Power plant boiler visual operation and maintenance auxiliary method based on machine learning and augmented reality

Abstract

The invention relates to the field of power plant boiler operation and maintenance. A power plant boiler visual operation and maintenance auxiliary method based on machine learning and augmented reality comprises the steps that a sensor is used for detecting the boiler operation state and internal parameters of a power plant boiler, data are generated and recorded according to the time sequence, and a data set is formed; a deep neural network model is established, the input of the deep neural network model is the data in the data set, and the output of the deep neural network model is the prediction of the boiler operation state and internal parameters of the power plant boiler in the future time period of the time period corresponding to the input data of the deep neural network model; establishing an augmented reality module; a sensor is used for detecting the current boiler operation state and internal parameters of a power plant boiler, data are generated and input into a deep neural network model and an augmented reality module, and when output data of the deep neural network model deviates from a normal range, the augmented reality module gives an alarm and generates adjustment parameters of the boiler according to the output data.

Description

Technical Field

[0001] The present invention relates to the field of operation and maintenance of power plant boilers. Background Art

[0002] Driven by the global high emphasis on the stability and security of power supply and the goal of improving the operation efficiency of power plants, the operation and maintenance technology of power plant boilers, especially the boiler operation and maintenance system applied to large thermal power plants, is facing higher requirements and development opportunities.

[0003] As the core equipment for power production, the accurate acquisition of internal parameters (such as temperature, pressure, water level, combustion status, etc.) of power plant boilers and the precise judgment of their operating states are crucial for the safe and efficient operation of boilers. However, the operation and maintenance management problems of boilers are gradually emerging, becoming the key factors restricting the improvement of boiler operation safety, efficiency, and stability. Traditional operation and maintenance means for power plant boilers are often relatively single. For example, relying on manual inspections to obtain parameter information is not only inefficient but also prone to untimely and inaccurate information acquisition. The reliance on instrument readings also has limitations. It is difficult for operation and maintenance personnel to understand complex internal conditions and quickly and accurately judge the operating state and potential failure risks of boilers from a large amount of instrument data. Some existing operation and maintenance systems lack the ability to deeply analyze and predict a large amount of internal parameter data and cannot meet the requirements of early warning and precise judgment of boiler operating states. Moreover, the low degree of information visualization during operation and maintenance is also a problem that needs to be solved. In addition, traditional operation and maintenance systems lack sufficient flexibility in dealing with complex and changeable operating conditions. For example, in the case of sudden load changes or slight abnormalities in equipment, existing operation and maintenance equipment is difficult to quickly make precise adjustments to maintain the best operating state of the boiler. And under different types of boilers and different operating conditions, the variation laws of internal parameters vary greatly, and it is difficult for traditional systems to perform refined operation and maintenance operations based on these differences.

[0004] The patent with Chinese Patent Number: CN202411542385.4 discloses a boiler energy-saving method based on IOT data analysis. By using Internet of Things sensors to collect the boiler operation parameter set in real time and bringing it into the parameter optimization model, multiple different boiler operation parameter combinations are obtained. Then, these multiple different boiler operation parameter combinations are brought into the boiler operation state prediction model, and thus multiple groups of prediction results corresponding one by one to the boiler operation parameter combinations are obtained. Based on the prediction results corresponding to each group of boiler operation parameter combinations, the fitness is evaluated, and the boiler operation parameter combination with the highest fitness is marked as the optimal boiler operation parameter, and corresponding boiler operation parameter optimization suggestions are generated based on the optimal boiler operation parameter. The patent with Chinese Patent Number: CN202410315377.X discloses a simulation and simulation-assisted decision-making control method and system for boiler power generation. First, the Granger causality test method is used to analyze the operation state of the boiler system to determine the corresponding characteristic variables and target variables. Then, the historical data of the operation state of the boiler system is used to train machine learning algorithms such as SVR, LSTM, and L1. Next, the trained network model is used to generate a new sample data set, and the training of the BPNN network is completed to obtain a network model that can predict the future trend of the state variables according to the characteristic variables. Finally, the network model is used to predict the future operation state of the boiler system according to the real-time operation data of the boiler system, and proportional adjustment is performed when there are potential safety hazards in the boiler system..

[0005] Machine learning algorithms have powerful data processing and analysis capabilities. For example, neural network algorithms can learn and model a large amount of data. Through the complex operations of multiple layers of neurons, the hidden patterns and rules in the data can be mined. Decision tree algorithms can classify and predict data in an intuitive tree-like structure, and have good effects on judging whether the internal parameters of the boiler are within the normal range and predicting the fault risk. Augmented reality (AR) technology can provide intuitive visual information assistance for operation and maintenance personnel. Through AR devices such as AR glasses or AR handheld terminals, virtual information can be superimposed on the real device view, enabling operation and maintenance personnel to more intuitively obtain and understand the internal parameter information and operation and maintenance prompt information of the device. This brings new ideas to the operation and maintenance of power plant boilers. The combination of the data processing ability of machine learning algorithms and the visualization ability of AR technology can analyze the internal parameters of the boiler according to requirements and present the results intuitively to operation and maintenance personnel, providing a theoretical basis for the design of an operation and maintenance system based on machine learning and AR. However, how to effectively integrate machine learning algorithms and AR technology into the power plant boiler operation and maintenance system to achieve precise operation and maintenance of the internal parameters of the boiler remains a technical challenge that needs to be solved. Summary of the Invention

[0006] The technical problem to be solved by the present invention is: how to achieve precise operation and maintenance of the internal parameters of the boiler, and improve the operation safety, efficiency and stability of the boiler.

[0007] The technical solution adopted by the present invention is: a visualization operation and maintenance assistance method for a power plant boiler based on machine learning and augmented reality, which is carried out according to the following steps Step 1: Use sensors to detect the boiler operation status and internal parameters of the power plant boiler, generate data and record it in chronological order to form a data set; Step 2: Establish a deep neural network model (machine learning algorithm module). The input of the deep neural network model is the data in the data set, and the output of the deep neural network model is the prediction of the boiler operation status and internal parameters of the power plant boiler in the future time period corresponding to the input data of the deep neural network model. Train the deep neural network model with the historical data in the data set (the historical data here is defined artificially, for example, the data generated by the boiler operation three days ago is defined as historical data) to obtain a trained deep neural network model; Step 3: Establish an augmented reality (AR) module, use the input data and output data of the deep neural network model in Step 2 to form a visualization model. When the output data of the deep neural network model deviates from the normal range, the augmented reality module issues an alarm and generates adjustment parameters for the boiler according to the output data (using existing technologies); Step 4: When the boiler is actually operating, the intelligent combustion optimization system (including the intelligent control part and the deep neural network model, the intelligent control part is an existing technology, which can intelligently monitor various parameters of the boiler and can operate the boiler) is started. Use the sensors in Step 1 to detect the current boiler operation status and internal parameters of the power plant boiler, generate data and input the data into the deep neural network model and the augmented reality (AR) module. When the output data of the deep neural network model deviates from the normal range, the augmented reality (AR) module issues an alarm and generates adjustment parameters for the boiler according to the output data, and then adjusts the power plant boiler with the adjustment parameters to assist the operation and maintenance personnel to precisely operate and make decisions on the boiler operation status and internal parameters of the power plant boiler.

[0008] The sensors described in Step 1 include thermocouple sensors and / or resistance temperature detectors for detecting the temperatures at different positions of the boiler combustion chamber, steam pipes, and water walls, piezoelectric pressure sensors and / or capacitive pressure sensors for detecting the steam pressure, feed water pressure, and steam-water separator pressure inside the boiler, differential pressure level sensors and / or ultrasonic level sensors for detecting the water levels of the steam drum and deaerator inside the boiler, oxygen sensors for detecting the oxygen content in the gas discharged from the boiler combustion chamber, and flame sensors for detecting the flame state (stability, intensity, etc.) of the boiler combustion chamber. All sensors form a sensor array.

[0009] In the deep neural network model in Step 2, decision trees are constructed according to different parameter intervals. If the output range of the deep neural network model deviates from the parameter interval, it indicates the existence of a fault risk. In the decision tree, data starts from the root node and gradually branches downward according to the attribute test conditions of the nodes, and finally reaches the leaf node. The result of the leaf node is the judgment of the current boiler state (such as whether the parameters are normal, whether there is a fault risk, etc.). When the output of the deep neural network model shows that the boiler is at high load or the parameters deviate from the normal range, the decision tree recalculates the information gain to adjust the branches. The decision tree recalculating the information gain to adjust the branches includes changing the attribute weights, adding new decision nodes or pruning.

[0010] In Step 3, the augmented reality module is established, and a visualization model is formed by using the input data and output data of the deep neural network model in Step 2. Specifically, the input data and output data of the deep neural network model are integrated and converted into a format that can be recognized and processed by the AR device. The AR device uses sensors to perform spatial positioning and mapping of the boiler environment, constructs a virtual three-dimensional space model, that is, the AR scene, and superimposes the information in the converted format onto the AR scene. When the maintenance personnel operate, the AR device obtains feedback information. If the operation meets the expectations, a confirmation prompt is given; if it does not meet the requirements, a correction prompt is given.

[0011] The beneficial effects of the present invention are as follows: The present invention proposes a visualization operation and maintenance assistance method for power plant boilers based on machine learning and augmented reality, aiming to achieve precise operation and maintenance of the internal parameters of the boiler through innovative technology applications and dynamic operation and maintenance management strategies, improve the operation safety, efficiency and stability of the boiler, and promote the progress and development of the operation and maintenance technology of power plant boilers.

[0012] The core of the present invention lies in: utilizing the analysis ability of machine learning algorithms for the internal parameter data of the boiler and the visualization ability of AR technology, designing a system that can actively provide comprehensive and accurate information and operation and maintenance operation guidance according to the operation requirements and internal parameters of the boiler, and optimizing the operation and maintenance process through intelligent operation and maintenance strategies to achieve the intelligentization and high efficiency of the power plant boiler operation and maintenance system.

[0013] The present invention can actively respond to the operation and maintenance requirements of the boiler and intelligently provide internal parameter information and operation and maintenance operation guidance. It is of great significance for improving the overall effect of power plant boiler operation and maintenance and coping with the operation and maintenance challenges under complex operation conditions and equipment differences. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 is the schematic flow chart of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0015] A visualization operation and maintenance assistance method for power plant boilers based on machine learning and augmented reality is carried out according to the following steps Step 1: Use sensors to detect the operating status and internal parameters of the power plant boiler, generate data and record it in chronological order to form a data set.

[0016] The sensors include thermocouple sensors and / or resistance temperature detectors for detecting the temperatures at different positions of the boiler combustion chamber, steam pipes, and water walls, piezoelectric pressure sensors and / or capacitive pressure sensors for detecting the steam pressure, feed water pressure, and steam-water separator pressure inside the boiler, differential pressure level sensors and / or ultrasonic level sensors for detecting the water levels of the steam drum and deaerator inside the boiler, oxygen sensors for detecting the oxygen content in the gas discharged from the boiler combustion chamber, and flame sensors for detecting the flame state (stability, intensity, etc.) of the boiler combustion chamber.

[0017] Step 2: Establish a deep neural network model (machine learning algorithm module). The input of the deep neural network model is the data in the data set, and the output of the deep neural network model is the prediction of the operating status and internal parameters of the power plant boiler in the future time period corresponding to the input data of the deep neural network model. Train the deep neural network model with the historical data in the data set (the historical data here is artificially defined, for example, the data generated by the boiler operation three days ago is defined as historical data) to obtain a trained deep neural network model.

[0018] In the deep neural network model in Step 2, decision trees are constructed according to different parameter intervals. If the output range of the deep neural network model deviates from the parameter interval, it indicates a potential fault risk. In the decision tree, data starts from the root node and gradually branches downward according to the attribute test conditions of the nodes, and finally reaches the leaf node. The result of the leaf node is the judgment of the current boiler state (such as whether the parameters are normal, whether there is a fault risk, etc.). When the output of the deep neural network model shows that the boiler is at high load or the parameters deviate from the normal range, the decision tree recalculates the information gain to adjust the branches. The decision tree recalculating the information gain to adjust the branches includes changing the attribute weights, adding new decision nodes, or pruning.

[0019] For the prediction and analysis of the internal parameters of the boiler, the decision tree algorithm can be used to judge whether the internal parameters of the boiler are within the normal range. For example, taking temperature as an example, a decision tree can be constructed according to different temperature intervals and other relevant parameters (such as pressure, water level, etc.). If the temperature is too high and the pressure is abnormal, it may indicate a certain fault risk, and the decision tree can accurately judge whether this situation belongs to a fault state or a potential dangerous state based on these characteristics.

[0020] In the prediction and analysis of the internal parameters of a boiler, a deep neural network model can be used to process two-dimensional data such as temperature fields and pressure fields. For example, the temperature distribution data on a certain plane inside the boiler (which can be regarded as a two-dimensional image data, where each pixel point represents the temperature value at a position) is used as the input of the deep neural network model. The convolutional layer can extract local features in the temperature field, such as local high-temperature regions, temperature gradients, etc. The pooling layer further compresses the data, and the fully connected layer finally outputs the prediction result of the entire temperature field, such as predicting the temperature field distribution at a future moment, or judging whether there is an abnormal situation in the current temperature field.

[0021] The specific methods for predicting and analyzing the internal parameters of a boiler include the following steps: Preprocess the raw data collected by sensors (such as temperature, pressure, water level, etc.). This includes data cleaning to remove outliers (for example, obviously unreasonable values caused by sensor failures or momentary interferences); data normalization to map parameter data of different magnitudes to the same interval, for example, normalizing temperature data, pressure data, etc. to the [0, 1] interval for the convenience of neural network training.

[0022] Conduct model training. For the decision tree algorithm, the labeled historical data (including temperature, pressure, water level and other parameter data in normal operating states and fault states) is used as the training set. By calculating metrics such as information gain, the best attribute is selected for node splitting to construct a decision tree model. For the convolutional neural network, the organized data of temperature fields, pressure fields, etc. (represented in the form of two-dimensional matrices) is used as the input, and the corresponding normal or abnormal state labels are used as the output. The backpropagation algorithm is adopted to adjust the weights of the network according to the loss function (such as the mean squared error loss function). After multiple rounds of iterative training, the model converges.

[0023] Prediction and analysis process. During real-time operation, the newly collected sensor data is preprocessed and then input into the trained model. For the decision tree model, the data starts from the root node and gradually branches down according to the attribute test conditions of the nodes, and finally reaches the leaf node. The result of the leaf node is the judgment of the current boiler state (such as whether the parameters are normal, whether there is a risk of failure, etc.). For the convolutional neural network, the input temperature field or pressure field data is processed through convolutional, pooling, and fully connected layers, and the prediction result of the temperature field or pressure field (such as the distribution situation at a future moment) or the judgment of whether the current state is abnormal is output. If the prediction result shows that the temperature in a certain area will exceed the normal range or there is abnormal fluctuation in the pressure field, a warning can be issued in a timely manner to provide a decision-making basis for the operation and maintenance personnel to take corresponding operation and maintenance operations to ensure the safe and efficient operation of the boiler.

[0024] When the boiler is at high load or the parameters deviate from the normal range, the decision tree algorithm recalculates the information gain to adjust the branches, such as changing the attribute weights. New decision nodes can also be added or pruning can be performed. For example, nodes can be added for new change patterns at high load, and redundant branches can be removed by pruning to improve the accuracy of fault judgment and the response speed. For the convolutional neural network, an adaptive learning rate adjustment strategy (such as Adagrad, Adam, etc.) can be used to adjust the weights. The number of network layers can also be increased or the neuron connection method can be adjusted. For example, a convolutional layer can be added or skip connections can be used to improve the prediction accuracy and the response speed, and potential fault risks can be detected in a timely manner.

[0025] Step 3: Establish an augmented reality (AR) module. Use the input data and output data of the deep neural network model in Step 2 to form a visualization model. When the output data of the deep neural network model deviates from the normal range, the AR module issues an alarm and generates adjustment parameters for the boiler based on the output data (using existing technologies). In Step 3, the establishment of the augmented reality (AR) module visualizes the input data and output data of the deep neural network model in Step 2 by integrating the input data and output data of the deep neural network model, converting them into a format that can be recognized and processed by the AR device. The AR device uses sensors to perform spatial positioning and mapping of the boiler environment, constructs a virtual three-dimensional space model, i.e., an AR scene, and superimposes the information in the converted format onto the AR scene. When the maintenance personnel operate, the AR device obtains feedback information. If the operation meets the expectations, a confirmation prompt is given; if it does not meet the requirements, a correction prompt is given.

[0026] First, after obtaining the prediction results and operation and maintenance guidance information from the machine learning algorithm module, they are integrated and then converted into a format that can be recognized and processed by the AR device. For example, three-dimensional temperature field data is converted into a format suitable for AR rendering, and text operation and maintenance guidance information is converted into a displayable or voice format.

[0027] Then, the AR device uses sensors to perform spatial positioning and mapping of the boiler environment and constructs a virtual three-dimensional space model. Next, the information in the converted format is superimposed onto the AR scene. For example, the temperature distribution is represented by different colors and superimposed on the virtual model of the boiler, and text and voice prompts for operation and maintenance guidance are displayed near the relevant components.

[0028] Finally, design an interaction function. When the maintenance personnel operate, the AR device obtains feedback information. If the operation meets the expectations, a confirmation prompt is given; if it does not meet the requirements, a correction prompt is given. Information query and extension functions are also provided to facilitate the maintenance personnel to query more detailed information for decision-making assistance.

[0029] Step 4: Use the sensors in Step 1 to detect the current operating status and internal parameters of the power plant boiler, generate data, and input the data into the deep neural network model and the augmented reality (AR) module. When the output data of the deep neural network model deviates from the normal range, the AR module issues an alarm, generates adjustment parameters for the boiler based on the output data, and then adjusts the power plant boiler with the adjustment parameters to assist the operation and maintenance personnel in precisely operating and making decisions on the operating status and internal parameters of the power plant boiler.

[0030] Data acquisition and transmission During the operation of the boiler, operating data is continuously collected through sensors (such as temperature sensors, pressure sensors, water level sensors, etc.) installed at various key parts of the boiler. These data are transmitted to the data processing center in real time through wired or wireless (such as ZigBee, Wi-Fi, etc.) communication methods. For example, the temperature sensor collects the temperature data inside the boiler every certain period (such as 10 seconds) and sends the data to the data processing center through the Wi-Fi module.

[0031] Fault diagnosis algorithm The data processing center analyzes the collected data using pre-set fault diagnosis algorithms. These algorithms can be rule-based, machine learning, or a combination of both. For rule-based algorithms, for example, when the water level is lower than a certain set value and the pressure is simultaneously higher than the safety upper limit, it is determined as a specific fault type. In terms of machine learning algorithms, such as using support vector machine (SVM) to train historical data and judge whether there is a fault risk based on real-time data. Once a fault is diagnosed, the AR module is triggered to provide corresponding operation guidance and troubleshooting solutions.

[0032] Association between the AR module and the boiler model The AR module pre-stores the three-dimensional model of the boiler, which accurately reflects each component of the boiler and its layout. When a fault is detected, the AR device (such as AR glasses or AR handheld terminal) identifies the boiler component currently faced through spatial positioning technology (such as SLAM technology) and associates the fault information with the corresponding component in the boiler model. For example, when a leakage fault is diagnosed at a certain pipe connection, the AR device can accurately identify the position of the pipe in the boiler three-dimensional model and highlight this part in the AR view.

[0033] Visualization of operation guidance and troubleshooting solutions Present operation guides and troubleshooting solutions in various forms such as text, images, and animations. For text, it is displayed in a clear and prominent font near the corresponding components in the AR view. For example, for valve faults, it shows "Rotate the valve 90 degrees clockwise to close the water flow". In terms of images, detailed structural schematic diagrams of the components can be provided, with key operation points marked, and voice prompts can also be provided.

[0034] Data acquisition layer Design a data acquisition module and install various sensors at key parts of the boiler, such as temperature sensors, pressure sensors, flow sensors, etc. These sensors can collect the operation status data of the boiler in real time, including internal parameters such as temperature, pressure, and flow. For example, install high-precision temperature sensors and pressure sensors at parts such as the superheater, reheater, and steam drum of the boiler to ensure comprehensive and accurate data acquisition.

[0035] Use a suitable communication protocol (such as Modbus, OPC, etc.) to transmit the collected data to the data processing center.

[0036] Data processing and analysis layer (machine learning algorithm part) Select suitable machine learning algorithms, such as decision tree algorithms for classifying fault types and neural network algorithms for predicting parameter change trends, etc.

[0037] For decision tree algorithms, first collect a large amount of historical operation data of the boiler (including data in normal and fault states) as the training set. For example, mark data with too high temperature and unstable pressure as a specific fault type, and then build a decision tree model. During operation, input the real-time collected data into the decision tree model, and the model classifies according to the pre-built decision rules to determine whether there is a fault and the fault type.

[0038] For neural network algorithms, build a multi-layer neural network structure. The input layer receives various parameters of the boiler, the hidden layer performs feature extraction and data processing, and the output layer predicts the change trend of the parameters (such as the rising trend of temperature in the future). Continuously adjust the weights of the neural network through the backpropagation algorithm to improve the prediction accuracy.

[0039] Establish a data storage and management system for storing historical data and real-time data for machine learning algorithms to train and optimize.

[0040] Visualization and interaction layer (AR technology part) Develop an AR scene construction module and use an AR development toolkit (such as Unity 3D combined with Vuforia, etc.) to build a three-dimensional virtual model of the boiler. This model accurately reflects the physical structure and component layout of the boiler.

[0041] Design a data interface to transmit the results processed by machine learning algorithms (such as fault diagnosis results, parameter prediction values, etc.) to the AR module. In the AR module, associate these results with the 3D virtual model of the boiler. For example, when the decision tree algorithm diagnoses a blockage fault in a certain pipeline, the AR module displays a fault prompt message at the corresponding pipeline part in the 3D model of the boiler.

[0042] Develop an interaction function to enable operation and maintenance personnel to interact with the system through AR devices (such as AR glasses or AR handheld terminals). For example, operation and maintenance personnel can query more detailed fault information or operation and maintenance guidance in the AR view through gesture operations (such as clicking, swiping, etc.), or obtain relevant information through voice commands.

[0043] Scheme for building an intelligent operation and maintenance management system: Sensor integration Sensor selection: Select appropriate sensors according to the key parameters of boiler operation. For example, for temperature monitoring, choose high-precision thermocouple or thermal resistance sensors; for pressure monitoring, select piezoresistive or capacitive pressure sensors; for flow monitoring, use electromagnetic flowmeters or vortex street flowmeters, etc. These sensors can accurately collect parameters such as temperature, pressure, and flow inside the boiler.

[0044] Sensor layout: Arrange sensors at key parts of the boiler. Install a water level sensor at the steam drum to accurately monitor the water level height; install temperature and pressure sensors on the steam pipeline to monitor the state of the steam; install oxygen sensors and fuel flow sensors near the burner to monitor the combustion situation. Ensure that the sensor layout is reasonable and can comprehensively reflect the operating state of the boiler.

[0045] Sensor network construction: Build a sensor network through wired (such as using shielded cables to connect sensors to data acquisition devices) or wireless (such as wireless communication technologies like ZigBee, LoRa, etc.) methods to transmit the collected data to the data processing center.

[0046] Intelligent control algorithm design Data processing algorithm: Use filtering algorithms (such as Kalman filtering) to process the raw data collected by sensors, remove noise interference, and improve the accuracy of the data. For example, for the data collected by temperature sensors, the Kalman filtering algorithm can smooth the data fluctuations and obtain data closer to the true temperature value.

[0047] State monitoring algorithm: Use threshold-based algorithms and trend analysis algorithms to monitor the operating state of the boiler. Set safety thresholds for parameters such as temperature, pressure, and water level, and determine an abnormal state when the parameters exceed the threshold range. At the same time, use trend analysis algorithms such as linear regression to predict the change trend of parameters and detect potential abnormal situations in advance.

[0048] Control strategy algorithm: Design an intelligent control strategy based on the operating status and parameter changes of the boiler. For example, when the load of the boiler changes, a fuzzy control algorithm is used to adjust the fuel supply and air supply of the burner to maintain stable steam pressure and temperature.

[0049] Function adjustment algorithm: Develop an algorithm to dynamically adjust the functions of the operation and maintenance assistance system. If it is detected that the boiler is operating at a high load and the temperature is rising rapidly, the algorithm will adjust the operation and maintenance assistance system to give priority to providing operation and maintenance guidance related to temperature control, such as highlighting the components of the cooling system in the AR view and providing detailed cooling operation guidance.

[0050] System integration and communication Integrate the sensor network and the intelligent control algorithm into a unified system. By writing software interfaces, the data collected by the sensors can be effectively processed by the intelligent control algorithm. For example, on an operating system platform based on Linux or Windows, use programming languages (such as C++, Python, etc.) to write programs to achieve the reading, processing of sensor data and the sending of control instructions.

[0051] Establish a reliable communication mechanism to ensure the accurate transmission of data between sensors, the data processing center and the operation and maintenance assistance system. Adopt communication technologies such as industrial Ethernet, fieldbus (such as Profibus, CAN bus, etc.) to achieve communication connections between different devices.

[0052] Compared with the prior art, the advantages of the technology of the present invention are: Intelligent operation and maintenance response mechanism: The present invention utilizes the data processing ability of machine learning algorithms and the visualization ability of AR technology to realize the intelligent and dynamic interaction between the boiler operating status and the operation and maintenance system. This can respond more precisely to the changes in operation and maintenance requirements than traditional manual inspection and single-parameter monitoring methods, and provide an immediate operation and maintenance management response.

[0053] Parameter selective regulation: Compared with traditional operation and maintenance adjustment methods, the operation and maintenance assistance system based on machine learning can more precisely select and regulate the internal parameters of the boiler, significantly improving the operation and maintenance management efficiency of the system. Especially in meeting the requirements of specific parameters for different boiler types and operating stages, it can effectively promote the safe operation of the boiler.

[0054] Dynamic operation and maintenance regulation: Through the dynamic analysis of the internal parameters of the boiler by machine learning algorithms and the real-time visualization of AR technology, the present invention can form a dynamic operation and maintenance environment according to the actual operation requirements of the boiler. This active regulation is superior to static operation and maintenance control methods and can adjust operation and maintenance strategies according to actual needs, thereby optimizing the operation and maintenance management performance.

[0055] Adaptive operation and maintenance management strategy: Through the intelligent algorithm function of the intelligent operation and maintenance management system, this technology realizes the adaptive adjustment of operation and maintenance management strategies, and can automatically optimize the operation and maintenance mode according to different boiler operating conditions. Compared with the operation and maintenance system with a fixed mode, it demonstrates higher flexibility and efficiency.

[0056] Simplifying system complexity and cost: By integrating key components such as machine learning algorithms, sensors, and AR technology, this invention reduces the need for additional operation and maintenance equipment (such as complex instruments and monitoring systems, etc.), simplifies the system structure, reduces manufacturing and maintenance costs, and improves economic feasibility.

[0057] The following combines the attached Figure 1 To introduce in detail the operation process of the intelligent combustion optimization system of the power plant boiler of the visualization operation and maintenance assistance method for power plant boilers proposed by this invention, which combines machine learning algorithms (including deep neural network models) and AR technology (augmented reality module), and further elaborates on this invention.

[0058] System initialization: When the boiler is put into operation, the intelligent combustion optimization system starts simultaneously, and the system begins to monitor key parameters such as the temperature field, pressure field, and water level inside the boiler.

[0059] Operation and monitoring of the intelligent combustion optimization system: System self-check: When the boiler is put into operation, the intelligent combustion optimization system starts and first conducts a self-check to check whether hardware and software modules such as sensors and data acquisition cards are working properly.

[0060] Data acquisition starts: After the self-check passes, sensors such as temperature, pressure, and water level start to collect data at a set frequency, such as once every 10 seconds.

[0061] Data transmission: The analog signals collected by the sensors are converted into digital signals by the data acquisition card and transmitted to the control center computer through industrial Ethernet or fieldbus.

[0062] Data processing: After the computer preprocesses the data (such as filtering), machine learning algorithms are used for analysis, including establishing combustion models, fault diagnosis, etc.

[0063] Decision generation: According to the analysis results, the combustion control sub-module generates control instructions, and the optimization strategy sub-module adjusts the strategy.

[0064] Monitoring method: Multi-point measurement and trend analysis of the temperature field: Construct the temperature field through multi-point temperature sensors and analyze the temperature change trend to judge the combustion situation.

[0065] Detection of Key Parts in the Pressure Field and Fluctuation Analysis: Measure the pressure at key parts and analyze the fluctuations to judge the ventilation and combustion stability.

[0066] Water Level Measurement and Range Judgment: Directly measure the water level and compare it with the safe range to take measures.

[0067] Combustion State Monitoring (Comprehensive Monitoring) and Visualization: Analyze multiple parameters comprehensively and use AR technology to visually present the combustion state.

[0068] Response of Machine Learning Algorithm: As the operating state of the boiler changes, the deep neural network model in the machine learning algorithm module starts to predict and analyze the internal parameters of the boiler based on its powerful data processing ability, making the operating state of the boiler enter a state that matches the optimization goal. The characteristics of the machine learning algorithm are affected by various factors. Among them, the parameters of the model affect the prediction accuracy, the type and structure of the algorithm change the response speed, and the scale and quality of the dataset affect the generalization ability of the model.

[0069] Process of the Machine Learning Algorithm Module Predicting and Analyzing the Internal Parameters of the Boiler: Data Input: The deep neural network model receives the internal parameter data of the boiler from the data acquisition system. These data include temperature field data (such as temperature values at different positions in the furnace), pressure field data (such as inlet and outlet pressures), and water level data, etc. For example, the temperature values collected by 10 temperature sensors at different positions in the furnace, the water level value of the steam drum, and the pressure difference between the inlet and outlet are used as input vectors and input into the deep neural network model.

[0070] Feature Extraction: In the hidden layer of the network, the input data is non-linearly transformed through the activation function of the neuron (such as the ReLU function) to extract the features in the data. For example, for temperature data, the gradient feature of the temperature may be extracted to reflect the change trend of the temperature at different positions; for pressure data, the correlation feature between the pressure difference and the combustion state is extracted.

[0071] Prediction Output: The output layer of the model predicts the operating state of the boiler based on the features extracted by the hidden layer. For example, the predicted value of the combustion efficiency, the predicted value of pollutant emissions (such as the prediction of nitrogen oxide emissions), and the judgment result of whether there is a potential fault (such as the probability of incomplete combustion fault) are output.

[0072] Factors Affecting the Characteristics of Machine Learning Algorithm: Influence of Model Parameters on Prediction Accuracy Weight Parameters: In a deep neural network, the weight parameters determine the influence degree of input features on the output result. Unreasonable settings (such as the weight between temperature and combustion efficiency) will cause the predicted value to deviate from the actual value, and the weight can be adjusted by backpropagation to improve the accuracy.

[0073] Bias parameter: Used to adjust the neuron activation threshold. An appropriate bias can enable the model to better fit the data, and improper setting may lead to misjudgment in fault determination.

[0074] Influence of algorithm type and structure on response speed Algorithm type: Different algorithms have different computational complexities. Deep neural networks have a higher computational complexity compared to support vector machines. A complex neural network structure may have a slow response when analyzing the internal parameters of a boiler and is difficult to respond promptly to changes in the operating state.

[0075] Structure level: A relatively shallow (3 - 5 layers) neural network has a small computational amount and a fast response, but may not be able to extract complex features; a relatively deep (more than 10 layers) network can extract complex features, but has a large computational amount and a slow response, and may miss key information in the rapid change of the temperature field.

[0076] Influence of the scale and quality of the dataset on the generalization ability of the model Dataset scale: A large-scale dataset contains more information on the operating state of the boiler, which helps to improve the generalization ability of the model. A large amount of data with different loads and fuel types can enable the model to better handle new scenarios, and a small-scale dataset may lead to overfitting.

[0077] Dataset quality: High-quality data should be accurate, complete, and representative. Data errors (such as inaccurate data caused by low sensor accuracy) will affect the training effect and reduce the generalization ability. For example, errors in water level data will cause deviations in the learning relationship of the model and affect prediction.

[0078] Combustion optimization: Parameter adjustment suggestions that meet the optimization goals are generated by the machine learning algorithm module and passed to the AR module for visual display. At the same time, the sensor array continuously monitors the parameter conditions in different areas inside the boiler to ensure that the operating state of the boiler remains within the optimal demand range.

[0079] Parameter adjustment suggestions Adjustment of fuel supply amount: Based on the analysis of the internal parameters of the boiler (such as temperature, pressure, oxygen content, etc.), if the machine learning algorithm module finds that the combustion efficiency is lower than the optimization goal, it generates suggestions to reduce or increase the fuel supply amount. This adjustment may be based on a predefined combustion efficiency model, which establishes the relationship between parameters such as temperature, pressure, oxygen content and the optimal fuel supply amount through a large amount of historical data and experimental data.

[0080] Adjustment of air supply volume: If the analysis finds that the oxygen content in the exhaust gas generated by combustion is too high or too low, the algorithm will generate suggestions to adjust the air supply volume. The optimal value is also determined through the analysis of historical operation data and the combustion theory model.

[0081] Combustion Temperature Adjustment: Combustion temperature is a key factor in reducing emissions of pollutants such as nitrogen oxides (NOx). If the machine learning algorithm detects that the NOx emissions are approaching or exceeding the permitted upper limit, it may recommend adjusting the combustion temperature. The adjustment strategy is based on the chemical reaction kinetics model between pollutant generation and combustion temperature, which is integrated into the machine learning algorithm.

[0082] Combustion Time Adjustment: In some cases, extending or shortening the combustion time of the fuel in the furnace can also affect pollutant emissions. Based on the type of fuel, combustion status, and pollutant emissions, the machine learning algorithm may recommend adjusting the operating mode of the burner (such as changing the combustion cycle or residence time) to optimize the combustion time.

[0083] Visualization Display Visualization of Parameter Adjustment: After the AR module receives the parameter adjustment suggestions generated by the machine learning algorithm, it presents this information to the operator in an intuitive way. For example, when the operator views the boiler through the AR device, on the virtual model of the boiler, a number will be shown next to the fuel supply pipe, indicating the recommended adjustment value of the fuel supply (e.g., "+10%" means increasing the fuel supply by 10%), along with an arrow indicating the adjustment direction.

[0084] Visualization of Combustion Status: In addition to the parameter adjustment suggestions, the AR module also shows the current combustion status of the boiler. Regarding the pollutant emissions, a pollution index bar is displayed near the exhaust gas outlet of the boiler, intuitively showing the relationship between the current pollutant emission level and the permitted upper limit.

[0085] Monitoring Role of the Sensor Array Temperature Monitoring: The sensor array installs temperature sensors at different positions inside the furnace to monitor the temperature distribution in the furnace. For example, thermocouple sensors are set at the center, around, top, and bottom of the furnace to collect temperature data in real time. These data can reflect the uniformity of combustion in the furnace.

[0086] Flame Monitoring: Monitor the characteristics of the flame through flame sensors, such as the shape, color, and intensity of the flame. The shape of the flame can reflect the stability of combustion, and the color of the flame is related to the combustion temperature and the degree of complete combustion of the fuel.

[0087] Pressure Monitoring: Install pressure sensors in the flue to monitor the pressure changes in the flue. The changes in flue pressure can reflect the ventilation situation during the combustion process and whether there is a blockage.

[0088] Gas Composition Monitoring: Use gas sensors to monitor the composition of the exhaust gas in the flue, such as the content of gases like oxygen, carbon dioxide, and NOx. By analyzing the composition of the exhaust gas, the degree of complete combustion and pollutant emissions can be evaluated.

[0089] Water level monitoring: Install a water level sensor on the steam drum to accurately monitor the water level height inside the steam drum. The sensor transmits the water level data to the control system in real time. Once the water level exceeds the safe range, the control system will take corresponding measures for adjustment.

[0090] The specific implementation steps are as follows: The first step: Material selection and preparation Select high-performance computing devices and AR devices as the key hardware for combustion optimization. Due to their powerful computing capabilities and visualization capabilities, they can play an important role in real-time data processing and display.

[0091] Build a machine learning model by combining a deep neural network with a reinforcement learning algorithm to achieve accurate prediction and optimization of the internal parameters of the boiler. The parameters of the model, the type and structure of the algorithm need to be determined according to actual needs during the preparation process. At the same time, the impact of the scale and quality of the dataset on the model performance should be considered.

[0092] Select appropriate sensors (such as high-precision temperature sensors, pressure sensors, etc.) to build the monitoring system of the boiler to ensure good data collection and stability.

[0093] The second step: System design and assembly Determine the assembly layout and key dimensions of the machine learning algorithm module, sensor array, intelligent controller and AR device inside the boiler, including the size of the algorithm server, the distribution position of the sensors, etc. Consider the impact of the characteristic adjustment factors of the machine learning algorithm on the system layout and performance. For example, different model parameters may require different server configurations to achieve the best prediction effect.

[0094] Install the algorithm server in the boiler control room and build the prepared machine learning model on each server.

[0095] Arrange the sensor array inside the boiler to ensure accurate monitoring of the key parameter conditions in different areas. Install the controller and electrically connect it to the sensor array to receive parameter data.

[0096] Connect the drive device, electrically connect it to the controller and connect it to the AR device to achieve parameter adjustment according to instructions.

[0097] Install the AR device, connect the AR glasses or AR handheld terminal to the algorithm server to ensure the visualization display of real-time data.

[0098] The third step: Control system development Integrate sensors to monitor parameters such as temperature, pressure, and water level in different areas of the boiler.

[0099] Write the control logic to automatically trigger the adjustment action of the drive device on the machine learning algorithm module based on the parameter signal. When writing the control logic, fully consider the adjustment factors of the machine learning algorithm, such as setting different prediction strategies according to different model parameters, setting the training cycle of the model according to the scale and quality of the data set, etc.

[0100] Through the above steps, the present invention realizes the intelligence and high efficiency of the intelligent combustion optimization system of the power plant boiler, ensures that the boiler maintains a good operating state, and improves the overall effect and safety of boiler operation and maintenance.

[0101] The following is a detailed description of the present invention taking a pulverized coal boiler as an example System initialization When the pulverized coal boiler is put into operation, the intelligent combustion optimization system is started accordingly. In the initial stage, the system begins to monitor key parameters such as the temperature field, pressure field, and water level inside the boiler. For example, in the initial stage of startup, the temperature field inside the boiler is about 300 - 500 °C, the pressure field is about 2 - 3 MPa, and the water level is near the lower limit value of the normal range. Assume the water level height is 1.5 meters (the total water level range is 1 - 3 meters).

[0102] Machine learning algorithm response Model construction and preparation: Select a high-performance computing device as the hardware support for the machine learning algorithm module. It has powerful computing capabilities and can meet the computing requirements of the combined model of deep neural network and reinforcement learning algorithm. In the established model, the number of layers of the deep neural network is set to 5 layers, and the number of neurons in each layer is determined according to the amount and complexity of input and output data. The reinforcement learning algorithm uses the Q-learning algorithm. When constructing the model, considering the operation data scale of the pulverized coal boiler (in one embodiment, about 100,000 historical operation data have been collected) and quality (the data has been cleaned and preprocessed to remove outliers and incorrect data), determine the initial parameters of the model, such as the learning rate is 0.001, the momentum factor is 0.9, etc., to ensure that the model has good generalization ability.

[0103] Reinforcement learning algorithm - Q-learning parameter description and substitution: Q-value update formula: In Q-learning, the Q-value update formula is Q ( s , a )= Q ( s , a )+ α r + γ max a ′ Q ( s ′,​a ′)− Q ( s , a )]。

[0104] Learning rate α Substitution when learning rate = 0.001: Learning rate α Determines the proportion of newly obtained information in updating the Q - value. When α = 0.001, it means that each time the Q - value is updated, the new estimate value ( r + γ max a ′ Q ( s ′, a ′)) will only have a small impact on the old Q - value. For example, in the operation scenario of a pulverized coal boiler, assuming the current state s and taking an action a such as adjusting a certain combustion parameter) to obtain a reward r and transferring to a new state s ′, when calculating the new Q - value, the old Q - value Q ( s , a ) will retain most of it, and the new estimated part r + γ max a ′ Q ( s ′, a ′)− Q ( s , a )] will only affect the final Q - value at a ratio of 0.1%.

[0105] Discount factor γ Substitution when the discount factor (assumed to be 0.9): Discount factor γ Is used to measure the influence degree of future rewards on the current decision - making. When γ = 0.9, it means that future rewards are relatively important compared to the current reward, but as the time step progresses, their importance will gradually decrease. In the example of the pulverized coal boiler, if the current action of adjusting the combustion parameter a results in a series of subsequent states s 1, s 2,⋯ and corresponding actions a 1, a 2,⋯, then when calculating the Q - value of the action s in the current state a , the maximum Q - value max a ′ Q ( s ′,a ') will affect the current Q - value calculation by 90%.

[0106] Initial Q - value setting: At the beginning of the model, initial Q - values need to be set for all state - action pairs ( s , a ). These initial values can be set based on experience or randomly. For example, all initial Q - values can be set to 0, and then as the algorithm iterates, the Q - values are gradually updated according to the actual reward feedback.

[0107] Mechanism of the algorithm: As the operating state of the boiler changes, for example, when the boiler load increases from 50% to 70%, the deep neural network model in the machine learning algorithm module begins to predict and analyze the internal parameters of the boiler based on its powerful data - processing ability. According to the input real - time temperature field, pressure field, and water - level data, the model predicts that at the current load, the temperature field may vary within the range of 1000 - 1200 °C, the pressure field will rise to 4 - 5 MPa, and the water level should be maintained between 1.8 - 2.2 meters to make the boiler operating state enter a state matching the optimization goal. At the same time, according to the characteristics of the algorithm type (Q - learning algorithm), the prediction strategy is quickly adjusted to adapt to the rapid changes in the boiler operating state and ensure the timely discovery of potential failure risks.

[0108] Combustion optimization Generation and visualization of parameter adjustment suggestions: Parameter adjustment suggestions meeting the optimization goal are generated by the machine learning algorithm module. For example, when it is predicted that the temperature field may be too high, it is recommended to reduce the fuel supply of the burner by 10%. These suggestions are passed to the AR module for visual display. The AR device (using AR glasses) will display them to the operation and maintenance personnel in an intuitive image and text form. For example, in the field of view of the AR glasses, a red warning sign is displayed at the burner part of the boiler, and the specific adjustment value and operation guidance are also shown.

[0109] Sensor monitoring guarantee: The sensor array continuously monitors the parameter conditions in different areas inside the boiler. High - precision temperature sensors are selected, with a measurement accuracy of up to ±1 °C, and are distributed at certain intervals in key parts such as the furnace and pipes of the boiler, with a total of 50 temperature sensors set; the pressure sensor has an accuracy of ±0.01 MPa, and 30 pressure sensors are set in different pressure sections; the water - level sensor uses a high - precision ultrasonic water - level sensor with an accuracy of ±0.05 meters and is installed at the water - level measurement part of the boiler. These sensors ensure that the boiler operating state is maintained within the best demand range.

[0110] System design and assembly Layout determination: The assembly layout of the machine learning algorithm module, sensor array, intelligent controller, and AR device inside the boiler is determined based on the boiler structure and algorithm characteristics. The algorithm server uses a standard 19-inch rack-mounted server and is placed in a dedicated computer room inside the boiler control room. The temperature of the computer room is controlled at 18 - 25°C, and the humidity is 40% - 60%. The sensor array is distributed according to the importance of different areas and the sensitivity of parameter changes. For example, temperature sensors are more densely distributed in the high-temperature area of the furnace, with one set every 1 - 2 meters; pressure sensors are mainly set at the inlets, outlets, and elbows of the pipelines; water level sensors are installed near the water level measuring cylinder.

[0111] Equipment installation and connection: Install the algorithm server inside the boiler control room, and establish a prepared machine learning model on each server. Arrange the sensor array inside the boiler to ensure accurate monitoring of key parameter conditions in different areas. Install the controller and electrically connect it to the sensor array to receive parameter data. Connect the drive devices (such as fuel supply adjustment devices, fan speed adjustment devices, etc.) so that they are electrically connected to the controller and connected to the AR device, thereby realizing parameter adjustment according to instructions. Install the AR device, connect the AR glasses to the algorithm server to ensure real-time visualization of data. The connection uses Gigabit Ethernet with a bandwidth of 1000 Mbps to ensure real-time data transmission.

[0112] Control system development Sensor integration: Integrate sensors such as temperature, pressure, and water level to monitor parameters in different areas of the boiler. Write control logic to automatically trigger the adjustment actions of the drive devices on the machine learning algorithm module based on parameter signals. For example, when the temperature sensor detects that the temperature exceeds 1200°C, the control logic will trigger the fuel supply adjustment device to reduce the fuel supply amount, and at the same time adjust the parameters in the machine learning algorithm module, such as reducing the learning rate to 0.0005, to improve the accuracy of the model's parameter prediction under high-temperature conditions. Set the training period of the model to once a day according to the scale of the dataset (about 100 new operation data per hour) to ensure that the model can learn the latest characteristics of the boiler operation data in a timely manner.

[0113] Through the above steps, the present invention realizes the intelligentization and high efficiency of the intelligent combustion optimization system of the pulverized coal boiler, ensures that the boiler maintains a good operating state, and improves the overall effect and safety of boiler operation and maintenance.

Claims

1. A visual operation and maintenance auxiliary method for power plant boilers based on machine learning and augmented reality, characterized in that: Follow the steps below Step 1: Use sensors to detect the boiler operating status and internal parameters of the power plant boiler, generate data and record them in chronological order to form a data set; Step 2: Establish a deep neural network model. The input of the deep neural network model is the data in the data set. The output of the deep neural network model is the prediction of the boiler operation status and internal parameters of the power plant boiler in the future time period corresponding to the time period of the input data of the deep neural network model. The deep neural network model is trained with the historical data in the data set to obtain a trained deep neural network model. Step 3: Establish an augmented reality module, and use the input data and output data of the deep neural network model in step 2 to form a visualization model. When the output data of the deep neural network model deviates from the normal range, the augmented reality module issues an alarm and generates adjustment parameters of the boiler based on the output data; Step 4: Use the sensor in step 1 to detect the current boiler operating status and internal parameters of the power plant boiler, generate data and input the data into the deep neural network model and augmented reality module. When the output data of the deep neural network model deviates from the normal range, the augmented reality module issues an alarm and generates the boiler adjustment parameters based on the output data. The power plant boiler is then adjusted using the adjustment parameters to assist operation and maintenance personnel in making precise operations and decisions on the boiler operating status and internal parameters of the power plant boiler.

2. According to claim 1, a power plant boiler visual operation and maintenance auxiliary method based on machine learning and augmented reality is characterized by: The sensors described in step one include thermocouple sensors and / or thermistor sensors for detecting the temperatures at different positions of the boiler combustion chamber, steam pipes, and water-cooled walls, piezoelectric pressure sensors and / or capacitive pressure sensors for detecting the steam pressure, feed water pressure, and steam-water separator pressure inside the boiler, differential pressure water level sensors and / or ultrasonic water level sensors for detecting the water levels of the steam drum and deaerator inside the boiler, oxygen sensors for detecting the oxygen content in the exhaust gas from the combustion chamber of the boiler, and flame sensors for detecting the flame state of the combustion chamber of the boiler.

3. The method for visual operation and maintenance of power plant boilers based on machine learning and augmented reality according to claim 1, characterized in that: In the deep neural network model in step 2, a decision tree is constructed according to different parameter intervals. If the output range of the deep neural network model deviates from the parameter interval, it indicates that there is a risk of failure. In the decision tree, the data starts from the root node and gradually branches downward according to the attribute test conditions of the node, and finally reaches the leaf node. The result of the leaf node is the judgment of the current boiler state. When the output of the deep neural network model shows that the boiler is under high load or the parameters deviate from the normal range, the decision tree recalculates the information gain adjustment branch, and the decision tree recalculates the information gain adjustment branch including changing the attribute weight, adding a new decision node or pruning.

4. The method for visual operation and maintenance of a power plant boiler based on machine learning and augmented reality according to claim 3 is characterized in that: In step three, an augmented reality module is established, and a visualization model is formed using the input data and output data of the deep neural network model in step two. Specifically, the input data and output data of the deep neural network model are integrated and converted into a format that can be recognized and processed by the AR device. The AR device uses sensors to spatially locate and map the boiler environment, and constructs a virtual three-dimensional space model, namely, an AR scene. The converted information is superimposed on the AR scene. When the operation and maintenance personnel operate, the AR device obtains feedback information, and a confirmation prompt is given if the operation meets the expectations, and a correction prompt is given if it does not meet the requirements.

Images