Intelligent remote automatic control method and system for ash pump room

Abstract

The invention relates to the technical field of remote equipment control, in particular to an intelligent remote automatic control method and system for an ash pump room. The method comprises the following steps: collecting multi-dimensional pump room operation parameters in real time based on a multi-source heterogeneous sensor, carrying out collaborative coding and multi-dimensional data state change perception, and constructing an ash pump room full-parameter perception map; environment monitoring data streams at multiple positions are collected, spatial deployment distribution identification is carried out, global environment monitoring calculation is carried out, and a three-dimensional environment intelligent monitoring matrix is constructed; the method comprises the following steps: acquiring an omnibearing monitoring video of an ash pump room, performing key scene element identification, performing three-dimensional topology point cloud modeling, and constructing an ash pump room three-dimensional topology model; according to the invention, by automatically adjusting the control parameters and the scheduling strategy, the dependence on manual intervention is greatly reduced, and the operation efficiency and the equipment stability of the ash pump room are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of remote equipment control, and in particular to an intelligent remote automation control method and system for an ash pump room. Background Art

[0002] With the continuous advancement of industrial automation and intelligent control technologies, traditional ash handling systems are gradually evolving towards higher efficiency, intelligence, and remote control. As a key auxiliary facility in thermal power plants, metallurgy, building materials, and other industries, ash pumphouses are primarily responsible for transporting and processing the large amounts of ash generated during boiler operation, playing a vital role in ensuring production continuity and environmental compliance. Driven by the concept of green production, the requirements for ash pumphouse operational efficiency, safety, and environmental performance are constantly increasing. Traditional manual operation and semi-automatic control methods are no longer able to meet the needs of modern industrial development.

[0003] Traditional ash pump room control systems mostly use a combination of PLC local control and manual inspections to manually monitor and operate equipment operating status, liquid level changes, pipeline patency, and valve switches. This method is not only labor-intensive and has low response efficiency, but also prone to response delays, misoperation, and information islands when faced with sudden failures or complex working conditions, which in turn affects the stability and safety of the entire ash conveying system. In addition, since the ash pump room environment is usually characterized by high temperature, high humidity, and high corrosion, traditional equipment is also prone to wear, corrosion, and blockage during long-term operation, further increasing the difficulty of operation and management. With the continuous maturity of advanced technologies such as the Internet of Things, artificial intelligence, and edge computing, building an intelligent remote automation control method for ash pump rooms based on intelligent perception, remote control, and data-driven has become an important direction of current technological innovation in the industry. Summary of the Invention

[0004] In order to solve the above technical problems, the present invention proposes an intelligent remote automation control method and system for an ash pump room to solve at least one of the above technical problems.

[0005] To achieve the above object, the present invention provides an intelligent remote automation control method for an ash pump room, comprising the following steps: Step S1: Based on multi-source heterogeneous sensors, multi-dimensional pump room operating parameters are collected in real time, and collaborative coding and multi-dimensional data state change perception are performed to construct a full parameter perception map of the ash pump room; Step S2: Collect environmental monitoring data streams from multiple locations, identify spatial deployment distribution, and perform global environmental monitoring calculations to construct a three-dimensional environmental intelligent monitoring matrix; Step S3: Obtain all-round monitoring video of the ash pump room, identify key scene elements and perform three-dimensional topological point cloud modeling to construct a three-dimensional topological model of the ash pump room; Step S4: mapping the full-parameter perception of the ash pump room to the 3D topological model of the ash pump room according to the full-parameter perception map of the ash pump room, and performing real-time monitoring status evolution updates based on the three-dimensional environment intelligent monitoring matrix to construct a digital twin model of the pump room; Step S5: Perform virtual full-process operation simulation on the pump room digital twin model, perform real-time equipment scheduling deviation detection, and mark the scheduling control deviation equipment; Step S6: Identify the signs of equipment failure in the dispatching control deviation, and then perform adaptive adjustment of the execution instructions to perform the remote automation control operation of the pump room.

[0006] In this specification, an intelligent remote automation control system for an ash pump room is provided, which is used to execute the intelligent remote automation control method for an ash pump room as described above, comprising: The operating parameter perception module is used to collect multi-dimensional pump room operating parameters in real time based on multi-source heterogeneous sensors, and perform collaborative encoding and multi-dimensional data state change perception to build a full parameter perception map of the ash pump room; The global environmental monitoring module is used to collect environmental monitoring data streams from multiple locations, identify spatial deployment distribution, and perform global environmental monitoring calculations to build a three-dimensional environmental intelligent monitoring matrix; The 3D point cloud modeling module is used to obtain all-round monitoring videos of the ash pump room, identify key scene elements, perform 3D topological point cloud modeling, and construct a 3D topological model of the ash pump room; The digital twin module is used to map the full-parameter perception of the ash pump room to the 3D topological model of the pump room based on the full-parameter perception map of the ash pump room, and to update the real-time monitoring status evolution based on the three-dimensional environment intelligent monitoring matrix to build a digital twin model of the pump room; The operation simulation module is used to simulate the virtual full-process operation of the pump room digital twin model, perform real-time equipment scheduling deviation detection, and mark the scheduling control deviation equipment; The adaptive adjustment module is used to identify the signs of equipment failure when the dispatching control deviates, and then adaptively adjust the execution instructions to perform remote automation control operations in the pump room.

[0007] The beneficial effects of the present invention are as follows: Ash pump rooms involve equipment such as motors, pumps, pipelines, valves, pressure vessels, and ash conveying channels. Operating parameters vary widely (such as flow rate, pressure, current, temperature, and vibration frequency), requiring comprehensive coverage using multiple sensor types. Collaborative encoding of heterogeneous sensor data eliminates data isolation between devices and achieves semantic unification. Time series analysis and state perception technologies are used to extract features and fusion model the changing trends of various parameters, enabling a deep understanding of operating conditions. A "pump room full-parameter perception map" is constructed, creating a structured, labeled representation of all operating conditions, providing high-dimensional information support for subsequent spatial mapping and anomaly identification. Environmental sensors (such as temperature and humidity, dust concentration, combustible / toxic gas sensors, and noise sensors) are deployed in different functional areas of the pump room (such as the pump area, electrical control room, pipe gallery, and ash bin) to achieve comprehensive data collection. A spatially deployed distribution recognition algorithm enables precise location of each monitoring point and analysis of its assigned area. The monitoring matrix not only provides current environmental conditions but also supports trend prediction, local anomaly detection, and analysis of the impact of environmental factors. Effectively identify environmental factors that affect equipment operation (e.g., high dust levels causing motor overheating, humidity affecting electrical systems), enhancing equipment protection capabilities. Linked with subsequent control systems, this allows for automatic adjustments to environmental conditions (e.g., ventilation, drainage, air conditioning, etc.), establishing a fusion of human, machine, and environment.

[0008] By combining high-definition video capture with AI image recognition technology, key scene elements (such as pump body numbers, valve types, and equipment layouts) are automatically identified. Key equipment is automatically labeled, achieving a fusion of "image and structure." 3D point cloud reconstruction technology (such as LiDAR or multi-angle video fusion) is used to create a 3D spatial model of the pump room, including equipment appearance, positional relationships, and workspace. This supports logical modeling of connections between devices (e.g., pump → pipeline → valve → outlet), building a "structured spatial logical chain." This provides spatial support for subsequent simulations, virtual inspections, and remote maintenance, realizing the concept of a "digital site" without on-site personnel. Various status data in the perception map (such as pump 1's vibration level, temperature, and startup frequency) are mapped and bound to corresponding entities in the 3D model, building a data-driven 3D dynamic system. Combined with environmental monitoring matrix data, this system synchronizes the evolution of equipment status, environmental conditions, and system load, enabling visualization of abnormal development trends and prediction of early warning lines. This system provides real-time visibility into changing pump room operating conditions, improving the timeliness of emergency response. It supports dynamic visualization, linkage analysis of key indicators, and playback and review of operating scenarios, becoming a virtual assistant and decision-making support platform for operators.

[0009] Without affecting actual operations, the system simulates the entire pump room operation process, including equipment startup sequence, pressure transmission paths, and control strategy execution results. It also simulates common abnormal conditions (such as water hammer, blockage, and excessive load) to preemptively verify the rationality of the scheduling plan. Actual operating results are compared with simulated expectations to identify issues such as equipment response delays, control execution failures, and policy mismatches. "Intelligent comparison + automatic tagging" is implemented to quickly locate critical equipment or execution links with problems, assisting with precise maintenance. Based on the deviation detection results from the previous step, control parameters or scheduling strategies (such as startup time, start-stop sequence, and control logic) are automatically adjusted, forming a data-driven adaptive feedback mechanism. This significantly reduces reliance on manual intervention and enhances the self-healing capabilities of the automatic control system. Remote dispatching commands are supported, enabling unmanned or reduced-staff operation of pump room equipment. When an abnormality occurs, an alternative strategy can be remotely switched to, accelerating emergency response. By optimizing control logic and adaptive scheduling, system operation is more coordinated, effectively avoiding inefficient equipment operation and energy waste, and achieving energy-saving operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 This is a schematic flow chart of the steps of an intelligent remote automation control method for an ash pump room according to the present invention; Figure 2 Detailed implementation flow chart of step S1; Figure 3 Detailed implementation flow chart of step S2; Figure 4 Schematic diagram of the detailed implementation steps of step S3. DETAILED DESCRIPTION

[0011] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0012] This application provides an intelligent remote automation control method and system for an ash pump room. The execution entities of the intelligent remote automation control method and system for an ash pump room include, but are not limited to, the following: mechanical equipment, data processing platforms, cloud server nodes, network upload devices, etc. equipped with the system, which can be regarded as general computing nodes of this application. The data processing platform includes, but is not limited to, at least one of an audio and image management system, an information management system, and a cloud data management system.

[0013] See also Figures 1 to 4 The present invention provides an intelligent remote automation control method for an ash pump room, comprising the following steps: Step S1: Based on multi-source heterogeneous sensors, multi-dimensional pump room operating parameters are collected in real time, and collaborative coding and multi-dimensional data state change perception are performed to construct a full parameter perception map of the ash pump room; Step S2: Collect environmental monitoring data streams from multiple locations, identify spatial deployment distribution, and perform global environmental monitoring calculations to construct a three-dimensional environmental intelligent monitoring matrix; Step S3: Obtain all-round monitoring video of the ash pump room, identify key scene elements and perform three-dimensional topological point cloud modeling to construct a three-dimensional topological model of the ash pump room; Step S4: mapping the full-parameter perception of the ash pump room to the 3D topological model of the ash pump room according to the full-parameter perception map of the ash pump room, and performing real-time monitoring status evolution updates based on the three-dimensional environment intelligent monitoring matrix to construct a digital twin model of the pump room; Step S5: Perform virtual full-process operation simulation on the pump room digital twin model, perform real-time equipment scheduling deviation detection, and mark the scheduling control deviation equipment; Step S6: Identify the signs of equipment failure in the dispatching control deviation, and then perform adaptive adjustment of the execution instructions to perform the remote automation control operation of the pump room.

[0014] In the embodiment of the present invention, see Figure 1 , is a schematic flow chart of the steps of an intelligent remote automation control method for an ash pump room of the present invention. In this example, the steps of the intelligent remote automation control method for an ash pump room include: Step S1: Based on multi-source heterogeneous sensors, multi-dimensional pump room operating parameters are collected in real time, and collaborative coding and multi-dimensional data state change perception are performed to construct a full parameter perception map of the ash pump room; In this embodiment, by deploying multi-source heterogeneous sensors, comprehensive perception of key operating elements of the ash pump room and tracking of dynamic state changes are achieved. A visual and structured "ash pump room full parameter perception map" is constructed based on a collaborative coding mechanism to support subsequent intelligent scheduling and automatic control. During the specific implementation process, multiple types of industrial-grade sensors are first deployed in multiple key locations in the pump room, including vibration acceleration sensors (such as IEPE type), bearing temperature thermocouples (Pt100), motor power factor measurement modules, current / voltage acquisition devices, and fluid pulsation pressure sensors, etc., to achieve multi-dimensional collection of pump body vibration spectrum, bearing thermal state, motor electrical characteristics, and fluid dynamic parameters. Various sensors use different communication protocols (such as Modbus RTU, CAN, EtherCAT, etc.) and are uniformly connected through edge computing nodes to build a standardized data channel.

[0015] To account for differences in sampling frequencies and data formats among sensors, the system utilizes multi-frequency synchronization mechanisms (e.g., frame synchronization based on the least common multiple sampling period) to time-align data, ensuring millisecond-level timestamp consistency between vibration and electrical data. Subsequently, the system introduces a collaborative encoding module. Based on multidimensional feature extraction results (such as frequency domain features, amplitude envelope, and parameter mean / extreme points), an improved sparse hash coding algorithm is used to uniformly encode all sensor data in a multimodal manner. This compresses heterogeneous data into feature vectors of equal length while maintaining temporal and structural correlations. To perceive data state changes, a sliding window mechanism is used for multi-timescale analysis (e.g., 2 seconds, 10 seconds, and 60 seconds). Clustering methods (e.g., K-means dynamic window clustering) are then combined to identify operational state changes and identify status levels such as normal, overload, minor anomaly, and fault premonition. Ultimately, the spatial deployment coordinates, collected data, status labels, and collaborative encoding vectors of all sensors are integrated into a unified data structure to construct a "full-parameter perception map." This map not only provides real-time visualization of the pump room's operating status, but also provides a solid foundation for subsequent risk control and scheduling. In experiments, the system was able to maintain a sensor data update cycle within 200ms, with an accuracy rate of 95.2% in detecting state changes.

[0016] Step S2: Collect environmental monitoring data streams from multiple locations, identify spatial deployment distribution, and perform global environmental monitoring calculations to construct a three-dimensional environmental intelligent monitoring matrix; In this implementation, high-resolution, multi-point, spatial monitoring modeling of the internal environmental conditions of the ash pump room is constructed to build a dynamic, three-dimensional intelligent environmental monitoring matrix, providing an environmentally sound basis for decision-making regarding pump room operational safety and intelligent scheduling. First, multiple types of environmental sensor units are deployed within the pump room, covering key operating areas and potential risk points. These sensors include industrial-grade temperature and humidity sensors (such as the DHT21), toxic gas concentration detectors (such as NH3, SO2, and CO concentration sensors), high-precision noise decibel meters (40-130dB measurement range), and water leak detection sensors (electrode or capacitive). Each sensor unit is deployed at multiple points within the pump room's functional areas, such as near the electrical control cabinet, slag discharge pipelines, the base of the water pump, and damp corners to ensure complete and representative environmental data coverage.

[0017] The data collection frequency of each environmental sensor is set between 1 and 5 Hz based on its sensing characteristics. Data is transmitted to edge data collection terminals via LoRa, Zigbee, or wired 485, and then consolidated into the cloud or local control center. The system first calculates the spatial coordinates of all collection points. Using laser ranging and combined with the pump room BIM model or on-site CAD drawings, a 3D model of the deployment location is constructed to generate the spatial coordinate vectors for each environmental monitoring point. The collected data streams are then modally unified and time-aligned, and time series curves are extracted for the same sensor data type. The system uses weighted sliding window averaging and Kalman filtering to denoise and extract trends from the collected data, generating multiple environmental state evolution trajectories, such as "temperature change curves" and "gas concentration change curves." Based on the spatial distribution coordinates of the collection points, the system incorporates a spatial deployment distribution recognition mechanism. Using Voronoi diagram partitioning and density-based spatial clustering (such as DBSCAN), the system intelligently identifies the spatial distribution of environmental parameters. Each spatial region is assigned a characteristic value for the environmental parameter, forming multiple "environmental monitoring sub-regions." On this basis, the system maps all sub-area parameters into the three-dimensional coordinate system of the pump room space, constructing a three-dimensional monitoring matrix with volumetric coverage and real-time data updates, enabling visual analysis of environmental conditions at any point in the pump room at any time. For example, in one experimental deployment, a total of 22 environmental nodes were deployed, covering an area of approximately 230 square meters. The system refreshed global environmental data within 3 seconds and accurately mapped the temperature, humidity, and air pressure distribution in each area, providing a critical basis for pump room equipment operation scheduling and safety warnings.

[0018] Step S3: Obtain all-round monitoring video of the ash pump room, identify key scene elements and perform three-dimensional topological point cloud modeling to construct a three-dimensional topological model of the ash pump room; In this embodiment, visual perception is used to fully model the internal structure and operating environment of the ash pump room, providing spatial recognition support for subsequent automated monitoring and control. First, by rationally deploying multi-angle high-definition surveillance cameras (e.g., 2 million pixels and above, supporting 30fps and low-light compensation functions) in the pump room, all-around video coverage of the main areas of the pump room is achieved. The collected video data is decoded and image stream preprocessed in real time by edge processing nodes, including frame rate standardization, image detail enhancement (using the CLAHE limit adaptive histogram equalization method), noise suppression, and edge sharpening operations, thereby constructing a basic delay-optimized monitoring video stream and providing high-quality image frame sequences for subsequent computing tasks. The system divides the optimized video stream into key areas of the pump room and uses a deep learning image segmentation model (such as one based on the U-Net or YOLOv8-seg structure) to semantically identify and annotate key scene elements in the video frames. Identification targets primarily include pump control cabinets and their indicator light status (green, red, and yellow represent run, stop, and alarm), electric doors and their open and closed status, water level scales (used to identify changes in liquid level), areas of flooded floor surfaces, and operation panel buttons. Indicator light status recognition utilizes color space separation and regional contrast recognition, combined with time series stability assessment to suppress false alarms.

[0019] To construct a 3D representation of the pump room's structure, the system performs feature matching and fusion processing on images from different viewpoints in surveillance video. Feature extraction methods such as SIFT and ORB are used to extract stable corner and edge points from multiple frames. A preliminary 3D structural model is then reconstructed using the SfM (Structure from Motion) algorithm. This model is then fused with the known 2D layout of the pump room or laser ranging data to achieve spatial scale correction. During the 3D reconstruction process, the system also distinguishes between static and dynamic elements in the scene, filtering out interfering factors such as operators and operating equipment arms, retaining only components with spatial reference significance. To further improve model accuracy, the system supplements the point cloud with RGB-D data or LiDAR point cloud data (such as the Velodyne VLP-16) to generate a high-density 3D point cloud. The system then uses MeshLab or PCL point cloud libraries for point cloud fusion, surface reconstruction, and texture mapping, ultimately constructing a complete 3D topological model that includes layout feature points and scene elements. During the test deployment, the system constructed a topological model with a point cloud density of 5mm through three-view video fusion and assisted laser scanning, keeping the 3D structure error within ±2cm. This accuracy is sufficient to meet the spatial recognition and interaction requirements of remote pump room control tasks. This 3D model can be used for subsequent equipment status mapping, navigation path planning, and digital twin system integration.

[0020] Step S4: mapping the full-parameter perception of the ash pump room to the 3D topological model of the ash pump room according to the full-parameter perception map of the ash pump room, and performing real-time monitoring status evolution updates based on the three-dimensional environment intelligent monitoring matrix to construct a digital twin model of the pump room; In this embodiment, the "full-parameter perception map of the ash pump room" constructed in step S1 serves as the foundation. This map includes high-frequency operating parameters such as vibration spectrum, temperature, current, pressure, and flow rate, collected from heterogeneous sensors across multiple sources, including the pump body, pipelines, motors, bearings, and fluid systems. This perception data includes metadata such as timestamps, unique device identifiers, and measurement location information, ensuring that each piece of data accurately corresponds to a specific device or operating node in the pump room. This perception map is mapped onto the 3D topological model reconstructed in step S3, forming a semantic binding relationship between the device and the space. Specifically, based on the geometric transformation relationship between the sensor deployment coordinate system and the 3D model's spatial coordinate system (typically achieved through an extrinsic parameter matrix or spatial registration), various real-time data streams are visually embedded into the corresponding model structure. For example, if the vibration value of a pump body is excessively high, the corresponding pump body in the 3D model will be highlighted in real time, and a vibration spectrum change graph will be displayed; a temperature anomaly in the control cabinet will simultaneously trigger a thermal map overlay. This mapping process utilizes a multidimensional perception-structure fusion engine that supports time series data caching, anomaly identification and tagging, and dynamic threshold adjustment. To enhance the digital twin model's sensitivity to external environmental changes, the system simultaneously integrates the "3D Environmental Intelligent Monitoring Matrix" constructed in step S2. This matrix provides multimodal environmental data, including temperature, humidity, noise, gas concentration, and water accumulation levels, from various spatial nodes within the pump room (e.g., pump room entrances and exits, operation areas, and liquid storage areas). Through spatial deployment identification and location coordinate mapping, these environmental parameters are projected into the 3D model in real time, enabling integrated "field-object-data" perception. After constructing a complete perception map, the system uses a "state evolution engine" to perform trend modeling and dynamic deduction of historical and current perception parameters. Specific methods include sliding window statistical analysis (e.g., 5-minute mean temperature change), time series anomaly detection (e.g., detecting parameter mutations using the ARIMA model), and correlation mining (e.g., whether increased humidity in a particular pump room area is associated with a decrease in motor power factor). All status changes are updated in real time within the digital twin model through dynamic color gradients, trend curves, and animated trajectories, enabling managers to intuitively perceive the overall status and changing trends of the pump room's operation on the platform. The digital twin model not only realizes the complete reconstruction of the physical space structure of the pump room, but also integrates the real-time mapping, dynamic evolution, interactive and predictable characteristics of all operating and environmental perception parameters.

[0021] Step S5: Perform virtual full-process operation simulation on the pump room digital twin model, perform real-time equipment scheduling deviation detection, and mark the scheduling control deviation equipment; In this embodiment, an engine simulates various aspects of the actual operation of an ash pumphouse, including pump start-stop sequences, motor load changes, fluid flow path switching, and valve control state responses. To ensure simulation accuracy, the system uses real historical operating data as initial values and reference sequences. It then constructs a multi-step dynamic simulation using a physical simulation model (e.g., based on the Navier-Stokes fluid dynamics model) and a control rule engine (e.g., a state transition matrix). For example, when simulating the slurry discharge process, the system simultaneously simulates the impact of slurry density changes on pump power load and maps this to a three-dimensional topological model. The simulation system outputs a "predicted operating parameter sequence," including motor power factor curves, vibration frequency trends, and pump start-stop response delays. This parameter sequence, serving as the "expected operating trajectory," is compared and analyzed with actual operating data. The system sets a 5-second scheduling synchronization window and allows a ±5% error tolerance. Based on this, a scheduling deviation detection model is constructed to compare the differences between predicted values and real-time collected values to identify any scheduling control deviations. If a device's real-time power factor deviates from the predicted value by more than a threshold (e.g., a set ±8%), an abnormal deviation alarm is triggered. To facilitate fault tracing and operational adjustments, the system highlights the deviating device in the 3D model. Each marker is accompanied by detailed deviation parameters (such as actual current value, predicted value, deviation percentage, and occurrence timestamp), which are automatically recorded in the device operation log database. Furthermore, this process incorporates device operating status tags (such as "active," "standby," and "load fluctuation") for contextual analysis to avoid misjudging scheduling deviations during device state transitions. In a laboratory environment, the simulation system can simultaneously simulate the full-process operation of more than 16 core pump units, achieving a root mean square error (RMS) prediction accuracy of less than 4% and a deviation detection latency of less than 1.2 seconds, meeting real-time monitoring requirements. By linking simulation output with real-time device scheduling information, the system can proactively identify potential scheduling issues, effectively improving the coordination and intelligence of the pump room's overall operation, and serving as a key support for remote automated control.

[0022] Step S6: Identify the signs of equipment failure in the dispatching control deviation, and then perform adaptive adjustment of the execution instructions to perform the remote automation control operation of the pump room.

[0023] In this embodiment, continuous status tracking is performed for identified devices with deviating schedules. Starting with the deviation marker, the system collects multi-dimensional operating parameters for the 30 seconds preceding and following the device, including bearing temperature (e.g., a 1Hz acquisition frequency), vibration frequency (e.g., using a high-precision accelerometer at 200Hz), power factor, current fluctuation, and the number of operating state transitions. This generates a time-based time series feature vector. Based on this data, the system employs a sliding window-based fault symptom identification model, for example, combining an LSTM (Long Short-Term Memory) network with a rule-matching model for parallel identification. The LSTM model detects periodic abnormal fluctuation trends, while the rule-based model performs static threshold comparisons against key domain values. For example, a continuous increase in the vibration frequency of a pump for more than four seconds, or a sharp jump in the power factor within the ±10% range for more than five times, triggers an "abnormal trend" label. After fault symptom identification, the system automatically compares the data against a device health reference model (typically established based on factory specifications or historical maintenance cycles) to calculate the device's current health index (e.g., a continuous value between 0 and 1, with values below 0.3 indicating high risk). The system then searches the "Risk Countermeasures Strategy Library" for corresponding equipment response strategies. For example, if pump A experiences high-frequency vibration and temperature rise, the strategy library might recommend load reduction, delayed shutdown, or switching to a backup pump. After determining the strategy, the system compares the original scheduling plan with the new strategy and builds an adaptive execution instruction adjustment engine. This engine optimizes the execution path based on the logical constraint graph (including the pump group linkage sequence, system load balancing rules, etc.) and equipment operational dependencies, ultimately generating a new set of execution control instructions (such as "Reduce the frequency of pump P3 to 70% of the rated value and simultaneously start pump P5 as a backup"). Before issuance, all instructions are quickly rehearsed through a simulation verification module to verify that the adjusted pump room still meets system load, flow, and safety boundary conditions. After the instructions are issued and executed, the system simultaneously enters the monitoring and feedback phase, observing in real time the impact of the adjustment results on overall system operation. If the risk factors are not alleviated or the scheduling chain is further disturbed, the system will trigger a second adaptive adjustment cycle. In a test scenario, this mechanism successfully identified and corrected three common equipment anomalies: excessive pump vibration, abnormal motor temperature rise, and operating power deviation. After adaptive adjustments, the overall load in the pump room was restored to a balanced state, and the average lead-time warning time for core equipment failures increased to 12 minutes, with no additional operational interruptions, demonstrating the high reliability of remote adaptive control. This step effectively ensures the stability, safety, and continuous operation of the pump room under intelligent control.

[0024] In this embodiment, refer to Figure 2 , is a flowchart of the detailed implementation steps of step S1. In this embodiment, the detailed implementation steps of step S1 include: Based on multi-source heterogeneous sensors, real-time collection of pump vibration spectrum, bearing temperature, motor power factor, and fluid pressure pulsation is carried out to obtain multi-dimensional pump room operating parameters; Calculate the multi-parameter acquisition frequency of multi-source heterogeneous sensors and extract the parameter acquisition frequency of different sensors; Performing multi-device signal synchronization optimization on the multi-dimensional pump room operating parameters according to the parameter acquisition frequency to generate time-series synchronized multi-dimensional parameters; Identify communication protocols based on multi-source heterogeneous sensors and extract the communication protocol of each sensor device; Perform protocol conflict analysis on the communication protocol of each sensor device and identify devices with data interaction conflicts; Perform protocol coordination conversion on devices with data interaction conflicts to obtain conflict resolution and conversion strategies; Based on the conflict resolution and conversion strategy, the time-series synchronization multi-dimensional parameters are collaboratively encoded, and the multi-dimensional data state change perception is performed to construct a full parameter perception map of the ash pump room.

[0025] To fully monitor the operating status of the ash pumphouse, this system deploys multi-source, heterogeneous industrial-grade sensors to collect key operating parameters in real time. These include pump vibration spectrum (using MEMS accelerometers and spectrum analyzers), bearing temperature (using thermocouple temperature sensors), motor power factor (using a power quality analyzer to collect the power factor of the three-phase AC power), and fluid pressure pulsation (using a high-sensitivity pressure transmitter). These sensors are installed at the pump bearings, motor power input, and pump inlet and outlet fluid pipelines, and transmit data back via a local area network or LoRa wireless link. The sampling frequency is: vibration spectrum at 500 Hz per second, bearing temperature every 5 seconds, power factor once per second, and pressure pulsation data at 100 Hz. By comparing against industry standards and combining experimental observations of the sensitivity of pumphouse operating status changes (for example, vibration and power factor changes are the primary indicators of a pump's transition from startup to load stabilization within 15 minutes), the sensor parameter configuration ensures that it meets dynamic monitoring requirements. The key to this step is high-frequency, multi-dimensional, and low-latency data acquisition, providing an accurate and detailed data foundation for subsequent pump room operating status perception and control logic. Frequency analysis is performed on data collected by different sensor types to clarify the update cycles of various sensor parameters, supporting subsequent synchronization. Since multi-source, heterogeneous sensors have varying sampling frequencies and data refresh mechanisms, to avoid data alignment errors caused by inconsistent sampling frequencies, the acquisition frequency of each parameter must first be calculated. The actual number of acquisitions per unit time is calculated by counting the number of data packets recorded within a sliding time window (e.g., 5 minutes). For example, a pressure sensor collects 30,000 data points in 5 minutes, indicating a sampling frequency of approximately 100 Hz. Meanwhile, a motor power factor sensor records only 300 times in the same time period, indicating a frequency of approximately 1 Hz. The system maps each frequency as a parameter into a frequency table, providing a calibration basis for subsequent synchronization. This process ensures that the time resolution conversion and interpolation strategies used in subsequent data fusion are based on the actual frequency, effectively avoiding data spurious synchronization or information loss. Inconsistent data timestamps due to varying sampling rates among different sensors contribute to this step's central purpose: to unify all sensor data streams onto the same timeline. The system adopts a synchronous interpolation and reconstruction method. For example, for low-frequency sampling data (such as temperature every 5 seconds), spline interpolation is used to fill in the intermediate time points; for high-frequency data (such as vibration spectrum), downsampling (average or median method) is used to make its time step consistent with other data.

[0026] The synchronization strategy is based on a master clock timestamp (e.g., using the system server as a unified benchmark), ensuring that all data points correspond to the pump room's operating status within the same acquisition window. System experiments have shown that by introducing a weighted multidimensional linear interpolation algorithm (WMLI), time alignment of data can improve multi-source data coordination by approximately 18.7%, effectively supporting subsequent multidimensional analysis models. Furthermore, the system packages all time-synchronized data into "time-synchronized multidimensional parameter frames," each containing all sensor data within a time slice, achieving structured unification of pump room data. Because the sensors used in the ash pump room vary in brand and model, and their communication protocols may include Modbus RTU, CAN, 4-20mA, RS485, ZigBee, and others, the system needs to automatically identify the sensor's communication method. In this step, the system utilizes a protocol frame header and footer recognition and communication handshake feature matching mechanism, combined with the driver layer protocol library, to capture handshake packets for each device during its network online phase and identify its protocol type and communication frame structure. Modbus devices begin with an address frame beginning with "0x01" and reply with a function code and CRC checksum. ZigBee modules use IEEE 802.15.4 protocol frames and transmit data in APDU format. The system automatically extracts these patterns and establishes a device-to-protocol mapping table, which is used for subsequent conflict resolution and conversion. This identification mechanism successfully identified protocols with an accuracy rate exceeding 96.5% in experiments, laying a solid foundation for data fusion between heterogeneous devices. Because sensor devices may have communication protocol incompatibilities, such as address conflicts or bus contention when Modbus RTU devices and CAN bus devices transmit simultaneously, the system needs to resolve potential data exchange conflicts between devices. The platform uses device mappings and communication resource usage tables to detect transmission channel overlap, master station polling conflicts, and frame format inconsistencies within the same time window. Upon detecting issues such as communication port occupancy conflicts (e.g., two serial devices bound to the same port) or timing data frame format mismatches (Modbus frames conflicting with ZigBee frames), the system flags the conflicting devices and initiates coordination strategy formulation. The system accurately identified 92.3% of transmission conflict points in experiments by detecting packet frequency and analyzing bandwidth usage in conflicting areas of data flows. Device communication conflicts are resolved through protocol conversion between hardware and software layers. The system deploys a protocol bridging module (e.g., using an MCU or embedded gateway) through middleware to convert protocols like Modbus and CAN into a unified message format (e.g., JSON over MQTT) for unified upload to the central processing platform.

[0027] The platform establishes device-specific conversion rules based on a conflict resolution strategy table. For example, if a CAN protocol device needs to share a data channel with an RS485 device, the system re-encapsulates the CAN frame data into RS485 format using a serial-to-parallel conversion module. The system also deploys an intelligent scheduling mechanism to fine-tune data upload intervals to avoid conflict windows and achieve communication load balancing. This strategy reduced device conflict interference by approximately 81.2% in a pump room test environment, significantly improving system stability. After all data is protocol-aligned and time-synchronized, a collaborative encoding method is used to fuse and model multidimensional data. The platform introduces a multidimensional state transition matrix to normalize parameters with different physical meanings (such as temperature, vibration frequency, and pressure amplitude) and map them into a unified state space. State transition analysis, anomaly trend detection, and clustering algorithms (such as K-Means dynamic time clustering) are then used to identify and annotate data state changes. The system creates a real-time perception map based on the operational indicator changes of each device. The parameter status is displayed as a multidimensional heat map, allowing managers to comprehensively visualize the pump room's operating status on a large screen.

[0028] In this embodiment, the specific steps of performing multi-device signal synchronization optimization on the multi-dimensional pump room operating parameters according to the parameter acquisition frequency to generate time-series synchronized multi-dimensional parameters are as follows: Obtain the preset main control system clock signal; Calculating a clock signal of each device of the multi-source heterogeneous sensor; Calculating the clock deviation of each device on the main control system clock signal according to the clock signal of each device to generate a clock deviation value for each device; Dynamically compensate for the deviation of multi-dimensional pump room operating parameters based on the clock deviation value of each device to obtain signal deviation compensated operating parameters; Performing a key device level assessment on the multi-source heterogeneous sensors to obtain priorities of multiple devices; Analyze the equipment signal change trend based on multi-dimensional pump room operating parameters to generate signal change trends for different equipment; Adaptive frequency acquisition adjustment is performed based on the parameter acquisition frequency, the priority of multiple devices and the signal change trend, and multi-device signal synchronization optimization is performed to generate timing synchronization multi-dimensional parameters.

[0029] In this embodiment, the master control system clock signal serves as the time reference for the entire system, and the timestamps of all devices and sensors are aligned with this clock signal. To ensure high-precision clock synchronization, the master control system clock can be obtained through a high-precision time synchronization protocol, such as NTP (Network Time Protocol) or PTP (Precision Time Protocol). During the experiment, to ensure the accuracy of the clock signal, the master control system clock is usually obtained from a standard time source (such as an atomic clock or GPS) and transmitted to all devices that need to be synchronized via the network. The accuracy of the master control system clock directly affects the accuracy of all subsequent synchronization operations. The selected clock source usually needs to meet accuracy requirements within 1ms, especially in complex industrial environments, where the accuracy requirements may be even higher. Therefore, the stability of the clock signal is key to system reliability and data consistency.

[0030] To ensure stable clock signal acquisition, a redundancy mechanism is required. If the master clock signal source fails, the system should automatically switch to a backup clock source, such as a backup GPS signal or a backup Network Time Protocol synchronization service. In a multi-source heterogeneous sensor system, the clocks of different devices are typically independent, and their accuracy and deviation may vary. Therefore, the key second step is to calculate and calibrate the clock signal of each device. Specifically, each sensor device compares its clock signal with the master system's clock signal to calculate its deviation from the master system's clock. To achieve this, synchronization algorithms, such as timestamp alignment or synchronization deviation correction algorithms, can be used to calculate each sensor's clock in real time. Timestamp alignment methods first align the device clock signal with the master clock signal by capturing the timestamps of the master system clock and the device clock. The clock deviation of each device can be calculated by calculating the difference between the timestamp of the device's data acquisition and the master clock signal. Typically, during experiments, device clock calculation is based on a time synchronization error model, with an accuracy requirement of within 1ms. During the calibration process, if the device clock deviation is large, regular adjustment or correction may be required to ensure high clock accuracy and stability. Each device periodically reports its collected data and timestamps. The master control system compares the clock signals of each device and calculates the clock deviation of each device based on the time difference. This process can use Kalman filtering or least squares method to optimize the error and reduce the impact of time error on the calculation results.

[0031] In an experiment, the clock signal of device A is 2 milliseconds ahead of the master control system clock, while the clock of device B is 1 millisecond behind. When calculating these clock offsets, the offset value can be recorded for each device. These offset values will be used for dynamic compensation in subsequent steps. Based on the clock offset values of each device, dynamic offset compensation is then performed on the multi-dimensional operating parameters of the pump room. When each sensor collects operating parameters, its data will have errors with the device clock. These errors may affect subsequent data analysis and decision-making. Therefore, a clock offset compensation algorithm is needed to correct these errors. This compensation algorithm generally uses linear interpolation or timing alignment to adjust the parameter timestamps of each device based on its clock offset. Specifically, assuming a device clock offset of Δt (e.g., 2 milliseconds), if the device's measurement value is collected at time T of the master control system clock, the device's measurement value can be adjusted to the data at time T + Δt. In a multi-source heterogeneous sensor system, the operational importance of different devices may vary. Therefore, a device ranking assessment is required to determine the priority of each device. The ranking of critical devices can be based on multiple dimensions, such as device importance, operating status, and impact on the overall system. Typically, a weighted scoring method (e.g., AHP, Analytical Hierarchy Process) is used to evaluate devices. Each device is assigned a different weight based on its role in the system. For example, a critical pump in a pump room might be rated high priority, while auxiliary devices such as temperature and humidity sensors might be rated low priority.

[0032] Based on device priority, the frequency of signal acquisition for critical equipment can be increased to enable more timely response to system needs. For example, the operating status of critical pumps determines the overall efficiency of the pump room, necessitating a high acquisition frequency and accuracy. When analyzing device signal trends, the key is to leverage the pump room's operating parameters to predict the patterns of change in each device's signals. This can be achieved through time series analysis methods such as ARIMA (Autoregressive Integrated Moving Average) or LSTM (Long Short-Term Memory) networks, which model the device's signal trends. Device signal trend analysis helps predict device failures, abnormal conditions, and future operating trends. For example, by analyzing historical data for signals such as pump speed and pressure, it is possible to predict the device's future operating status and provide early warnings.

[0033] Based on experimental data, device trend analysis helps optimize device scheduling and improve system responsiveness. For example, if the main pump in a pump room shows signs of pressure fluctuation, the system can preemptively activate a backup pump to prevent a failure. Adaptive frequency acquisition is adjusted based on each device's priority, acquisition frequency requirements, and signal trends to ensure that signals from different devices are synchronized on the same timeline. This process aims to optimize data acquisition frequency to balance system resource consumption and data accuracy. First, for critical devices, the acquisition frequency can be dynamically adjusted based on the device's signal trends and priority. For example, if a device shows signs of failure, the acquisition frequency can be increased; for less important devices, the acquisition frequency can be appropriately reduced to conserve system resources. During this process, an adaptive sampling algorithm is used to adjust the sampling frequency in real time based on the device signal's changing trends. During the multi-device signal synchronization optimization process, a timing synchronization algorithm is used to calculate the clock offset and priority of each device to ensure that all device signals are aligned within the same time window. The resulting multi-dimensional timing synchronization parameters ensure accurate, real-time, and synchronized operating data from the pump room.

[0034] In this embodiment, refer to Figure 3 , is a flowchart of the detailed implementation steps of step S2. In this embodiment, the detailed implementation steps of step S2 include: The ash pump room is equipped with multiple environmental monitoring sensors, including temperature and humidity sensors, gas concentration detectors, noise monitors, and water leakage detection sensors; Collecting environmental monitoring data streams at multiple locations based on the environmental monitoring sensor; Performing multimodal environmental change tracking on the environmental monitoring data stream to generate multiple environmental data change curves; Calculating the spatial deployment position of the environmental monitoring sensor and extracting the position coordinates of multiple environmental monitoring nodes; Based on the location coordinates of multiple environmental monitoring nodes, the spatial deployment distribution of multiple environmental data change curves is identified, and global environmental monitoring calculations are performed to construct a three-dimensional environmental intelligent monitoring matrix.

[0035] In this example, a multi-point deployment strategy was implemented based on typical operating scenarios in an ash pump room to comprehensively monitor environmental changes. The specific sensor types and deployment methods are as follows: Industrial-grade SHT35 temperature and humidity sensors, with high accuracy of ±0.1°C and ±1.5%RH, are installed around the pump body, at ventilation openings, and inside the control cabinet to monitor heat and humidity accumulation. Gas concentration detectors, including MQ-135 and PID photoelectric sensors, detect ammonia, methane, and volatile organic compounds (VOCs), respectively. These sensors are primarily located in leak-prone areas such as the pump reservoir, maintenance access, and power distribution room. Noise monitors, equipped with A-weighted electroacoustic sensors, are installed on the pump room's roof beams, walkways, and control room entrances to identify abnormal changes in mechanical noise. Water leak detection sensors, including conductive water intrusion probes, are mounted on the pump base and in cable ducts. All sensors are connected to an RS485 bus network and assigned distinct node numbers. They work with remote DTU modules to upload data to a monitoring host. The multi-point deployment design takes into account coverage, sensing accuracy, ease of maintenance, and electromagnetic interference. This ensures that each environmental variable has a dedicated monitoring point, enabling comprehensive environmental awareness in the pump room. A sensor array deployed on-site enables continuous data collection from multiple physical measurement points. The system maintains a unified acquisition frequency of 1Hz to 5Hz, adjustable on demand. The acquisition platform uses edge computing terminals (such as Raspberry Pi 4 or ARM Cortex-A53 industrial boards) for preliminary data buffering and standardization. Each sensor regularly reports collected data, including acquisition timestamp, measurement value, and device ID. Data is transmitted using JSON or MQTT protocols and transmitted back to the central platform via a local area network or 4G / 5G communication module. Data streams generated by multiple sensors at different locations form a multidimensional input data matrix, such as temperature and humidity (T / H), gas concentration (GAS), noise decibels (DB), and water leak alarms (LEAK). The system analyzes the data stream structure and stores it in separate channels, retaining both raw and cleaned data files for subsequent analysis and traceability. To prevent data collection errors from affecting the results, we also introduced sliding window averaging, data median filtering, and outlier removal algorithms to ensure the stability and continuity of the input data. The data streams collected simultaneously from multiple locations provide high-resolution, structured raw input for subsequent environmental change modeling.

[0036] The system performs modal classification on the diverse data streams generated by multiple sensors, and constructs change curves for each environmental factor using a time series reconstruction approach. Change tracking utilizes a sliding window processing mechanism, extracting sequences with a 30-second window width. The system then applies trend filtering to the data using an exponential smoothing algorithm (EMA), extracting dynamic temporal trends for each environmental variable. For temperature and humidity data, the system analyzes intraday fluctuations; for gas concentrations, it focuses on identifying short-term spikes or threshold-crossing events. Noise data is analyzed using a Fourier transform to identify changes in frequency domain energy distribution to identify potential mechanical anomalies. Water leak alarms present binary step signals, and transition detection is used to identify state mutation points. Ultimately, all data is unified into a two-dimensional time-variable matrix and visualized as multiple curves, each representing the change trajectory of an environmental variable at a specific monitoring point. The system also generates a list of environmental mutation events to identify the start and end times and intensity levels of abnormal changes, laying the data foundation for subsequent spatial fusion and anomaly response. Accurately determining the deployment coordinates of each sensor is essential for spatial correlation analysis of monitoring data. In this step, the system establishes a three-dimensional spatial coordinate system for the pump room (with the center of the pump body as the origin, in centimeters) by combining the BIM pump room architectural model with on-site surveying information. After each sensor is deployed, its X / Y / Z coordinates are recorded and annotated in the architectural model. During sensor installation, a laser rangefinder and a total station are used for positioning measurement, with measurement accuracy controlled within ±2cm. The system establishes a "device location table" in JSON format, including information such as device number, location coordinates, device type, and deployment area (such as the west pump area and the east control area). In addition, to improve the efficiency of automated deployment, some sensors support UWB or BLE positioning chips, and their relative positions can be dynamically collected by the base station, making them suitable for mobile or portable monitoring devices. Ultimately, a complete spatial mapping of the environmental monitoring network is constructed, providing a basic coordinate system for subsequent spatial distribution analysis and multi-dimensional field map visualization.

[0037] After constructing the environmental variable curve and locating node coordinates, the system integrates the time series data with the spatial distribution to construct a three-dimensional spatiotemporal distribution model. Using node coordinates as the base point, the system uses spatial interpolation algorithms (such as inverse distance weighted (IDW) or the Kriging method) to transform the data distribution at each moment into a continuous spatial field. This method constructs temperature, humidity, and gas concentration fields, respectively. This continuous spatial data evolves over time to form a multidimensional matrix structure, denoted as E(x, y, z, t), representing the environmental state at each location within the pump room at any given moment. The system performs dynamic analysis on this matrix, including global anomaly detection (based on multivariate statistical control charts) and spatial gradient calculation (to identify localized leak sources or heat accumulation). Furthermore, based on sensor signal strength and deployment density, blind spot compensation assessments are performed to ensure that there are no significant gaps in the global environmental status. Ultimately, a "three-dimensional environmental intelligent monitoring matrix" is constructed, which can drive the pump room's automatic ventilation, exhaust, noise reduction, and fault warning modules in real time, enabling precise control and automated response to environmental risks, significantly enhancing the system's safety and automation capabilities.

[0038] In this embodiment, reference Figure 4 The above is a flowchart of the detailed implementation steps of step S3. In this embodiment, the detailed implementation steps of step S3 include: Obtain a comprehensive surveillance video of the ash pump room; enhance image details and optimize timing frame delay in the comprehensive surveillance video of the ash pump room to construct a delay-optimized surveillance video; The delay-optimized surveillance video is used to divide the pump room area and identify key scene elements. The indicator lights of the pump control cabinet, the switch status of the electric door, the water level scale, and the water accumulation area are marked to obtain multiple scene element features. Identify regional building layout feature points on delay-optimized surveillance videos and mark multiple layout feature points; Perform three-dimensional topological point cloud modeling on multiple scene element features and multiple layout feature points to construct a three-dimensional topological model of the ash pump room.

[0039] In this embodiment, high-definition wide-angle cameras are deployed (it is recommended to use industrial IP cameras that support 1080p and above resolutions and 60fps frame rates) to achieve blind-angle monitoring of the pump room from all sides. The cameras are centrally connected to the monitoring host through PoE power supply, and streaming media access is completed through the ONVIF protocol. After the video is acquired, it enters the image enhancement stage, and an image enhancement algorithm based on Retinex theory is used to restore details of the image in the low-light area to improve the image contrast and edge clarity. In order to cope with problems such as inconsistent data frame rates of different cameras and network transmission delays, the system introduces a timing frame delay optimization mechanism. This mechanism is based on the buffer frame reordering and timestamp resynchronization method to keep all images displayed without tearing on a unified timeline, and the frame delay is controlled within 50ms, ensuring the consistency and timeliness of the video stream in subsequent image recognition tasks. Through this step, the system completes the clarity and timing standardization of the original video, laying a solid visual data foundation for building accurate scene semantic analysis and three-dimensional modeling. The latency-optimized video is used to divide the pump room into zones based on field of view, such as the pump operation area, electrical control area, access and exit area, and drainage inspection area. This division is based on BIM architectural model data and image scene segmentation models (such as DeepLabV3+). The system generates contour boundaries for each zone based on segmentation masks. Based on this, YOLOv7 or CenterNet object detection models are used to identify and mark specific scene objects. The pump control cabinet indicator lights are identified through color channel separation and brightness analysis to determine their status (e.g., solid red indicates alarm, green indicates operation). Electric door status is identified using inter-frame differencing and edge detection to determine door position changes. Water level markings are identified using an optical character recognition (OCR) text recognition model to extract the markings. Waterlogged areas are identified using a combination of image texture analysis and morphological processing, identifying suspected waterlogging areas based on ground reflections and shadow boundary changes. All recognition results are annotated as bounding boxes or polygonal ROIs, recording their pixel coordinates, status values, and recognition confidence within the image. The data is then structured to prepare for subsequent 3D mapping.

[0040] The mapping relationship between the image and the real space is established through feature point annotation. First, feature points are extracted from key frames in the video using the SIFT (Scale-Invariant Feature Transform) and SURF (Speeded Up Robust Features) algorithms. Stable structural areas, such as wall corners, beam-column intersections, doorframe edges, and pipe joints, are specifically selected as layout anchor points. These feature points not only have excellent spatial recognition but also exhibit strong scale and angle invariance, making them suitable for subsequent multi-frame registration and spatial modeling. Subsequently, the system uses camera intrinsic parameters (focal length, principal point position, distortion coefficient) and viewpoint parameters, combined with the motion trajectories of the feature points across multiple frames, to estimate the initial coordinates of these points in 3D space using SfM (Structure from Motion) and SLAM (Simultaneous Localization and Mapping) methods. Furthermore, to improve recognition efficiency and accuracy, the system combines manual pre-annotation with self-supervised model training to expand the sample library of feature points and construct a unique pump room spatial structure recognition model. Ultimately, each layout feature point contains both image coordinates and estimated 3D coordinates, enabling visualization and digital fusion of spatial structure in the image. By structured integration of the aforementioned image features, the system transforms 2D images into 3D topological modeling. Specifically, the system first uniformly maps multiple scene elements (such as indicator lights, water levels, and waterlogged areas) and spatial layout feature points into a world coordinate system, generating a basic skeleton using a sparse point cloud approach. Subsequently, multi-frame image reconstruction techniques (such as Colmap or OpenMVS) are used to restore the dense point cloud, constructing a high-density point cloud model encompassing multiple structural surfaces, such as the pump body, walls, passageways, and control cabinets. The system also maps each object recognition result in the image into 3D space using optical flow registration, generating semantically annotated 3D point sets (e.g., "electric door status point" or "waterlogged point set"). Each point includes its location, element type, timestamp, and status information. Furthermore, to support subsequent real-time visualization and interaction, the point cloud model is converted into an Octree structure for hierarchical storage, which can be visualized in Unity3D or WebGL. The final constructed three-dimensional topological model of the ash pump room can not only accurately present the physical structure of the pump room, but also map the status changes of each monitoring element in real time, forming a truly visual intelligent monitoring and control basic platform.

[0041] In this embodiment, the steps of obtaining a full-scale monitoring video of the ash pump room, enhancing image details of the full-scale monitoring video of the ash pump room, and optimizing the timing frame delay are as follows: Obtain all-round monitoring video of the ash pump room; Segment the ash pump room's all-around monitoring video into multiple areas to obtain multiple different video areas; Based on the multi-scale convolutional neural network, multi-convolutional layer feature extraction is performed on multiple different video regions to obtain different scale image features of each region; Performing texture detail requirement analysis on the image features of different scales and performing adaptive detail enhancement to obtain an adaptive detail enhanced video; Perform global pixel optical flow calculation frame by frame on the adaptive detail enhanced video to obtain the pixel motion direction and speed of each frame image; Perform deep optical flow change evolution analysis based on the pixel motion direction and speed to generate pixel optical flow motion vector change information for each frame; Generate a transition frame between two frames of image according to the pixel optical flow motion vector change information; Inter-frame interpolation optimization is performed based on the transition frames, and dynamic smoothing processing is performed to construct a delay-optimized monitoring video.

[0042] In this example, multiple high-definition surveillance cameras must be deployed. These cameras should cover every critical area of the pump room, including the entrances and exits, pipelines, valves, and equipment operating status. A variety of camera types are available, such as wide-angle and fisheye cameras, which can capture a 360-degree panoramic view of the entire pump room, ensuring comprehensive monitoring. For dimly lit or smoky environments, cameras with infrared imaging and low-light sensitivity are also recommended. Surveillance video is transmitted via the network to a video management system for centralized management. Real-time video streams can be stored, backed up, and scheduled via a cloud platform. Depending on the system's bandwidth and processing power, the video stream can be compressed and stored at different resolutions. Generally, 1080p (1920×1080) is a common resolution. For higher detail, 4K (3840×2160) resolution can be considered for video recording. The video capture frequency should be 30 frames per second (FPS) to ensure smooth recording of the pump room's dynamics. The surveillance video is segmented into multiple regions. The goal of segmentation is to divide the video footage into multiple independent sub-regions based on the pump room's functional areas and equipment layout, facilitating subsequent regional feature extraction and analysis. This step primarily relies on image segmentation technology, employing deep learning-based object detection and segmentation algorithms such as U-Net and Mask R-CNN. Specifically, the object detection algorithm first identifies various objects within the pump room, such as equipment, pipes, valves, and operators, and divides them into distinct regions. For example, a pump room may contain several key areas, such as the control room, machinery area, and discharge area, which require separate processing. Each segmented sub-region serves as an independent input to the subsequent deep learning model, enabling more accurate analysis. To ensure segmentation accuracy, a threshold can be set to filter out low-confidence regions and further refine the segmentation results. Each video region is input to a CNN, where it undergoes layer-by-layer feature extraction through multiple convolutional layers. At each layer, the convolution operation extracts a local feature map, which is then downsampled by a pooling layer to retain the most important features. In this process, feature maps of varying scales help the network capture regional details at a more granular level. In order to further enhance the performance of the network, a pre-trained model (such as ResNet, VGG) can be used as the basic model and fine-tuned on it so that the network can better adapt to the special conditions in the ash pump room environment.

[0043] During feature extraction, multiple convolutional layers can be configured, with kernel sizes in each layer ranging from 3×3, 5×5, to 7×7, ensuring the network can extract feature information at different scales. Furthermore, the network depth can be adjusted based on task requirements to better extract both high-level semantic information and detailed information. Texture analysis algorithms, such as gray-level co-occurrence matrix (GLCM) and local binary pattern (LBP), are used to analyze the texture features of different regions. These analyses can identify image regions where texture details are blurred or lost and require enhancement. Next, an adaptive enhancement algorithm (such as a detail enhancement network based on a convolutional neural network) is used to enhance the image. The key to adaptive enhancement is dynamic adjustment based on local texture requirements. For example, strong detail enhancement can be applied around equipment and pipelines, while reducing detail enhancement in background areas. Adaptive detail enhancement can enhance image texture clarity and more intuitively present details such as equipment operating status and pipeline pressure fluctuations. To avoid noise caused by excessive enhancement, a regularization term can be added to control the strength and range of the enhancement. When selecting parameters, the enhancement factor can be set between 1.2 and 2.0 to maintain a natural appearance of details. Common methods for optical flow calculation include the classic Lucas-Kanade method, the Horn-Schunck method, and deep learning-based optical flow estimation algorithms. First, the motion direction and velocity of each pixel are calculated using two consecutive image frames. Specifically, based on the image gradient information, the optical flow algorithm can estimate the motion vector of each pixel. The motion vector of each pixel is determined by its displacement between the two frames and includes both the direction and velocity of motion. To ensure the accuracy of optical flow calculations, high-precision optical flow estimation algorithms can be used, such as the FlowNet model based on convolutional neural networks or the multi-level pyramid optical flow algorithm. These methods are more adaptable to complex scenes and dynamic changes. By comparing the optical flow fields between consecutive frames, the change pattern of the optical flow vector can be extracted. For example, if the optical flow vector of a certain area shifts significantly over several consecutive frames, it may indicate that the device in that area is in operation or being operated. On the other hand, if the optical flow vector changes slightly, it may indicate that the device is stationary.

[0044] This optical flow variation provides valuable information for subsequent image transition frame generation. Deep learning models, such as convolutional LSTMs (Long Short-Term Memory), can be used to model the temporal dependencies of these optical flow variations, generating information about the optical flow motion vector changes for each frame. Based on this information, transition frames can be generated between two frames. Transition frames are generated to enhance video smoothness. In dynamic scenes, in surveillance videos, smooth transition frames help eliminate jitter or sudden changes, enhancing the visual quality. Interpolation algorithms are typically used to generate transition frames, with the most common method being optical flow-based interpolation. When generating transition frames, the calculated optical flow information is used to shift the pixels between the previous and next frames, and a new intermediate frame is generated based on the resulting motion trajectory. This method effectively smooths the transition between frames, avoiding abrupt scene changes. Finally, based on the generated transition frames, interframe interpolation optimization and dynamic smoothing are performed to construct latency-optimized surveillance videos. Dynamic smoothing can be achieved through a variety of methods, including filter-based smoothing algorithms (such as Gaussian filtering and mean filtering) or more advanced image enhancement networks based on deep learning. Through interframe interpolation and dynamic smoothing, video continuity and fluidity are improved, especially in high-dynamic-range video content. This effectively reduces motion blur and image jumps, making surveillance video clearer and more stable, thereby enhancing the intelligent remote control capabilities and automated management level of the ash pump room.

[0045] In this embodiment, step S4 includes the following steps: Based on the characteristics of multiple scene elements, the sensor parameters of the ash pump room full parameter perception map are matched, and time series monitoring synchronization processing is performed to generate visual monitoring synchronization parameters; Based on the visual monitoring synchronization parameters, the ash pump room three-dimensional topological model is mapped to all pump room parameters, and the pump room operation parameter mapping model is constructed; Perform multi-point environmental monitoring position registration based on the three-dimensional environmental intelligent monitoring matrix to obtain accurate environmental monitoring registration data; Based on the precise registration data from environmental monitoring, the pump room operation parameter mapping model is used to render environmental status changes, and real-time monitoring status evolution is updated to build a digital twin model of the pump room.

[0046] In this embodiment, effective correspondence and synchronous matching between pump room visual recognition data and multi-source sensor parameters are achieved, establishing a data mapping channel between visual perception and physical monitoring. First, the system extracts the spatial locations and identification labels of key scene elements (such as electric doors, control cabinets, water level gauges, and waterlogged areas) from a previously constructed three-dimensional topological model of the ash pump room. It also obtains corresponding sensor device information, including sensor ID, measurement type (temperature and humidity, vibration, water leakage, current, etc.), sampling frequency, and data format. Using spatial registration algorithms, such as those based on the Iterative Closest Point (ICP) method and the fusion of spatial vector direction constraints, high-precision matching between image feature points and physical sensor deployment locations is achieved. The system then aligns the sampling frequencies of each sensor and synchronizes the time series using interpolation resampling and Kalman filtering to ensure consistency between the sensor data and the image frame sequence timeline. Based on this, the system constructs "visual monitoring synchronization parameters." This parameter set includes the momentary state value of each scene element, the corresponding spatial sensor value, signal strength, and timestamp label. This completes the mapping between visual recognition, sensor monitoring, and system time, providing structured data support for the next step of 3D perception fusion. Multimodal sensor information is mapped in real time to the 3D pump room structure, enabling state visualization and data-driven operational analysis of each key component in the space. After achieving spatiotemporal synchronization between vision and sensors, the system embeds the sensor data corresponding to each scene element into its geometric point cloud structure, using the 3D topological model of the ash pump room as the primary benchmark model. Using an object-oriented 3D data annotation method, each model component is assigned a data binding structure, including the device ID, latest state value, state history queue, and anomaly markers. State transitions are represented through visual mapping and real-time material changes (such as color change and flashing). During the mapping process, the system utilizes a multi-threaded concurrent architecture to handle the synchronization of multiple data streams (such as temperature, current, and liquid level) injected into the model simultaneously. A short-period sliding window algorithm is also introduced to filter the data for timeliness, ensuring that the current mapping results reflect the real-time state. Finally, the system constructed a "pump room operation parameter mapping model", which is a three-dimensional digital carrier that integrates real-time, spatial topology, and multi-parameter and multi-source fusion. It can realize the holographic presentation of key operating data inside the pump room on the control platform.

[0047] The 3D intelligent environmental monitoring matrix is a multidimensional distributed network constructed based on the deployment of multiple environmental sensor nodes (temperature and humidity sensors, gas concentration detectors, noise monitors, water leak detectors, etc.). The system first reads the deployment coordinates of each sensor (obtained from BIM drawings or RTK calibration) and performs error correction using laser ranging or indoor UWB positioning to ensure that all node coordinates have an accuracy better than ±5 cm. The system then spatially registers this node location data with the 3D pump room model, embedding all environmental nodes within the 3D topology using spatial grid projection and minimum error fitting. Furthermore, precise spatial location information is appended to the monitoring data collected by each node to form "environmental monitoring precise registration data," including location ID, spatial coordinates, monitoring type, data timestamp, and numerical entity. This step ensures that the environmental monitoring data not only has a numerical reference dimension but also has clear spatial orientation, a critical prerequisite for subsequent state rendering and real-time variation tracking. The pump room operating parameter mapping model constructed in step S2 is integrated with the precise environmental monitoring registration data generated in step S3. The system then performs dynamic rendering of each surface area in the 3D model based on physical data. Rendering uses a state material replacement mechanism based on rule mapping: for example, red high-temperature materials are rendered in areas with excessive temperature and humidity, a flashing effect is triggered in areas where water leakage is detected, and a vibration layer is used to represent areas where noise exceeds the standard. The rendering process is driven by OpenGL or Unity3D real-time graphics engines, and is linked to real-time monitoring data streams through the WebSocket protocol to ensure that the screen presentation is updated synchronously with the sensor data. In addition, the system also introduces a state evolution tracking mechanism, which compares the current and historical state differences based on sliding window time series data, scores the degree of state variation, and drives the dynamic change process of the corresponding area. The final "pump room digital twin model" has multiple capabilities such as complete topological structure, precise physical parameter mapping, real-time visualization of environmental status, and perception of abnormal evolution, providing a powerful and intuitive data foundation and operable interface for remote automation control, abnormal warning, and decision-making scheduling of the pump room.

[0048] In this embodiment, the specific steps of step S5 are: Conduct virtual full-process operation simulation on the pump room digital twin model to generate pump room operation simulation data; Calculate the desulfurization tower slurry density, flue gas flow rate, and limestone consumption based on the pump room operation simulation data, and obtain upstream process parameters by fitting; Perform sliding window ash generation prediction on upstream process parameters to generate ash generation prediction values for different window periods; Based on the ash production prediction value, dynamic start-stop timing allocation of multiple pumps is performed, and intelligent scheduling of operating load parameters is performed to generate a remote scheduling decision execution instruction set; Perform real-time remote drive control of the pump room based on remote dispatch decision execution instruction sets and collect all equipment operating parameters; Extracting preset equipment operating parameters in each execution table based on the remote scheduling decision execution instruction set; Based on the preset equipment operating parameters, real-time equipment status monitoring is performed on all equipment operating parameters, and scheduling deviation detection is performed to mark the scheduling control deviation equipment.

[0049] In this embodiment, a dynamic physical behavior model is constructed based on the three-dimensional topology of the pump room, sensor physical characteristics, and operating condition records. Using a discrete event-driven approach coupled with continuous state equations, the system simulates various typical operating processes, including pump start-up and shutdown, motor power load changes, water pressure fluctuations, and ash concentration variations. By setting virtual operating condition inputs, such as pump start-up sequence, slurry concentration changes, and external input parameter perturbations, the system runs hundreds of simulations to generate "pump room operation simulation data" with timeliness and random perturbation factors. This data includes pump operation curves, electrical power consumption fluctuations, liquid level trends, and pipeline flow dynamics. To ensure data credibility, the system uses historical operating data for simulation calibration and introduces a parameter tuning mechanism based on Bayesian optimization to keep the simulation output within an error rate of ±3%, ensuring engineering applicability for subsequent upstream process parameter derivation and scheduling model training. Key parameters of the upstream processes served by the ash pump room, such as the desulfurization tower, are reverse-engineered and modeled to extract information about their operating conditions. The system first analyzes variables related to the ash pump in the simulation data, such as liquid flow rate, concentration changes, motor load, and fluid temperature. Using multivariate regression and physical causal modeling, it reconstructs the core parameters required for the desulfurization tower slurry treatment process, including slurry density (ρ), flue gas flow (Q), and limestone consumption (L). Slurry density is inferred from its concentration distribution using a sonic conduction model, flue gas flow is predicted using wind pressure differential modeling, and limestone consumption is estimated by combining slurry pH changes and ash output. The system then fits these parameters to actual operating process indicators and generates a dynamic upstream process parameter set using a sliding fitting window (window size ranging from 10 minutes to 1 hour). This enables "digital deduction of the process environment" from pump room status to the desulfurization tower process, providing a data interface for predictive control and cross-process collaboration. Dynamic features extracted from the upstream process parameters are used as input to perform short-term prediction and trend analysis of ash production using a sliding window mechanism. Specifically, the system uses variables such as historical slurry density, flue gas flow rate, and limestone dosage to form an input feature matrix, and employs a multi-model fusion prediction mechanism (such as support vector regression (SVR), LSTM neural network, and XGBoost regression) to model and predict ash production at different time scales (e.g., 5 minutes, 15 minutes, 30 minutes, and 1 hour). The sliding window setting enables the prediction system to have real-time update and rapid response capabilities, and can adapt to fluctuations in production caused by sudden changes in pump room operating conditions. The final output is a set of time series prediction values used to describe the ash production trend over several future time periods. After model cross-validation, the prediction accuracy remains within the range of MAPE ≤ 6%. The prediction results will serve as an important basis for subsequent pump group start-up and shutdown strategies and load scheduling decisions.

[0050] An intelligent scheduling engine enables adaptive optimization of pump group operating strategies. The scheduling engine first identifies ash load trends within the forecast period and constructs a multi-objective optimization model based on parameters such as the rated power, start-stop time response characteristics, and service life status of different pumps. With the goals of minimizing energy consumption, balancing loads, and minimizing pump wear, a hybrid strategy of particle swarm optimization (PSO) and genetic algorithms (GA) is used to generate a multi-pump start-stop schedule. The system then further incorporates pipeline fluid dynamics simulation to predict the balance of liquid level and flow rate within each time period, adjusting parameters such as the operating frequency and current load of each pump. Ultimately, it outputs a "remote scheduling decision execution instruction set" that includes multiple items such as start-stop time points, power settings, upper and lower current limits, and standby pump pre-start time. This instruction set is highly real-time and flexible, enabling proactive pump room management based on ash prediction.

[0051] In this embodiment, the specific steps of step S6 are: Conduct deviation state time series statistics on the dispatch control deviation equipment and generate time series deviation state parameter curve; Identify fault signs based on the time series deviation state parameter curve, and perform equipment abnormality risk assessment and prediction to obtain the equipment abnormality risk assessment value; Make risk countermeasure decisions based on the equipment abnormality risk assessment value and build a risk countermeasure strategy; Adaptively adjust the execution instructions of the remote scheduling decision execution instruction set based on the risk countermeasure strategy, and build intelligent adaptive adjustment instructions; Based on intelligent adaptive adjustment instructions to perform remote automation control operations in the pump room.

[0052] In this embodiment, time-series statistics are performed on the continuous state data of devices experiencing deviated dispatch control to identify trends in their operating parameters, providing a foundation for subsequent fault symptom identification. The system first extracts key operating parameters from devices marked as "deviated dispatch control," such as current, frequency response, fluid output pressure, vibration spectrum, and operating temperature. These parameters are sampled and statistically analyzed over rolling time windows (e.g., 5-minute, 15-minute, and 30-minute periods). For each parameter, the system constructs a standard deviation curve, a rate of change curve, and a cumulative deviation curve, using an exponential moving average (EMA) algorithm to smooth noise and maintain trends. Furthermore, the coefficient of variation (CV) is introduced to uniformly quantify fluctuation intensity, ensuring statistical consistency across different parameters. Finally, a "time-series deviation state parameter curve" is plotted to visualize the persistent, intermittent, or sudden behavior of device deviations, laying the data foundation for further abnormality risk modeling and prediction. This step ensures early detection of abnormal trends in devices and improves the proactive response of the overall control system. Based on these time-series deviation state parameter curves, potential fault symptoms are identified and the operational abnormality risk faced by the device is assessed. The system analyzes the curve shape by constructing a "fault precursor identification model." This model incorporates characteristic patterns of typical fault samples, such as sustained current rise, frequency response hysteresis, and sudden drop in output pressure. It then uses a combination of a support vector machine (SVM) and a decision tree classifier to match and identify the current deviation curve. Each fault type (such as motor winding overheating, impeller blockage, and bearing fatigue) is assigned a corresponding characteristic threshold range. The system calculates the similarity between the current curve shape and each fault template and outputs a classification result. Based on fault symptom identification, a Bayesian dynamic risk assessment model is further employed. This model combines the current deviation level, duration, historical equipment health status, and prior probability of failure to output a device abnormality risk assessment value between 0 and 1. The closer the assessment value is to 1, the more likely the equipment will experience a serious failure or operational instability in the near future. This step achieved an accuracy rate exceeding 93% on the experimental dataset, providing a quantitative basis for subsequent risk prevention and control.

[0053] After equipment anomaly risks are identified and quantified, targeted countermeasures are developed based on the risk level to reduce the risk of equipment failure and maintain overall system stability. The system sets risk thresholds, such as low (<0.3), medium (0.3–0.6), and high (>0.6), corresponding to corresponding countermeasures of varying strength. In the medium-risk range, the system initiates lightweight measures, such as adjusting operating frequency and reducing flow pressure. In high-risk conditions, the system triggers stronger measures, such as switching on backup pumps, slowing down primary equipment, and initiating alert notifications. Strategies are developed based on a multi-objective control optimization model, which optimizes for "minimizing load fluctuations, minimizing equipment switching, and minimizing system energy consumption." Simulations verify system stability after strategy execution. Each risk-mitigation strategy includes information such as the device number, target adjustment parameters, execution time, and whether to trigger backup equipment, all output as a strategy set. This step ensures that the system can proactively intervene before equipment risk signs appear, forming a feedforward control loop and improving the robustness and autonomous regulation capabilities of the pump room system. This system is further integrated into the original remote dispatch control logic to form "intelligent adaptive adjustment instructions" that can adjust in real time as device status changes. First, the system analyzes the structure of the currently issued remote dispatch execution instruction set to identify the target device's current control parameters and operating status. Subsequently, the adjustment module partially replaces or dynamically weights and modifies core parameters within the instruction (such as start / stop timing, current limit, and flow rate setting). For example, if the device's abnormality risk value reaches 0.7, the system reduces its operating hours by 10% and introduces a dynamic load transfer mechanism for backup pumps. The entire adjustment process relies on a reinforcement learning model to optimize execution, ensuring that the implemented strategy achieves better control than the original solution. Intelligent adaptive instructions support a dynamic rollback mechanism, allowing system stability to decline after implementation, restoring the system to the previous optimal state. Finally, the intelligent instruction set is pushed to the device control terminal using standard Modbus TCP or OPC UA protocols to ensure execution compatibility and real-time execution.

[0054] After constructing and validating the intelligent adaptive adjustment instructions, the system immediately remotely executes them, ensuring that the entire pump room control system dynamically responds to the current risk status. This step consists of three parts: instruction issuance, execution verification, and operational monitoring. The system distributes the intelligent instruction set in real time to each control device (such as the inverter, PLC, and remote I / O module) via the edge control node. This delivery process is confirmed via a two-way handshake protocol between the main control platform and the device layer. After the device executes the adjusted operating parameters, the system collects its response feedback in real time and verifies its consistency with the expected parameters. Simultaneously, the monitoring system updates the device operating data stream in 1-second increments and assesses the actual impact of the current adjustment strategy on system load, energy consumption, and ash conveying efficiency in real time. If any deviation is detected, an iterative strategy optimization process is initiated. This mechanism implements a closed-loop remote automated control process with learning, adaptive, and self-correcting capabilities, significantly improving the intelligent level of pump room operation, safety and fault tolerance, and maintenance efficiency.

[0055] In this embodiment, an intelligent remote automation control system for an ash pump room is provided, which is used to execute the intelligent remote automation control method for an ash pump room as described above, including: The operating parameter perception module is used to collect multi-dimensional pump room operating parameters in real time based on multi-source heterogeneous sensors, and perform collaborative encoding and multi-dimensional data state change perception to build a full parameter perception map of the ash pump room; The global environmental monitoring module is used to collect environmental monitoring data streams from multiple locations, identify spatial deployment distribution, and perform global environmental monitoring calculations to build a three-dimensional environmental intelligent monitoring matrix; The 3D point cloud modeling module is used to obtain all-round monitoring videos of the ash pump room, identify key scene elements, perform 3D topological point cloud modeling, and construct a 3D topological model of the ash pump room; The digital twin module is used to map the full-parameter perception of the ash pump room to the 3D topological model of the pump room based on the full-parameter perception map of the ash pump room, and to update the real-time monitoring status evolution based on the three-dimensional environment intelligent monitoring matrix to build a digital twin model of the pump room; The operation simulation module is used to simulate the virtual full-process operation of the pump room digital twin model, perform real-time equipment scheduling deviation detection, and mark the scheduling control deviation equipment; The adaptive adjustment module is used to identify the signs of equipment failure when the dispatching control deviates, and then adaptively adjust the execution instructions to perform remote automation control operations in the pump room.

[0056] The present invention is therefore intended to be illustrative and non-restrictive in all respects, with the scope of the invention being defined by the appended claims rather than the foregoing description, and all changes that come within the meaning and range of equivalents of the application documents are intended to be embraced therein.

[0057] The foregoing description is intended only to provide specific embodiments of the present invention, which are intended to enable those skilled in the art to understand and implement the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not intended to be limited to the embodiments shown herein, but is to be construed in the widest manner consistent with the principles and novel features disclosed herein.

Claims

1. An intelligent remote automation control method for ash pump room, characterized in that: The following steps are involved: Step S1: Based on multi-source heterogeneous sensors, multi-dimensional pump room operating parameters are collected in real time, and collaborative coding and multi-dimensional data state change perception are performed to construct a full parameter perception map of the ash pump room; Step S2: Collect environmental monitoring data streams from multiple locations, identify spatial deployment distribution, and perform global environmental monitoring calculations to construct a three-dimensional environmental intelligent monitoring matrix; Step S3: Obtain all-round monitoring video of the ash pump room, identify key scene elements and perform three-dimensional topological point cloud modeling to construct a three-dimensional topological model of the ash pump room; Step S4: mapping the full-parameter perception of the ash pump room to the 3D topological model of the ash pump room according to the full-parameter perception map of the ash pump room, and performing real-time monitoring status evolution updates based on the three-dimensional environment intelligent monitoring matrix to construct a digital twin model of the pump room; Step S5: Perform virtual full-process operation simulation on the pump room digital twin model, perform real-time equipment scheduling deviation detection, and mark the scheduling control deviation equipment; Step S6: Identify the signs of equipment failure in the dispatching control deviation, and then perform adaptive adjustment of the execution instructions to perform the remote automation control operation of the pump room.

2. The intelligent remote automation control method for ash pump room according to claim 1, characterized in that: The specific steps of step S1 are: Based on multi-source heterogeneous sensors, real-time collection of pump vibration spectrum, bearing temperature, motor power factor, and fluid pressure pulsation is carried out to obtain multi-dimensional pump room operating parameters; Calculate the multi-parameter acquisition frequency of multi-source heterogeneous sensors and extract the parameter acquisition frequency of different sensors; Performing multi-device signal synchronization optimization on the multi-dimensional pump room operating parameters according to the parameter acquisition frequency to generate time-series synchronized multi-dimensional parameters; Identify communication protocols based on multi-source heterogeneous sensors and extract the communication protocol of each sensor device; Perform protocol conflict analysis on the communication protocol of each sensor device and identify devices with data interaction conflicts; Perform protocol coordination conversion on devices with data interaction conflicts to obtain conflict resolution and conversion strategies; Based on the conflict resolution and conversion strategy, the time-series synchronization multi-dimensional parameters are collaboratively encoded, and the multi-dimensional data state change perception is performed to construct a full parameter perception map of the ash pump room.

3. The intelligent remote automation control method for ash pump room according to claim 2, characterized in that: The specific steps of performing multi-device signal synchronization optimization on the multi-dimensional pump room operating parameters according to the parameter acquisition frequency to generate time-series synchronized multi-dimensional parameters are as follows: Obtain the preset main control system clock signal; Calculating a clock signal of each device of the multi-source heterogeneous sensor; Calculating the clock deviation of the main control system clock signal for each device according to the clock signal of each device to generate a clock deviation value for each device; Dynamically compensate for the deviation of multi-dimensional pump room operating parameters based on the clock deviation value of each device to obtain signal deviation compensated operating parameters; Performing a key device level assessment on the multi-source heterogeneous sensors to obtain priorities of multiple devices; Analyze the equipment signal change trend based on multi-dimensional pump room operating parameters to generate signal change trends for different equipment; Adaptive frequency acquisition adjustment is performed based on the parameter acquisition frequency, the priority of multiple devices and the signal change trend, and multi-device signal synchronization optimization is performed to generate timing synchronization multi-dimensional parameters.

4. The intelligent remote automation control method for ash pump room according to claim 1, characterized in that: The specific steps of step S2 are: The ash pump room is equipped with multiple environmental monitoring sensors, including temperature and humidity sensors, gas concentration detectors, noise monitors, and water leakage detection sensors; Collecting environmental monitoring data streams at multiple locations based on the environmental monitoring sensor; Performing multimodal environmental change tracking on the environmental monitoring data stream to generate multiple environmental data change curves; Calculating the spatial deployment position of the environmental monitoring sensor and extracting the position coordinates of multiple environmental monitoring nodes; Based on the location coordinates of multiple environmental monitoring nodes, the spatial deployment distribution of multiple environmental data change curves is identified, and global environmental monitoring calculations are performed to construct a three-dimensional environmental intelligent monitoring matrix.

5. The intelligent remote automation control method for ash pump room according to claim 1, characterized in that: The specific steps of step S3 are: Obtain a comprehensive surveillance video of the ash pump room; enhance image details and optimize timing frame delay in the comprehensive surveillance video of the ash pump room to construct a delay-optimized surveillance video; The delay-optimized surveillance video is used to divide the pump room area and identify key scene elements. The indicator lights of the pump control cabinet, the switch status of the electric door, the water level scale, and the water accumulation area are marked to obtain multiple scene element features. Identify regional building layout feature points on delay-optimized surveillance videos and mark multiple layout feature points; Perform three-dimensional topological point cloud modeling on multiple scene element features and multiple layout feature points to construct a three-dimensional topological model of the ash pump room.

6. The intelligent remote automation control method for ash pump room according to claim 5, characterized in that: The specific steps of obtaining a full-scale monitoring video of the ash pump room, enhancing image details of the full-scale monitoring video of the ash pump room, and optimizing the timing frame delay to construct a delay-optimized monitoring video are as follows: Obtain all-round monitoring video of the ash pump room; Segment the ash pump room's all-around monitoring video into multiple areas to obtain multiple different video areas; Based on the multi-scale convolutional neural network, multi-convolutional layer feature extraction is performed on multiple different video regions to obtain different scale image features of each region; Performing texture detail requirement analysis on the image features of different scales and performing adaptive detail enhancement to obtain an adaptive detail enhanced video; Perform global pixel optical flow calculation frame by frame on the adaptive detail enhanced video to obtain the pixel motion direction and speed of each frame image; Perform deep optical flow change evolution analysis based on the pixel motion direction and speed to generate pixel optical flow motion vector change information for each frame; Generate a transition frame between two frames of image according to the pixel optical flow motion vector change information; Inter-frame interpolation optimization is performed based on the transition frames, and dynamic smoothing processing is performed to construct a delay-optimized monitoring video.

7. The intelligent remote automation control method for ash pump room according to claim 1, characterized in that: The specific steps of step S4 are: Based on the characteristics of multiple scene elements, the sensor parameters of the ash pump room full parameter perception map are matched, and time series monitoring synchronization processing is performed to generate visual monitoring synchronization parameters; Based on the visual monitoring synchronization parameters, the ash pump room three-dimensional topological model is mapped to all pump room parameters, and the pump room operation parameter mapping model is constructed; Perform multi-point environmental monitoring position registration based on the three-dimensional environmental intelligent monitoring matrix to obtain accurate environmental monitoring registration data; Based on the precise registration data from environmental monitoring, the pump room operation parameter mapping model is used to render environmental status changes, and real-time monitoring status evolution is updated to build a digital twin model of the pump room.

8. The intelligent remote automation control method for ash pump room according to claim 1, characterized in that: The specific steps of step S5 are: Conduct virtual full-process operation simulation on the pump room digital twin model to generate pump room operation simulation data; Calculate the desulfurization tower slurry density, flue gas flow rate, and limestone consumption based on the pump room operation simulation data, and obtain upstream process parameters by fitting; Perform sliding window ash generation prediction on upstream process parameters to generate ash generation prediction values for different window periods; Based on the ash production prediction value, dynamic start-stop timing allocation of multiple pumps is performed, and intelligent scheduling of operating load parameters is performed to generate a remote scheduling decision execution instruction set; Perform real-time remote drive control of the pump room based on remote dispatch decision execution instruction sets and collect all equipment operating parameters; Extracting preset equipment operating parameters in each execution table based on the remote scheduling decision execution instruction set; Based on the preset equipment operating parameters, real-time equipment status monitoring is performed on all equipment operating parameters, and scheduling deviation detection is performed to mark the scheduling control deviation equipment.

9. The intelligent remote automation control method for ash pump room according to claim 1, characterized in that: The specific steps of step S6 are: Conduct deviation state time series statistics on the dispatch control deviation equipment and generate time series deviation state parameter curve; Identify fault signs based on the time series deviation state parameter curve, and perform equipment abnormality risk assessment and prediction to obtain the equipment abnormality risk assessment value; Make risk countermeasure decisions based on the equipment abnormality risk assessment value and build a risk countermeasure strategy; Adaptively adjust the execution instructions of the remote scheduling decision execution instruction set based on the risk countermeasure strategy, and build intelligent adaptive adjustment instructions; Based on intelligent adaptive adjustment instructions to perform remote automation control operations in the pump room.

10. An intelligent remote automation control system for an ash pump room, characterized in that: The method for executing the intelligent remote automation control method for an ash pump room according to claim 1 comprises: The operating parameter perception module is used to collect multi-dimensional pump room operating parameters in real time based on multi-source heterogeneous sensors, and perform collaborative encoding and multi-dimensional data state change perception to build a full parameter perception map of the ash pump room; The global environmental monitoring module is used to collect environmental monitoring data streams from multiple locations, identify spatial deployment distribution, and perform global environmental monitoring calculations to build a three-dimensional environmental intelligent monitoring matrix; The 3D point cloud modeling module is used to obtain all-round monitoring videos of the ash pump room, identify key scene elements, perform 3D topological point cloud modeling, and construct a 3D topological model of the ash pump room; The digital twin module is used to map the full-parameter perception of the ash pump room to the 3D topological model of the pump room based on the full-parameter perception map of the ash pump room, and to update the real-time monitoring status evolution based on the three-dimensional environment intelligent monitoring matrix to build a digital twin model of the pump room; The operation simulation module is used to simulate the virtual full-process operation of the pump room digital twin model, perform real-time equipment scheduling deviation detection, and mark the scheduling control deviation equipment; The adaptive adjustment module is used to identify the signs of equipment failure when the dispatching control deviates, and then adaptively adjust the execution instructions to perform remote automation control operations in the pump room.

Images