Cooperative control system for civil engineering concrete construction equipment of thermal power plant

Abstract

The invention relates to the technical field of control and regulation, in particular to a cooperative control system for civil engineering concrete construction equipment of a thermal power plant, which is characterized in that concrete slump compensation data of a mixing plant, a hydraulic pressure analysis value of a delivery pump and space trajectory calibration data of a vibrating rod are acquired in real time through a dynamic self-calibration data acquisition module; the hybrid optimization collaborative analysis module constructs an equipment communication time delay and control coupling degree weighted adjacency matrix based on a graph neural network, generates a pump and station collaborative parameter Pareto solution set by adopting a non-dominated sorting genetic algorithm, and screens matching parameter groups of the rotating speed of a conveying pump and the discharging rate of a mixing station to form a collaborative optimization instruction. A closed-loop control chain for data acquisition, decision optimization, fault-tolerant execution and feedback updating is established, and the problem of flow state instability caused by dynamic response hysteresis between the concrete delivery pump and the mixing plant is solved.

Description

Technical Field

[0001] The present invention relates to the field of control and regulation technology, and in particular to a collaborative control system for civil construction concrete equipment in a thermal power plant. Background Art

[0002] In the civil construction of thermal power plants, the construction of concrete structures relies on a suite of specialized equipment working in concert to handle the high-load demands of critical components such as large-scale foundations, load-bearing walls, and towers. Concrete mixing plants use automated metering systems to mix aggregate, cement, and water to create a homogenous mixture. Truck-mounted or stationary pumping equipment delivers the mixture to the pouring site, overcoming long distances or high-altitude obstacles. Vibrators are inserted into the slurry to remove air bubbles, improving structural density and overall durability. Formwork systems provide precise geometric constraints to ensure component dimensional accuracy. This equipment suite integrates advanced technologies to reduce human error and enhance construction controllability. It is particularly well-suited for the high-volume concrete operations typical of thermal power plants, such as large boiler foundations and cooling tower construction, enabling efficient, safe, and design-compliant infrastructure construction.

[0003] In the civil concrete construction of thermal power plants, the multi-device coordinated control system suffers from real-time synchronization hysteresis problems, which is mainly attributed to the lack of an efficient information exchange mechanism between the controllers of different equipment, and the failure to achieve continuous adaptive adjustment of the operation process. Logical reasoning points out that the status changes of independent operating units such as mixing stations, pumping devices and vibrating equipment are difficult to be captured and responded to immediately by the central monitoring system, which may cause the concrete flow to lose control or structural defects. For example, when the pumping rate demand suddenly increases, the mixing station discharge control fails to match the change in time, resulting in material supply interruption or blockage of the conveying pipeline, which not only affects construction efficiency but also increases the risk of material waste. Summary of the Invention

[0004] In view of the shortcomings of the existing technology, the present invention provides a collaborative control system for civil engineering concrete construction equipment in a thermal power plant to solve the problem of flow instability caused by dynamic response hysteresis between the concrete delivery pump and the mixing station.

[0005] In order to solve the above technical problems, the specific technical solutions of the present invention are as follows: The collaborative control system for civil construction concrete equipment in a thermal power plant provided by the present invention comprises: Dynamic self-calibration data acquisition module to obtain concrete flow data of the mixing station, hydraulic pressure data of the delivery pump and motion trajectory data of the vibrator; A hybrid optimization collaborative analysis module, which inputs the data acquired by the dynamic self-calibration data acquisition module and analyzes the device interaction relationship to generate device collaborative optimization instructions; A reinforcement learning execution strategy module inputs the collaborative optimization instructions generated by the hybrid optimization collaborative analysis module and compiles them into executable instructions; A fault-tolerant closed-loop control module inputs the executable instructions compiled by the reinforcement learning execution strategy module, outputs them to the target device for execution, and captures execution exception data; Feeding back the execution exception data to the hybrid optimization collaborative analysis module to trigger iterative updates of device interaction relationships; The collaborative optimization instructions include matching adjustment parameters between the conveying pump speed and the mixing station discharge rate.

[0006] Furthermore, in the collaborative control system for civil construction concrete equipment in a thermal power plant according to the present invention, the dynamic self-calibration data acquisition module includes: A multi-spectral fusion unit inputs the concrete phase change signal and aggregate distribution data at the mixing station outlet, and outputs the slump compensation data to the mixing optimization collaborative analysis module as the concrete flow state data of the mixing station; A dual-mode strain sensing unit inputs a temperature-compensated fiber Bragg grating array signal and outputs the analyzed arm strain value to the hybrid optimization collaborative analysis module as the hydraulic pressure data of the delivery pump; An inertial navigation calibration unit inputs 9-axis inertial measurement unit and global positioning system data, and outputs vibrator calibration trajectory data to the hybrid optimization collaborative analysis module as the vibrator motion trajectory data; The multi-spectral fusion unit, dual-modal strain sensing unit and inertial navigation calibration unit also input the execution abnormality data fed back by the fault-tolerant closed-loop control module to trigger the sensor calibration operation.

[0007] Furthermore, in the collaborative control system for civil construction concrete equipment in a thermal power plant according to the present invention, the hybrid optimization collaborative analysis module includes: A graph neural network topology modeling unit inputs the flow state data, hydraulic pressure data, and motion trajectory data transmitted by the dynamic self-calibration data acquisition module to construct a weighted adjacency matrix of communication delay and control coupling degree of the mixing station, delivery pump, and vibrator; When the communication delay of the delivery pump node in the weighted adjacency matrix of the control coupling degree exceeds a threshold, outputting a mesh relay switching instruction to the fault-tolerant closed-loop control module; The execution exception data fed back by the fault-tolerant closed-loop control module is input to update the weight of the control coupling degree weighted adjacency matrix.

[0008] Furthermore, in the collaborative control system for civil construction concrete equipment in a thermal power plant according to the present invention, the hybrid optimization collaborative analysis module further comprises: A multi-objective evolutionary algorithm optimization unit inputs the weighted adjacency matrix generated by the graph neural network topology modeling unit and generates a Pareto solution set of pump and station coordination parameters through a non-dominated sorting genetic algorithm; Filter the Pareto solution of the pump and station collaborative parameters to concentrate the matching parameter group of the conveying pump speed and the mixing station discharge rate, and output it as an equipment collaborative optimization instruction to the reinforcement learning execution strategy module; The execution abnormality data fed back by the fault-tolerant closed-loop control module is input to update the flow state instability risk entropy weight coefficient of the non-dominated sorting genetic algorithm.

[0009] Furthermore, in the collaborative control system for civil construction concrete equipment in a thermal power plant according to the present invention, the multi-objective evolutionary algorithm optimization unit is further used to: Input the concrete flow state data of the dynamic self-calibration data acquisition module, calculate the deviation rate according to the slump measured value and the preset design calibration value, and generate the flow state risk entropy value; The weighted adjacency matrix output by the graph neural network topology modeling unit is used to generate the Pareto solution set of pump-station coordination parameters through non-dominated sorting. An ε-constraint method screening operation is performed on the Pareto solution set of the pump-station coordination parameters, wherein the constraints include: A communication delay threshold preset at the pump control node; The tolerance range of flow risk entropy value is preset based on the concrete mix ratio; The filtered conveying pump speed and mixing station discharge rate matching parameter group is output to the reinforcement learning execution strategy module to form equipment collaborative optimization instructions.

[0010] Furthermore, in the collaborative control system for civil construction concrete equipment in a thermal power plant according to the present invention, the reinforcement learning execution strategy module includes: A collaborative parameter execution unit inputs a parameter group matching the conveying pump speed and the mixing station discharge rate in the equipment collaborative optimization instruction; a deep deterministic policy gradient decision unit, which generates a screw conveyor speed increment and a conveying pump motor frequency offset based on the matching parameter group; A fuzzy proportional-integral-differential controller inputs the flow state risk entropy value generated by the hybrid optimization collaborative analysis module, performs a differential operation on the flow state risk entropy value to obtain a rate of change, and dynamically adjusts the proportional band width according to the amplitude of the rate of change; The actuator unit converts the speed increment and frequency offset output by the deep deterministic policy gradient decision unit into physical control signals, and outputs them to the fault-tolerant closed-loop control module.

[0011] Furthermore, in the collaborative control system for civil construction concrete equipment in a thermal power plant according to the present invention, the reinforcement learning execution strategy module further includes: An intermediate language compiling unit inputs the screw conveyor speed increment instruction and the conveying pump motor frequency offset instruction output by the deep deterministic policy gradient decision unit; Decomposing the speed increment instruction and the frequency offset instruction into operator nodes and operand leaf nodes through an abstract syntax tree parser, and constructing an intermediate representation in a static single assignment form; A logic compiler framework, which loads a preset programmable logic controller instruction mapping rule library, converts the operator nodes in the intermediate representation into corresponding function block instructions, and generates programmable logic controller-specific code; When semantic checking detects an undefined operator or type conflict, a compilation error code including an error location identifier is generated; If there is a compilation error code, trigger an instruction regeneration request to the deep deterministic policy gradient decision unit; If the compilation is successful, the programmable logic controller-specific code encapsulated as a structured data packet is output to the fault-tolerant closed-loop control module.

[0012] Furthermore, in the collaborative control system for civil construction concrete construction equipment in a thermal power plant according to the present invention, the fault-tolerant closed-loop control module includes: a time-sensitive network transmission unit, which parses the transmission priority tag in the header of the programmable logic controller-specific code packet output by the reinforcement learning execution strategy module and allocates preemptive transmission time slots according to the priority level; The dedicated code after the time slot is allocated is encapsulated into a time-sensitive network frame structure and output to the target device actuator via industrial Ethernet; Real-time monitoring of the 5G base station uplink packet loss rate. When the monitored value exceeds a preset proportional coefficient of the pump node communication delay threshold defined by the graph neural network topology modeling unit, the LoRa radio frequency module is activated to establish a mesh relay communication link, and Reed-Solomon forward error correction coding is loaded at the data link layer. The structured exception handling unit captures the duration of packet loss events and the peak packet loss rate, and generates an exception report including temporal and spatial feature data; The abnormality report is fed back to the graph neural network topology modeling unit to trigger the calculation of the communication reliability attenuation factor and the update of the weighted adjacency matrix.

[0013] Furthermore, in the collaborative control system for civil construction concrete construction equipment in a thermal power plant according to the present invention, the fault-tolerant closed-loop control module further comprises: The vibration and template coupling unit inputs the six-degree-of-freedom trajectory data of the vibrating rod output by the inertial navigation calibration unit and the flow state risk entropy value generated by the hybrid optimization collaborative analysis module; Initialize the particle swarm position vector to the current spatial coordinate of the vibrator, and the velocity vector to the gradient of the flow risk entropy value; By iteratively solving the Newton and Euler motion differential equations, the optimal insertion angle solution set that satisfies the minimum axial resistance and the highest concrete density is obtained; Outputs the spatial optimal solution including the pitch angle, azimuth angle and insertion depth to the multi-axis motion controller of the vibrator actuator; When the fiber Bragg grating sensor arranged at the template bearing point detects that the local strain value exceeds 70% of the material yield strength, the strain gradient direction and amplitude characteristics of the over-limit area are extracted, and the inertia weight parameter of the algorithm is dynamically adjusted according to the cosine similarity between the gradient direction and the particle swarm convergence direction.

[0014] Furthermore, in the collaborative control system for civil construction and concrete construction equipment in a thermal power plant according to the present invention, the logic compiler framework in the reinforcement learning execution strategy module is configured as follows: Capturing a standardized abnormal execution code through a diagnostic interface of a target device, the code including an error type identifier and fault location information; Triggering the third-level rollback operation includes retrieving the previous dedicated code package marked as verified in the instruction cache queue, verifying the code integrity signature, and loading it into the target device executor to overwrite the current abnormal state; Call the preset error type and deviation mapping rule library to interpret communication timeout errors as negative device coupling deviation values ​​and control misalignment errors as positive device coupling deviation values; Generate a structured correction parameter package including deviation direction and amplitude; Inputting the structured correction parameter package into the graph neural network topology modeling unit; The control coupling strength factor of the weighted adjacency matrix is ​​updated according to the deviation direction. Negative deviation reduces the weight of the associated edge, while positive deviation increases the weight. Normalization is performed after the update.

[0015] Beneficial effects of the present invention: The present invention realizes the fusion collection of multi-source heterogeneous data of concrete flow state, equipment load and spatial trajectory through a dynamic self-calibration data acquisition module, eliminating the time lag error of traditional manual detection; the equipment interaction model constructed based on the graph neural network quantifies the relationship between communication delay and control coupling degree, and combines the Pareto solution set generated by the non-dominated sorting genetic algorithm to realize the optimal matching decision between the conveying pump speed and the mixing station discharge rate, avoiding the flow state out of control caused by asynchronous equipment action; the reinforcement learning execution strategy module compiles the optimization instructions into the underlying equipment executable code in real time, and ensures the deterministic delay execution of the control instructions through the time-sensitive network transmission mechanism; the fault-tolerant closed-loop control module captures execution exception data and triggers a three-level rollback operation, synchronously generates a deviation parameter package and feeds it back to the graph neural network topology modeling unit, dynamically updates the control coupling strength factor of the weighted adjacency matrix, and forms a closed-loop adaptive control chain of data acquisition, optimization decision-making, instruction execution and feedback update, continuously optimizes the dynamic response synchronization between equipment, and effectively solves the problem of concrete flow instability. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solution of the present invention, the following is a brief introduction to the drawings required for use in the embodiments. Obviously, for ordinary technicians in this field, other drawings can be obtained based on the drawings without paying any creative labor.

[0017] Figure 1 This is a system architecture diagram of a collaborative control system for civil engineering concrete construction equipment in a thermal power plant provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solutions and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with the specific embodiments of the present invention and the corresponding drawings. Obviously, the embodiments described 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 work are within the scope of protection of the present invention. The technical solutions provided by each embodiment of the present invention are described in detail below in conjunction with the drawings. In order to better understand the purpose of the present invention, the present invention is further described in detail below.

[0019] See also Figure 1 The present invention provides a collaborative control system for civil construction concrete equipment in a thermal power plant, comprising: Dynamic self-calibration data acquisition module to obtain concrete flow data of the mixing station, hydraulic pressure data of the delivery pump and motion trajectory data of the vibrator; A hybrid optimization collaborative analysis module, which inputs the data acquired by the dynamic self-calibration data acquisition module and analyzes the device interaction relationship to generate device collaborative optimization instructions; A reinforcement learning execution strategy module inputs the collaborative optimization instructions generated by the hybrid optimization collaborative analysis module and compiles them into executable instructions; A fault-tolerant closed-loop control module inputs the executable instructions compiled by the reinforcement learning execution strategy module, outputs them to the target device for execution, and captures execution exception data; Feeding back the execution exception data to the hybrid optimization collaborative analysis module to trigger iterative updates of device interaction relationships; The collaborative optimization instructions include matching adjustment parameters between the conveying pump speed and the mixing station discharge rate.

[0020] The dynamic self-calibration data acquisition module uses a multispectral fusion unit to analyze the phase change signals and aggregate distribution data of the concrete at the mixing station outlet, generating slump compensation values ​​as concrete flow data. The dual-modal strain sensing unit processes temperature-compensated fiber Bragg grating array signals and outputs boom strain resolution values ​​as hydraulic pressure data for the delivery pump. The inertial navigation calibration unit integrates data from a 9-axis inertial measurement unit and a global positioning system to provide spatial trajectory calibration data for the vibrator. These units collect key parameters in real time, providing basic input for subsequent modules and triggering sensor calibration operations through a feedback mechanism to maintain data accuracy.

[0021] The hybrid optimization collaborative analysis module receives the output data from the dynamic self-calibration data acquisition module. The graph neural network topology modeling unit constructs a weighted adjacency matrix of communication delay and control coupling between the mixing station, conveying pump, and vibrator. Based on this matrix, the multi-objective evolutionary algorithm optimization unit uses a non-dominated sorting genetic algorithm to generate a Pareto solution set of the conveying pump and mixing station collaborative parameters. Using the ε constraint method, it selects a matching parameter set for the conveying pump speed and the mixing station discharge rate to form equipment collaborative optimization instructions. This process considers constraints such as flow risk entropy and communication delay thresholds to achieve quantitative analysis of equipment interactions. Feedback data is then used to update the weighted adjacency matrix weights and the flow instability risk entropy weight coefficients to optimize the decision logic.

[0022] The reinforcement learning execution strategy module inputs the collaborative optimization instructions generated by the hybrid optimization collaborative analysis module into the collaborative parameter execution unit. The deep deterministic policy gradient decision unit then generates the screw conveyor speed increment and the conveying pump motor frequency offset. A fuzzy proportional-integral-derivative controller dynamically adjusts the proportional band width based on the rate of change of the flow risk entropy value. The actuator unit converts the speed increment and frequency offset into physical control signals. The intermediate language compilation unit uses an abstract syntax tree parser to decompose instructions into a static single-assignment intermediate representation. The logic compiler framework loads the programmable logic controller instruction mapping rule library and converts it into specialized code. It also triggers instruction regeneration if semantic verification fails, ensuring that the compiled output is an executable structured data package.

[0023] The fault-tolerant closed-loop control module receives executable instructions compiled by the reinforcement learning execution strategy module. The time-sensitive network transmission unit parses the packet header priority tag, allocates preemptive transmission time slots, and encapsulates them into time-sensitive network frames, which are then output to the target device actuator via industrial Ethernet. The 5G uplink packet loss rate is monitored in real time. When the rate exceeds a preset proportionality factor, the LoRa radio frequency module is activated to establish a mesh relay link and load forward error correction code. The structured exception handling unit captures the spatiotemporal characteristics of packet loss events and generates an exception report, which is fed back to the graph neural network topology modeling unit to trigger the calculation of the communication reliability attenuation factor and the update of the weighted adjacency matrix. The vibration and template coupling unit combines the vibration rod trajectory data and the flow risk entropy value to initialize the particle swarm position vector and velocity vector. It then iteratively solves the optimal insertion angle solution set through the differential equation of motion and outputs it to the multi-axis motion controller. When the template strain exceeds the limit, the inertia weight parameter is adjusted to maintain convergence.

[0024] Execution exception data is captured by the fault-tolerant closed-loop control module and fed back to the hybrid optimization collaborative analysis module, triggering iterative updates of device interaction relationships. The logic compiler framework captures standardized exception codes through the diagnostic interface, performs a three-level rollback operation to restore the previously verified code, and calls the error type and deviation mapping rule library to generate a correction parameter package input graph neural network topology modeling unit. The weighted adjacency matrix control coupling strength factor is updated according to the deviation direction and normalized, forming an adaptive closed-loop control chain to solve the problem of dynamic response hysteresis between devices.

[0025] Specifically, in the collaborative control system for civil construction concrete construction equipment in a thermal power plant according to the present invention, the dynamic self-calibration data acquisition module includes: A multi-spectral fusion unit inputs the concrete phase change signal and aggregate distribution data at the mixing station outlet, and outputs the slump compensation data to the mixing optimization collaborative analysis module as the concrete flow state data of the mixing station; A dual-mode strain sensing unit inputs a temperature-compensated fiber Bragg grating array signal and outputs the analyzed arm strain value to the hybrid optimization collaborative analysis module as the hydraulic pressure data of the delivery pump; An inertial navigation calibration unit inputs 9-axis inertial measurement unit and global positioning system data, and outputs vibrator calibration trajectory data to the hybrid optimization collaborative analysis module as the vibrator motion trajectory data; The multi-spectral fusion unit, dual-modal strain sensing unit and inertial navigation calibration unit also input the execution abnormality data fed back by the fault-tolerant closed-loop control module to trigger the sensor calibration operation.

[0026] The dynamic self-calibration data acquisition module uses a multispectral fusion unit to collect visible and near-infrared spectral data of concrete at the mixing station outlet. Combined with the aggregate particle size distribution characteristics, it establishes a mapping model between phase change signals and aggregate spatial position. This unit uses a convolutional neural network to extract the slurry fluidity feature vector from the multispectral image, compares it with the preset slump design value, and generates a dynamic compensation value. This slump compensation data, representing the concrete flow state, is output to the mixed optimization collaborative analysis module. This process enables non-contact flow state monitoring, providing real-time material characteristic parameters for subsequent collaborative control.

[0027] A dual-modal strain sensing unit is deployed at key bearing points on the pump boom to collect wavelength offsets from the fiber Bragg grating array reflections. This unit applies a temperature compensation coefficient matrix to perform modal decomposition on the original signal, separating the strain component caused by hydraulic pressure from the ambient temperature drift component. The boom strain is obtained by solving the elastic constitutive equation and transmitted to the hybrid optimization collaborative analysis module as an equivalent representation of the pump hydraulic pressure data. This approach eliminates the impact of temperature interference on pressure monitoring and improves the accuracy of hydraulic system status perception.

[0028] The inertial navigation calibration unit integrates a 9-axis inertial measurement unit (IMU) and a GPS receiver to collect the vibrator's three-axis acceleration, angular velocity, and geodetic coordinate data in real time. Using a Kalman filter algorithm, the unit fuses the IMU's high-frequency update characteristics with the GPS's absolute positioning capabilities to compensate for accumulated errors and output six-degree-of-freedom spatial trajectory calibration data, which is then fed into the hybrid optimization collaborative analysis module as the vibrator's trajectory data. This technology addresses the trajectory distortion problem inherent in existing single-position positioning methods in high-vibration environments.

[0029] When the fault-tolerant closed-loop control module reports abnormal data, the multispectral fusion unit triggers the visible spectrum sensor's white balance calibration process based on the flow deviation identifier in the abnormality code. The dual-modal strain sensing unit recalibrates the fiber Bragg grating array's reference wavelength parameters based on the strain threshold exceeded in the abnormality data packet. The inertial navigation calibration unit resets the Kalman filter covariance matrix based on the positioning drift recorded in the abnormality report. This feedback mechanism forms a closed-loop sensor self-calibration mechanism, maintaining the long-term measurement reliability of the data acquisition unit.

[0030] Specifically, in the collaborative control system for civil construction concrete equipment in a thermal power plant according to the present invention, the hybrid optimization collaborative analysis module includes: A graph neural network topology modeling unit inputs the flow state data, hydraulic pressure data, and motion trajectory data transmitted by the dynamic self-calibration data acquisition module to construct a weighted adjacency matrix of communication delay and control coupling degree of the mixing station, delivery pump, and vibrator; When the communication delay of the delivery pump node in the weighted adjacency matrix of the control coupling degree exceeds a threshold, outputting a mesh relay switching instruction to the fault-tolerant closed-loop control module; The execution exception data fed back by the fault-tolerant closed-loop control module is input to update the weight of the control coupling degree weighted adjacency matrix.

[0031] The graph neural network topology modeling unit receives the concrete slump compensation data, boom strain analysis value and vibrator spatial trajectory calibration data transmitted by the dynamic self-calibration data acquisition module, and constructs an equipment topology network with the mixing station control cabinet, conveying pump controller and vibrator operation as nodes; this unit maps each equipment node to a vertex in the graph neural network, and its attribute vector contains the real-time status parameters of the corresponding equipment; directed edge connections are established between vertices, and the edge weight contains a binary pair of communication delay and control coupling degree, where the communication delay value is obtained by calculating the timestamp difference, and the control coupling degree is quantitatively generated based on the command response frequency and action correlation between devices, forming a control coupling degree weighted adjacency matrix to represent the dynamic interaction relationship of the system.

[0032] When the communication delay value corresponding to the delivery pump node in the weighted adjacency matrix exceeds the preset threshold, the graph neural network topology modeling unit determines that the current communication link is insufficiently reliable; the unit generates a mesh relay switching instruction containing a sequence of backup relay nodes by analyzing the path redundancy between nodes, and outputs it to the fault-tolerant closed-loop control module to activate the multi-path data transmission mechanism to avoid control inaccuracy caused by a single link failure.

[0033] After the execution anomaly data fed back by the fault-tolerant closed-loop control module is fed into the graph neural network topology modeling unit, this unit analyzes the device interaction failure event type in the anomaly report. If the anomaly is a communication timeout, the control coupling component in the associated edge weight is reduced; if the anomaly is a control misalignment, the control coupling component in the associated edge weight is increased. The updated weighted adjacency matrix redistributes the coordination priorities between devices, achieving adaptive iterative optimization of the network topology.

[0034] Specifically, in the collaborative control system for civil construction concrete equipment in a thermal power plant according to the present invention, the hybrid optimization collaborative analysis module further includes: A multi-objective evolutionary algorithm optimization unit inputs the weighted adjacency matrix generated by the graph neural network topology modeling unit and generates a Pareto solution set of pump and station coordination parameters through a non-dominated sorting genetic algorithm; Filter the Pareto solution of the pump and station collaborative parameters to concentrate the matching parameter group of the conveying pump speed and the mixing station discharge rate, and output it as an equipment collaborative optimization instruction to the reinforcement learning execution strategy module; The execution abnormality data fed back by the fault-tolerant closed-loop control module is input to update the flow state instability risk entropy weight coefficient of the non-dominated sorting genetic algorithm.

[0035] The multi-objective evolutionary algorithm optimization unit receives the control coupling degree weighted adjacency matrix generated by the graph neural network topology modeling unit and uses it as the population initialization parameter of the non-dominated sorting genetic algorithm; the unit uses the conveying pump speed and the mixing station discharge rate as decision variables, and takes the minimization of flow instability risk entropy and the optimization of equipment collaborative energy consumption as parallel objective functions, and generates a non-dominated solution set through crossover and mutation operations; the solution set is stratified using the fast non-dominated sorting mechanism, and the Pareto solution set of the collaborative parameters of the conveying pump and mixing station is screened out in combination with the congestion calculation to characterize the optimal collaborative strategy combination under different working conditions.

[0036] An ε-constrained screening operation is performed on the generated Pareto solution set, setting two constraints: the first constraint is the communication delay threshold of the pump nodes defined by the graph neural network topology modeling unit, eliminating solutions with excessive delays; the second constraint is the flow risk entropy tolerance interval preset based on the concrete mix parameters, eliminating solutions with excessive entropy values. A gradient optimization search is performed on the solution set that meets the constraints, selecting the parameter set with the highest match between the pump speed and the mixing plant discharge rate. This generates equipment collaborative optimization instructions and transmits them to the reinforcement learning execution strategy module.

[0037] When the fault-tolerant closed-loop control module reports abnormal execution data, the multi-objective evolutionary algorithm optimization unit analyzes the flow instability event characteristics in the abnormal report. If the flow state mutation is caused by conveying pressure fluctuation, the weight coefficient of the flow instability risk entropy in the objective function is increased; if the material interruption is caused by discharge rate mismatch, the weight coefficient of the equipment collaborative energy consumption is reduced. This dynamic weight adjustment mechanism enables the non-dominated sorting genetic algorithm to adapt to actual operating conditions and optimize the engineering applicability of the parameter solution set.

[0038] Specifically, in the collaborative control system for civil construction concrete equipment in a thermal power plant according to the present invention, the multi-objective evolutionary algorithm optimization unit is further used to: Input the concrete flow state data of the dynamic self-calibration data acquisition module, calculate the deviation rate according to the slump measured value and the preset design calibration value, and generate the flow state risk entropy value; The weighted adjacency matrix output by the graph neural network topology modeling unit is used to generate the Pareto solution set of pump-station coordination parameters through non-dominated sorting. An ε-constraint method screening operation is performed on the Pareto solution set of the pump-station coordination parameters, wherein the constraints include: A communication delay threshold preset at the pump control node; The tolerance range of flow risk entropy value is preset based on the concrete mix ratio; The filtered conveying pump speed and mixing station discharge rate matching parameter group is output to the reinforcement learning execution strategy module to form equipment collaborative optimization instructions.

[0039] The multi-objective evolutionary algorithm optimization unit receives the control coupling degree weighted adjacency matrix output by the graph neural network topology modeling unit and maps its edge weights into initial population distribution parameters. Using the conveying pump speed and the mixing station discharge rate as decision variables, this unit establishes a dual-objective function space for minimizing the risk entropy of flow instability and optimizing the coordinated energy consumption of equipment. Parent individuals are selected through a tournament selection mechanism, and the offspring population is generated using simulated binary crossover and polynomial mutation operators. A fast non-dominated sorting hierarchical strategy is used to divide the Pareto hierarchy. The solution set distribution is maintained by combining crowding distance calculations. Finally, the Pareto frontier solution set for the pump and station coordination parameters is output, covering optimization solutions under different weight configurations.

[0040] During the Pareto solution set screening phase, the unit applies pre-set constraints and performs ε filtering. Unreliable network solutions are eliminated based on the communication delay threshold of the pump control node, and solutions with insufficient stability are eliminated based on the flow risk entropy tolerance interval set by the concrete mix parameters. A gradient sensitivity analysis is performed on the remaining solution set, calculating the eigenvalues ​​of the covariance matrix between the pump speed and the mixing station discharge rate. The matching parameter set whose eigenvector direction is closest to the optimal collaborative efficiency curve is selected, generating equipment collaborative optimization instructions that are transmitted to the reinforcement learning execution strategy module.

[0041] When the fault-tolerant closed-loop control module reports execution anomaly data, this unit analyzes the event classification identifiers contained in the anomaly code. For flow regime mutations caused by conveying pressure fluctuations, the unit increases the weight of the flow instability risk entropy in the objective function; for control inaccuracies caused by material supply interruptions, the unit reduces the weight of equipment collaborative energy consumption. After the weights are updated, the non-dominated sorting grading criteria are recalculated to adapt the algorithm's evolutionary direction to actual operating conditions.

[0042] Specifically, in the collaborative control system for civil construction concrete equipment in a thermal power plant according to the present invention, the reinforcement learning execution strategy module includes: A collaborative parameter execution unit inputs a parameter group matching the conveying pump speed and the mixing station discharge rate in the equipment collaborative optimization instruction; a deep deterministic policy gradient decision unit, which generates a screw conveyor speed increment and a conveying pump motor frequency offset based on the matching parameter group; A fuzzy proportional-integral-differential controller inputs the flow state risk entropy value generated by the hybrid optimization collaborative analysis module, performs a differential operation on the flow state risk entropy value to obtain a rate of change, and dynamically adjusts the proportional band width according to the amplitude of the rate of change; The actuator unit converts the speed increment and frequency offset output by the deep deterministic policy gradient decision unit into physical control signals, and outputs them to the fault-tolerant closed-loop control module.

[0043] The multi-objective evolutionary algorithm optimization unit receives slump compensation data from the dynamic self-calibration data acquisition module and calculates the flow pattern volatility based on the deviation between the measured concrete slump value and the preset design calibration value. This unit quantifies flow pattern uncertainty using the Shannon entropy model and maps the volatility into a flow pattern risk entropy value, representing the stability risk level of the concrete delivery process. This metric, as one of the core optimization objectives, provides a quantitative basis for subsequent parameter screening.

[0044] This unit loads the control coupling weighted adjacency matrix generated by the graph neural network topology modeling unit and interprets its edge weights as equipment coordination constraints. In the decision space, the conveying pump speed and the mixing station discharge rate are used as coordinate axes, and a non-dominated sorting genetic algorithm is used to initialize the population: the offspring solution vector is generated by simulating binary crossover operations, and the population diversity is maintained by combining polynomial mutation operators. The fast non-dominated sorting mechanism is applied to divide the Pareto hierarchy layers, and the congestion distance is calculated to retain the distribution breadth of the frontier solution set. The Pareto solution set of pump station coordination parameters covering different weight configurations is output.

[0045] An ε-constrained screening operation is performed on the Pareto solution set, setting dual hard constraint boundaries: the first boundary is the communication delay threshold of the pump control node, derived from the delay weight component of the weighted adjacency matrix, which is used to eliminate solutions with unreliable network transmission; the second boundary is based on the flow risk entropy tolerance interval preset by the concrete mix parameters, filtering out solutions that exceed the material tolerance range. A gradient sensitivity analysis is performed on the solution set that meets the constraints, calculating the eigenvalues ​​of the covariance matrix of the pump speed and discharge rate parameters. The matching parameter group with the eigenvector direction closest to the optimal equipment collaborative efficiency curve is selected, and the equipment collaborative optimization instructions are generated and transmitted to the reinforcement learning execution strategy module. This screening mechanism maximizes collaborative efficiency while ensuring system stability.

[0046] When the fault-tolerant closed-loop control module receives feedback on execution anomaly data, it dynamically adjusts the optimization strategy based on the event type identified by the anomaly code: for sudden flow changes, the weight coefficient of the flow risk entropy in the objective function is increased; for equipment response hysteresis events, the weight coefficient of the control coupling is strengthened. After the weight update, the genetic algorithm population distribution is reinitialized to adapt the evolution direction of the Pareto solution set to the actual operating conditions, forming a closed-loop optimization decision-making mechanism.

[0047] Specifically, in the collaborative control system for civil construction concrete equipment in a thermal power plant according to the present invention, the reinforcement learning execution strategy module further includes: An intermediate language compiling unit inputs the screw conveyor speed increment instruction and the conveying pump motor frequency offset instruction output by the deep deterministic policy gradient decision unit; Decomposing the speed increment instruction and the frequency offset instruction into operator nodes and operand leaf nodes through an abstract syntax tree parser, and constructing an intermediate representation in a static single assignment form; A logic compiler framework, which loads a preset programmable logic controller instruction mapping rule library, converts the operator nodes in the intermediate representation into corresponding function block instructions, and generates programmable logic controller-specific code; When semantic checking detects an undefined operator or type conflict, a compilation error code including an error location identifier is generated; If there is a compilation error code, trigger an instruction regeneration request to the deep deterministic policy gradient decision unit; If the compilation is successful, the programmable logic controller-specific code encapsulated as a structured data packet is output to the fault-tolerant closed-loop control module.

[0048] The intermediate language compilation unit receives the screw conveyor speed increment instruction and the conveyor pump motor frequency offset instruction output by the deep deterministic policy gradient decision unit as control parameter inputs. This unit performs lexical analysis on the instructions using an abstract syntax tree parser, decomposing the speed increment instruction into arithmetic operator nodes and speed variable operand leaf nodes, and the frequency offset instruction into logical operator nodes and frequency variable operand leaf nodes. It then constructs an abstract syntax tree hierarchy based on operator precedence and converts it into a static single-assignment intermediate representation, eliminating ambiguity in duplicate variable assignments and providing standardized intermediate layer data for subsequent code generation.

[0049] The logic compiler framework loads a preset programmable logic controller instruction mapping rule library, which defines the mapping relationship between operator nodes and programmable logic controller function block instructions; maps arithmetic operator nodes in the intermediate representation to arithmetic operation function blocks, logical operator nodes to comparison function blocks, and operand leaf nodes to register addresses; and generates programmable logic controller-specific code that meets preset standards through the instruction synthesis engine, realizing the conversion of high-level control instructions into executable code for the underlying device.

[0050] The semantic checker performs type checking and context verification on the generated PLC-specific code. When undefined operator mappings are detected, a compile error code is generated, indicating the incorrect operator position. When an operand type conflicts with a function block input parameter type, an operand type conflict error code is generated. If a compile error code is found, an instruction regeneration request is sent back to the deep deterministic policy gradient decision unit, requesting the regeneration of speed increment and frequency offset instructions that conform to the semantic rules.

[0051] Once the PLC-specific code passes semantic verification, the logic compiler framework encapsulates the code into a structured data packet, adding a transmission priority tag and a checksum field to the packet header. This packet is then output to the time-sensitive network transmission unit of the fault-tolerant closed-loop control module, forming the final carrier of the device's executable instructions. The feedback signal of successful compilation simultaneously triggers a policy gradient update in the reinforcement learning execution strategy module, optimizing the subsequent control instruction generation strategy.

[0052] Specifically, the collaborative control system for civil construction concrete construction equipment in a thermal power plant according to the present invention, wherein the fault-tolerant closed-loop control module includes: a time-sensitive network transmission unit, which parses the transmission priority tag in the header of the programmable logic controller-specific code packet output by the reinforcement learning execution strategy module and allocates preemptive transmission time slots according to the priority level; The dedicated code after the time slot is allocated is encapsulated into a time-sensitive network frame structure and output to the target device actuator via industrial Ethernet; Real-time monitoring of the 5G base station uplink packet loss rate. When the monitored value exceeds a preset proportional coefficient of the pump node communication delay threshold defined by the graph neural network topology modeling unit, the LoRa radio frequency module is activated to establish a mesh relay communication link, and Reed-Solomon forward error correction coding is loaded at the data link layer. The structured exception handling unit captures the duration of packet loss events and the peak packet loss rate, and generates an exception report including temporal and spatial feature data; The abnormality report is fed back to the graph neural network topology modeling unit to trigger the calculation of the communication reliability attenuation factor and the update of the weighted adjacency matrix.

[0053] The Time-Sensitive Network (TSN) transmission unit parses the header fields of the PLC-specific code packets output by the reinforcement learning execution strategy module to extract the embedded transmission priority tag. Based on a priority level mapping table defined by a pre-defined standard, it allocates preemptive transmission slots to high-priority instructions and periodic polling slots to regular instructions. The resulting slot allocations are then encapsulated into the scheduling control field of the TSN frame, adding a timestamp and sequence number to form a complete frame structure. The frame is then sent to the target device actuator via the Industrial Ethernet physical layer, achieving deterministic transmission latency control.

[0054] This unit continuously monitors the service quality indicators of the 5G base station uplink. When the real-time packet loss rate exceeds the preset proportional coefficient of the communication delay threshold of the delivery pump node defined by the graph neural network topology modeling unit, it determines that the reliability of the main communication link is insufficient; activates the LoRa radio frequency module to switch to the mesh relay mode, builds a self-organizing network topology at the MAC layer, and loads the Reed-Solomon forward error correction coding block at the data link layer. It improves the anti-burst interference capability through redundant check bits and maintains the complete transmission of control instructions.

[0055] The structured exception handling unit, deployed in the deep packet inspection layer of the network protocol stack, captures the duration of packet loss events, the peak packet loss rate, and the spatial coordinates of the event in real time. It then flows the raw exception data through a feature extraction engine, outputting a spatiotemporal feature dataset containing a timestamp sequence, geographic tags, and network topology paths, generating a standardized exception report. This report is transmitted via a feedback bus to the graph neural network topology modeling unit, triggering the recalculation of the communication reliability attenuation factor.

[0056] After anomaly reports are fed into the graph neural network topology modeling unit, it analyzes link failure patterns in the spatiotemporal feature dataset. It calculates the communication reliability attenuation factor based on the packet loss duration and updates the delay weight components of the corresponding edges in the weighted adjacency matrix based on the network topology path. This updated matrix redefines the coordination priorities between device nodes, enabling online self-optimization of the control network topology and forming a closed-loop mechanism for handling transmission anomalies.

[0057] Specifically, in the collaborative control system for civil construction concrete construction equipment in a thermal power plant according to the present invention, the fault-tolerant closed-loop control module further includes: The vibration and template coupling unit inputs the six-degree-of-freedom trajectory data of the vibrating rod output by the inertial navigation calibration unit and the flow state risk entropy value generated by the hybrid optimization collaborative analysis module; Initialize the particle swarm position vector to the current spatial coordinate of the vibrator, and the velocity vector to the gradient of the flow risk entropy value; By iteratively solving the Newton and Euler motion differential equations, the optimal insertion angle solution set that satisfies the minimum axial resistance and the highest concrete density is obtained; Outputs the spatial optimal solution including the pitch angle, azimuth angle and insertion depth to the multi-axis motion controller of the vibrator actuator; When the fiber Bragg grating sensor arranged at the template bearing point detects that the local strain value exceeds 70% of the material yield strength, the strain gradient direction and amplitude characteristics of the over-limit area are extracted, and the inertia weight parameter of the algorithm is dynamically adjusted according to the cosine similarity between the gradient direction and the particle swarm convergence direction.

[0058] The vibrator-to-formwork coupling unit receives the vibrator's six-degree-of-freedom spatial coordinate data from the inertial navigation calibration unit and simultaneously acquires the real-time flow risk entropy value calculated by the hybrid optimization collaborative analysis module. This unit initializes the particle swarm algorithm's position vector to the vibrator's current three-dimensional geodetic coordinates, maps the gradient of the flow risk entropy value to the directional component of the velocity vector, and establishes a vibration path optimization model. This process converts concrete state parameters into vibrator motion constraints, achieving coupled modeling of material properties and mechanical action.

[0059] A dual objective function is defined within the optimization model: the axial resistance of the vibrator is calculated based on Newton's laws of motion, minimizing the resistance component; and the concrete density function is derived based on Euler's dynamic equations, maximizing the density component. The saddle point solution space of the objective function is iteratively solved using implicit differential equations, outputting the optimal insertion angle solution set that satisfies the dual objective constraints. This solution set includes parameter combinations for pitch angle, azimuth angle, and insertion depth. This solution set is transmitted to the multi-axis motion controller of the vibrator actuator, driving the hydraulic servo system for precise insertion trajectory control.

[0060] When the fiber Bragg grating sensors at the template's bearing points detect local strain exceeding the material's yield strength threshold, the strain feature extraction engine analyzes the strain gradient direction vector and amplitude characteristics in the exceeding region. This unit calculates the cosine similarity between the gradient direction and the particle swarm's current convergence direction. When the similarity falls below a preset threshold, the algorithm's inertia weight parameter is increased to enhance global search capabilities; when the similarity approaches consistency, the inertia weight parameter is reduced to improve local convergence accuracy. This mechanism dynamically balances path exploration with convergence stability, avoiding the risk of template structure overload.

[0061] During the vibration process, the flow risk entropy value is updated in real time and input into the particle swarm velocity vector calculation module, forming a closed-loop control chain for concrete flow, vibration action, and formwork strain. The iterative results of the particle swarm position vector are synchronously fed back to the inertial navigation calibration unit to correct for accumulated trajectory measurement errors and maintain spatial positioning accuracy in coordination with the formwork protection mechanism.

[0062] Specifically, in the collaborative control system for civil construction and concrete construction equipment in a thermal power plant according to the present invention, the logic compiler framework in the reinforcement learning execution strategy module is configured as follows: Capturing a standardized abnormal execution code through a diagnostic interface of a target device, the code including an error type identifier and fault location information; Triggering the third-level rollback operation includes retrieving the previous dedicated code package marked as verified in the instruction cache queue, verifying the code integrity signature, and loading it into the target device executor to overwrite the current abnormal state; Call the preset error type and deviation mapping rule library to interpret communication timeout errors as negative device coupling deviation values ​​and control misalignment errors as positive device coupling deviation values; Generate a structured correction parameter package including deviation direction and amplitude; Inputting the structured correction parameter package into the graph neural network topology modeling unit; The control coupling strength factor of the weighted adjacency matrix is ​​updated according to the deviation direction. Negative deviation reduces the weight of the associated edge, while positive deviation increases the weight. Normalization is performed after the update.

[0063] The logic compiler framework collects abnormal code execution streams in real time through the target device's standard diagnostic interface (such as the OPC UA protocol). This stream includes error type identifiers and fault location register addresses defined by pre-defined standards. The capture mechanism uses an event-driven model. When the device controller triggers an exception interrupt, it automatically extracts the error classification code and physical location coordinates from the exception code to form a structured exception data packet, providing input for subsequent processing.

[0064] A three-level rollback sequence is triggered: the first level retrieves the most recently timestamped, validated historical dedicated code package from the instruction cache queue; the second level verifies the integrity signature of the code package using a cryptographic hash function to verify that the data has not been tampered with; the third level loads the verified code package into the target device actuator via the device firmware flash interface, overwriting the current abnormal control state. This mechanism enables rapid and secure recovery of control instructions, reducing system downtime.

[0065] A pre-defined error type and deviation mapping rule library is invoked to perform semantic parsing. When the exception code identifier matches the communication timeout error class, it is interpreted as a negative device coupling deviation value, indicating that internode coordination strength needs to be weakened. When it matches the control misalignment error class, it is interpreted as a positive device coupling deviation value, indicating that internode coordination strength needs to be strengthened. The deviation amplitude is calculated based on the failure frequency recorded in the fault location register, forming a structured correction parameter package containing the deviation direction identifier and quantized amplitude.

[0066] After the structured correction parameter package is input into the graph neural network topology modeling unit, the unit updates the weighted adjacency matrix based on the deviation direction: negative deviations reduce the control coupling strength factor weight of the associated edges, while positive deviations increase the weight. After the weight update, matrix normalization is performed to maintain a constant sum of all edge weights, preserving the stability of the network topology. The updated adjacency matrix redefines the interaction strength between device nodes, enabling online self-correction of the control logic.

[0067] The corrected weighted adjacency matrix is ​​simultaneously input into the multi-objective evolutionary algorithm optimization unit, triggering the recalculation of the Pareto solutions of the collaborative parameters. The entropy weighting coefficient for the flow instability risk is dynamically adjusted based on the direction of the deviation: negative deviations reduce the entropy weighting, while positive deviations increase it. This closed-loop update mechanism enables the equipment coordination strategy to adapt to changing field conditions in real time, forming a two-way linkage between exception handling and system optimization.

[0068] In large-scale foundation concrete construction scenarios at thermal power plants, the existing equipment coordination system fails to dynamically match the mixing station discharge rate with the delivery pump speed, resulting in concrete flow instability. During implementation of the present invention, the dynamic self-calibration data acquisition module uses a multi-spectral fusion unit to capture the aggregate distribution characteristics at the mixing station outlet in real time, combines near-infrared spectroscopy to analyze slurry fluidity, and outputs a slump compensation value; the dual-modal strain sensing unit decouples the temperature drift component of the delivery pump boom based on the fiber grating array signal and extracts the pure hydraulic pressure strain value; the inertial navigation calibration unit fuses the 9-axis inertial measurement unit with satellite positioning data to eliminate trajectory drift in strong vibration environments and output the six-degree-of-freedom calibration coordinates of the vibrator. This data acquisition mechanism provides a high-precision working condition perception foundation for collaborative control.

[0069] The hybrid optimization collaborative analysis module constructs a three-layer closed-loop decision chain: the graph neural network topology modeling unit maps the mixing station controller, conveying pump PLC, and vibrator operating terminal as vertices, generates control coupling weights based on the command response frequency between devices, and generates communication delay weights based on the network round-trip delay, forming a weighted adjacency matrix. When the conveying pump node delay exceeds the limit, the mesh relay switching instruction is triggered. The multi-objective evolutionary algorithm optimization unit uses a non-dominated sorting genetic algorithm to generate the Pareto solution set of pump station collaborative parameters with the goals of minimizing flow risk entropy and optimizing energy consumption. The ε constraint method is used to load the communication delay threshold and the flow risk entropy tolerance interval for dual filtering to select the parameter group that matches the conveying pump speed and the mixing station discharge rate. This decision-making process achieves precise collaborative planning of equipment actions.

[0070] The reinforcement learning execution strategy module converts optimization instructions into executable code: a deep deterministic policy gradient decision unit parses and matches parameter sets, outputting the screw conveyor speed increment and pump frequency offset. An abstract syntax tree parser decomposes instructions into operator nodes and operand leaf nodes, and a logic compiler framework maps these instructions into function block instructions based on a pre-defined quasi-map, generating a PLC-specific code package. The fault-tolerant closed-loop control module monitors the 5G uplink packet loss rate in real time by allocating transmission slots based on priority via a time-sensitive network transmission unit. When the packet loss rate exceeds a preset proportionality factor, it activates the LoRa mesh relay link and loads forward error correction code. Execution of abnormal data triggers a three-level rollback: retrieving historical verification code packages, verifying integrity signatures, and overwriting abnormal states. It also generates a deviation parameter package to update the weighted adjacency matrix weight coefficients. The vibration and template coupling unit initializes the particle swarm position vector to the real-time coordinates of the vibrator rod, associates the velocity vector with the flow state entropy gradient, and solves for the optimal insertion angle solution to drive the multi-axis controller. When the template strain exceeds a limit, the particle swarm inertia weight is dynamically adjusted based on the strain gradient direction. This closed-loop execution mechanism enables continuous self-optimization of dynamic responses between devices.

[0071] This invention addresses dynamic response hysteresis by constructing a closed-loop control chain. A dynamic self-calibration data acquisition module acquires real-time concrete slump compensation data, pump hydraulic pressure analysis, and vibrator spatial trajectory calibration data, providing comprehensive awareness of material flow, equipment load, and spatial positioning. This module utilizes a multispectral fusion unit to analyze aggregate distribution characteristics, a dual-modal strain sensing unit to decouple temperature interference, and an inertial navigation calibration unit to compensate for positioning drift. This module provides a high-precision, real-time data source for collaborative control, eliminating the time lag inherent in existing manual detection.

[0072] The hybrid optimization collaborative analysis module uses a graph neural network to construct a weighted adjacency matrix of equipment communication delay and control coupling, quantifying the interactions between the mixing station, delivery pump, and vibrator. A non-dominated sorting genetic algorithm is used to generate a Pareto solution of pump-station coordination parameters, and an ε-constraint method is used to select matching parameter sets for delivery pump speed and mixing station discharge rate. This process simultaneously integrates the dual constraints of flow risk entropy tolerance and communication delay threshold to achieve precise matching of equipment actions while ensuring system stability, thus avoiding flow instability caused by decision lag.

[0073] The reinforcement learning execution strategy module compiles coordination parameters into executable instructions and generates speed increments and frequency offsets through deep deterministic policy gradient decision-making. The fault-tolerant closed-loop control module uses a time-sensitive network transmission mechanism to ensure real-time execution of instructions and captures execution anomaly data, feeding it back to the graph neural network topology modeling unit. This feedback data triggers iterative updates of the weighted adjacency matrix weights, dynamically adjusting the device control coupling strength factors to form a closed-loop adaptive control chain, continuously optimizing the dynamic response synchronization between devices.

Claims

1. A collaborative control system for civil construction concrete equipment in a thermal power plant, characterized by: include: Dynamic self-calibration data acquisition module to obtain concrete flow data of the mixing station, hydraulic pressure data of the delivery pump and motion trajectory data of the vibrator; A hybrid optimization collaborative analysis module, which inputs the data acquired by the dynamic self-calibration data acquisition module and analyzes the device interaction relationship to generate device collaborative optimization instructions; A reinforcement learning execution strategy module inputs the collaborative optimization instructions generated by the hybrid optimization collaborative analysis module and compiles them into executable instructions; A fault-tolerant closed-loop control module inputs the executable instructions compiled by the reinforcement learning execution strategy module, outputs them to the target device for execution, and captures execution exception data; Feeding back the execution exception data to the hybrid optimization collaborative analysis module to trigger iterative updates of device interaction relationships; The collaborative optimization instructions include matching adjustment parameters between the conveying pump speed and the mixing station discharge rate.

2. The collaborative control system for civil construction concrete equipment in a thermal power plant according to claim 1, characterized in that: The dynamic self-calibration data acquisition module includes: A multi-spectral fusion unit inputs the concrete phase change signal and aggregate distribution data at the mixing station outlet, and outputs the slump compensation data to the mixing optimization collaborative analysis module as the concrete flow state data of the mixing station; A dual-mode strain sensing unit inputs a temperature-compensated fiber Bragg grating array signal and outputs the analyzed arm strain value to the hybrid optimization collaborative analysis module as the hydraulic pressure data of the delivery pump; An inertial navigation calibration unit, which inputs 9-axis inertial measurement unit and global positioning system data, and outputs vibrator calibration trajectory data to the hybrid optimization collaborative analysis module as the vibrator motion trajectory data; The multi-spectral fusion unit, dual-modal strain sensing unit and inertial navigation calibration unit also input the execution abnormality data fed back by the fault-tolerant closed-loop control module to trigger the sensor calibration operation.

3. The collaborative control system for civil construction concrete equipment in a thermal power plant according to claim 2, characterized in that: The hybrid optimization collaborative analysis module includes: A graph neural network topology modeling unit inputs the flow state data, hydraulic pressure data and motion trajectory data transmitted by the dynamic self-calibration data acquisition module to construct a weighted adjacency matrix of communication delay and control coupling degree of the mixing station, delivery pump and vibrator; When the communication delay of the delivery pump node in the weighted adjacency matrix of the control coupling degree exceeds a threshold, outputting a mesh relay switching instruction to the fault-tolerant closed-loop control module; The execution exception data fed back by the fault-tolerant closed-loop control module is input to update the weight of the control coupling degree weighted adjacency matrix.

4. The collaborative control system for civil construction concrete equipment in a thermal power plant according to claim 3, characterized in that: The hybrid optimization collaborative analysis module also includes: A multi-objective evolutionary algorithm optimization unit inputs the weighted adjacency matrix generated by the graph neural network topology modeling unit and generates a Pareto solution set of pump and station coordination parameters through a non-dominated sorting genetic algorithm; Filter the Pareto solution of the pump and station collaborative parameters to concentrate the matching parameter group of the conveying pump speed and the mixing station discharge rate, and output it as an equipment collaborative optimization instruction to the reinforcement learning execution strategy module; The execution abnormality data fed back by the fault-tolerant closed-loop control module is input to update the flow state instability risk entropy weight coefficient of the non-dominated sorting genetic algorithm.

5. The collaborative control system for civil construction concrete equipment in a thermal power plant according to claim 4, characterized in that: The multi-objective evolutionary algorithm optimization unit is further used for: Input the concrete flow state data of the dynamic self-calibration data acquisition module, calculate the deviation rate according to the slump measured value and the preset design calibration value, and generate the flow state risk entropy value; The weighted adjacency matrix output by the graph neural network topology modeling unit is used to generate the Pareto solution set of pump-station coordination parameters through non-dominated sorting. An ε-constraint method screening operation is performed on the Pareto solution set of the pump-station coordination parameters, wherein the constraints include: A communication delay threshold preset at the pump control node; The tolerance range of flow risk entropy value is preset based on the concrete mix ratio; The filtered conveying pump speed and mixing station discharge rate matching parameter group is output to the reinforcement learning execution strategy module to form equipment collaborative optimization instructions.

6. The collaborative control system for civil construction concrete equipment in a thermal power plant according to claim 5, characterized in that: The reinforcement learning execution strategy module includes: A collaborative parameter execution unit inputs a parameter group matching the conveying pump speed and the mixing station discharge rate in the equipment collaborative optimization instruction; a deep deterministic policy gradient decision unit, which generates a screw conveyor speed increment and a conveying pump motor frequency offset based on the matching parameter group; A fuzzy proportional-integral-differential controller inputs the flow state risk entropy value generated by the hybrid optimization collaborative analysis module, performs a differential operation on the flow state risk entropy value to obtain a rate of change, and dynamically adjusts the proportional band width according to the amplitude of the rate of change; The actuator unit converts the speed increment and frequency offset output by the deep deterministic policy gradient decision unit into physical control signals, and outputs them to the fault-tolerant closed-loop control module.

7. The collaborative control system for civil construction concrete equipment in a thermal power plant according to claim 6, characterized in that: The reinforcement learning execution strategy module also includes: An intermediate language compiling unit inputs the screw conveyor speed increment instruction and the conveying pump motor frequency offset instruction output by the deep deterministic policy gradient decision unit; Decomposing the speed increment instruction and the frequency offset instruction into operator nodes and operand leaf nodes through an abstract syntax tree parser, and constructing an intermediate representation in a static single assignment form; A logic compiler framework, which loads a preset programmable logic controller instruction mapping rule library, converts the operator nodes in the intermediate representation into corresponding function block instructions, and generates programmable logic controller-specific code; When semantic checking detects an undefined operator or type conflict, a compilation error code including an error location identifier is generated; If there is a compilation error code, trigger an instruction regeneration request to the deep deterministic policy gradient decision unit; If the compilation is successful, the programmable logic controller-specific code encapsulated as a structured data packet is output to the fault-tolerant closed-loop control module.

8. The collaborative control system for civil construction concrete equipment in a thermal power plant according to claim 7, characterized in that: The fault-tolerant closed-loop control module includes: a time-sensitive network transmission unit, which parses the transmission priority tag in the header of the programmable logic controller-specific code packet output by the reinforcement learning execution strategy module and allocates preemptive transmission time slots according to the priority level; The dedicated code after the time slot is allocated is encapsulated into a time-sensitive network frame structure and output to the target device actuator via industrial Ethernet; Real-time monitoring of the 5G base station uplink packet loss rate. When the monitored value exceeds a preset proportional coefficient of the pump node communication delay threshold defined by the graph neural network topology modeling unit, the LoRa radio frequency module is activated to establish a mesh relay communication link, and Reed-Solomon forward error correction coding is loaded at the data link layer. The structured exception handling unit captures the duration of packet loss events and the peak packet loss rate, and generates an exception report including temporal and spatial feature data; The abnormality report is fed back to the graph neural network topology modeling unit to trigger the calculation of the communication reliability attenuation factor and the update of the weighted adjacency matrix.

9. The collaborative control system for civil construction concrete equipment in a thermal power plant according to claim 8, characterized in that: The fault-tolerant closed-loop control module also includes: The vibration and template coupling unit inputs the six-degree-of-freedom trajectory data of the vibrating rod output by the inertial navigation calibration unit and the flow state risk entropy value generated by the hybrid optimization collaborative analysis module; Initialize the particle swarm position vector to the current spatial coordinate of the vibrator, and the velocity vector to the gradient of the flow risk entropy value; By iteratively solving the Newton and Euler motion differential equations, the optimal insertion angle solution set that satisfies the minimum axial resistance and the highest concrete density is obtained; Outputs the spatial optimal solution including the pitch angle, azimuth angle and insertion depth to the multi-axis motion controller of the vibrator actuator; When the fiber Bragg grating sensor arranged at the template bearing point detects that the local strain value exceeds 70% of the material yield strength, the strain gradient direction and amplitude characteristics of the over-limit area are extracted, and the inertia weight parameter of the algorithm is dynamically adjusted according to the cosine similarity between the gradient direction and the particle swarm convergence direction.

10. The collaborative control system for civil construction concrete equipment in a thermal power plant according to claim 9, characterized in that: The logic compiler framework in the reinforcement learning execution strategy module is configured as follows: Capturing a standardized abnormal execution code through a diagnostic interface of a target device, the code including an error type identifier and fault location information; Triggering the third-level rollback operation includes retrieving the previous dedicated code package marked as verified in the instruction cache queue, verifying the code integrity signature, and loading it into the target device executor to overwrite the current abnormal state; Call the preset error type and deviation mapping rule library to interpret communication timeout errors as negative device coupling deviation values ​​and control misalignment errors as positive device coupling deviation values; Generate a structured correction parameter package including deviation direction and amplitude; Inputting the structured correction parameter package into the graph neural network topology modeling unit; The control coupling strength factor of the weighted adjacency matrix is ​​updated according to the deviation direction. Negative deviation reduces the weight of the associated edge, while positive deviation increases the weight. Normalization is performed after the update.

Images