An unattended burner control system and method

Abstract

The present application relates to the technical field of unattended burner control, and particularly relates to an unattended burner control system and method; in the design, a model operation unit deploys a combustion dynamic model constructed by a long short-term memory neural network, and performs online fine tuning after offline training, can predict the furnace temperature in real time, dynamically calculate the optimal air valve opening, and realize accurate matching of the air-fuel ratio; an internet of things communication unit is connected with a cloud end through an industrial internet of things protocol, uploads data and receives optimization strategies; a data center unit stores and analyzes historical and real-time data; a global optimization engine unit runs a multi-objective particle swarm optimization algorithm, generates an optimal combustion strategy in combination with a production plan and an electricity price policy; a device health diagnosis unit compares data to evaluate the degree of health decline, and a predictive maintenance engine unit generates a maintenance work order and suggestions by using machine learning to build a life prediction model, so that intelligent operation and predictive maintenance are realized.

Description

Technical Field

[0001] This invention relates to the field of unattended burner control technology, and in particular to an unattended burner control system and method. Background Technology

[0002] In current industrial heating applications, such as furnace drying, crude oil heating, and fuel oil injection, automated control technology for burners has been widely adopted. Existing technologies have achieved basic unmanned operation of burners: rapid deployment of equipment is achieved through detachable support frames; combustion status is monitored in real time using flame sensors and pressure switches, and automatic ignition, flameout protection, and remote fault alarms are implemented through controllers; simultaneously, combined with temperature sensors and simple PID control algorithms, the opening of fuel valves and dampers can be automatically adjusted according to process requirements to maintain stable temperature. These technologies effectively reduce the intensity and safety risks of manual operation, reduce on-site construction work, and initially realize remote monitoring and automated operation of burners.

[0003] However, the control logic of existing technical solutions is mostly preset or based on simple feedback linear regulation. Once the PID parameters are tuned, they remain fixed. When faced with fluctuations in fuel calorific value, changes in ambient temperature, changes in furnace resistance characteristics, or nonlinear drift caused by long-term equipment operation, the fixed control model is difficult to adapt, and problems such as regulation lag, large overshoot, or even system oscillation are likely to occur, resulting in decreased combustion efficiency and poor control accuracy. In addition, existing control systems are mostly independent information islands, which can only adjust according to a single preset target and cannot interact with upper-level production management systems or enterprise resource planning systems. It is difficult to perform global optimization based on production plans, energy consumption indicators, or equipment health status, and it is impossible to achieve truly intelligent operation and predictive maintenance. Summary of the Invention

[0004] The purpose of this invention is to provide an unattended burner control system and method, aiming to solve the problems of existing technology control logic, which is mostly preset or simple feedback linear adjustment with fixed PID parameters. Faced with various situations such as fuel calorific value fluctuations, fixed control models are difficult to adapt, resulting in problems such as adjustment lag, leading to decreased combustion efficiency and control accuracy. Furthermore, existing control systems are often information silos, adjusting only according to a single objective, unable to interact with upper-level systems, making it difficult to achieve global optimization, intelligent operation, and predictive maintenance.

[0005] To achieve the above objectives, the present invention provides an unattended burner control method, comprising the following steps: Based on the current ambient temperature and furnace pressure, the matching ignition parameters are retrieved from historical combustion characteristics, the ignition electrode discharge is controlled, and the valves of the fuel branch are opened in sequence. At the same time, the amount of combustion air is adjusted and the flame intensity is detected in real time. If ignition fails, purging is automatically performed and the ignition parameters are corrected according to the reason for failure and the test is repeated. During the stable combustion stage, real-time data on furnace temperature, flue gas oxygen content, fuel flow rate, and furnace pressure are collected. Then, based on the above data, the trend of furnace temperature change is predicted in real time. When the set temperature changes or is disturbed by external factors, the optimal air valve opening is calculated online and the control signal is output to achieve precise dynamic matching of the air-fuel ratio. The system continuously monitors the cumulative number of ignitions and breakdown voltage of the ignition electrode, the operating current and vibration spectrum of the wind turbine, and the action response time of the valve. The real-time data is then compared with the health baseline to assess the health degradation of key components. When the assessment value exceeds the warning threshold, predictive maintenance suggestions are automatically sent. When a non-critical fault is detected, the system automatically switches to redundant equipment or adjusts the operating mode until the fault self-heals. Based on regularly uploaded combustion efficiency, energy consumption data, and equipment health reports, combined with enterprise production plans and peak-valley electricity pricing policies, the optimal combustion strategy is generated through global optimization, issued and executed, and operating parameters are adjusted to achieve the best match between energy consumption and production needs.

[0006] The ignition parameters include the valve opening degree, ignition duration, and purging time during ignition. The historical combustion characteristics store the environmental parameters and corresponding ignition parameter combinations for each successful ignition.

[0007] The predicted furnace temperature is based on a data-driven model constructed using a long short-term memory neural network. This data-driven model is trained offline using historical operating data and fine-tuned online using transfer learning techniques based on the latest operating data during actual operation.

[0008] The method for establishing the health baseline is to collect operating data of key components for more than one week during the initial stable operation period after the equipment is first put into operation or after a major overhaul, and then take the statistical characteristic values ​​as the benchmark after data cleaning.

[0009] The self-healing mechanism includes automatically switching to the redundant flame sensor signal when the flame sensor malfunctions, and switching to the parallel bypass valve for fine control when the fuel valve is stuck.

[0010] The global optimization is a multi-objective particle swarm optimization algorithm, which uses production plan completion rate, comprehensive energy consumption cost and equipment loss rate as optimization objectives to generate a Pareto optimal solution set for cloud platform decision-making.

[0011] This invention also provides an unattended burner control system applied to the aforementioned unattended burner control method. The unattended burner control system includes a field control layer module, an edge intelligent control module, a human-machine interface terminal, and a cloud-based collaborative management platform. The field control layer module includes a burner unit, a fan unit, a fuel pipeline unit, an air pipeline unit, a sensing unit, and an execution unit. The sensing unit includes a flame sensor subunit, a temperature sensor subunit, a pressure sensor subunit, and a flue gas oxygen content sensor subunit. The execution unit includes a fuel regulating valve subunit, an air regulating valve subunit, and a fuel branch solenoid valve subunit. The edge intelligent control module includes a data acquisition unit, a storage unit, a model calculation unit, an equipment health diagnosis unit, and an IoT communication unit. The cloud-based collaborative management platform includes a data middleware unit, a global optimization engine unit, a predictive maintenance engine unit, and a human-machine interface unit. The burner unit serves as a combustion reaction vessel for generating an initial flame. The fan unit is used to provide combustion air; The fuel pipeline unit is used to provide a fuel passage and dynamically transmit the regulated fuel flow rate in combustion control; The air duct unit serves as the physical channel for air delivery and works in conjunction with the fan unit to execute airflow adjustment commands in combustion control. The flame sensor subunit is used to confirm whether a flame has been established. The temperature sensor subunit is used to calculate the deviation from the set temperature and determine the adjustment amount; The pressure sensor subunit is used to detect whether the gas pressure is within a safe range during self-testing and to monitor the furnace pressure. The flue gas oxygen content sensor subunit is used for adaptive combustion control; The fuel regulating valve subunit is used to establish the initial fuel flow and acts as the main actuator for regulating the fuel flow in the combustion control MPC algorithm. At the same time, it can switch to the bypass valve when the fuel valve is stuck. The air conditioning valve subunit is used in adaptive combustion control to precisely adjust the air volume in conjunction with the fan frequency converter and perform dynamic air-fuel ratio matching. The fuel branch solenoid valve subunit is used as a safety shut-off valve for the ignition line or main line. The data acquisition unit is used to collect temperature, pressure, flame, and oxygen content data from the sensor unit, as well as feedback signals from the actuator, in real time during self-testing, ignition, combustion control, and health assessment. The storage unit is used to store historical data on combustion characteristics, health baseline, and health assessment. The model computing unit is used to deploy and run the combustion dynamic model and the model prediction controller, and is responsible for predicting the temperature change trend in real time and calculating the optimal fuel valve and air valve opening commands online in a rolling optimization manner. The equipment health diagnosis unit is used to compare the real-time collected ignition electrode breakdown voltage, fan vibration spectrum, valve response time with the health baseline in the storage unit to assess the degree of health decline, and generate an early warning or trigger a self-healing action when the threshold is exceeded. The IoT communication unit serves as a bridge between the edge and the cloud, establishing an encrypted connection with the cloud platform through the Industrial Internet of Things protocol. It is responsible for uploading data and receiving optimization strategies from the cloud. In step four, it is also responsible for pushing early warning information to the cloud. The data platform unit is used to store and analyze historical and real-time data uploaded by all access devices; The global optimization engine unit is used to run the global optimization algorithm, combine macro data such as enterprise production plans and peak-valley electricity pricing policies, generate the optimal combustion strategy, and send it to the edge intelligent control module for execution; The predictive maintenance engine unit is used to establish a more macroscopic equipment life prediction model based on a large amount of historical equipment data stored in the data platform and machine learning algorithms. When it receives a health warning from the edge module, it generates specific predictive maintenance work orders and suggestions by combining factors such as spare parts inventory and production plans. The human-machine interface unit is used to visually display equipment status, energy efficiency analysis reports, health trends and early warning information to managers, and provides an entry point for managers to view Pareto optimal solution sets, issue production plans or confirm optimization instructions.

[0012] The sensing unit communicates with the cloud-based collaborative management platform via a 4G / 5G or industrial Wi-Fi network. When the network is interrupted, the sensing unit activates a local caching and decision-making mode, and automatically resumes data transmission after the network is restored.

[0013] This invention discloses an unattended burner control system and method. The model computation unit in this design deploys a combustion dynamic model based on a long short-term memory neural network, performs offline training using historical operating data, and fine-tunes the model parameters online during actual operation using transfer learning technology. When fuel calorific value fluctuates, ambient temperature changes, furnace resistance characteristics change, or long-term equipment operation causes nonlinear drift in the system, the model computation unit can predict the furnace temperature change trend in real time and dynamically calculate the optimal air valve opening, achieving precise dynamic matching of the air-fuel ratio. Furthermore, the IoT communication unit establishes an encrypted connection with the cloud platform via an industrial IoT protocol, responsible for uploading combustion efficiency, energy consumption data, and equipment health status. The system reports and receives optimization strategies from the cloud. The data platform unit stores and analyzes historical and real-time data uploaded by all connected devices. The global optimization engine unit runs a multi-objective particle swarm optimization algorithm, combines enterprise production plans with peak-valley electricity pricing policies, generates the optimal combustion strategy, and issues it for execution. The equipment health diagnosis unit compares the real-time collected ignition electrode breakdown voltage, fan vibration spectrum, and valve response time with the health baseline to assess the degree of health degradation. The predictive maintenance engine unit, based on a large amount of historical equipment data stored in the data platform, uses machine learning algorithms to establish an equipment life prediction model, generates specific predictive maintenance work orders and suggestions, thereby achieving truly intelligent operation and predictive maintenance. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 This is a schematic diagram of the process before real-time acquisition of furnace temperature, flue gas oxygen content, fuel flow rate and furnace pressure data in an unattended burner control method of the present invention.

[0016] Figure 2 This is a schematic diagram of the process after real-time acquisition of furnace temperature, flue gas oxygen content, fuel flow rate and furnace pressure data in an unattended burner control method of the present invention.

[0017] Figure 3 This is a flowchart of the steps of an unattended burner control method according to the present invention.

[0018] Figure 4 This is a flowchart illustrating the steps of adjusting the amount of combustion air in an unattended burner control method according to the present invention.

[0019] Figure 5 This is a flowchart of the flame intensity detection steps in an unattended burner control method of the present invention.

[0020] Figure 6 This is a flowchart of the online rolling optimization calculation steps in an unattended burner control method of the present invention.

[0021] Figure 7 This is a flowchart of the offline training steps in an unattended burner control method of the present invention.

[0022] Figure 8 This is a schematic diagram of the principle of an unattended burner control system according to the present invention.

[0023] In the diagram: 1-Unattended burner control system, 2-Field control layer module, 3-Edge intelligent control module, 4-Human-machine interaction terminal, 5-Cloud collaborative management platform, 6-Burner unit, 7-Fan unit, 8-Fuel pipeline unit, 9-Air pipeline unit, 10-Sensing unit, 11-Actuation unit, 12-Flame sensor sub-unit, 13-Temperature sensor sub-unit, 14-Pressure sensor sub-unit, 15-Flue gas oxygen content sensor sub-unit, 16-Fuel regulating valve sub-unit, 17-Air regulating valve sub-unit, 18-Fuel branch solenoid valve sub-unit, 19-Data acquisition unit, 20-Storage unit, 21-Model calculation unit, 22-Equipment health diagnosis unit, 23-IoT communication unit, 24-Data platform unit, 25-Global optimization engine unit, 26-Predictive maintenance engine unit, 27-Human-machine interaction interface unit. Detailed Implementation

[0024] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0025] Please see Figures 1-7 This invention provides an unattended burner control method, comprising the following steps: S100: Based on the current ambient temperature and furnace pressure, it retrieves matching ignition parameters from historical combustion characteristics, controls the ignition electrode to discharge, and sequentially opens the valves of the fuel branch. At the same time, it adjusts the amount of combustion air and monitors the flame intensity in real time. If ignition fails, it automatically performs purging and corrects the ignition parameters according to the cause of failure for retry. In this embodiment, the historical combustion characteristics refer to the data set continuously recorded and stored by the data acquisition unit 19 during the operation of the burner system, which can characterize the correlation between the operating status of the burner under different operating conditions and the ignition results, including environmental parameters, ignition parameters and ignition results.

[0026] In this embodiment, the method for adjusting the amount of combustion air is as follows: S101: The air conditioning valve subunit 17 changes the valve core opening according to the control command; S102: The fan unit 7 adjusts the motor speed through a frequency converter, changing the fan outlet pressure and flow rate as the power source for air supply; S103: When the required air volume changes little, the opening of the air regulating valve is adjusted first. When the required air volume changes significantly, the fan speed and valve opening are adjusted simultaneously.

[0027] In the above steps, S101 achieves fine control of airflow, with an adjustment accuracy of less than 1%; in S102, the frequency conversion adjustment range is typically 30% to 100% of the rated speed; and in S103, both rapid response and inefficient operation of the fan are ensured.

[0028] In this embodiment, the flame intensity detection process is as follows: S111: Before the ignition operation begins, the data acquisition unit 19 reads the reference signal value of the flame sensor subunit 12 in the absence of flame. The equipment health diagnosis unit 22 compares the reference value with the preset absence of flame threshold. If the reference value is abnormally high, it indicates that the flame sensor subunit 12 may be faulty or interfered with by ambient light. The system suspends ignition and issues an alarm. S112: When the fuel branch solenoid valve subunit 18 is opened and the ignition electrode is discharged, the data acquisition unit 19 continuously acquires the output signal of the flame sensor subunit 12 at millisecond intervals. The model calculation unit 21 compares the real-time signal value with the flame establishment threshold. If the signal value continues to exceed the threshold within the set ignition safety time, it is determined that the flame has been successfully established and the system enters the stable combustion stage. If the signal value never reaches the threshold, it is determined that the ignition has failed and the system immediately cuts off the fuel and performs furnace purging. S113: During the stable combustion stage, the data acquisition unit 19 continuously acquires the original signal of the flame sensor subunit 12 at a millisecond cycle, and obtains a stable flame intensity value through a filtering algorithm. The model calculation unit 21 compares the filtered flame intensity value with a preset multi-level threshold and divides the combustion state into a normal combustion zone, a fluctuating combustion zone, and a dangerous critical zone. S114: When the flame intensity value is in the fluctuating combustion zone, the model calculation unit 21 combines the data from the flue gas oxygen content sensor subunit 15 to comprehensively determine the root cause of the flame fluctuation. If the flame intensity is low and the flue gas oxygen content is high, it indicates that there is excess air and insufficient fuel. The model prediction controller prioritizes increasing the opening of the fuel regulating valve subunit 16. If the flame intensity is low and the flue gas oxygen content is low, it indicates that there is insufficient air and excess fuel. In this case, the opening of the air regulating valve subunit 17 is prioritized or the speed of the fan unit 7 is increased. The flame intensity is restored to the normal combustion zone through dynamic adjustment. S115: When the signal value of the flame sensor subunit 12 remains below the flameout protection threshold for a set delay time, the model calculation unit 21 determines that a flameout fault has occurred. The fuel branch solenoid valve subunit 18 and the fuel regulating valve subunit 16 immediately cut off the fuel supply, while the blower unit 7 continues to operate for forced purging. The data acquisition unit 19 records key data at the moment of flameout, and the IoT communication unit 23 pushes the fault information to the cloud platform. If the flameout fault is determined to be a non-critical fault, the system automatically retrieves matching ignition parameters from the historical combustion characteristic database of the storage unit 20 after purging and executes the re-ignition process.

[0029] In the above steps, the safety, reliability, and intelligence of the burner system are significantly improved through a closed-loop control system encompassing pre-ignition self-check, ignition establishment judgment, continuous monitoring during the stabilization phase, dynamic adjustment of the air-fuel ratio, and flameout protection and automatic retry. The flame intensity detection process effectively avoids false ignition caused by sensor failure during the ignition phase, greatly improving the ignition success rate. During the stable combustion phase, millisecond-level continuous monitoring and dynamic adjustment achieve precise control of flame intensity, effectively avoiding combustion instability. During the fault protection phase, emergency cutoff, forced purging, and automatic retry mechanisms ensure equipment safety and enable rapid recovery after non-critical faults. The flame intensity detection process organically combines real-time monitoring, intelligent evaluation, dynamic adjustment, and fault self-healing, fundamentally solving the problems of lagging flame detection and insufficient adjustment capabilities in traditional burner control systems, providing a core guarantee for the high reliability and high efficiency operation of unattended burners.

[0030] S200: During the stable combustion stage, it collects furnace temperature, flue gas oxygen content, fuel flow rate and furnace pressure data in real time. Then, based on the above data, it predicts the change trend of furnace temperature in real time. When the set temperature changes or is disturbed by external factors, it calculates the optimal air valve opening online and outputs control signals to achieve precise dynamic matching of air-fuel ratio. In this embodiment, the process of online rolling optimization calculation is as follows: S201: The data acquisition unit 19 acquires the furnace temperature, fuel flow rate, air flow rate, furnace pressure and flue gas oxygen content data in real time at millisecond intervals, and the model calculation unit 21 uses these data as the initial state of the model prediction controller. S202: The model operation unit 21 is based on the deployed long short-term memory neural network combustion dynamic model. Starting from the current state, it predicts the trajectory of furnace temperature change step by step within the preset prediction time domain. The prediction process adopts a rolling method, and the predicted value of each time step is used as the input for the next prediction, forming a complete future temperature change prediction sequence. In the above steps, the forward propagation calculation formula in the long short-term memory neural network combustion dynamic model is as follows: ① Forward propagation in long short-term memory neural networks: Forget Gate Calculation: f t = σ ( W f [ h t -1, x t ]+ b f ) Input gate calculation: i t = σ ( W i [ h t -1, x t ]+ b i ) c t =tanh( W C [ h t -1, x t ]+ b C ) Cell status update: Ct = f t ⊙ C t -1+ i t ⊙c t Output gate calculation: o t = σ ( W o [ h t -1, x t ]+ b o ) h t = o t ⊙tanh( C t ) ② Calculation of predicted output: t + k = Wy h t + b y in, t + k For the future k Predicted furnace temperature at each time step ht This represents the current hidden state.

[0031] ③ Rolling calculation of the prediction step: In the prediction time domain Np Internally, rolling forecasts are made step-by-step over time: t +1= fLSTM ( x t +1, h t , C t ) t +2= fLSTM ( x t +2, h t +1, C t +1)

[0032] t + N p =fLSTM( x t + N p , ht + N p -1, C t + N p -1) .

[0033] S203: The model operation unit 21 solves the optimization problem in the control time domain with the goal of minimizing the deviation between the set temperature and the actual predicted temperature, while taking into account fuel consumption and equipment constraints, and calculates the optimal air regulating valve opening sequence and fuel regulating valve opening sequence through online optimization algorithm; In the above steps, the optimization problem is solved using a model predictive controller, which involves the following formulas for calculating the objective function and constraints: ① Optimize the objective function

[0034] The model predictive controller aims to minimize the deviation between the setpoint temperature and the actual predicted temperature, while also considering fuel consumption. The objective function is calculated as follows: in, J To optimize the target value, N p To predict the length of the time domain, t + k For the future k The predicted temperature of the step, r t + k For the future k The set temperature of the step, Q The output error weight matrix is... Nc To control the time domain length, Δu t + j For the first j Step control increment, R To control the incremental weight matrix.

[0035] ② Control Constraints The formulas for calculating the opening constraints of the fuel regulating valve and the air regulating valve are as follows: u min ≤ u t ≤ umax in, u t This represents the control quantity (valve opening) at the current moment. u min and u max These represent the minimum and maximum opening degrees of the valve, respectively.

[0036] ③ Control Incremental Constraints The formula for calculating the control quantity change rate constraint is: Δ u min ≤Δ u t ≤Δ u max Where, Δ u t Δ is the control increment at the current moment. u min and Δ u max These represent the minimum and maximum allowable rates of change, respectively.

[0037]

[0038] ④ Wind-fuel ratio constraint The formula for calculating the wind-fuel ratio constraint is: in, u air , t For the opening degree of the air regulating valve, u fuel,t For the fuel regulating valve opening, AFR min and AFR max These are the minimum and maximum air-fuel ratios, respectively.

[0039] ⑤ Furnace pressure constraint The formula for calculating furnace pressure constraint is: P min ≤ P t ≤ P max in, P t The furnace pressure at the current moment. P min and P max These are the lower and upper limits of the furnace pressure, respectively.

[0040] ⑥ Optimization Solution

[0041] The model predictive controller uses a quadratic programming algorithm to solve for the optimal control sequence. The solution formula is as follows: in, U This is the optimal control sequence. U For the control variable vector.

[0042] S204: The model calculation unit 21 will send the air regulating valve opening command and the fuel regulating valve opening command of the first control time step obtained by optimization calculation to the air regulating valve subunit 17 and the fuel regulating valve subunit 16 respectively through the output module for execution; S205: When the next sampling time arrives, the model operation unit 21 re-collects the current state data, rolls the prediction time domain and control time domain forward by one sampling cycle, and repeats steps one to four.

[0043] S300: Continuously monitors the cumulative number of ignitions and breakdown voltage of the ignition electrode, the operating current and vibration spectrum of the wind turbine, and the action response time of the valve. Then compares the real-time data with the health baseline to assess the health degradation of key components. When the assessment value exceeds the warning threshold, it automatically sends predictive maintenance suggestions. When a non-critical fault is detected, it automatically switches to redundant equipment or adjusts the operating mode to self-heal the fault. S400: Based on regularly uploaded combustion efficiency, energy consumption data and equipment health reports, combined with enterprise production plans and peak-valley electricity pricing policies, it generates the optimal combustion strategy through global optimization, issues it for execution, and adjusts operating parameters to achieve the best match between energy consumption and production needs.

[0044] Furthermore, the ignition parameters include the valve opening degree during ignition, the ignition duration, and the purging time. The historical combustion characteristics store the environmental parameters and corresponding ignition parameter combinations for each successful ignition.

[0045] In this embodiment, by establishing a mapping relationship between environmental parameters and ignition parameters, the ignition process is no longer an open-loop control with fixed parameters, but rather an intelligent matching based on historical successful experience. When the ambient temperature or furnace pressure changes, the method can automatically retrieve the ignition parameter combination that best matches the current environment, thereby achieving rapid, stable, and reliable ignition under different operating conditions. This avoids ignition failure or insufficient ignition caused by environmental changes, further improving the reliability and adaptability of unattended operation. The physical meaning of valve opening is as follows: when the valve is fully closed, the opening is 0%, and fluid cannot pass through; when the valve is fully open, the opening is 100%, and fluid passes through at maximum flow rate; when the valve is in the intermediate position, the opening is between 0% and 100%, and the fluid flow rate has a certain non-linear relationship with the opening. For the fuel regulating valve subunit 16, the valve opening determines the amount of fuel gas entering the burner; for the air regulating valve subunit 17, the valve opening determines the amount of combustion air supplied. Among them, the safety constraints on valve opening are: ① Minimum opening constraint: The opening of the fuel regulating valve shall not be lower than the set minimum value (usually 5% to 10%) to avoid flame instability or backfire due to excessive opening; ② Maximum opening constraint: The opening of the fuel regulating valve shall not exceed the set maximum value (usually 90% to 95%) to avoid losing the regulating margin when the valve is close to fully open; ③ Air-fuel ratio constraint: The ratio of the air regulating valve opening to the fuel regulating valve opening must be maintained within a safe range (usually 8:1 to 12:1) to ensure complete combustion and not produce excessive harmful emissions; ④ Rate of change constraint: The rate of change of valve opening is limited (usually no more than 10% to 20% per second) to avoid violent fluctuations in combustion due to excessively rapid adjustment.

[0046] Furthermore, the predicted furnace temperature is based on a data-driven model constructed using a long short-term memory neural network. This data-driven model is trained offline using historical operating data and fine-tuned online based on the latest operating data through transfer learning techniques during actual operation.

[0047] In this embodiment, the offline training process includes: S211: The data acquisition unit 19 extracts historical running data for model training from the historical combustion feature database of the storage unit 20. The model operation unit 21 performs outlier detection and removal, removes invalid data caused by sensor failure or communication interruption, fills in missing values, and uses linear interpolation or forward filling methods to fill in data gaps. Finally, it performs data smoothing filtering to eliminate high-frequency noise interference and obtain a high-quality training dataset. In the above steps, outlier detection uses a standard deviation-based outlier detection method, calculated using the following formula: | X i - X |> k σ in, X iFor the i-th data point, X The mean of the data. σ Standard deviation, k This is the threshold coefficient (usually set to 3). When the deviation of a data point from the mean exceeds... k Values ​​exceeding one standard deviation are considered outliers and removed.

[0048] In the above steps, missing value imputation uses linear interpolation to fill missing data, and the calculation formula is as follows:

[0049] in, X t Let t be the time to be filled. t Data values, X t1 and X t2 These are the nearest valid data values ​​before and after the missing point, respectively.

[0050]

[0051] In the above steps, the data smoothing filter uses a moving average filtering method to eliminate high-frequency noise. The calculation formula is as follows: Where, x t The smoothed data values, m The length of the sliding window. X i The original data value within the window.

[0052] S212: The model operation unit 21 uses furnace temperature as the prediction target label and fuel flow rate, air flow rate, furnace pressure, flue gas oxygen content, and historical temperature values ​​from multiple past time steps as input features. The model operation unit 21 uses a sliding window method to construct time series samples. Each sample contains the input feature sequence from several past time steps and the output label for the corresponding future time step. After construction, the model operation unit 21 divides the dataset into training set, validation set, and test set according to the proportion. S213: The model operation unit 21 constructs a combustion dynamic model architecture based on a long short-term memory neural network. The model input layer receives multi-dimensional time series features constructed by a sliding window. The intermediate layer contains multiple long short-term memory network layers, each layer is set with several hidden units to extract deep features of the time series. The output layer uses a fully connected layer to map the hidden state to the predicted value of the furnace temperature. The model operation unit 21 initializes the model parameters, uses the Xavier initialization method to assign initial values ​​to the weight matrix, and initializes the bias term to zero. S214: The model operation unit 21 inputs the sample data of the training set into the model for forward propagation calculation. The model executes the calculation of the forget gate, input gate, cell state update and output gate in the order of time steps, passing it layer by layer and finally outputting the predicted value of the furnace temperature. The model operation unit 21 substitutes the predicted value and the actual historical value into the loss function to calculate the mean squared error loss value, and introduces the L2 regularization term to prevent overfitting and obtain the total loss value. The model operation unit 21 also calculates the loss value on the validation set. S215: The model operation unit 21 calculates the gradient of the loss function with respect to the parameters of each layer of the model based on the calculated total loss value through the time backpropagation algorithm. The model operation unit 21 adopts the Adam optimization algorithm, combines the first moment estimation and second moment estimation of the gradient, adaptively adjusts the learning rate, and iteratively updates the model weights and biases. S216: After training, the model operation unit 21 inputs the test set data into the model for prediction, calculates evaluation indicators such as root mean square error, mean absolute percentage error and coefficient of determination, and comprehensively verifies the prediction accuracy and generalization ability of the model. If the model performance meets the preset requirements, the model operation unit 21 packages and saves the trained model parameters, network architecture information and normalization parameters, and stores them in the model library of the storage unit 20.

[0053] Furthermore, the method for establishing the health baseline is to collect operating data of key components for more than one week during the initial stable operation period after the equipment is first put into operation or after a major overhaul, and then take the statistical characteristic values ​​as the benchmark after data cleaning.

[0054] In this embodiment, by using the equipment's own optimal operating state as the health assessment benchmark, rather than adopting a universal fixed threshold, the health diagnosis becomes more personalized and accurate. The data collection cycle of more than one week ensures the statistical stability and representativeness of the baseline, while the data cleaning step removes the interference of outliers. The method of establishing the health baseline provides a scientific and reliable reference standard for subsequent equipment health assessment, enabling the assessment results to truly reflect the degree of degradation of the equipment relative to its own optimal state, and providing an accurate decision-making basis for predictive maintenance.

[0055] Furthermore, the fault self-healing includes automatically switching the system to the redundant flame sensor signal when the flame sensor malfunctions; and switching the system to the parallel bypass valve for fine control when the fuel valve is stuck.

[0056] In this embodiment, the redundant design of key components and the automatic fault switching mechanism enable the system to continue operating even in the event of non-critical failures, avoiding system shutdown due to a single component failure. The redundant switching of the flame sensor ensures the continuity of flame monitoring and eliminates safety hazards caused by sensor failure. The fuel valve bypass switching mechanism ensures that fuel flow control is not affected by main valve jamming, maintaining stable combustion. The fault self-healing function significantly improves the availability and robustness of the system, reduces the number of unplanned shutdowns, and truly achieves high-reliability operation under unattended conditions.

[0057] Furthermore, the global optimization is a multi-objective particle swarm optimization algorithm, which uses production plan completion rate, comprehensive energy consumption cost and equipment loss rate as optimization objectives to generate a Pareto optimal solution set for cloud platform decision-making.

[0058] In this embodiment, the calculation formula of the multi-objective particle swarm optimization algorithm is as follows: ① Particle swarm initialization formula: Particle position initialization: X i (0) = X min + r 1 ( X max - X min ) in, X i (0) Let be the initial position vector of the i-th particle. X min and X max These are the lower and upper limits of the decision variable (such as the range of fuel flow rate, the range of air flow rate, etc.). r 1 is a random number within the interval [0,1].

[0059] Particle velocity initialization: V i (0) = V min + r 2 ( V max - V min ) in, V i (0)Let be the initial velocity vector of the i-th particle. V min and V max These are the lower and upper limits of the speed, respectively. r 2 is a random number in the interval [0,1].

[0060] ② Formula for multi-objective fitness function:

[0061] The objective function for production plan completion is to minimize the deviation between actual and planned output. The calculation formula is as follows: in, f 1( X Let Q be the production plan deviation rate. actual (X) represents the predicted output under the current combustion strategy X, and Q is... plan The target output for the production plan.

[0062] The overall energy cost objective function aims to minimize the sum of fuel and electricity costs. The calculation formula is as follows: f 2( X )= Cfuel F ( X )+ Cpower P ( X ) in, f 2( X The total energy cost is... Cfuel For fuel unit price, F ( X ( ) represents the fuel consumption under the current strategy. Cpower Electricity price (taking into account peak and off-peak time differences). P ( X () represents the electricity consumption under the current strategy.

[0063] The objective function for equipment loss rate is to minimize the cumulative loss of critical components.

[0064] The calculation formula is: in, f 3( X The overall equipment loss rate is denoted as . N cycle ( X () represents the number of ignitions under the current strategy. Ncycle,max This represents the maximum permissible number of ignition cycles for the ignition electrode. T valve ( X This represents the cumulative number of valve actions under the current strategy. T valve,max This represents the maximum permissible number of valve actuations. I fan ( X The integral of the cumulative operating current of the wind turbine under the current strategy is calculated. I fan,max The integral of the maximum allowable current of the wind turbine, α α β β γ γ The weighting coefficients satisfy α + β + γ = 1 α + β + γ =1.

[0065] Please see Figure 8 The present invention also provides an unattended burner control system, applied to the unattended burner control method described above. The unattended burner control system 1 includes a field control layer module 2, an edge intelligent control module 3, a human-machine interaction terminal 4, and a cloud collaborative management platform 5. The field control layer module 2 includes a burner unit 6, a fan unit 7, a fuel pipeline unit 8, an air pipeline unit 9, a sensing unit 10, and an execution unit 11. The sensing unit 10 includes a flame sensor subunit 12, a temperature sensor subunit 13, a pressure sensor subunit 14, and a flue gas oxygen content sensor subunit 15. The execution unit 11 includes a fuel regulating valve subunit 16, an air regulating valve subunit 17, and a fuel branch solenoid valve subunit 18. The edge intelligent control module 3 includes a data acquisition unit 19, a storage unit 20, a model calculation unit 21, an equipment health diagnosis unit 22, and an IoT communication unit 23. The cloud collaborative management platform 5 includes a data middleware unit 24, a global optimization engine unit 25, a predictive maintenance engine unit 26, and a human-machine interaction interface unit 27. The burner unit 6 serves as a combustion reaction vessel for generating an initial flame. The fan unit 7 is used to provide combustion air; The fuel pipeline unit is used to provide a fuel passage and dynamically transmit the regulated fuel flow rate in combustion control. The air duct unit 9 serves as the physical channel for air delivery and works in conjunction with the fan unit 7 to execute air volume adjustment commands in combustion control. The flame sensor subunit 12 is used to confirm whether a flame has been established; The temperature sensor subunit 13 is used to calculate the deviation from the set temperature and determine the adjustment amount; The pressure sensor subunit 14 is used to detect whether the gas pressure is within the safe range during self-testing and to monitor the furnace pressure. The flue gas oxygen content sensor subunit 15 is used for adaptive combustion control; The fuel regulating valve subunit 16 is used to establish the initial fuel flow and acts as the main actuator for regulating the fuel flow in the combustion control MPC algorithm. At the same time, it can switch to the bypass valve when the fuel valve is stuck. The air conditioning valve subunit 17 is used in adaptive combustion control to precisely adjust the air volume in conjunction with the fan frequency converter and perform dynamic air-fuel ratio matching. The fuel branch solenoid valve subunit 18 is used as a safety shut-off valve for the ignition line or main line. The data acquisition unit 19 is used to collect temperature, pressure, flame, oxygen content data and feedback signals from the actuator in real time during self-testing, ignition, combustion control and health assessment. The storage unit 20 is used to store historical data on combustion characteristics, health baseline, and health assessment. The model calculation unit 21 is used to deploy and run the combustion dynamic model and the model prediction controller, and is responsible for predicting the temperature change trend in real time and calculating the optimal fuel valve and air valve opening commands online. The equipment health diagnosis unit 22 is used to compare the real-time collected ignition electrode breakdown voltage, fan vibration spectrum, valve response time with the health baseline in the storage unit 20, assess the degree of health decline, and generate an early warning or trigger a self-healing action when the threshold is exceeded. The IoT communication unit 23 is used as a bridge between the edge and the cloud. It establishes an encrypted connection with the cloud platform through the Industrial Internet of Things protocol, and is responsible for uploading data and receiving optimization strategies issued by the cloud. At the same time, in step four, it is responsible for pushing early warning information to the cloud. The data platform unit 24 is used to store and analyze historical and real-time data uploaded by all access devices; The global optimization engine unit 25 is used to run the global optimization algorithm, combine macro data such as enterprise production plan and peak-valley electricity price policy, generate the optimal combustion strategy, and send it to the edge intelligent control module 3 for execution; The predictive maintenance engine unit 26 is used to establish a more macroscopic equipment life prediction model based on a large amount of historical equipment data stored in the data platform and using machine learning algorithms. When it receives a health warning from the edge module, it generates specific predictive maintenance work orders and suggestions by combining factors such as spare parts inventory and production plans. The human-machine interface unit 27 is used to visually display equipment status, energy efficiency analysis reports, health trends and early warning information to managers, and provides an entry point for managers to view Pareto optimal solution sets, issue production plans or confirm optimization instructions.

[0066] In this embodiment, a layered architecture consisting of the field control layer module 2, the edge intelligent control module 3, the cloud collaborative management platform 5, and the human-machine interaction terminal 4 provides complete hardware and software support. The various units and sub-units within the field control layer module 2 have clearly defined roles and work collaboratively to form a complete physical execution and sensing system. The edge intelligent control module 3, through the organic cooperation of the data acquisition unit 19, storage unit 20, model calculation unit 21, equipment health diagnosis unit 22, and IoT communication unit 23, brings intelligent computing capabilities closer to the data source, achieving low-latency real-time control and local decision-making. The cloud collaborative management platform 5, through the collaborative operation of the data middle platform unit 24, global optimization engine unit 25, predictive maintenance engine unit 26, and human-machine interaction interface unit 27, achieves cross-device data aggregation, global optimization, and advanced maintenance management. This three-layer architecture ensures millisecond-level real-time control response and achieves enterprise-level global optimization, completely breaking the information silo limitations of traditional control systems.

[0067] Furthermore, the sensing unit 10 communicates with the cloud-based collaborative management platform 5 via a 4G / 5G or industrial Wi-Fi network. When the network is interrupted, the sensing unit 10 activates a local caching and decision-making mode, and automatically resumes data transmission after the network is restored.

[0068] In this embodiment, the adoption of high-speed wireless communication technology significantly reduces on-site wiring costs and construction difficulties, making system deployment more flexible and convenient. The introduction of local caching and decision-making modes enables the system to maintain normal operation even when facing network fluctuations or interruptions. The edge intelligent control module 3 can continue to make independent decisions based on locally stored data and models, avoiding control interruptions caused by network problems. The automatic resume transmission function after network recovery ensures data integrity, enabling the cloud platform to obtain complete historical data for subsequent analysis and model optimization. The design of the communication mechanism gives the system stronger anti-interference capabilities and operational reliability in complex industrial environments.

[0069] The above description discloses only one preferred embodiment of the present invention, and should not be construed as limiting the scope of the present invention. Those skilled in the art will understand that all or part of the processes of the above embodiments can be implemented, and equivalent changes made in accordance with the claims of the present invention are still within the scope of the invention.

Claims

1. A method for controlling an unattended burner, characterized in that, Includes the following steps: Based on the current ambient temperature and furnace pressure, the matching ignition parameters are retrieved from historical combustion characteristics, the ignition electrode discharge is controlled, and the valves of the fuel branch are opened in sequence. At the same time, the amount of combustion air is adjusted and the flame intensity is detected in real time. If ignition fails, purging is automatically performed and the ignition parameters are corrected according to the reason for failure and the test is repeated. During the stable combustion stage, real-time data on furnace temperature, flue gas oxygen content, fuel flow rate, and furnace pressure are collected. Then, based on the above data, the trend of furnace temperature change is predicted in real time. When the set temperature changes or is disturbed by external factors, the optimal air valve opening is calculated online and the control signal is output to achieve precise dynamic matching of the air-fuel ratio. The system continuously monitors the cumulative number of ignitions and breakdown voltage of the ignition electrode, the operating current and vibration spectrum of the wind turbine, and the action response time of the valve. The real-time data is then compared with the health baseline to assess the health degradation of key components. When the assessment value exceeds the warning threshold, predictive maintenance suggestions are automatically sent. When a non-critical fault is detected, the system automatically switches to redundant equipment or adjusts the operating mode until the fault self-heals. Based on regularly uploaded combustion efficiency, energy consumption data, and equipment health reports, combined with enterprise production plans and peak-valley electricity pricing policies, the optimal combustion strategy is generated through global optimization, issued and executed, and operating parameters are adjusted to achieve the best match between energy consumption and production needs.

2. The unattended burner control method as described in claim 1, characterized in that, The ignition parameters include the valve opening degree, ignition duration, and purging time during ignition. The historical combustion characteristics store the environmental parameters and corresponding ignition parameter combinations for each successful ignition.

3. The unattended burner control method as described in claim 2, characterized in that, The predicted furnace temperature is based on a data-driven model constructed using a long short-term memory neural network. This data-driven model is trained offline using historical operating data and fine-tuned online using transfer learning techniques based on the latest operating data during actual operation.

4. The unattended burner control method as described in claim 3, characterized in that, The method for establishing the health baseline is to collect operating data of key components for more than one week during the initial stable operation period after the equipment is first put into operation or after a major overhaul, and then take the statistical characteristic values ​​as the benchmark after data cleaning.

5. The unattended burner control method as described in claim 4, characterized in that, The fault self-healing includes automatically switching to the redundant flame sensor signal when the flame sensor malfunctions; and switching to the parallel bypass valve for fine control when the fuel valve is stuck.

6. The unattended burner control method as described in claim 5, characterized in that, The global optimization is a multi-objective particle swarm optimization algorithm, which uses production plan completion rate, comprehensive energy consumption cost and equipment loss rate as optimization objectives to generate a Pareto optimal solution set for cloud platform decision-making.

7. An unattended burner control system, applied to the unattended burner control method as described in claim 6, characterized in that, The unattended burner control system includes a field control layer module, an edge intelligent control module, a human-machine interface terminal, and a cloud-based collaborative management platform. The field control layer module includes a burner unit, a fan unit, a fuel pipeline unit, an air pipeline unit, a sensing unit, and an execution unit. The sensing unit includes a flame sensor subunit, a temperature sensor subunit, a pressure sensor subunit, and a flue gas oxygen content sensor subunit. The execution unit includes a fuel regulating valve subunit, an air regulating valve subunit, and a fuel branch solenoid valve subunit. The edge intelligent control module includes a data acquisition unit, a storage unit, a model calculation unit, an equipment health diagnosis unit, and an IoT communication unit. The cloud-based collaborative management platform includes a data middleware unit, a global optimization engine unit, a predictive maintenance engine unit, and a human-machine interface unit. The burner unit serves as a combustion reaction vessel for generating an initial flame. The fan unit is used to provide combustion air; The fuel pipeline unit is used to provide a fuel passage and dynamically transmit the regulated fuel flow rate in combustion control; The air duct unit serves as the physical channel for air delivery and works in conjunction with the fan unit to execute airflow adjustment commands in combustion control. The flame sensor subunit is used to confirm whether a flame has been established. The temperature sensor subunit is used to calculate the deviation from the set temperature and determine the adjustment amount; The pressure sensor subunit is used to detect whether the gas pressure is within a safe range during self-testing and to monitor the furnace pressure. The flue gas oxygen content sensor subunit is used for adaptive combustion control; The fuel regulating valve subunit is used to establish the initial fuel flow and acts as the main actuator for regulating the fuel flow in the combustion control MPC algorithm. At the same time, it can switch to the bypass valve when the fuel valve is stuck. The air conditioning valve subunit is used in adaptive combustion control to precisely adjust the air volume in conjunction with the fan frequency converter and perform dynamic air-fuel ratio matching. The fuel branch solenoid valve subunit is used as a safety shut-off valve for the ignition line or main line. The data acquisition unit is used to collect temperature, pressure, flame, and oxygen content data from the sensor unit, as well as feedback signals from the actuator, in real time during self-testing, ignition, combustion control, and health assessment. The storage unit is used to store historical data on combustion characteristics, health baseline, and health assessment. The model computing unit is used to deploy and run the combustion dynamic model and the model prediction controller, and is responsible for predicting the temperature change trend in real time and calculating the optimal fuel valve and air valve opening commands online in a rolling optimization manner. The equipment health diagnosis unit is used to compare the real-time collected ignition electrode breakdown voltage, fan vibration spectrum, valve response time with the health baseline in the storage unit to assess the degree of health decline, and generate an early warning or trigger a self-healing action when the threshold is exceeded. The IoT communication unit serves as a bridge between the edge and the cloud, establishing an encrypted connection with the cloud platform through the Industrial Internet of Things protocol. It is responsible for uploading data and receiving optimization strategies from the cloud. In step four, it is also responsible for pushing early warning information to the cloud. The data platform unit is used to store and analyze historical and real-time data uploaded by all access devices; The global optimization engine unit is used to run the global optimization algorithm, combine macro data such as enterprise production plans and peak-valley electricity pricing policies, generate the optimal combustion strategy, and send it to the edge intelligent control module for execution; The predictive maintenance engine unit is used to establish a more macroscopic equipment life prediction model based on a large amount of historical equipment data stored in the data platform and machine learning algorithms. When it receives a health warning from the edge module, it generates specific predictive maintenance work orders and suggestions by combining factors such as spare parts inventory and production plans. The human-machine interface unit is used to visually display equipment status, energy efficiency analysis reports, health trends and early warning information to managers, and provides an entry point for managers to view Pareto optimal solution sets, issue production plans or confirm optimization instructions.

8. The unattended burner control method as described in claim 7, characterized in that, The sensing unit communicates with the cloud-based collaborative management platform via a 4G / 5G or industrial Wi-Fi network. When the network is interrupted, the sensing unit activates local caching and decision-making mode, and automatically resumes data transmission after the network is restored.

Images