Real-time status monitoring system and method of intelligent operation and maintenance robots based on artificial intelligence

Abstract

The present invention discloses a real-time status monitoring system and method for an intelligent operation and maintenance robot based on artificial intelligence, belonging to the field of artificial intelligence technology. In the present invention, the intelligent operation and maintenance robot arrives at each equipment monitoring point in a substation, communication room, or data center to collect multi-dimensional data; upon arriving at the monitoring point, it obtains real-time electrical parameter data, fuses it with the multi-dimensional data, and obtains a data set; uses a convolutional neural network combined with a long-short-term memory network to build a main analysis model, inputs the trained main analysis model, and outputs the current status evaluation result of the equipment; measures the cooling medium parameters, calculates the enthalpy difference change during the cooling process, and quantifies the heat exchange efficiency of the cooling system; uses a support vector regression algorithm to build a transformer cooling system performance evaluation model, and outputs the current heat dissipation capacity coefficient of the cooling system; combines the current heat dissipation capacity coefficient of the cooling system with the current status evaluation result of the equipment to determine whether the cooling system meets the heat dissipation requirements.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a real-time status monitoring system and method for an intelligent operation and maintenance robot based on artificial intelligence. Background Art

[0002] Substations are critical hubs in the power system, responsible for voltage conversion, power distribution, and transmission. The safe and stable operation of their equipment is directly related to the reliability of the power grid. Any equipment failure can cause a power outage, impacting industrial production, commercial activities, and residents' lives. Communications rooms are the core facilities of the communications network, hosting a large number of servers, switches, routers, and other communications equipment. These devices ensure data transmission, exchange, and storage. Data centers are critical infrastructure in today's digital age, centrally storing and processing massive amounts of data, providing computing resources and data services for various applications such as cloud computing, big data, and artificial intelligence.

[0003] For example, traditional manual inspections of substations make it difficult to achieve real-time, continuous monitoring of substation equipment. Due to limited manpower, inspection intervals are often long, making it impossible to promptly detect unexpected anomalies during equipment operation. Under traditional O&M models, the recording and organization of test data is primarily manual, resulting in inconsistent data formats and decentralized storage, making large-scale data integration and deep mining difficult. When manual inspections discover equipment anomalies, they rely primarily on the experience and expertise of O&M personnel for fault diagnosis and decision-making, making it difficult to quickly and accurately determine the cause of the fault and develop the optimal repair plan. Summary of the Invention

[0004] The purpose of the present invention is to provide a real-time status monitoring system and method for an intelligent operation and maintenance robot based on artificial intelligence to solve the problems raised in the prior art.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A real-time status monitoring method for an intelligent operation and maintenance robot based on artificial intelligence, the method comprising the following steps:

[0007] The intelligent operation and maintenance robot arrives at each equipment monitoring point according to the preset inspection route and time interval; activates the sensor array to collect multi-dimensional data;

[0008] Acquire real-time electrical parameter data from the equipment's own monitoring system, and fuse the real-time electrical parameter data with multi-dimensional data to obtain a substation dataset;

[0009] The substation dataset is preprocessed, and a primary analysis model is built using a convolutional neural network combined with a long-short-term memory network. The convolutional neural network is used to automatically extract local features from image and sensor time series data, and the long-short-term memory network is used to capture temporal dependencies in the data sequence. The primary analysis model is trained and optimized, and the preprocessed substation dataset is input into the trained primary analysis model to output an assessment of the current equipment status.

[0010] Measure cooling medium parameters, calculate enthalpy difference changes during the cooling process, and quantify the cooling system's heat exchange efficiency. Based on thermodynamic principles and historical operating data, a support vector regression algorithm is used to construct a transformer cooling system performance evaluation model. The model inputs the cooling system's heat exchange efficiency, ambient temperature, and transformer load factor, and outputs the cooling system's current heat dissipation capacity coefficient.

[0011] The current heat dissipation capacity coefficient of the cooling system is combined with the current status evaluation result of the equipment to determine whether the cooling system meets the heat dissipation requirements and issue a fault warning.

[0012] Specifically, the A* algorithm is used to grid the equipment distribution map. Each grid node is assigned a travel cost. The starting point is set as the robot charging and standby area, and the end point is the monitoring point of each equipment. The A* algorithm is used to calculate the route with the lowest overall cost, ensuring that the robot can traverse all equipment in the shortest possible path while avoiding live hazardous areas and equipment maintenance work sites.

[0013] The user sets the time interval. When the robot arrives at each equipment monitoring point, the lidar scans the surrounding environment contours in real time and matches them with the pre-stored map features. The visual odometry assists in correcting the displacement deviation, allowing the robot to monitor the position of the high-voltage switch cabinet door.

[0014] The robot activates its sensor array to capture panoramic and detailed close-up images of the device from different angles. Multispectral imaging is used to capture infrared and ultraviolet images of the device. Infrared imaging is used to locate heating areas, while ultraviolet imaging is used to detect ultraviolet light generated by partial discharge.

[0015] The robot uses temperature and humidity sensors to obtain real-time information about the temperature and humidity of the environment in which the device is located. It also uses gas sensors to obtain real-time information about the concentration of gases in the environment in which the device is located. It also uses ultrasonic sensors to transmit and receive ultrasonic waves, which penetrate the device's casing to detect the integrity of the internal structure.

[0016] The robot integrates the collected data to form multi-dimensional data.

[0017] The robot has pre-built-in drivers for several substation equipment communication interfaces. Through electrical signal feature matching and handshake protocol interaction, it determines the interface type, loads the corresponding driver, and establishes a physical link connection.

[0018] Specifically, the robot integrated protocol parsing engine, for IEC 61850 protocols, parses the service model from the abstract communication service interface layer, obtains the real-time data model object of the device, and actively obtains real-time electrical parameter data, including voltage, current instantaneous values, power factor, and circuit breaker opening and closing position signals, at a set period through report services and read services. For private protocols, it reversely parses the byte stream data format based on the protocol documents provided by the equipment manufacturer, locates and extracts the real-time electrical parameter data.

[0019] The network time protocol is used to synchronize with the clock source in the substation, and the clock accuracy is controlled at the millisecond level. The real-time electrical parameter data is fused with the multi-dimensional data based on the timestamp to obtain the substation data set.

[0020] The substation dataset was preprocessed, including data cleaning and normalization. A convolutional neural network was constructed, using a multi-layer convolutional architecture to automatically extract low-level texture and edge features from images and sensor time series data curves. As the network deepened, pooling layers were introduced to reduce data dimensionality and computational complexity. For time series data, the convolutional neural network's one-dimensional convolution kernel slid along the time dimension to capture correlation features from data at different time intervals.

[0021] Specifically, the convolutional neural network is connected to a long short-term memory network. In the long short-term memory network, the input gate, forget gate, and output gate use a sigmoid activation function to control the flow of information. The memory unit state is updated based on the tanh function. Several layers of long short-term memory networks are stacked, each layer containing several hidden units, so that the model can capture the gradual change trend of operating parameters in the time series.

[0022] The preprocessed substation dataset is randomly divided into training, validation, and test sets in proportion. The model is trained using a mini-batch stochastic gradient descent algorithm, which is adjusted according to the dataset size and computing resources. During the training process, a learning rate decay strategy is used to gradually adjust the model parameters as the number of training rounds increases.

[0023] Specifically, the accuracy of classification tasks, which include normal equipment and equipment failure, is evaluated based on the cross-entropy loss function. The mean square error loss is combined with the deviation of the quantitative prediction to measure the severity of the failure, and a composite loss function is formed to drive model training. An early stopping method is introduced to continuously monitor the loss value of the validation set. If the loss value does not decrease after several consecutive epochs, the training is terminated early to prevent overfitting and lock in the optimal generalization state of the model. L2 regularization is used to impose a constraint penalty on the model weights. A grid search strategy is used to traverse and adjust the number of convolution kernels, size, number of long and short-term memory network layers, and number of hidden units of the convolutional neural network to select the optimal model configuration based on the performance indicators of the validation set.

[0024] The trained and optimized main analysis model is deployed in the intelligent operation and maintenance robot. When the robot collects a new batch of substation data, it is input into the model in real time after preprocessing, and the model starts the forward propagation calculation; the model outputs the current status assessment results of the equipment, including the equipment health score, fault probability value and qualitative judgment of the fault type.

[0025] In the transformer cooling system of the substation, for the cooling oil, the inlet temperature T of the cooling oil is obtained in real time. oil，in and the outlet temperature of the cooling oil T oil,out , using Coriolis mass flowmeter to measure the mass flow of cooling oil m oil The measuring range depends on the scale of the transformer cooling system. For cooling air, the inlet temperature of the cooling air is measured. air,in and the outlet temperature of the cooling air T air,out ; Use a hot wire anemometer to measure the cooling air velocity, calculate the volume flow rate based on the cross-sectional area of the air duct, and convert it into the mass flow rate of the cooling air m according to the standard state equation air ;

[0026] According to the principles of thermodynamics, the calculation formula for the change in enthalpy of cooling oil is:

[0027] ΔH oil =m oil *c p,oil *(T oil,out -T oil,in );

[0028] Where ΔH oil is the change in cooling oil enthalpy, c p，oil is the specific heat capacity of the cooling oil at constant pressure;

[0029] The formula for the change in enthalpy of cooling air is:

[0030] ΔH air =m air *(h air,out -h air,in );

[0031] Where ΔH air is the change in cooling air enthalpy, h air，in is the specific enthalpy of the cooling air at the inlet, h air，out is the specific enthalpy of the cooling air outlet, h air，in =c p，air *T air，in +d,h air,out =c p,air *T air,out +d,c p,air is the specific heat capacity of air at constant pressure, d is a constant term related to the reference enthalpy setting;

[0032] For the enthalpy difference change of the entire cooling system ΔH system Based on the law of conservation of energy, the heat released by the cooling oil should be equal to the heat absorbed by the cooling air, expressed as: ΔH system =ΔH oil =ΔH air

[0033] The heat exchange efficiency η of the cooling system is defined as the ratio of the actual heat exchange capacity to the theoretical maximum heat exchange capacity, and the calculation formula is:

[0034]

[0035] Where ΔH actual It is the enthalpy difference change achieved by the cooling system in actual operation, and its value is ΔH system ; ΔH max The theoretical limit enthalpy difference change is achieved when the temperature difference between the inlet and outlet of the cooling medium reaches the maximum possible temperature difference. For the oil-air cooling system, assuming that the cooling air temperature is constant, ΔH max It is approximately the change in enthalpy difference when the cooling oil is cooled from the initial temperature to the cooling air temperature. The formula is expressed as:

[0036] ΔH max =m oil *c p,oil *(T oil，in -T air,in ).

[0037] In the support vector regression algorithm, the radial basis function is selected as the kernel function. During the model hyperparameter tuning stage, a grid search combined with a cross-validation strategy is used to traverse the penalty parameters and kernel functions. The optimal combination is locked based on the minimum mean square error criterion of the validation set, so that the model simulates the dynamic performance of the cooling system. The historical data is divided into a training set and a test set in proportion, and model training is started. During the training process, the model gradually learns the fluctuation law of the heat exchange efficiency of the cooling system under different combinations of ambient temperature and load rate, and constructs a mapping relationship from the input parameter space to the output space of the heat dissipation capacity coefficient. After training is completed, the heat exchange efficiency of the cooling system, the ambient temperature and the transformer load rate are input into the model, and the current heat dissipation capacity coefficient of the cooling system is output.

[0038] At the same time, the current heat dissipation capacity coefficient of the cooling system and the current status evaluation result of the equipment are captured; based on the substation equipment design manual and the statistical analysis of historical operation and maintenance data, the basic threshold range of the heat dissipation capacity coefficient is set for typical load conditions in different seasons and time periods, and the threshold is dynamically corrected in combination with the oil temperature index in the current status evaluation result of the equipment.

[0039] Specifically, when the cooling system's current heat dissipation capacity coefficient is within A1 of the lower threshold, the equipment health score is between B2 and B3, the failure probability is between C1% and C2%, and there are no critical fault types, a level 1 warning is triggered. Upon receiving the warning instruction, the intelligent operation and maintenance robot automatically shortens the inspection cycle and pushes the warning information to the operation and maintenance personnel's mobile terminal.

[0040] If the cooling system's current heat dissipation coefficient falls below the thresholds A1-A2, the equipment health score drops to B1-B2, and the failure probability rises to C2%-C3%, or a fault type that affects equipment operation but does not yet threaten power supply safety occurs, a Level 2 alert is triggered. In addition to monitoring robots to increase the frequency of inspections, the O&M system also issues a pop-up alert on the large screen of the substation monitoring center and sends an emergency plan to the O&M team.

[0041] When the heat dissipation coefficient is lower than the threshold value A2, the equipment health score is lower than B1, the failure probability exceeds C3%, or a high-risk fault type is identified, the third-level warning is activated. If conditions permit, the substation automatic control system will reduce the load of the equipment. If conditions do not permit, a local power outage will be carried out to avoid risks. An alarm will pop up on the large screen of the substation monitoring center, the sound and light prompts will be strengthened, and an emergency plan will be sent to the operation and maintenance team. Among them, A1, A2, B1, B2, B3, C1, C2 and C3 are all positive decimals and are automatically set by the system.

[0042] The AI-based real-time status monitoring system for intelligent operation and maintenance robots includes:

[0043] Inspection and data acquisition module: includes: a path planning and navigation unit and a sensor array activation and control unit; the path planning and navigation unit enables the intelligent operation and maintenance robot to reach each equipment monitoring point according to the preset inspection path and time interval; the sensor array activation and control unit activates the sensor array and collects multi-dimensional data;

[0044] Data fusion and preprocessing module: includes: electrical parameter acquisition and adaptation unit, data fusion unit and preprocessing unit; wherein, the electrical parameter acquisition and adaptation unit obtains real-time electrical parameter data from the equipment's own monitoring system, the data fusion unit fuses the real-time electrical parameter data with multi-dimensional data to obtain a substation data set, and the preprocessing unit preprocesses the substation data set;

[0045] Model building and analysis module: includes: a model building unit, a model training and optimization unit, and a real-time evaluation unit; the model building unit uses a convolutional neural network combined with a long short-term memory network to build a main analysis model. The convolutional neural network is used to automatically extract local features from image and sensor time series data, and the long short-term memory network is used to capture the time dependency in the data sequence; the model training and optimization unit trains and optimizes the main analysis model, and the real-time evaluation unit inputs the preprocessed substation data set into the trained main analysis model and outputs the current status evaluation results of the equipment;

[0046] Heat dissipation coefficient calculation module: includes: enthalpy difference calculation and efficiency evaluation unit and heat dissipation coefficient calculation unit; the enthalpy difference calculation and efficiency evaluation unit measures the cooling medium parameters, calculates the enthalpy difference change during the cooling process, and quantifies the heat exchange efficiency of the cooling system; the heat dissipation coefficient calculation unit uses the support vector regression algorithm based on thermodynamic principles and historical operating data to construct a transformer cooling system performance evaluation model, inputs the cooling system heat exchange efficiency, ambient temperature and transformer load rate, and outputs the current heat dissipation capacity coefficient of the cooling system;

[0047] Fault warning and decision support module: includes: a comprehensive judgment unit and a warning push unit; wherein, the comprehensive judgment unit combines the current heat dissipation capacity coefficient of the cooling system with the current status evaluation result of the equipment, and the warning push unit determines whether the cooling system meets the heat dissipation requirements and issues a fault warning.

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] 1. The intelligent operation and maintenance robot of the present invention can autonomously shuttle within the substation based on preset inspection paths and time intervals, accurately arrive at each equipment monitoring point according to the set time intervals, activate the sensor array, and quickly and synchronously collect multi-dimensional data.

[0050] 2. This paper uses a convolutional neural network combined with a long short-term memory network to build a main analysis model to conduct in-depth analysis of the massive substation data set collected. The convolutional neural network can automatically extract local features in image and sensor time series data, while the long short-term memory network is good at capturing the time dependency in the data sequence.

[0051] 3. Based on the analysis results of the artificial intelligence algorithm, the present invention can provide intelligent decision-making support for substation operation and maintenance. The system automatically determines whether the equipment has a failure risk and whether the cooling system meets the heat dissipation requirements based on the evaluation results of the current status of the equipment and the performance of the cooling system, and provides graded warnings based on the severity of the failure. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 This is a schematic diagram of the steps of the real-time status monitoring method of the intelligent operation and maintenance robot based on artificial intelligence of the present invention;

[0053] Figure 2 This is a system structure diagram of the real-time status monitoring system of the intelligent operation and maintenance robot based on artificial intelligence of the present invention. DETAILED DESCRIPTION

[0054] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0055] Example: Figure 1-Figure 2 As shown, the present invention provides a technical solution.

[0056] According to one embodiment of the present invention, Figure 1 The steps of the method for real-time status monitoring of an intelligent operation and maintenance robot based on artificial intelligence are shown in the schematic diagram. The method for real-time status monitoring of an intelligent operation and maintenance robot based on artificial intelligence includes the following steps:

[0057] The intelligent operation and maintenance robot arrives at each equipment monitoring point according to the preset inspection route and time interval; activates the sensor array to collect multi-dimensional data;

[0058] Acquire real-time electrical parameter data from the equipment's own monitoring system, and fuse the real-time electrical parameter data with multi-dimensional data to obtain a substation dataset;

[0059] The substation dataset is preprocessed, and a primary analysis model is built using a convolutional neural network combined with a long-short-term memory network. The convolutional neural network is used to automatically extract local features from image and sensor time series data, and the long-short-term memory network is used to capture temporal dependencies in the data sequence. The primary analysis model is trained and optimized, and the preprocessed substation dataset is input into the trained primary analysis model to output an assessment of the current equipment status.

[0060] Measure cooling medium parameters, calculate enthalpy difference changes during the cooling process, and quantify the cooling system's heat exchange efficiency. Based on thermodynamic principles and historical operating data, a support vector regression algorithm is used to construct a transformer cooling system performance evaluation model. The model inputs the cooling system's heat exchange efficiency, ambient temperature, and transformer load factor, and outputs the cooling system's current heat dissipation capacity coefficient.

[0061] The current heat dissipation capacity coefficient of the cooling system is combined with the current status evaluation result of the equipment to determine whether the cooling system meets the heat dissipation requirements and issue a fault warning.

[0062] Specifically, the A* algorithm is used to grid the equipment distribution map. Each grid node is assigned a travel cost. The starting point is set as the robot charging and standby area, and the end point is the monitoring point of each equipment. The A* algorithm is used to calculate the route with the lowest overall cost, ensuring that the robot can traverse all equipment in the shortest possible path while avoiding live hazardous areas and equipment maintenance work sites.

[0063] The user sets the time interval. When the robot arrives at each equipment monitoring point, the lidar scans the surrounding environment contours in real time and matches them with the pre-stored map features. The visual odometry assists in correcting the displacement deviation, allowing the robot to monitor the position of the high-voltage switch cabinet door.

[0064] The robot activates its sensor array to capture panoramic and detailed close-up images of the device from different angles. Multispectral imaging is used to capture infrared and ultraviolet images of the device. Infrared imaging is used to locate heating areas, while ultraviolet imaging is used to detect ultraviolet light generated by partial discharge.

[0065] The robot uses temperature and humidity sensors to obtain real-time information about the temperature and humidity of the environment in which the device is located. It also uses gas sensors to obtain real-time information about the concentration of gases in the environment in which the device is located. It also uses ultrasonic sensors to transmit and receive ultrasonic waves, which penetrate the device's casing to detect the integrity of the internal structure.

[0066] The robot integrates the collected data to form multi-dimensional data.

[0067] The robot has pre-built-in drivers for several substation equipment communication interfaces. Through electrical signal feature matching and handshake protocol interaction, it determines the interface type, loads the corresponding driver, and establishes a physical link connection.

[0068] Specifically, the robot integrated protocol parsing engine, for IEC 61850 protocols, parses the service model from the abstract communication service interface layer, obtains the real-time data model object of the device, and actively obtains real-time electrical parameter data, including voltage, current instantaneous values, power factor, and circuit breaker opening and closing position signals, at a set period through report services and read services. For private protocols, it reversely parses the byte stream data format based on the protocol documents provided by the equipment manufacturer, locates and extracts the real-time electrical parameter data.

[0069] The network time protocol is used to synchronize with the clock source in the substation, and the clock accuracy is controlled at the millisecond level. The real-time electrical parameter data is fused with the multi-dimensional data based on the timestamp to obtain the substation data set.

[0070] The substation dataset was preprocessed, including data cleaning and normalization. A convolutional neural network was constructed, using a multi-layer convolutional architecture to automatically extract low-level texture and edge features from images and sensor time series data curves. As the network deepened, pooling layers were introduced to reduce data dimensionality and computational complexity. For time series data, the convolutional neural network's one-dimensional convolution kernel slid along the time dimension to capture correlation features from data at different time intervals.

[0071] Specifically, the convolutional neural network is connected to a long short-term memory network. In the long short-term memory network, the input gate, forget gate, and output gate use a sigmoid activation function to control the flow of information. The memory unit state is updated based on the tanh function. Several layers of long short-term memory networks are stacked, each layer containing several hidden units, so that the model can capture the gradual change trend of operating parameters in the time series.

[0072] In this embodiment, a convolutional neural network (CNN) is constructed. The first-layer convolution kernel is set to a size of 3×3, a step size of 1, and a number of 32 convolution kernels. ReLU is used as the activation function to preliminarily extract low-level texture and edge features of the image and sensor time series data curves. As the network deepens, the size of the second-layer convolution kernel is adjusted to 5×5, and the number is increased to 64. The ReLU activation function is also used. After that, a maximum pooling layer is introduced with a pooling kernel of 2×2 and a step size of 2 to reduce the data dimension and the amount of calculation. For time series data, the one-dimensional convolution kernel of CNN slides along the time dimension. For example, the convolution kernel size is set to 3 and the step size is 1 to capture the correlation features of data in different time periods.

[0073] The CNN is connected to a long short-term memory (LSTM) network. In the LSTM, the input, forget, and output gates use sigmoid activation functions to control the flow of information, and memory cell state updates are based on the tanh function. Two LSTM layers, each containing 128 hidden units, are stacked, enabling the model to capture the gradual changes in operating parameters in time series.

[0074] The model was trained using a mini-batch stochastic gradient descent algorithm. The preprocessed substation dataset was randomly split into training, validation, and test sets in a ratio of 70%, 20%, and 10%. Based on the dataset size and computing resources, the mini-batch size was set to 64. The initial learning rate was set to 0.001, and a learning rate decay strategy was used during training. After every five epochs, the learning rate was decayed by 0.9 times the original value, and the model parameters were gradually adjusted with each training round.

[0075] The accuracy of the classification task (classifying normal and faulty devices) was evaluated using the cross-entropy loss function, combined with the mean squared error loss to measure the fault severity and quantify the prediction bias. This composite loss function was used to drive model training. An early stopping algorithm was introduced to continuously monitor the validation set loss. If the loss did not decrease after 10 consecutive epochs, training was terminated early to prevent overfitting and lock in the model's optimal generalization state. L2 regularization was applied to constrain the model weights, with a regularization coefficient of 0.01. A grid search strategy was used to iteratively adjust the number of convolutional kernels (ranging from 16 to 64 with a step size of 16), the size (3×3, 5×5), the number of LSTM layers (1 to 3), and the number of hidden units (64 to 256 with a step size of 64) of the CNN. The optimal model configuration was selected based on validation set performance metrics (accuracy, recall, F1 score, root mean squared error, etc.).

[0076] The model is trained using the training set. During the training process, as the number of epochs increases, the loss value of the model on the training set gradually decreases, and the performance indicators on the validation set also change.

[0077] During the grid search process, a comprehensive evaluation of various hyperparameter combinations was conducted. The model achieved the best overall performance on the validation set when the number of CNN convolution kernels was 32, the size was 3×3, the number of LSTM layers was 2, and the number of hidden units was 128. The model achieved an accuracy of 92%, a recall of 90%, an F1 value of 0.91, and a root mean square error of 0.12. These indicators demonstrate that the model can accurately classify normal and faulty equipment states and achieve high accuracy in quantitatively predicting fault severity.

[0078] Specifically analyzing the model's output of the equipment's current status assessment results, the model accurately assigns a health score based on the equipment's multi-dimensional data. For example, a main transformer in good operating condition is assessed to have a health score of 90 points. Equipment with potential hazards (such as slightly elevated oil temperature or slightly abnormal vibration) has a health score between 60 and 80 points. Equipment showing obvious signs of failure (such as partial discharge or insulation aging) has a health score below 60 points.

[0079] In terms of numerical prediction of fault probabilities, the model can provide corresponding probability estimates for different types of faults. For example, for a certain switchgear, the model predicted a 15% probability of a poor contact failure. Further inspection revealed slight oxidation at the switchgear's connections, resulting in a slight increase in contact resistance, validating the accuracy of the model's prediction.

[0080] The model can accurately identify a variety of common fault types. For example, when monitoring a transformer, the model identified the fault type as a "core multi-point grounding fault." Disassembling the transformer revealed that the core indeed had a multi-point grounding problem, further demonstrating the model's reliability in fault diagnosis.

[0081] In the transformer cooling system of the substation, for the cooling oil, the inlet temperature T of the cooling oil is obtained in real time. oil，in and the outlet temperature of the cooling oil T oil，out , using Coriolis mass flowmeter to measure the mass flow of cooling oil m oil The measuring range depends on the scale of the transformer cooling system. For cooling air, the inlet temperature of the cooling air is measured. air，in and the outlet temperature of the cooling air T air，out ; Use a hot wire anemometer to measure the cooling air velocity, calculate the volume flow rate based on the cross-sectional area of the air duct, and convert it into the mass flow rate of the cooling air m according to the standard state equation air ;

[0082] According to the principles of thermodynamics, the calculation formula for the change in enthalpy of cooling oil is:

[0083] ΔH oil =m oil *c p,oil *(T oil,out -T oil,in );

[0084] Where ΔH oil is the change in cooling oil enthalpy, c p，oil is the specific heat capacity of the cooling oil at constant pressure;

[0085] The formula for the change in enthalpy of cooling air is:

[0086] ΔH air =m air *(h air,out -h air,in );

[0087] Where ΔH air is the change in cooling air enthalpy, h air，in is the specific enthalpy of the cooling air at the inlet, h air，out is the specific enthalpy of the cooling air outlet, h air，in =c p,air *T air，in +d,h air,out =c p,air *T air,out +d,c p,air is the specific heat capacity of air at constant pressure, d is a constant term related to the reference enthalpy setting;

[0088] For the enthalpy difference change of the entire cooling system ΔH system Based on the law of conservation of energy, the heat released by the cooling oil should be equal to the heat absorbed by the cooling air, expressed as: ΔH system =ΔH oil =ΔH air ;

[0089] The heat exchange efficiency η of the cooling system is defined as the ratio of the actual heat exchange capacity to the theoretical maximum heat exchange capacity, and the calculation formula is:

[0090]

[0091] Where ΔH actual It is the enthalpy difference change achieved by the cooling system in actual operation, and its value is ΔH system ; ΔH max The theoretical limit enthalpy difference change is achieved when the temperature difference between the inlet and outlet of the cooling medium reaches the maximum possible temperature difference. For the oil-air cooling system, assuming that the cooling air temperature is constant, ΔH max It is approximately the change in enthalpy difference when the cooling oil is cooled from the initial temperature to the cooling air temperature. The formula is expressed as:

[0092] ΔH max =m oil *c p,oil *(T oil，in -T air,in ).

[0093] In the support vector regression algorithm, the radial basis function is selected as the kernel function. During the model hyperparameter tuning stage, a grid search combined with a cross-validation strategy is used to traverse the penalty parameters and kernel functions. The optimal combination is locked based on the minimum mean square error criterion of the validation set, so that the model simulates the dynamic performance of the cooling system. The historical data is divided into a training set and a test set in proportion, and model training is started. During the training process, the model gradually learns the fluctuation law of the heat exchange efficiency of the cooling system under different combinations of ambient temperature and load rate, and constructs a mapping relationship from the input parameter space to the output space of the heat dissipation capacity coefficient. After training is completed, the heat exchange efficiency of the cooling system, the ambient temperature and the transformer load rate are input into the model, and the current heat dissipation capacity coefficient of the cooling system is output.

[0094] At the same time, the current heat dissipation capacity coefficient of the cooling system and the current status evaluation result of the equipment are captured; based on the substation equipment design manual and the statistical analysis of historical operation and maintenance data, the basic threshold range of the heat dissipation capacity coefficient is set for typical load conditions in different seasons and time periods, and the threshold is dynamically corrected in combination with the oil temperature index in the current status evaluation result of the equipment.

[0095] Specifically, when the cooling system's current heat dissipation capacity coefficient is within A1 of the lower threshold, the equipment health score is between B2 and B3, the failure probability is between C1% and C2%, and there are no critical fault types, a level 1 warning is triggered. Upon receiving the warning instruction, the intelligent operation and maintenance robot automatically shortens the inspection cycle and pushes the warning information to the operation and maintenance personnel's mobile terminal.

[0096] If the cooling system's current heat dissipation coefficient falls below the thresholds A1-A2, the equipment health score drops to B1-B2, and the failure probability rises to C2%-C3%, or a fault type that affects equipment operation but does not yet threaten power supply safety occurs, a Level 2 alert is triggered. In addition to monitoring robots to increase the frequency of inspections, the O&M system also issues a pop-up alert on the large screen of the substation monitoring center and sends an emergency plan to the O&M team.

[0097] When the heat dissipation coefficient is lower than the threshold value A2, the equipment health score is lower than B1, the failure probability exceeds C3%, or a high-risk fault type is identified, the third-level warning is activated. If conditions permit, the substation automatic control system will reduce the load of the equipment. If conditions do not permit, a local power outage will be carried out to avoid risks. An alarm will pop up on the large screen of the substation monitoring center, the sound and light prompts will be strengthened, and an emergency plan will be sent to the operation and maintenance team. Among them, A1, A2, B1, B2, B3, C1, C2 and C3 are all positive decimals and are automatically set by the system.

[0098] According to another embodiment of the present invention, Figure 2 As shown in the system structure diagram of the real-time status monitoring system of the intelligent operation and maintenance robot based on artificial intelligence, the real-time status monitoring system of the intelligent operation and maintenance robot based on artificial intelligence includes:

[0099] Inspection and data acquisition module: includes: a path planning and navigation unit and a sensor array activation and control unit; the path planning and navigation unit enables the intelligent operation and maintenance robot to reach each equipment monitoring point according to the preset inspection path and time interval; the sensor array activation and control unit activates the sensor array and collects multi-dimensional data;

[0100] Data fusion and preprocessing module: includes: electrical parameter acquisition and adaptation unit, data fusion unit and preprocessing unit; wherein, the electrical parameter acquisition and adaptation unit obtains real-time electrical parameter data from the equipment's own monitoring system, the data fusion unit fuses the real-time electrical parameter data with multi-dimensional data to obtain a substation data set, and the preprocessing unit preprocesses the substation data set;

[0101] Model building and analysis module: includes: a model building unit, a model training and optimization unit, and a real-time evaluation unit; the model building unit uses a convolutional neural network combined with a long short-term memory network to build a main analysis model. The convolutional neural network is used to automatically extract local features from image and sensor time series data, and the long short-term memory network is used to capture the time dependency in the data sequence; the model training and optimization unit trains and optimizes the main analysis model, and the real-time evaluation unit inputs the preprocessed substation data set into the trained main analysis model and outputs the current status evaluation results of the equipment;

[0102] Heat dissipation coefficient calculation module: includes: enthalpy difference calculation and efficiency evaluation unit and heat dissipation coefficient calculation unit; the enthalpy difference calculation and efficiency evaluation unit measures the cooling medium parameters, calculates the enthalpy difference change during the cooling process, and quantifies the heat exchange efficiency of the cooling system; the heat dissipation coefficient calculation unit uses the support vector regression algorithm based on thermodynamic principles and historical operating data to construct a transformer cooling system performance evaluation model, inputs the cooling system heat exchange efficiency, ambient temperature and transformer load rate, and outputs the current heat dissipation capacity coefficient of the cooling system;

[0103] In this embodiment, the enthalpy difference change is calculated based on the collected cooling medium parameters and the heat exchange efficiency is evaluated. For example, a data set includes a cooling oil inlet temperature of 60°C, an outlet temperature of 50°C, and a mass flow rate of 100 kg / s; and cooling air inlet temperature of 25°C, an outlet temperature of 35°C, and a mass flow rate of 500 kg / s. Assuming the cooling oil's constant-pressure specific heat capacity within this temperature range is 2 kJ / (kg·°C), the cooling oil enthalpy change formula yields the following: 100 * 2 * (50 - 60) = -2000 kJ.

[0104] Assuming the specific heat capacity of air at constant pressure is 1.005 kJ / (kg·°C), and using an approximate calculation for dry air, the change in enthalpy of the cooling air is: 500 * 1.005 * (35 - 25) = 5025 kJ. Due to the law of conservation of energy, the actual heat exchange capacity should be the smaller of the two absolute values, i.e., 2000 kJ.

[0105] Assuming the theoretical maximum heat exchange capacity is 2500 kJ, the heat exchange efficiency at this moment is: 2000 / 2500*100%=80%.

[0106] Across the entire data set, the calculated heat exchange efficiency ranged from 60% to 95%, with an average efficiency of approximately 75%. However, when the ambient temperature was high (above 35°C) and the transformer load factor was greater than 80%, the heat exchange efficiency dropped significantly, even falling below 65% in some periods. This indicates that the cooling system faces significant pressure under high temperature and high load conditions.

[0107] The calculated heat exchange efficiency, along with the corresponding ambient temperature and transformer load factor data, was input into the constructed SVR model to calculate the heat dissipation coefficient. Grid search and cross-validation determined that the model performed best when the penalty parameter C = 1 and the kernel parameter γ = 0.1.

[0108] The dataset was divided into training set, validation set and test set in the proportion of 70%, 20% and 10%. After training the model on the training set, it was evaluated on the validation set. The mean square error (MSE) of the model on the validation set was 0.025, and the coefficient of determination (R 2 ) is 0.85. This shows that the model can fit the data well and has a high accuracy in predicting the heat dissipation capacity coefficient.

[0109] Testing on the test data set showed that the average error between the model's predicted and calculated heat dissipation coefficients was 0.03, with a relative error within 5%. For example, for a set of test data, the actual heat dissipation coefficient was 0.78, while the model predicted it to be 0.80.

[0110] By analyzing test data under different operating conditions, we found that when the ambient temperature is low (less than 20°C) and the load factor is less than 50%, the model's predicted heat dissipation coefficient is relatively stable and highly consistent with the actual value. Even when the ambient temperature and load factor fluctuate drastically, the model can still track the changing trend of the heat dissipation coefficient well, although the error is slightly larger, but still within an acceptable range. This demonstrates that the constructed SVR model can effectively calculate the heat dissipation coefficient based on the cooling system's heat exchange efficiency, ambient temperature, and transformer load factor, providing a reliable basis for determining whether the cooling system meets heat dissipation requirements.

[0111] Fault warning and decision support module: includes: a comprehensive judgment unit and a warning push unit; wherein, the comprehensive judgment unit combines the current heat dissipation capacity coefficient of the cooling system with the current status evaluation result of the equipment, and the warning push unit determines whether the cooling system meets the heat dissipation requirements and issues a fault warning.

[0112] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above and that the invention can be embodied in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered in all respects as illustrative and non-restrictive, and the scope of the invention is defined by the appended claims, not the foregoing description, and all variations within the meaning and range of equivalents of the claims are intended to be included therein. Any reference sign in a claim should not be construed as limiting the claim to which it relates.

Claims

1. A real-time status monitoring method for intelligent operation and maintenance robots based on artificial intelligence, characterized in that: The method comprises the following steps: The intelligent operation and maintenance robot arrives at each equipment monitoring point according to the preset inspection route and time interval; activates the sensor array to collect multi-dimensional data; Acquire real-time electrical parameter data from the equipment's own monitoring system, and fuse the real-time electrical parameter data with multi-dimensional data to obtain a substation dataset; The substation dataset is preprocessed, and a primary analysis model is built using a convolutional neural network combined with a long-short-term memory network. The convolutional neural network is used to automatically extract local features from image and sensor time series data, and the long-short-term memory network is used to capture temporal dependencies in the data sequence. The primary analysis model is trained and optimized, and the preprocessed substation dataset is input into the trained primary analysis model to output an assessment of the current equipment status. Measure cooling medium parameters, calculate enthalpy difference changes during the cooling process, and quantify the cooling system's heat exchange efficiency. Based on thermodynamic principles and historical operating data, a support vector regression algorithm is used to construct a transformer cooling system performance evaluation model. The model inputs the cooling system's heat exchange efficiency, ambient temperature, and transformer load factor, and outputs the cooling system's current heat dissipation capacity coefficient. In the transformer cooling system of the substation, for the cooling oil, the inlet temperature of the cooling oil is obtained in real time and cooling oil outlet temperature , using a Coriolis mass flow meter to measure the mass flow of cooling oil The measuring range depends on the scale of the transformer cooling system; for cooling air, the inlet temperature of the cooling air is measured. and the outlet temperature of the cooling air ; Use a hot wire anemometer to measure the cooling air flow rate, calculate the volume flow rate based on the cross-sectional area of the air duct, and convert it into the mass flow rate of the cooling air according to the standard state equation ; According to the principles of thermodynamics, the calculation formula for the change in enthalpy of cooling oil is: ； in, is the change in cooling oil enthalpy, is the specific heat capacity of the cooling oil at constant pressure; The formula for the change in enthalpy of cooling air is: ； in, is the change in cooling air enthalpy, is the specific enthalpy of the cooling air inlet, is the specific enthalpy of the cooling air outlet, , , is the specific heat capacity of air at constant pressure, d is a constant term related to the reference enthalpy setting; For the enthalpy difference change of the entire cooling system Based on the law of conservation of energy, the heat released by the cooling oil should be equal to the heat absorbed by the cooling air, which can be expressed as: ; Cooling system heat exchange efficiency It is defined as the ratio of actual heat exchange capacity to theoretical maximum heat exchange capacity, and the calculation formula is: ； in, is the enthalpy difference change achieved by the cooling system in actual operation, and its value is ; The theoretical limit enthalpy difference change is achieved when the temperature difference between the inlet and outlet of the cooling medium reaches the maximum possible temperature difference. For the oil-air cooling system, assuming that the cooling air temperature is constant, It is approximately the change in enthalpy difference when the cooling oil is cooled from the initial temperature to the cooling air temperature. The formula is expressed as: ； The current heat dissipation capacity coefficient of the cooling system is combined with the current status evaluation result of the equipment to determine whether the cooling system meets the heat dissipation requirements and issue a fault warning.

2. The method for real-time status monitoring of an intelligent operation and maintenance robot based on artificial intelligence according to claim 1 is characterized in that: Using the A* algorithm, the equipment distribution map is gridded, and each grid node is assigned a travel cost. The starting point is set as the robot charging and standby area, and the end point is the monitoring point of each equipment. The A* algorithm calculates the route with the lowest overall cost, ensuring that the robot can traverse all equipment in the shortest possible path while avoiding live hazardous areas and equipment maintenance work sites. The user sets the time interval. When the robot arrives at each equipment monitoring point, the lidar scans the surrounding environment contours in real time and matches them with the pre-stored map features. The visual odometry assists in correcting the displacement deviation, allowing the robot to monitor the position of the high-voltage switch cabinet door.

3. The method for real-time status monitoring of an intelligent operation and maintenance robot based on artificial intelligence according to claim 2 is characterized in that: The robot activates its sensor array to capture panoramic and detailed close-up images of the device from different angles. Multispectral imaging is used to capture infrared and ultraviolet images of the device. The infrared imaging is used to locate heating areas, while the ultraviolet imaging is used to detect ultraviolet light generated by partial discharge. The robot uses temperature and humidity sensors to obtain real-time information about the temperature and humidity of the environment in which the device is located. It also uses gas sensors to obtain real-time information about the concentration of gases in the environment in which the device is located. It also uses ultrasonic sensors to transmit and receive ultrasonic waves, which penetrate the device's casing to detect the integrity of the internal structure. The robot integrates the collected data to form multi-dimensional data.

4. The method for real-time status monitoring of an intelligent operation and maintenance robot based on artificial intelligence according to claim 3 is characterized in that: The robot has pre-built-in drivers for several substation equipment communication interfaces. Through electrical signal feature matching and handshake protocol interaction, it determines the interface type, loads the corresponding driver, and establishes a physical link connection. The robot integrated protocol parsing engine parses the service model of the IEC 61850 protocol from the abstract communication service interface layer, obtains the real-time data model object of the device, and actively obtains real-time electrical parameter data according to the set period through report service and read service, including voltage, current instantaneous value, power factor and circuit breaker opening and closing position signal; For private protocols, reverse parse the byte stream data format based on the protocol document provided by the equipment manufacturer to locate and extract the real-time electrical parameter data; The network time protocol is used to synchronize with the clock source in the substation, and the clock accuracy is controlled at the millisecond level. The real-time electrical parameter data is fused with the multi-dimensional data based on the timestamp to obtain the substation data set.

5. The method for real-time status monitoring of an intelligent operation and maintenance robot based on artificial intelligence according to claim 1 is characterized in that: The substation dataset was preprocessed, including data cleaning and normalization. A convolutional neural network was constructed, using a multi-layer convolutional architecture to automatically extract low-level texture and edge features from images and sensor time series data curves. As the network deepened, pooling layers were introduced to reduce data dimensionality and computational complexity. For time series data, the convolutional neural network's one-dimensional convolution kernel slid along the time dimension to capture correlation features from data at different time intervals. The convolutional neural network is connected to a long short-term memory network. In the long short-term memory network, the input gate, forget gate, and output gate use a sigmoid activation function to control the flow of information. The memory unit state is updated based on the tanh function. Several layers of long short-term memory networks are stacked, each layer containing several hidden units, so that the model can capture the gradual change trend of operating parameters in the time series.

6. The method for real-time status monitoring of an intelligent operation and maintenance robot based on artificial intelligence according to claim 5 is characterized in that: The preprocessed substation dataset is randomly divided into training, validation, and test sets in proportion. The model is trained using a mini-batch stochastic gradient descent algorithm, which is adjusted according to the dataset size and computing resources. During the training process, a learning rate decay strategy is used to gradually adjust the model parameters as the number of training rounds increases. The accuracy of classification tasks, which include both normal and faulty equipment, is evaluated based on the cross-entropy loss function. The mean squared error loss is used to measure the severity of the fault and quantify the deviation in the prediction, resulting in a composite loss function that drives model training. An early stopping method is introduced to continuously monitor the validation set loss value. If the loss value does not decrease after several consecutive epochs, training is terminated early to prevent overfitting and lock in the model's optimal generalization state. L2 regularization is used to impose a constraint penalty on the model weights. A grid search strategy is used to traverse and adjust the number of convolutional kernels, sizes, long-short-term memory network layers, and number of hidden units in the convolutional neural network, selecting the optimal model configuration based on the validation set performance indicators. The trained and optimized main analysis model is deployed in the intelligent operation and maintenance robot. When the robot collects a new batch of substation data, it is input into the model in real time after preprocessing, and the model starts the forward propagation calculation; the model outputs the current status assessment results of the equipment, including the equipment health score, fault probability value and qualitative judgment of the fault type.

7. The method for real-time status monitoring of an intelligent operation and maintenance robot based on artificial intelligence according to claim 1 is characterized in that: In the support vector regression algorithm, the radial basis function is selected as the kernel function. During the model hyperparameter tuning stage, a grid search combined with a cross-validation strategy is used to traverse the penalty parameters and kernel functions. The optimal combination is locked based on the minimum mean square error criterion of the validation set, so that the model simulates the dynamic performance of the cooling system. The historical data is divided into a training set and a test set in proportion, and model training is started. During the training process, the model gradually learns the fluctuation law of the heat exchange efficiency of the cooling system under different combinations of ambient temperature and load rate, and constructs a mapping relationship from the input parameter space to the output space of the heat dissipation capacity coefficient. After training is completed, the heat exchange efficiency of the cooling system, the ambient temperature and the transformer load rate are input into the model, and the current heat dissipation capacity coefficient of the cooling system is output.

8. The method for real-time status monitoring of an intelligent operation and maintenance robot based on artificial intelligence according to claim 1, characterized in that: The current heat dissipation capacity coefficient of the cooling system and the current status evaluation result of the equipment are captured at the same timestamp; based on the substation equipment design manual and statistical analysis of historical operation and maintenance data, a basic threshold range of the heat dissipation capacity coefficient is set for typical load conditions in different seasons and time periods, and the threshold is dynamically corrected in combination with the oil temperature indicator in the current status evaluation result of the equipment; When the cooling system's current heat dissipation coefficient is within A1 of the lower threshold, the equipment health score is between B2 and B3, the failure probability is between C1% and C2%, and there are no critical fault types, a Level 1 warning is triggered. Upon receiving the warning command, the intelligent operation and maintenance robot automatically shortens the inspection cycle and pushes the warning information to the operation and maintenance personnel's mobile terminal. If the cooling system's current heat dissipation coefficient falls below the thresholds A1-A2, the equipment health score drops to B1-B2, and the failure probability rises to C2%-C3%, or a fault type that affects equipment operation but does not yet threaten power supply safety occurs, a Level 2 alert is triggered. In addition to monitoring robots to increase the frequency of inspections, the O&M system also displays a pop-up alert on the large screen of the substation monitoring center and sends an emergency plan to the O&M team. When the heat dissipation coefficient is lower than the threshold value A2, the equipment health score is lower than B1, the failure probability exceeds C3%, or a high-risk fault type is identified, a three-level warning is activated. The substation automatic control system will reduce the load of the equipment if conditions permit. If conditions do not permit, a local power outage will be carried out to avoid risks. An alarm will pop up on the large screen of the substation monitoring center, the sound and light prompts will be strengthened, and an emergency plan will be sent to the operation and maintenance team. Among them, A1, A2, B1, B2, B3, C1, C2 and C3 are all positive decimals and are automatically set by the system.

9. A real-time status monitoring system for an intelligent operation and maintenance robot based on artificial intelligence, using the real-time status monitoring method for an intelligent operation and maintenance robot based on artificial intelligence according to any one of claims 1 to 8, characterized in that: include: Inspection and data acquisition module: includes: a path planning and navigation unit and a sensor array activation and control unit; the path planning and navigation unit enables the intelligent operation and maintenance robot to reach each equipment monitoring point according to the preset inspection path and time interval; the sensor array activation and control unit activates the sensor array and collects multi-dimensional data; Data fusion and preprocessing module: includes: electrical parameter acquisition and adaptation unit, data fusion unit and preprocessing unit; wherein, the electrical parameter acquisition and adaptation unit obtains real-time electrical parameter data from the equipment's own monitoring system, the data fusion unit fuses the real-time electrical parameter data with multi-dimensional data to obtain a substation data set, and the preprocessing unit preprocesses the substation data set; Model building and analysis module: includes: a model building unit, a model training and optimization unit, and a real-time evaluation unit; the model building unit uses a convolutional neural network combined with a long short-term memory network to build a main analysis model. The convolutional neural network is used to automatically extract local features from image and sensor time series data, and the long short-term memory network is used to capture the time dependency in the data sequence; the model training and optimization unit trains and optimizes the main analysis model, and the real-time evaluation unit inputs the preprocessed substation data set into the trained main analysis model and outputs the current status evaluation results of the equipment; Heat dissipation coefficient calculation module: includes: enthalpy difference calculation and efficiency evaluation unit and heat dissipation coefficient calculation unit; the enthalpy difference calculation and efficiency evaluation unit measures the cooling medium parameters, calculates the enthalpy difference change during the cooling process, and quantifies the heat exchange efficiency of the cooling system; the heat dissipation coefficient calculation unit uses the support vector regression algorithm based on thermodynamic principles and historical operating data to construct a transformer cooling system performance evaluation model, inputs the cooling system heat exchange efficiency, ambient temperature and transformer load rate, and outputs the current heat dissipation capacity coefficient of the cooling system; Fault warning and decision support module: includes: a comprehensive judgment unit and a warning push unit; wherein, the comprehensive judgment unit combines the current heat dissipation capacity coefficient of the cooling system with the current status evaluation result of the equipment, and the warning push unit determines whether the cooling system meets the heat dissipation requirements and issues a fault warning.

Images