understand the integrity of state changes. Through retraining, the neural network can identify which signal changes are short-term, harmless anomalies, and which changes may indicate real changes in the equipment state. This learning based on long-term features can improve the network's sensitivity and accuracy to state changes. Mutation signals may not always mean equipment failure, and sometimes may be caused by short-term environmental changes, measurement errors, or other harmless events. By incorporating normal signals before and after the mutation into the training set, the neural network can learn to "tolerate" these short-term anomalies, thereby reducing false alarms in actual operation. In this way, the system learns when to ignore the mutation signal and when to be alert and take action, thereby improving the robustness of the overall system. The complexity of the equipment operating environment requires the neural network to have strong adaptability. By continuously incorporating the mutation signals encountered in actual operation and the normal signals before and after them into the training set, the neural network can continuously adjust its prediction model to adapt to the state changes of the equipment under different operating conditions. This dynamic retraining mechanism enables the network to better respond to new signal patterns and reduce misjudgments caused by equipment aging, environmental changes or other factors. Equipment state changes are often not instantaneous, but are gradually reflected through a series of gradual signal changes. By incorporating multimodal signals before and after the state change and during the mutation into the training, the neural network can learn the characteristics of this gradual change. This not only helps to improve the network's ability to predict state changes, but also can provide early warning before the state actually changes, providing longer response time for maintenance and repair.

[0100] Finally, the generated fusion result is used as the final recognition result and output to the front end; at the same time, a data record of the multimodal signal is generated.

[0101] On the other hand, this embodiment also provides an AI-based intelligent identification system for equipment status in a distribution station area, which includes:

[0102] The acquisition unit collects data from sensors on the device to obtain multi-modal signals.

[0103] The prediction unit inputs the multimodal signal into a neural network for state prediction.

[0104] The recognition unit collects the monitoring image of the device in real time through the image acquisition device, and recognizes the device status in the image.

[0105] The generation unit fuses the image recognition result and the prediction result of the neural network to generate the final recognition result.

[0106] If the above functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0107] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0108] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0109] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0110] Example 2 is an embodiment of the present invention, which provides an AI-based intelligent identification method for equipment status in a distribution station area. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0111] The experimental site is set up in a typical urban distribution substation, equipped with various types of electrical equipment, including transformers, switchgear, and capacitor banks. In order to achieve real-time monitoring and status recognition, the experiment installed a variety of sensors on these devices, such as current sensors, voltage sensors, temperature sensors, vibration sensors, etc., and set up image acquisition equipment to obtain real-time monitoring images of the equipment.

[0112] During the experiment, the sensor collects the current, voltage, temperature, vibration and other data of the equipment in real time, forming a multi-modal signal that is input into the neural network for state prediction. The neural network uses a convolutional neural network (CNN) for feature extraction, and then fuses the features of different modes through a multi-head attention mechanism, and finally inputs the fully connected layer for state prediction. In addition, the camera installed in the station area collects the monitoring images of the equipment in real time, and uses CNN to extract the spatial features of the image after standardization and denoising, and uses a long short-term memory network (LSTM) to analyze these features to detect whether the equipment has periodic behavior.

[0113] In each state recognition process, the system first generates a state prediction result of a multimodal signal, and at the same time generates a possible device state set through image analysis. If the prediction result of the multimodal signal is in the state set generated by image recognition, the system directly outputs the prediction result of the multimodal signal as the final recognition result. If it is not in the state set, the system will generate an error and further verify and train.

[0114] The following is a summary of the data recorded during the experiment, which includes the prediction results of multimodal signals, the set of possible states of image recognition, and the final comprehensive recognition results.

[0115] Table 1 Data record table

[0116]

[0117]

[0118] It can be seen from the experimental data that the present invention realizes efficient recognition of the equipment status in the distribution station area through comprehensive analysis of multimodal signals and image recognition. Specifically, the combination of multimodal signal prediction and image recognition enables the system to accurately determine the equipment status under various operating conditions. The following is an analysis of each experimental scenario:

[0119] Transformer 1 and Transformer 2: In both scenarios, the results of multimodal signals and image recognition are normal operation or cyclic operation. Transformer 1's multimodal signal predicts normal operation, the state set of image recognition includes normal operation and cyclic operation, and the final comprehensive recognition result is normal operation; while Transformer 2's multimodal signal predicts cyclic operation, the state set of image recognition also includes cyclic operation, and the final result is cyclic operation. This shows that the system has high accuracy and consistency in judging the normal operation and cyclic operation of the transformer.

[0120] Switchgear 1 and Switchgear 2: Switchgear 1 was identified as operating normally in both detection methods, reflecting that the operating status of the equipment is stable and the system recognition results are consistent. The multimodal signal of switchgear 2 is predicted to be an abnormal state, and the state set of image recognition also includes abnormal states, which shows that the system can effectively identify the abnormal state of switchgear 2. Through this cross-validation of multimodal and image recognition, false alarms can be effectively avoided and the reliability of equipment fault detection can be improved.

[0121] Capacitor Banks 1, 2, and 3: For capacitor banks 1 and 2, the multimodal signal prediction and image recognition results are consistent, and can accurately identify the normal operating state and periodic operating state of the equipment. The multimodal signal prediction of capacitor bank 3 is an abnormal state, and the image recognition results also confirm this, and the final comprehensive recognition result is an abnormal state. This shows that when the system processes complex equipment states, it can achieve higher recognition accuracy by combining multimodal signals and images.

[0122] In summary, the experimental results show the advantages of the method of the present invention in the consistency and accuracy of multimodal signals and image recognition results. When the multimodal signal and image recognition results are consistent, the system can quickly output the final result; when the two results are inconsistent, the system can further optimize its own performance through error reporting and retraining mechanisms. This method effectively improves the accuracy and robustness of equipment status recognition, significantly enhances the intelligence level of equipment status monitoring in distribution substations, and provides a strong guarantee for the stable operation of smart grids. By combining and analyzing the data before and after the change, the system can avoid misjudgments caused by short-term mutations, thereby improving the reliability and adaptability of status recognition.

[0123] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should be included in the scope of the claims of the present invention.

Claims

1. The AI-based intelligent identification method for equipment status in distribution station area is characterized by: include: Collect data from sensors on the device to obtain multi-modal signals; Inputting the multimodal signal into a neural network for state prediction; Through the image acquisition device, the monitoring image of the device is collected in real time, and the device status in the image is identified; The recognition results of the image and the prediction results of the neural network are combined to generate the final recognition results.

2. The AI-based intelligent identification method for equipment status in a distribution station area according to claim 1, characterized in that: The sensors include current sensors, voltage sensors, temperature sensors, vibration sensors, humidity sensors, air pressure sensors, electromagnetic sensors, gas sensors, optical sensors, and position sensors; According to the monitorable data of the device, a sensor for acquiring the monitorable data is configured on the device, and a multi-modal signal is obtained through the sensor.

3. The AI-based intelligent identification method for equipment status in a distribution station area according to claim 2, characterized in that: The neural network includes, for different devices, constructing a training set and a validation set based on a multimodal signal set obtained by combining sensors matched with the devices; Suppose there are N sensors, and the collected multimodal signals are represented by X = {x1, x2, …, x N }; Among them, each x i Represents the data of the i-th sensor; Use convolutional neural network for preliminary feature extraction. The feature extraction process is expressed as: F i =CNN i (x i ) Among them, F i Indicates that from x i Extracted features; CNN i represents the convolutional neural network for the i-th sensor signal; i = 1, 2, ..., N; Fusion of features from different modalities through multi-head attention mechanism: F fusion =Attention(F1,F2,…,F N ) Among them, Attention represents a multi-head attention mechanism, which is used for the interaction and fusion of different features; The fused features are input into a fully connected layer for state prediction: y=σ(WF fusion +b) Among them, W represents the weight matrix, b represents the bias vector, and σ represents the activation function ReLU.

4. The AI-based intelligent identification method for equipment status in a distribution station area according to claim 3 is characterized in that: The monitoring image of the real-time acquisition device includes all the picture information monitored by the image acquisition device, and the device state is identified by learning the characteristics of the device motion state; The device motion state includes the actions of the device when it is working and performing different operations; if multiple device states show the same motion state, then in the process of image recognition, it is identified as a set of all possible device states, and each element in the set represents a possible device state; if one device state only matches one motion state, then the device state set generated by recognition only contains one element.

5. The AI-based intelligent identification method for equipment status in a distribution station area according to claim 4 is characterized in that: The device status in the image is identified by: t After standardization and denoising, it is expressed as: I pre (t)=Preprocess(I t ) Among them, Preprocess represents the preprocessing process, I pre (t) represents the processed image; Use CNN to extract the spatial features of each frame of image; Using a long short-term memory network, the extracted feature sequence is input into a gated recurrent unit, and time series modeling is used to detect whether there is a periodic behavior pattern: P cyclic =σ(LSTM(F seq (1),F seq (2),…,F seq (T))) Among them, P cyclic It is a binary classification output, indicating whether the behavior is periodic. The output value 1 indicates periodic behavior, and 0 indicates non-periodic behavior. T indicates the length of the image sequence. If it is not a periodic behavior, it is identified as a suspected change in the device state; if it is a periodic behavior, a periodic behavior is extracted, a complete action cycle is extracted and its features are constructed: Cycle boundary identification: Identify the starting point of the action cycle through time series analysis start and the end point t end ; Extracting periodic features: Input the image sequence within the period into LSTM to extract the periodic behavior features F cycle ; F cycle =LSTM cycle (F seq (t start ),F seq (t start +1),…,F seq (t end )) Among them, LSTM cycle is an LSTM model specifically used for period feature extraction. seq (t start ) indicates t start The spatial characteristics of the moment; Use the trained decision tree model DecisionTreeClassifier and input the local feature F cycle , output device state set ε: ε=DecisionTreeClassifier(F cycle ) When the device status is suspected to have changed, the last device status identification result ε is obtained. 上 , obtain the next identification device status result, if the device status is still suspected to have changed, re-obtain the next identification device status result until the device status result ε can be identified 下 Until; and change the identification result of suspected device status change during this period to: 上 to ε 下 Status changes.

6. The AI-based intelligent identification method for equipment status in a distribution station area according to claim 5, characterized in that: The fusion includes, when the device state set ε is identified, if the state prediction result through the multimodal signal is in the device state set ε, directly generating a fusion result as the state prediction result of the multimodal signal; if the state prediction result through the multimodal signal is not in the device state set ε, generating an error; When the device state is suspected to have changed, the state prediction result of the multimodal signal is output first, and the recognition result of the image is changed to ε 上 to ε 下 When the state of ε changes, the state prediction result of the multimodal signal is verified. If the state prediction result of the priority output is not within ε 上 or ε 下 Then, a multimodal signal with a time length of T before and after the state change and a multimodal signal identified as a period of suspected device state change are obtained and put into the training set of the neural network for training.

7. The AI-based intelligent identification method for equipment status in a distribution station area according to claim 6, characterized in that: The generated fusion result is used as the final recognition result and output to the front end; At the same time, a data record of the multimodal signal is generated.

8. An AI-based intelligent identification system for equipment status in a distribution station area using the method according to any one of claims 1 to 7, characterized in that: A collection unit collects data from sensors on the device to obtain multi-modal signals; A prediction unit, inputting the multimodal signal into a neural network for state prediction; The recognition unit collects the monitoring image of the device in real time through the image acquisition device, and recognizes the device status in the image; The generation unit fuses the image recognition result and the prediction result of the neural network to generate the final recognition result.

9. A computer device comprising: Memory and processor; The memory stores a computer program, which is characterized in that when the processor executes the computer program, it implements the steps of the AI-based intelligent identification method for the status of equipment in a distribution station area.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for intelligently identifying the status of equipment in a distribution station area based on AI are implemented.

Images