A circulating fluidized bed boiler intelligent temperature control system and method

Abstract

The present invention relates to the technical field of circulating fluidized bed boilers. More specifically, the present invention provides an intelligent temperature control system and method for a circulating fluidized bed boiler. Through multi-algorithm collaborative control and an intelligent compensation mechanism, precise regulation of boiler operating parameters is achieved. A model prediction control unit constructs a multivariable control matrix based on an ARX / FIR model, which can predict the dynamic trajectory of boiler bed temperature, pressure and oxygen content parameters in real time, and converts the multivariable coupling problem that is difficult to handle with traditional PID control into an optimization problem, thereby improving control accuracy. A neural network optimization control unit dynamically adjusts the ratio of coal feed and air volume through a reinforcement learning mechanism, thereby improving the thermal efficiency of the boiler. A fuzzy control unit converts the operator's experience rules into compensation control quantities under nonlinear boundary conditions, effectively suppressing bed temperature overshoot and oxygen content mutation, and maintaining the stability of the combustion process under variable operating conditions.

Description

Technical Field

[0001] The present invention relates to the technical field of circulating fluidized bed boilers, and more particularly to an intelligent temperature control system and method for a circulating fluidized bed boiler. Background Art

[0002] Circulating fluidized bed (CFB) boilers are highly efficient, low-pollution coal-fired boilers with advantages such as wide fuel adaptability, high combustion efficiency, and low pollutant emissions. They are widely used in the power, chemical, and heating sectors. However, the operation of CFB boilers is complex and is affected by a variety of factors, such as coal type, load, and air volume. This can lead to unstable operation and large fluctuations in bed temperature, impacting the safety and economic efficiency of the boilers.

[0003] As a highly efficient and clean coal-fired equipment, the precise control of the operating parameters of a circulating fluidized bed boiler directly affects combustion efficiency, pollutant emissions, and operational safety. Existing temperature control systems mostly rely on a single control algorithm, such as traditional PID control (the most basic feedback control algorithm in the field of automatic control) or basic fuzzy control, which has problems such as dynamic response lag, insufficient multi-variable coupling compensation, and poor adaptability to operating conditions. For example, there is a strong nonlinear correlation between the bed temperature, pressure, and oxygen content parameters, and a single algorithm is difficult to decouple and optimize the control quantity in real time; at the same time, the traditional system lacks an effective processing mechanism for sensor noise, data redundancy, and transmission delays, resulting in reduced prediction accuracy and untimely early warning responses. In addition, there is a lack of a safety verification mechanism when operators manually intervene, and the adjustment of the control strategy relies on offline configuration, which makes it difficult to meet the real-time optimization needs of complex operating conditions.

[0004] Therefore, developing an intelligent temperature control system for circulating fluidized bed boilers that integrates multiple intelligent control algorithms, has an adaptive learning mechanism, and can achieve multi-modal intelligent control is of great significance for improving the stability, safety, and economy of boiler operation. Summary of the Invention

[0005] The object of the present invention is to provide an intelligent temperature control system for a circulating fluidized bed boiler and a method thereof, so as to solve the problems raised in the above background technology.

[0006] To achieve the above objectives, the present invention provides the following technical solutions: including a data acquisition module, an intelligent control module, an execution module, a monitoring and early warning module, and a human-computer interaction module:

[0007] The data acquisition module is used to collect boiler operating parameters in real time and perform standardized processing, and synchronously transmit the standardized data to the intelligent control module through the redundant communication bus;

[0008] The intelligent control module consists of a model prediction control unit, a fuzzy control unit, and a neural network optimization control unit. It is used to receive standardized data transmitted by the data acquisition module in real time, integrate the three control algorithms of model prediction control, fuzzy control, and neural network optimization to perform dynamic calculations to generate the final control instructions, and transmit them synchronously to the execution module. At the same time, it generates prediction data and transmits it to the monitoring and early warning module.

[0009] The execution module includes the coal feeder, primary fan, secondary fan and valves. It receives the final control instructions from the intelligent control module to execute equipment actions, and feeds back the execution status data to the data acquisition module. At the same time, it uploads the action parameters to the monitoring and early warning module.

[0010] The monitoring and early warning module processes the boiler operating parameters of the data acquisition module, the prediction data of the intelligent control module, and the action parameters of the execution module in parallel. It triggers the three-level early warning mechanism through pattern matching to generate early warning information. The early warning information is simultaneously pushed to the human-computer interaction module and triggers the protection logic of the intelligent control module.

[0011] The human-computer interaction module uses three-dimensional visualization technology to display the boiler's operating status, final control instruction parameters and early warning information in real time. It integrates a two-way control channel to support operators to manually correct the final control instruction parameters. The corrected instructions are directly transmitted to the intelligent control module after authority verification.

[0012] A further technical solution of the present application is as follows: the boiler operating parameters include boiler bed temperature, pressure and oxygen content parameters;

[0013] The standardized data contains a triple field of timestamp, sensor ID and raw measurement value. The intelligent control module automatically matches the corresponding multivariable control matrix column vector by parsing the sensor ID in the field;

[0014] The model prediction control unit constructs a multivariable control matrix based on the ARX model and the FIR model, predicts the future trajectory of the boiler operating parameters in real time, and generates a multi-parameter coupled reference control quantity including a coal feed speed reference and a damper opening reference;

[0015] The fuzzy control unit adopts the TS fuzzy model to convert the operator's experience rules into compensation control quantities under nonlinear boundary conditions. The compensation weight is dynamically adjusted by monitoring the bed temperature overshoot and oxygen content mutation rate, and is fused with the multi-parameter coupling benchmark control quantity to generate a fusion control instruction.

[0016] The neural network optimization control unit adopts a parallel fast learning network structure, dynamically adjusts the set value of the coal feed rate and air volume ratio through the reinforcement learning mechanism, and generates optimized control instructions;

[0017] The fusion control instructions and the optimization control instructions are dynamically calculated to generate the final control instructions.

[0018] A further technical solution of the present application is: the final control instruction generated by the intelligent control module is encapsulated in JSON format, including four-dimensional information of device address code, action type, target value and priority. The execution module triggers the control program of the corresponding device by parsing the device address code.

[0019] A further technical solution of the present application is: when the execution status data fed back by the execution module is transmitted via the redundant communication bus of the data acquisition module, a digital signature and a timestamp are synchronously attached, and the monitoring and early warning module calculates the transmission delay compensation amount by comparing the timestamp with the local clock difference.

[0020] A further technical solution of this application: the three-level warning mechanism includes: the warning information output by the monitoring and warning module is transmitted using a hierarchical queue, the red warning directly triggers the sound and light alarm interface of the human-computer interaction module, the orange warning is pushed to the operation log pending queue, and the yellow warning is displayed as an interface floating prompt.

[0021] A further technical solution of the present application is: the correction instruction includes an operator identity authentication token and a parameter change trajectory. After the intelligent control module verifies the validity of the token, it stores the parameter change trajectory in the control strategy version library.

[0022] A further technical solution of the present application is that the prediction data generated by the intelligent control module includes real-time predicted future trajectories of boiler bed temperature, pressure and oxygen content parameters. The prediction data is exchanged in real time with the monitoring and early warning module using shared memory to avoid repeated transmission of data on redundant communication buses.

[0023] A further technical solution of the present application is: the neural network optimization control unit is also provided with a training data buffer for receiving the original data of the boiler operating parameters transmitted by the data acquisition module. When the amount of data in the training data buffer reaches a preset threshold, the offline retraining process of the neural network optimization control is automatically triggered.

[0024] An intelligent temperature control method for an intelligent temperature control system of a circulating fluidized bed boiler, the method comprising the following steps:

[0025] S1. The data acquisition module acquires boiler operating parameters in real time and generates standardized data;

[0026] S2. The intelligent control module receives standardized data and generates final control instructions through multi-algorithm fusion calculations;

[0027] S3. The execution module adjusts the device action according to the final control instruction and feedbacks the execution status data;

[0028] S4. The monitoring and warning module triggers a three-level warning mechanism through pattern matching to generate warning information and simultaneously pushes the warning information to the human-computer interaction module;

[0029] S5. The human-computer interaction module supports manual modification of the final control instruction parameters. The modified instructions are passed to the intelligent control module after authority verification.

[0030] The present application further provides a technical solution: the multi-algorithm fusion calculation in step S2 is specifically three control algorithms: model predictive control, fuzzy control and neural network optimization, including a dynamic credibility allocation mechanism, which uses the entropy weight method to evaluate the output credibility of each control algorithm in real time, wherein the model predictive control weight coefficient is negatively correlated with the rate of change of the operating condition;

[0031] The execution status data in step S3 includes a digital signature and a timestamp, and the monitoring and warning module calculates the transmission delay compensation amount by comparing the timestamp with the local clock difference;

[0032] The warning information in step S4 adopts a hierarchical queue transmission mechanism. The red warning directly triggers the sound and light alarm interface of the human-computer interaction module, the orange warning is pushed to the operation log processing queue, and the yellow warning is displayed as an interface floating prompt.

[0033] The present invention realizes precise control of boiler operating parameters through multi-algorithm collaborative control and intelligent compensation mechanism. The model prediction control unit is based on the multivariable control matrix constructed by the ARX / FIR model, which can predict the dynamic trajectory of the boiler bed temperature, pressure and oxygen content parameters in real time, and convert the multivariable coupling problem that is difficult to handle with traditional PID control into an optimization problem, thereby improving the control accuracy. The neural network optimization control unit dynamically adjusts the ratio of coal feed and air volume through the reinforcement learning mechanism to improve the thermal efficiency of the boiler. The fuzzy control unit converts the operator's experience rules into compensation control quantities under nonlinear boundary conditions, effectively suppressing bed temperature overshoot and oxygen content mutation, and maintaining the stability of the combustion process under variable operating conditions.

[0034] The present invention constructs a full-link closed-loop monitoring from data acquisition to execution feedback through a three-level safety control system. The data acquisition module ensures data reliability through redundant communication buses and digital signature technology, and the three-level instruction fusion mechanism of the intelligent control module improves control robustness. The innovative hierarchical queue transmission mechanism of the monitoring and early warning module shortens the red warning response time, and cooperates with the sound and light alarm interface to achieve immediate intervention in dangerous working conditions. In addition, the human-computer interaction module based on three-dimensional visualization supports the rapid loading of fault plans, and combines the authority verification mechanism to ensure the safety of manual intervention, thereby reducing fault recovery time. The system's built-in energy-saving mode automatically reduces the fan speed when operating under partial load, and cooperates with the intelligent control algorithm to achieve energy efficiency optimization under all working conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0036] The following will be combined with the accompanying drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention. The present invention is further described below in conjunction with the embodiments.

[0037] See also Figure 1 In one embodiment of the present application, a circulating fluidized bed boiler intelligent temperature control system includes a data acquisition module, an intelligent control module, an execution module, a monitoring and warning module, and a human-computer interaction module. The data acquisition module includes a bed temperature sensor, a pressure sensor, a flow meter, and an oxygen concentration sensor. The data acquisition module collects boiler operating parameters in real time and synchronously transmits standardized data to the intelligent control module via a redundant communication bus. The standardized data includes multi-source heterogeneous parameters with aligned timestamps.

[0038] The data acquisition module integrates multiple sensor types to achieve comprehensive sensing of boiler operating parameters. The bed temperature sensor utilizes an armored thermocouple inserted vertically into the center of the dense phase zone, equipped with a mounting bracket and compressed air purge device to prevent ash clogging. A pressure sensor is installed horizontally in the cyclone separator inlet flue, equipped with a condenser to reduce the medium temperature. A flow meter is installed in the primary air main duct, equipped with a rectifier to ensure flow stability. The oxygen concentration sensor is installed obliquely in the tail flue, equipped with an electric heater and a compressed air back-purge system to maintain probe cleanliness. The data acquisition card utilizes synchronous sampling mode, a built-in digital filter to suppress high-frequency noise, and an isolation amplifier and low-pass filter in the signal conditioning circuit to enhance signal quality. The communication bus utilizes a dual-redundant network, encapsulating timestamps, sensor IDs, and raw data according to standard protocols. A hardware clock synchronization module ensures time skew within 1ms. The redundant design utilizes a triple sensor-level configuration, a dual-bus architecture at the communication level, and a hot standby module at the power level to ensure high reliability and real-time data acquisition. Standardized data is transmitted to the intelligent control module via redundant communication buses, providing a complete and accurate input foundation for the system.

[0039] The intelligent control module consists of a model prediction control unit, a fuzzy control unit, and a neural network optimization control unit. The model prediction control unit constructs a multivariable control matrix based on the ARX model (AutoRegressive with eXogenous inputs) and the FIR model (Finite Impulse Response). It predicts the future trajectory of boiler operating parameters in real time and generates multi-parameter coupled reference control variables, including a coal feed speed reference and a damper opening reference.

[0040] The fuzzy control unit uses an improved TS fuzzy model (Takagi-Sugeno fuzzy model) to convert the operator's experience rules into compensation control variables under nonlinear boundary conditions. The compensation weight is dynamically adjusted by monitoring the bed temperature overshoot and oxygen content mutation rate, and is then integrated with the baseline control variable to generate a fusion control instruction.

[0041] The neural network optimization control unit adopts a parallel fast learning network structure, dynamically adjusts the set values of the coal feed and air volume ratio through a reinforcement learning mechanism, and generates optimized control instructions.

[0042] The intelligent control module's specific implementation utilizes a multi-algorithm collaborative control architecture, integrating three control algorithms: model predictive control (MPC), fuzzy control, and neural network optimization. The MPC constructs a multivariable control matrix based on the ARX and FIR models. Model parameters are updated online using recursive least squares methods to provide real-time predictions of the future dynamic trajectories of the boiler bed temperature, pressure, and oxygen content. The prediction horizon is set at 15 steps, while the control horizon is 8 steps. A rolling horizon optimization algorithm is used to solve the constrained optimal control increments, generating multi-parameter coupled control variables for the coal feed speed reference and the damper opening reference, addressing the optimization control problem under variable coupling and constraint conditions.

[0043] The fuzzy control unit utilizes a TS fuzzy model, converting operator experience rules into compensation control variables under nonlinear boundary conditions. The input variables are bed temperature deviation and oxygen content change rate, and the output is the compensation control variable. A Gaussian distribution is used as the membership function, and the compensation weight is dynamically adjusted by monitoring the real-time bed temperature overshoot and oxygen content mutation rate. The weight function is designed to be positively correlated with the deviation amplitude. The compensation control variable is then nonlinearly fused and superimposed with the baseline control variable to generate a fused control command, enhancing the system's robustness under variable operating conditions.

[0044] The neural network optimization control unit utilizes a parallel, fast-learning network structure. The main network generates real-time optimization setpoints, while the auxiliary network provides historical experience as a reference. The network comprises four layers (an 8-node input layer, two hidden layers with 16 nodes each, and a 4-node output layer). The activation function uses a combination of Reluctant Unit (ReLU) and Sigmoid. The online optimization module dynamically adjusts the setpoints for the coal feed and air volume ratio using a reinforcement learning mechanism. The reward function comprehensively considers thermal efficiency, NOx emissions, and pressure fluctuations, and the exploration rate uses exponential decay to balance exploration and utilization. After safety verification of the optimized parameters, the optimized control instructions are generated to achieve global energy efficiency optimization.

[0045] The three-level command fusion mechanism achieves multi-command coordination through dynamic weight allocation. The baseline command weight decreases linearly with bed temperature deviation, the compensation command weight is positively correlated with the rate of change of oxygen content, and the optimization command weight is dynamically adjusted based on historical optimization results. The fusion formula uses a weighted summation algorithm to ensure smooth transition and real-time performance of control commands. The final output is denormalized and transmitted to the execution module, forming a closed-loop control system.

[0046] The intelligent control module outputs the final control instruction through a three-level instruction fusion mechanism;

[0047] The execution module includes the coal feeder, primary fan, secondary fan and valves. It receives the final control instructions from the intelligent control module to execute equipment actions, and feeds back the execution status to the data acquisition module. At the same time, it uploads the action parameters to the monitoring and early warning module.

[0048] The monitoring and early warning module processes the boiler operating parameters of the data acquisition module, the prediction data of the intelligent control module, and the action parameters of the execution module in parallel. It triggers the three-level early warning mechanism through pattern matching to generate early warning information. The early warning information is simultaneously pushed to the human-computer interaction module and triggers the protection logic of the control module.

[0049] The human-computer interaction module uses three-dimensional visualization technology to display the boiler's operating status, final control instruction parameters and early warning information in real time. It integrates a two-way control channel to support operators to manually correct the final control instruction parameters. The corrected instructions are directly transmitted to the intelligent control module after authority verification.

[0050] The execution module's specific implementation utilizes multi-device collaboration to physically execute the final control command. The coal feeder utilizes variable frequency control for speed regulation, the primary and secondary fans are equipped with blade adjustment mechanisms, and the valves utilize electric actuators for opening control. After receiving the final control command, the equipment transmits actual speed, damper opening, and vibration status to the data acquisition module via Profibus-DP, simultaneously uploading the action parameters to the monitoring and early warning module. The execution module incorporates redundant control circuits, automatically switching to a backup circuit in the event of a primary circuit failure to ensure operational reliability.

[0051] The monitoring and early warning module uses a stream processing architecture to analyze multi-source data in parallel, leveraging the Apache Flink engine to set up sliding windows for real-time pattern matching. The early warning rule base includes three levels of fault signatures. Red alerts match critical events such as sudden bed temperature changes and pressure limits, triggering audible and visual alarms that are pushed to the human-machine interface. Orange alerts correspond to abnormal operating conditions such as oxygen deviation and air volume imbalance, storing them in a pending queue and marking them in the operation log. Yellow alerts identify warning conditions such as parameter fluctuations and equipment preheating, alerting operators with a floating prompt. When an alert is triggered, a protection command is simultaneously sent to the intelligent control module, activating the equipment safety interlock.

[0052] The human-computer interaction module features a 3D visualization interface built on the Unity3D engine and integrates the HoloLens 2 for augmented reality interaction. The main interface renders a 3D boiler model in real time, with the temperature field intuitively displayed using gradient shading. Historical trend curves support multi-parameter timeline comparison, and the control panel provides sliding adjustments for key parameters. A bidirectional control channel integrates an authorization verification module, allowing operators to enter correction commands through gestures or voice input. After dual authentication using fingerprint and dynamic tokens, standardized commands are directly transmitted to the intelligent control module, and correction traces are automatically stored in an audit log for traceability.

[0053] See also Figure 1 As a preferred embodiment of the present application, the standardized data output by the data acquisition module contains triple fields of timestamp, sensor ID and original measurement value, and the intelligent control module automatically matches the corresponding multivariable control matrix column vector by parsing the sensor ID in the field.

[0054] This embodiment is implemented as follows: the standardized data frames output by the data acquisition module are transmitted via a redundant CANopen bus. Upon receipt, the intelligent control module first parses the data frame structure. The data frame contains three fields: a timestamp (UTC format, 1ms accuracy), a sensor ID (2-byte hexadecimal encoding), and a raw measurement value (32-bit floating point). The sensor ID uses a predefined encoding rule, for example, 0x01 for a bed temperature sensor, 0x02 for a pressure sensor, 0x03 for a flow meter, and 0x04 for an oxygen concentration sensor, ensuring a one-to-one correspondence with the column vectors of the control matrix.

[0055] The model predictive control unit maintains a multivariable control matrix, with the matrix column vectors arranged in order by sensor ID. Once a sensor ID is resolved, it is directly mapped to the matrix column index via a table lookup. For example, bed temperature data with ID 0x01 corresponds to the first column of the matrix, pressure data with ID 0x02 corresponds to the second column, and so on. The raw measurement values, after range conversion and normalization, are then populated into the corresponding row of the column vector at the current moment.

[0056] Timestamps are used for data alignment and synchronization compensation. The intelligent control module uses a sliding time window mechanism (window width 200ms) to update the matrix for data with the same timestamp. If timestamps are out of order (e.g., due to network latency), a linear interpolation algorithm is used to estimate the current value to ensure real-time control of the matrix.

[0057] For redundant sensor data (such as triple-redundant bed temperature sensors), the raw values are first processed through a median filter before being mapped to the control matrix. If there is a sensor ID conflict or the data is abnormal (outside the physical range), the data is marked as invalid, triggering a secondary alert in the monitoring and warning module and temporarily freezing updates to the corresponding matrix column until the fault is resolved.

[0058] After the control matrix is constructed, the model prediction control unit combines the ARX model with the FIR filter algorithm to perform multi-step predictions. The ARX model calculates the future trajectory based on the current matrix column vectors, and the FIR filter performs noise compensation on the prediction results. Ultimately, it generates a multi-parameter coupled control variable, including a coal feed speed reference and a damper opening reference, which is then passed to the command fusion mechanism.

[0059] See also Figure 1 As a preferred embodiment of the present application, the final control instruction generated by the intelligent control module is encapsulated in JSON format, including four-dimensional information of device address code, action type, target value and priority. The execution module triggers the control program of the corresponding device by parsing the device address code.

[0060] This embodiment is implemented as follows: the final control instruction is encapsulated in JSON format and contains a four-dimensional information field. The device address code is encoded in 4-digit hexadecimal, where the first digit identifies the device type (such as 0x1 for a coal feeder and 0x2 for a primary fan), and the last three digits are the device serial number. The action type field is defined as an enumeration type, including speed (speed regulation), opening (opening adjustment), position (position control), etc. The target value field sets the physical range according to the device type, such as the speed range of the coal feeder is 0-1200r / min, and the damper opening is 0% to 100%. The priority field is identified by an integer from 1 to 5, with 1 being the highest priority and 5 being the lowest, and the default value is 3.

[0061] During the generation process, the intelligent control module queries the physical address code from the device mapping table based on the fused control variable. The action type is automatically determined based on the control variable dimension, such as speed regulation for the coal feeder and damper opening adjustment. The target value is converted to physical units after filtering and limiting. Priority is dynamically assigned based on the control mode, and emergency conditions automatically elevate the command priority.

[0062] After receiving instructions via the TCP / IP protocol, the execution module first verifies the integrity of the JSON format and the CRC checksum. Device address code parsing uses a segmented matching algorithm, with the first four digits extracting the device type and the last three digits matching the specific device. The action type triggers the corresponding control program. Speed commands output PWM signals to the inverter via the PID controller, while opening commands drive the closed-loop position control of the electric actuator. The target value is smoothed using linear interpolation to avoid sudden step changes, and the execution cycle is kept within 50ms.

[0063] The priority handling mechanism uses preemptive scheduling. When a new instruction has a higher priority than the currently executing instruction, the system immediately interrupts and saves the current state, executing the higher-priority operation. After the instruction is executed, a status feedback packet containing the actual value, execution error, and timestamp is generated. This packet is transmitted back to the data acquisition module via a redundant bus and simultaneously uploaded to the monitoring module to record the operation log.

[0064] Regarding safety mechanisms, if a command timeout exceeds 200ms without a response, a retransmission is automatically triggered. After three cumulative failures, the system downgrades to the backup control channel. In the event of a device failure, the system automatically switches to a safe state (e.g., coal feeder shutdown, damper fully open) and issues a red alert to the human-machine interface, forming a complete closed-loop control process.

[0065] See also Figure 1 As a preferred embodiment of the present application, the execution status data fed back by the execution module is synchronously attached with a digital signature and a timestamp when it is transmitted via the redundant communication bus of the data acquisition module. The monitoring and early warning module calculates the transmission delay compensation amount by comparing the timestamp with the local clock difference.

[0066] This embodiment is implemented as follows: When the execution module generates a status data packet, it first generates a digital signature using the AES-CMAC algorithm. The device's unique identifier is selected as key-associated data, and a hash operation is performed on the status data payload (including the device address, action type, and actual value fields). This generates a 128-bit signature value and appends it to the end of the packet. The timestamp is generated using a hardware clock synchronized with the PTPv2 protocol, with an accuracy of 1μs and a 64-bit Unix timestamp (UTC) format. The redundant communication bus utilizes two CANopen channels for independent transmission. Data packets on each channel carry the same signature and timestamp. The receiving end verifies transmission integrity by comparing the data consistency between the two channels.

[0067] When the monitoring and warning module receives data, it first extracts the timestamp and calculates the difference between it and the local PTP clock (synchronization accuracy ±500ns) to obtain the transmission delay compensation Δt. This compensation algorithm uses a sliding average filter (window width 100 cycles) to dynamically adjust the time base of the prediction model. Digital signature verification uses a pre-stored key library to match the device identifier. If verification fails, a secondary warning is triggered and the packet is discarded. The timestamp Δt of the valid packet is used to correct for real-time deviations in the prediction model. For example, in bed temperature prediction, this can compensate for state estimation errors caused by network latency.

[0068] The exception handling mechanism includes: starting the local clock calibration process when the timestamp jump exceeds 1ms, and resynchronizing with the data acquisition module through the three-way handshake protocol; marking the device as an untrusted node if the signature verification fails for three consecutive cycles, and switching to the backup control channel; automatically enabling the buffer queue for timing reconstruction when the transmission delay compensation exceeds the threshold (such as Δt>50ms) to ensure the timing rationality of the control instructions.

[0069] This implementation method ensures data credibility through cryptographic signatures, uses high-precision clock synchronization to achieve nanosecond timestamp alignment, and combines a redundant channel verification mechanism to ensure the integrity, real-time and reliability of status feedback data during transmission, providing an accurate basis for dynamic compensation of the monitoring module.

[0070] See also Figure 1 As a preferred embodiment of the present application: the warning information output by the monitoring and warning module adopts a hierarchical queue transmission mechanism, the red warning directly triggers the sound and light alarm interface of the human-computer interaction module, the orange warning is pushed to the operation log processing queue, and the yellow warning is displayed as an interface floating prompt.

[0071] This embodiment is implemented as follows: First, the warning classification and generation logic is implemented. The real-time data stream is pattern matched by the Flink stream processing engine. The rule library predefines three levels of feature thresholds: Red warning: Emergency events such as sudden changes in bed temperature exceeding 50°C / s and pressure exceeding ±10% of the rated value are triggered immediately after matching;

[0072] Orange warning: abnormal operating conditions such as oxygen content continuously deviating from the set value by ±3%, air volume imbalance exceeding 15%, etc. are generated for three sampling cycles;

[0073] Yellow warning: Warnings such as parameter fluctuations exceeding the normal range but not reaching the threshold, equipment preheating status, etc. are generated in real time.

[0074] Secondly, the queue transmission mechanism, the red warning queue: uses a memory priority queue (maximum length 100). After the warning is generated, it is directly pushed to the human-computer interaction module through a TCP long connection, triggering the sound and light alarm protocol (including buzzer control instructions and red light flashing frequency encoding).

[0075] Orange warning queue: Uses Kafka message queue for temporary storage. Messages contain device ID, warning type, timestamp, and original data fragments. They are pushed to the operation log interface and marked as pending, allowing operators to view them by priority.

[0076] Yellow warning queue: The warning information is pushed to the front-end in real time via WebSocket. The warning information is encapsulated as a JSON object (including device identification, parameter values, and recommended operations) and displayed in a floating prompt box that automatically fades out after 5 seconds.

[0077] Among them, to ensure transmission reliability, the red alert adopts a three-way handshake confirmation mechanism. If no confirmation frame is received within 3 seconds, it switches to UDP broadcast mode to ensure that the alarm is not missed; the orange alert message enables persistent storage (retained for 72 hours), and the consumer end confirms the processing status through the ACK mechanism; the yellow alert integrates heartbeat detection, and automatically caches it to the local IndexedDB when the network is interrupted, and resends it after recovery.

[0078] Then comes the linkage control logic. When a red alert is triggered, a protection instruction (such as emergency coal stop, full-open air damper) is synchronously sent to the intelligent control module, and the instruction priority is forcibly set to level 1. After the orange alert is activated, the monitoring module automatically retrieves the historical similar case library, generates a list of recommended operations and pushes it to the human-machine interface. If a yellow alert is triggered three times on the same device, it will automatically upgrade to an orange alert and generate an equipment maintenance work order.

[0079] Finally, the interface interaction design is involved. The human-machine interface integrates an audio-visual alarm hardware interface (supporting RS485 / Modbus protocols). When a red alert occurs, red light flows around the screen borders, and a buzzer emits a rapid pulse tone. The operation log interface adopts a B / S architecture, with orange alert entries displayed with a red border and pinned to the top, supporting batch processing operations (confirm / ignore / transfer work orders).

[0080] The floating prompt is rendered using HTML5 Canvas, supports dragging and dropping, and expands the detailed parameter trend chart after clicking.

[0081] Ultimately, differentiated processing of warning information is achieved through hierarchical queues, and transmission reliability is ensured by combining the stream processing engine and message middleware. The linkage control logic and interface interaction design form a complete closed loop to ensure the timeliness of warning response and convenient operation.

[0082] See also Figure 1 As a preferred embodiment of the present application: the correction instruction includes an operator identity authentication token and a parameter change trajectory. After the intelligent control module verifies the validity of the token, it stores the parameter change trajectory in the control strategy version library.

[0083] The specific implementation is as follows: Correction instructions are transmitted via a bidirectional control channel using the HTTPS protocol and encapsulated in JSON format. These instructions contain three fields: the operator authentication token (JWT format), a parameter change log (an array of key-value pairs), and a digital signature. The authentication token, generated by the rights management system, contains the operator ID, role level (e.g., administrator / operator), and expiration date, and is signed using RSA asymmetric encryption. The digital signature is generated using the HMAC-SHA256 algorithm, using the token key as the secret key, to ensure tamper-proof instructions.

[0084] After receiving a command, the intelligent control module first performs a security check: it parses the JWT token to verify the signature validity and expiration date, and queries a pre-stored token blacklist. The digital signature is then compared with the locally calculated value. If the check fails, the command is discarded and an audit log is triggered. After passing the check, the parameter change trajectory is parsed. The trajectory field contains parameter IDs, such as the coal feed PID parameter, the original value, the new value, and the change timestamp. This supports batch modifications, such as adjusting multiple control loop parameters simultaneously. The control strategy version library uses the Git-LFS distributed storage architecture, with the master library deployed on redundant servers and slave libraries synchronized to edge nodes in real time. When a parameter changes, the system automatically generates a new version branch. The version number follows semantic specifications, and the change record is appended to the version log, including the operator ID, the hash value of the change content, and the timestamp. If a parameter conflict is detected, such as a new value that exceeds physical constraints, the conflict resolution mechanism is triggered, prompting the operator to choose to overwrite, roll back, or cancel the operation.

[0085] The storage process consists of four steps: 1. Write-Ahead Logging (WAL) records the intended change; 2. Execute parameter updates to the in-memory control matrix; 3. Persistently store the changes in the master repository; 4. Synchronize with the slave repository and commit the transaction. Version rollback supports rapid recovery of specified historical versions, using a diff algorithm to calculate parameter differences and perform batch updates.

[0086] The monitoring module monitors repository change events in real time and generates audit reports containing the operator, changed parameters, scope of impact, and effective date. It supports multi-dimensional searches by time, operator, and parameter type. Regarding security mechanisms, modifications to key parameters require secondary authentication, such as dynamic verification codes, and token-based permissions control the depth of modification. For example, operators can only adjust setpoint offsets, while administrators can modify the control algorithm structure.

[0087] See also Figure 1 As a preferred embodiment of the present application: the prediction data output by the intelligent control module is exchanged in real time with the monitoring and early warning module using a shared memory method to avoid repeated transmission of data on a redundant communication bus.

[0088] The specific implementation is as follows: a multi-segment shared memory pool architecture is employed. The main memory area is divided into a ring buffer (capacity 1024 data blocks). Each block contains a 64-bit timestamp, a 16-bit prediction parameter header (identifying bed temperature, pressure, oxygen level, etc.), a floating-point prediction array (adjustable dimension, default 128 points), and a checksum (CRC16). An additional 128-byte control information segment is allocated to store read / write pointer positions, a version number (32-bit integer, supporting semantic versioning), and synchronization semaphores. After the model prediction control unit generates prediction data, it writes it directly to the ring buffer via a memory-mapped file (mmap). A mutex (pthread_mutex) is acquired before writing to ensure atomic operations. Data is stored in little-endian order, and the prediction array is compressed using SIMD instruction set optimizations to reduce memory usage. The timestamp is calibrated using the TSC counter and hardware clock, with an accuracy of 0.1μs. After the write is complete, the write pointer and version number in the control information segment are updated, the mutex is released, and a condition variable (pthread_cond_signal) is triggered to notify the monitoring module. The monitoring module polls shared memory via an independent thread, using the difference between two pointers (read / write) to determine data validity. When the version number parity changes, the full data is copied to the local cache. The read process utilizes lock-free programming (CAS atomic operations), directly mapping memory to user space and avoiding kernel-mode switching overhead. Data parsing utilizes SIMD instruction parallel processing to improve throughput. For synchronization and fault tolerance, initial synchronization uses the PTPv2 protocol to calibrate both clocks, and the offset is stored in the control information segment. During runtime, dynamic time warping (DTW) is used to compensate for minor clock drift. Checksum errors automatically roll back to the previous version of the data. Three consecutive errors trigger a bus redundancy mechanism to ensure critical data is not lost. Performance optimization measures include write-merging to reduce cache line invalidations and pre-allocating a memory pool in the ring buffer to avoid dynamic allocation overhead. The monitoring module implements zero-copy technology, directly accessing GPU memory through memory registration to accelerate visualization rendering.

[0089] See also Figure 1 As a preferred embodiment of the present application: the raw data of the boiler operating parameters transmitted by the data acquisition module are simultaneously injected into the training data buffer of the intelligent control module. When the amount of buffer data reaches a preset threshold, the offline retraining process of the neural network optimization control is automatically triggered.

[0090] Specifically, the intelligent control module automatically matches the column vector of the multivariable control matrix by parsing the standardized data (including triple fields of timestamp, sensor ID, and original measurement value) output by the data acquisition module, and uses a three-level instruction fusion mechanism to generate the final control instruction including dynamic weight benchmark instruction, compensation instruction and optimization instruction. The instruction encapsulates the four-dimensional information of device address code, action type, target value and priority in JSON format, and is transmitted to the execution module via TCP / IP protocol to trigger the action execution of coal feeder, primary fan, secondary fan and valve. The execution status is attached with digital signature and timestamp when it is transmitted back through redundant communication bus. The monitoring and early warning module uses stream processing architecture to analyze multi-source data in parallel, and triggers the three-level early warning mechanism through pattern matching. The red warning directly triggers the sound and light alarm interface of the human-computer interaction module and synchronously transmits the protection instruction to the control module. The orange warning is pushed to the operation log pending queue, and the yellow warning is displayed as a floating prompt. The human-computer interaction module builds a three-dimensional visualization interface based on the Unity3D engine and integrates HoloLens 2. Augmented reality interaction is achieved, allowing operators to input authorized correction instructions through a bidirectional control channel. The corrected trajectory is stored in the control strategy version library, triggering the offline retraining process of the neural network optimization control. The training data buffer uses a double-buffered queue architecture to inject raw data. When the data volume reaches a threshold, the asynchronous training process is automatically initiated. After the new model passes verification, it is hot-switched to the control system. The standardized data output by the data acquisition module is simultaneously injected into the training data buffer of the intelligent control module. The monitoring module calculates the transmission delay compensation by comparing the timestamp with the local clock difference. The execution status data feedback from the execution module is synchronously attached with a digital signature and timestamp when transmitted via the redundant communication bus. This forms a fully closed-loop control architecture that includes data acquisition, intelligent control, equipment execution, monitoring and early warning, and human-machine interaction. Each module uses shared memory to achieve real-time exchange of predicted trend data. The AES-CMAC algorithm and digital signatures are used to ensure data transmission security. The PTPv2 protocol achieves nanosecond clock synchronization to ensure the real-time and reliability of control instructions. The overall system adopts an edge computing and cloud collaborative architecture, supports dynamic configuration of final control instruction parameters and model structure, and realizes efficient and stable operation and intelligent operation and maintenance management of industrial boilers.

[0091] A method for an intelligent temperature control system for a circulating fluidized bed boiler, the method comprising the following steps:

[0092] S1. The data acquisition module acquires boiler operating parameters in real time and generates standardized data;

[0093] S2. The intelligent control module receives standardized data and generates final control instructions through multi-algorithm fusion calculations;

[0094] S3. The execution module adjusts the device action according to the final control instruction and feedbacks the execution status data;

[0095] S4. The monitoring and warning module triggers a three-level warning mechanism through pattern matching to generate warning information and simultaneously pushes the warning information to the human-computer interaction module;

[0096] S5. The human-computer interaction module supports manual modification of the final control instruction parameters. The modified instructions are passed to the intelligent control module after authority verification.

[0097] Furthermore, the multi-algorithm fusion calculation in step S2 specifically includes three control algorithms: model predictive control, fuzzy control, and neural network optimization, and includes a dynamic credibility allocation mechanism. The output credibility of each control algorithm is evaluated in real time through the entropy weight method, wherein the model predictive control weight coefficient is negatively correlated with the rate of change of the operating condition;

[0098] The execution status data in step S3 includes a digital signature and a timestamp, and the monitoring and warning module calculates the transmission delay compensation amount by comparing the timestamp with the local clock difference;

[0099] The warning information in step S4 adopts a hierarchical queue transmission mechanism. The red warning directly triggers the sound and light alarm interface of the human-computer interaction module, the orange warning is pushed to the operation log processing queue, and the yellow warning is displayed as an interface floating prompt.

[0100] Specifically, the data acquisition module acquires boiler operating parameters (including bed temperature, pressure, oxygen content, coal feed rate, etc.) in real time through a redundant sensor network. After limiting and filtering the raw data and normalizing it, it generates standardized data containing three fields: timestamp, sensor ID, and measurement value.

[0101] After receiving standardized data, the intelligent control module uses a multi-algorithm fusion computing framework (integrating model predictive control, fuzzy PID control, and an expert rule base) to evaluate the credibility of each algorithm's output in real time through a dynamic credibility allocation mechanism. The entropy weight method dynamically adjusts the weight coefficient based on historical prediction errors, operating condition fluctuations, and other characteristics. The model predictive control weight is negatively correlated with the operating condition change rate to ensure a balance between steady-state accuracy and dynamic response.

[0102] After the execution module parses the final control instruction (equipment address code, action type, target value, and priority encapsulated in JSON format), it drives the coal feeder inverter, damper actuator, and other regulating devices to operate, and feeds back execution status data with an AES-CMAC digital signature and timestamp.

[0103] The monitoring and warning module uses the Flink stream processing engine to analyze multi-source data streams in real time and triggers a three-level warning mechanism through pattern matching. Red warnings (such as sudden changes in bed temperature exceeding 50°C / s) directly trigger the audio and visual alarm interface of the human-computer interaction module and simultaneously send protection instructions to the intelligent control module. Orange warnings (such as oxygen levels continuously deviating from the set value) are pushed to the operation log queue for processing. Yellow warnings (such as parameter fluctuations not reaching the threshold) are displayed as floating prompts.

[0104] The human-computer interaction module builds a three-dimensional visualization interface based on the Unity3D engine, supporting operators to interact with augmented reality through HoloLens 2. Manual correction instructions must be authenticated by JWT tokens and verified by digital signatures before being passed to the intelligent control module to trigger the Git-LFS distributed update of the control strategy version library. At the same time, when executing status data feedback, the monitoring module compares the timestamp with the local clock synchronized with PTPv2, and uses a sliding average filtering algorithm to dynamically calculate the transmission delay compensation amount to ensure the real-time performance of the prediction model and control instructions.

[0105] In summary, the present invention realizes precise control of boiler operating parameters through multi-algorithm collaborative control and intelligent compensation mechanism. The model prediction control unit is based on the multivariable control matrix constructed by the ARX / FIR model, which can predict the dynamic trajectory of the boiler bed temperature, pressure and oxygen content parameters in real time, and convert the multivariable coupling problem that is difficult to handle with traditional PID control into an optimization problem, thereby improving the control accuracy. The neural network optimization control unit dynamically adjusts the ratio of coal feed and air volume through the reinforcement learning mechanism, thereby improving the thermal efficiency of the boiler. The fuzzy control unit converts the operator's experience rules into compensation control quantities under nonlinear boundary conditions, effectively suppressing bed temperature overshoot and oxygen content mutation, and maintaining the stability of the combustion process under variable operating conditions.

[0106] The above is a schematic description of the present invention and its embodiments, which is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. Therefore, if a person skilled in the art is inspired by this and, without departing from the purpose of the present invention, designs a structure and embodiment similar to this technical solution without inventiveness, they shall fall within the scope of protection of the present invention.

[0107] In addition, it should be understood that although this specification is described in terms of implementation methods, not every implementation method contains only one independent technical solution. This narrative method of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment can also be appropriately combined to form other implementation methods that can be understood by those skilled in the art.

Claims

1. An intelligent temperature control system for a circulating fluidized bed boiler, comprising a data acquisition module, an intelligent control module, an execution module, a monitoring and early warning module, and a human-computer interaction module, characterized in that: The data acquisition module is used to collect boiler operating parameters in real time and perform standardized processing, and synchronously transmit the standardized data to the intelligent control module through the redundant communication bus; The intelligent control module consists of a model prediction control unit, a fuzzy control unit, and a neural network optimization control unit. It is used to receive standardized data transmitted by the data acquisition module in real time, integrate the three control algorithms of model prediction control, fuzzy control, and neural network optimization to perform dynamic calculations to generate the final control instructions, and transmit them synchronously to the execution module. At the same time, it generates prediction data and transmits it to the monitoring and early warning module. The execution module includes the coal feeder, primary fan, secondary fan and valves. It receives the final control instructions from the intelligent control module to execute equipment actions, and feeds back the execution status data to the data acquisition module. At the same time, it uploads the action parameters to the monitoring and early warning module. The monitoring and early warning module processes the boiler operating parameters of the data acquisition module, the prediction data of the intelligent control module, and the action parameters of the execution module in parallel. It triggers the three-level early warning mechanism through pattern matching to generate early warning information. The early warning information is simultaneously pushed to the human-computer interaction module and triggers the protection logic of the intelligent control module. The human-computer interaction module uses 3D visualization technology to display the boiler's operating status, final control instruction parameters, and early warning information in real time. It integrates a two-way control channel to support operators to manually modify the final control instruction parameters. The modified instructions are directly transmitted to the intelligent control module after authority verification. The boiler operating parameters include boiler bed temperature, pressure and oxygen content parameters; The standardized data contains a triple field of timestamp, sensor ID and raw measurement value. The intelligent control module automatically matches the corresponding multivariable control matrix column vector by parsing the sensor ID in the field; The model prediction control unit constructs a multivariable control matrix based on the ARX model and the FIR model, predicts the future trajectory of the boiler operating parameters in real time, and generates a multi-parameter coupled reference control quantity including a coal feed speed reference and a damper opening reference; The fuzzy control unit adopts the TS fuzzy model to convert the operator's experience rules into compensation control quantities under nonlinear boundary conditions. The compensation weight is dynamically adjusted by monitoring the bed temperature overshoot and oxygen content mutation rate, and is fused with the multi-parameter coupling benchmark control quantity to generate a fusion control instruction. The neural network optimization control unit adopts a parallel fast learning network structure, dynamically adjusts the set value of the coal feed rate and air volume ratio through the reinforcement learning mechanism, and generates optimized control instructions; The fusion control instructions and the optimization control instructions are dynamically calculated to generate the final control instructions.

2. The intelligent temperature control system for a circulating fluidized bed boiler according to claim 1, characterized in that: The final control instruction generated by the intelligent control module is encapsulated in JSON format, including four-dimensional information of device address code, action type, target value and priority. The execution module triggers the control program of the corresponding device by parsing the device address code.

3. The intelligent temperature control system for a circulating fluidized bed boiler according to claim 1, characterized in that: When the execution status data fed back by the execution module is transmitted via the redundant communication bus of the data acquisition module, a digital signature and a timestamp are synchronously added, and the monitoring and early warning module calculates the transmission delay compensation amount by comparing the timestamp with the local clock difference.

4. The intelligent temperature control system for a circulating fluidized bed boiler according to claim 1, characterized in that: The three-level early warning mechanism includes: The warning information output by the monitoring and warning module is transmitted using a hierarchical queue. The red warning directly triggers the sound and light alarm interface of the human-computer interaction module, the orange warning is pushed to the operation log processing queue, and the yellow warning is displayed as a floating prompt on the interface.

5. The intelligent temperature control system for a circulating fluidized bed boiler according to claim 1, characterized in that: The correction instruction includes an operator identity authentication token and a parameter change trajectory. After the intelligent control module verifies the validity of the token, it stores the parameter change trajectory in the control strategy version library.

6. The intelligent temperature control system for a circulating fluidized bed boiler according to claim 1, characterized in that: The prediction data generated by the intelligent control module includes real-time predictions of the future trajectory of boiler bed temperature, pressure and oxygen content parameters, and the prediction data is exchanged in real time with the monitoring and early warning module using a shared memory method to avoid repeated transmission of data on redundant communication buses.

7. The intelligent temperature control system for a circulating fluidized bed boiler according to claim 1, characterized in that: The neural network optimization control unit is also provided with a training data buffer for receiving the original data of the boiler operating parameters transmitted by the data acquisition module. When the data volume in the training data buffer reaches a preset threshold, the offline retraining process of the neural network optimization control is automatically triggered.

8. The intelligent temperature control method applied to the intelligent temperature control system of a circulating fluidized bed boiler according to any one of claims 1 to 7, characterized in that: The method comprises the following steps: S1. The data acquisition module acquires boiler operating parameters in real time and generates standardized data; S2. The intelligent control module receives standardized data and generates final control instructions through multi-algorithm fusion calculations; S3. The execution module adjusts the device action according to the final control instruction and feedbacks the execution status data; S4. The monitoring and warning module triggers a three-level warning mechanism through pattern matching to generate warning information and simultaneously pushes the warning information to the human-computer interaction module; S5. The human-computer interaction module supports manual correction of the final control command parameters. The correction command is passed to the intelligent control module after authorization verification; The multi-algorithm fusion calculation in step S2 specifically includes three control algorithms: model predictive control, fuzzy control, and neural network optimization, and includes a dynamic credibility allocation mechanism. The output credibility of each control algorithm is evaluated in real time through the entropy weight method, where the model predictive control weight coefficient is negatively correlated with the rate of change of the operating condition; The execution status data in step S3 includes a digital signature and a timestamp, and the monitoring and warning module calculates the transmission delay compensation amount by comparing the timestamp with the local clock difference; The warning information in step S4 adopts a hierarchical queue transmission mechanism. The red warning directly triggers the sound and light alarm interface of the human-computer interaction module, the orange warning is pushed to the operation log processing queue, and the yellow warning is displayed as an interface floating prompt.

Images