Multi-sensor fusion monitoring method and system for setting machine waste heat recovery system

Abstract

This invention relates to the field of waste heat recovery technology in textile machinery, and more particularly to a multi-sensor fusion monitoring method and system for waste heat recovery systems in stenters. The method acquires data from multiple sensors and performs spatiotemporal alignment and fusion calculations to obtain a comprehensive system status assessment result. Based on this result, anomaly identification and multi-objective optimization are performed to generate a control execution scheme. After the scheme is executed, response data is collected to analyze control effect deviations, and relevant parameters are adaptively adjusted accordingly. This invention achieves precise monitoring and intelligent control of the waste heat recovery system status, improving system operational stability and energy efficiency.

Description

Technical Field

[0001] This invention relates to the field of waste heat recovery technology in textile machinery, and in particular to a multi-sensor fusion monitoring method and system for waste heat recovery systems of setting machines. Background Technology

[0002] In the textile printing and dyeing industry, stenters are among the major energy-consuming pieces of equipment, and the high-temperature exhaust gases they emit contain a significant amount of waste heat. To improve energy efficiency and reduce production costs, waste heat recovery systems are widely used in stenter exhaust gas treatment processes. These systems typically recover heat energy from the exhaust gases using devices such as heat exchangers, which can then be used to heat fresh air, process water, or for other purposes, thereby achieving energy recycling.

[0003] To ensure the stable and efficient operation of waste heat recovery systems and prevent equipment failures, real-time status monitoring is standard practice in the industry. Existing monitoring technologies mainly rely on deploying several independent temperature, pressure, and flow sensors at key nodes of the system, such as the air inlet, outlet, heat exchanger surface, and circulation pipelines. These sensors each collect local parameter data at their monitoring points and transmit it to the central control system for centralized display and threshold alarms.

[0004] However, this conventional monitoring method based on single-point independent sensing and simple threshold judgment has significant drawbacks. Because sensors are physically distributed and data acquisition is asynchronous, the acquired data is inherently multi-source, heterogeneous, and spatiotemporally discrete. It cannot effectively integrate these scattered and asynchronous data points to construct a global, coherent panoramic view of the system's thermal state. This results in a one-sided and fragmented assessment of the system's operating status, making it difficult to accurately reflect the complex dynamic coupling relationships within the system.

[0005] Furthermore, when a monitored parameter exceeds a preset threshold, conventional solutions typically only trigger isolated alarms, failing to accurately pinpoint the root cause of the anomaly or provide a systematic optimization control strategy. This is because an anomaly in a single parameter may be caused by multiple interrelated factors. The lack of integrated analysis of the physical constraints between multiple parameters results in low accuracy in fault diagnosis, and control adjustments often rely on operator experience, employing a "trial-and-error" approach to single-parameter adjustments, making it difficult to achieve synergistic optimization of the system's overall energy efficiency and stability. This lagging and one-sided monitoring method restricts further performance improvements and long-term reliable operation of waste heat recovery systems. Summary of the Invention

[0006] The present invention provides a multi-sensor fusion monitoring method and system for a waste heat recovery system of a stenter, which can solve the problems in the prior art.

[0007] A first aspect of the present invention provides a multi-sensor fusion monitoring method for a waste heat recovery system of a stenter, comprising:

[0008] Multi-source heterogeneous sensing data characterizing the thermal state of the waste heat recovery system of the stenter is acquired by multiple sensors distributed at different spatial locations. Based on the spatial location information and data acquisition time information of each sensor, the multi-source heterogeneous sensing data is spatiotemporally aligned to obtain spatiotemporally aligned multi-source sensing data.

[0009] Based on the physical correlation constraints between the spatiotemporally aligned multi-source sensor data, the spatiotemporally aligned multi-source sensor data are fused and calculated to obtain the system comprehensive state evaluation result;

[0010] Based on the comprehensive system status assessment results and the preset system normal operation status boundary, deviation is calculated to determine abnormal feature description information. Based on the abnormal feature description information and the comprehensive system status assessment results, multi-objective optimization is performed to generate a control execution scheme for the waste heat recovery system.

[0011] After executing the control execution scheme, system response data is collected. The system response data is compared and analyzed with the expected target of the control execution scheme to obtain control effect deviation information. The control effect deviation information is used to adaptively adjust the parameters in the fusion calculation and the multi-objective optimization solution.

[0012] Multi-source heterogeneous sensing data characterizing the thermal state of the waste heat recovery system of the stenter is acquired by multiple sensors distributed at different spatial locations. Based on the spatial location information and data acquisition time information of each sensor, the multi-source heterogeneous sensing data is spatiotemporally aligned to obtain spatiotemporally aligned multi-source sensing data, including:

[0013] When acquiring multi-source heterogeneous sensing data characterizing the thermal state of the system using multiple sensors distributed at different spatial locations in the waste heat recovery system of the stenter, the spatial coordinate position information and data acquisition time information of each sensor are recorded simultaneously.

[0014] Based on the spatial coordinate location information of each sensor and the pipeline topology information of the waste heat recovery system, the spatial distribution relationship characteristics of the sensors are determined. These characteristics describe the spatial distance relationships and heat transfer path relationships between the sensors.

[0015] Based on the spatial distribution characteristics of the sensors and the data acquisition time information, the time delay compensation between sensor data at different spatial locations is calculated;

[0016] The timestamp correction process is performed on the multi-source heterogeneous sensing data based on the time delay compensation amount, and the spatial coordinate normalization process is performed on the timestamp-corrected multi-source heterogeneous sensing data according to the spatial distribution relationship characteristics of the sensors, so as to obtain spatiotemporally aligned multi-source sensing data.

[0017] Based on the physical correlation constraints between the spatiotemporally aligned multi-source sensor data, the spatiotemporally aligned multi-source sensor data are fused and calculated to obtain the system comprehensive state evaluation result, including:

[0018] Based on the energy transfer coupling relationship between the spatiotemporally aligned multi-source sensor data, a state evolution relationship describing the thermal energy flow process of the waste heat recovery system is determined. This state evolution relationship represents the constraint effect of physical correlation constraints on the system state changes.

[0019] The state evolution relationship is used to perform state estimation calculation on the spatiotemporally aligned multi-source sensing data to predict the theoretical state value of each sensor's monitoring position at the current moment, thus obtaining predicted state data;

[0020] The spatiotemporally aligned multi-source sensor data and the predicted state data are subjected to deviation analysis to calculate the deviation between the actual measured values ​​of each sensor and the predicted state data.

[0021] The measurement reliability of each sensor's measurement data is quantitatively evaluated based on the aforementioned deviation, resulting in a reliability evaluation index for each sensor.

[0022] A dynamic weight allocation strategy is determined based on the credibility assessment index, and the spatiotemporally aligned multi-source sensor data is fused and calculated according to the dynamic weight allocation strategy to obtain the system comprehensive state assessment result.

[0023] A dynamic weight allocation strategy is determined based on the reliability assessment index, and the spatiotemporally aligned multi-source sensor data is fused and calculated according to the dynamic weight allocation strategy to obtain a comprehensive system state assessment result, including:

[0024] Based on the credibility assessment index, it is determined that there are sensor data with measurement anomalies in the spatiotemporally aligned multi-source sensing data.

[0025] The reliability evaluation index corresponding to the sensor data with measurement anomalies is suppressed using a nonlinear decay function to obtain a corrected reliability evaluation index;

[0026] Based on the revised reliability assessment index, the fusion weight coefficient of each sensor data is calculated, and the dynamic weight allocation strategy is determined through the fusion weight coefficient.

[0027] According to the dynamic weight allocation strategy, the spatially adjacent sensor data in the spatiotemporally aligned multi-source sensing data are locally weighted and fused to obtain the regional state value. The fusion weight coefficient of the local weighted fusion is adjusted according to the spatial position relationship of the sensors.

[0028] A global weighted fusion of all regional state values ​​is performed to obtain the comprehensive system state assessment result. The fusion weight coefficient of the global weighted fusion is adjusted according to the data integrity of the regional state values.

[0029] Based on the comprehensive system status assessment results and the preset normal system operating status boundaries, deviation calculations are performed to determine abnormal feature description information. Based on the abnormal feature description information and the comprehensive system status assessment results, multi-objective optimization is performed to generate a control execution scheme for the waste heat recovery system, including:

[0030] The system comprehensive status assessment result is compared with the preset system normal operation status boundary, the deviation value of the system comprehensive status assessment result from the system normal operation status boundary is calculated, and the abnormal feature description information is determined based on the deviation value;

[0031] Based on the description of the abnormal features, the key influencing factors causing state deviation in the waste heat recovery system are identified, and a state callback path planning strategy is determined for the key influencing factors.

[0032] Based on the state callback path planning strategy and the system comprehensive state evaluation results, a multi-objective optimization problem is determined, which includes a system energy efficiency optimization objective function and a state deviation minimization objective function. The system energy efficiency optimization objective function takes maximizing waste heat recovery efficiency as the optimization direction, and the state deviation minimization objective function takes minimizing the deviation value as the optimization direction.

[0033] The multi-objective optimization problem is solved to obtain the optimal values ​​of control variables that satisfy the comprehensive optimality of multiple objective functions. Based on these control variable values, a control execution scheme for the waste heat recovery system is generated.

[0034] Based on the anomaly characteristic description information, the key influencing factors causing state deviation in the waste heat recovery system are identified, and a state callback path planning strategy is determined for the key influencing factors, including:

[0035] The abnormal feature description information is parsed to determine the feature vector describing the system state deviation characteristics. The feature vector includes the spatial distribution characteristics of the state deviation and the temporal evolution characteristics of the state deviation.

[0036] Based on the causal correlation analysis between the eigenvectors and the control variables of the waste heat recovery system, the key influencing factors that play a dominant role in state deviation are identified;

[0037] For the aforementioned key influencing factors, a dynamic transmission relationship between the control variable adjustment action and the system state change response is determined. This dynamic transmission relationship represents the process by which the system state evolves from the current deviation state to the boundary of the normal operating state after the control variable is adjusted.

[0038] Based on the dynamic transmission relationship, multiple candidate adjustment paths are determined to restore the system state from the current deviation state to the normal operating state boundary. The multiple candidate adjustment paths are evaluated, and the adjustment path with the shortest state recovery time and the smallest adjustment magnitude of the control variable is selected as the state callback path planning strategy.

[0039] After executing the control execution scheme, system response data is collected. The system response data is then compared and analyzed with the expected target of the control execution scheme to obtain control effect deviation information, including:

[0040] After executing the control execution scheme, system response data of the waste heat recovery system is collected within a preset response observation time window;

[0041] Temporal features are extracted from the system response data to obtain transient response features and steady-state response features.

[0042] Based on the transient response characteristics and the steady-state response characteristics, a system response integrity evaluation index is determined. The system response integrity evaluation index is used to quantify the completeness of the system response data in representing the control execution process.

[0043] The transient response characteristics are compared with the dynamic performance indicators in the expected target to calculate the transient process deviation, and the steady-state response characteristics are compared with the steady-state performance indicators in the expected target to calculate the steady-state result deviation.

[0044] The transient process deviation and the steady-state result deviation are fused and calculated based on the system response integrity evaluation index to obtain the control effect deviation information.

[0045] A second aspect of the present invention provides a multi-sensor fusion monitoring system for a waste heat recovery system of a stenter, comprising:

[0046] The data alignment unit is used to acquire multi-source heterogeneous sensing data characterizing the thermal state of the system through multiple sensors distributed in different spatial locations of the waste heat recovery system of the stenter. Based on the spatial location information and data acquisition time information of each sensor, the multi-source heterogeneous sensing data is spatiotemporally aligned to obtain spatiotemporally aligned multi-source sensing data.

[0047] The data fusion unit is used to perform fusion calculations on the spatiotemporally aligned multi-source sensor data according to the physical association constraints between the spatiotemporally aligned multi-source sensor data, and obtain the system comprehensive state evaluation result.

[0048] The optimization control unit is used to calculate the deviation between the system comprehensive state assessment result and the preset system normal operation state boundary to determine the abnormal feature description information, and to perform multi-objective optimization solution based on the abnormal feature description information and the system comprehensive state assessment result to generate a control execution scheme for the waste heat recovery system.

[0049] The feedback adjustment unit is used to collect system response data after executing the control execution scheme, compare and analyze the system response data with the expected target of the control execution scheme, and obtain control effect deviation information. The control effect deviation information is used to adaptively adjust the parameters in the fusion calculation and the multi-objective optimization solution.

[0050] A third aspect of the present invention provides an electronic device, comprising:

[0051] processor;

[0052] Memory used to store processor-executable instructions;

[0053] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0054] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0055] This invention employs multi-sensor fusion technology to monitor the waste heat recovery system of a stenter, significantly improving the comprehensiveness and accuracy of system status perception. Multiple sensors distributed at different spatial locations within the system can simultaneously collect multi-source heterogeneous thermal state data. By performing spatiotemporal alignment processing on this data, information bias caused by differences in sensor location and acquisition timing is effectively eliminated, constructing a unified spatiotemporal foundation for system status observation and providing reliable data support for subsequent precise analysis.

[0056] This invention, based on spatiotemporally aligned multi-source sensor data, performs fusion calculations according to their inherent physical correlation constraints, generating more accurate and robust comprehensive system status assessment results. This process overcomes the limitations of single-sensor information being one-sided and susceptible to interference, achieving in-depth insight and quantitative assessment of the overall system operation, making the determination of system energy efficiency, stability, and potential faults more scientific and reliable.

[0057] This invention, by comparing the comprehensive state assessment results with preset normal operating boundaries, can quickly identify and quantify abnormal deviations in the system state, thereby generating information describing the abnormal characteristics. Based on this characteristic information and the comprehensive assessment results, multi-objective optimization solutions can be performed to automatically generate control execution schemes that take into account multiple objectives such as energy efficiency improvement, operational safety, and equipment lifespan. This realizes a closed-loop link from state perception to optimized control, significantly enhancing the system's adaptive regulation capability in the face of complex operating conditions and abnormal states.

[0058] This invention, after executing the control scheme, collects system response data and compares it with the expected target to obtain control effect deviation information. This deviation information is then used to adaptively adjust key parameters in data fusion and optimization, forming a dynamic closed loop of continuous learning and optimization. This enables the monitoring system to continuously adapt to dynamic factors such as equipment aging and environmental changes, maintaining high-precision state assessment and efficient control performance over the long term, ultimately achieving long-term, stable, and efficient optimal operation of the waste heat recovery system. Attached Figure Description

[0059] Figure 1 This is a flowchart illustrating the multi-sensor fusion monitoring method for the waste heat recovery system of the stenter according to an embodiment of the present invention.

[0060] Figure 2 This is a schematic diagram of the process for determining the overall system status assessment result according to an embodiment of the present invention. Detailed Implementation

[0061] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0062] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0063] Figure 1 This is a flowchart illustrating the multi-sensor fusion monitoring method for the waste heat recovery system of the stenter according to an embodiment of the present invention. Figure 1 As shown, the multi-sensor fusion monitoring method for the waste heat recovery system of the stenter includes:

[0064] Multi-source heterogeneous sensing data characterizing the thermal state of the waste heat recovery system of the stenter is acquired by multiple sensors distributed at different spatial locations. Based on the spatial location information and data acquisition time information of each sensor, the multi-source heterogeneous sensing data is spatiotemporally aligned to obtain spatiotemporally aligned multi-source sensing data.

[0065] Based on the physical correlation constraints between the spatiotemporally aligned multi-source sensor data, the spatiotemporally aligned multi-source sensor data are fused and calculated to obtain the system comprehensive state evaluation result;

[0066] Based on the comprehensive system status assessment results and the preset system normal operation status boundary, deviation is calculated to determine abnormal feature description information. Based on the abnormal feature description information and the comprehensive system status assessment results, multi-objective optimization is performed to generate a control execution scheme for the waste heat recovery system.

[0067] After executing the control execution scheme, system response data is collected. The system response data is compared and analyzed with the expected target of the control execution scheme to obtain control effect deviation information. The control effect deviation information is used to adaptively adjust the parameters in the fusion calculation and the multi-objective optimization solution.

[0068] Multi-source heterogeneous sensing data characterizing the thermal state of the waste heat recovery system of the stenter is acquired by multiple sensors distributed at different spatial locations. Based on the spatial location information and data acquisition time information of each sensor, the multi-source heterogeneous sensing data is spatiotemporally aligned to obtain spatiotemporally aligned multi-source sensing data, including:

[0069] When acquiring multi-source heterogeneous sensing data characterizing the thermal state of the system using multiple sensors distributed at different spatial locations in the waste heat recovery system of the stenter, the spatial coordinate position information and data acquisition time information of each sensor are recorded simultaneously.

[0070] Based on the spatial coordinate location information of each sensor and the pipeline topology information of the waste heat recovery system, the spatial distribution relationship characteristics of the sensors are determined. These characteristics describe the spatial distance relationships and heat transfer path relationships between the sensors.

[0071] Based on the spatial distribution characteristics of the sensors and the data acquisition time information, the time delay compensation between sensor data at different spatial locations is calculated;

[0072] The timestamp correction process is performed on the multi-source heterogeneous sensing data based on the time delay compensation amount, and the spatial coordinate normalization process is performed on the timestamp-corrected multi-source heterogeneous sensing data according to the spatial distribution relationship characteristics of the sensors, so as to obtain spatiotemporally aligned multi-source sensing data.

[0073] During the operation of the waste heat recovery system in the stenter, multiple sensors distributed at different spatial locations within the system collect data reflecting the system's thermal state in real time. These sensors include temperature sensors, pressure sensors, flow sensors, and humidity sensors, which are installed at key locations such as the stenter's flue outlet, heat exchanger inlet and outlet, circulating water pipelines, and steam generator. When each sensor collects data, the acquisition module automatically records the sensor's spatial coordinate position and the time of data acquisition. The spatial coordinate position information is represented using a three-dimensional coordinate system, with a fixed point on the stenter as the origin, and the x, y, and z coordinate values ​​of each sensor's installation location are obtained by measurement. The time of data acquisition uses a unified clock source synchronized with timestamps to ensure that the time reference of all sensors is consistent, with timestamp accuracy reaching the millisecond level.

[0074] After acquiring the spatial coordinates of the sensors, the analysis is performed in conjunction with the pipeline topology information of the waste heat recovery system. This topology information includes physical structural parameters such as flue gas routing, heat exchanger layout, pipe connection methods, and valve distribution. By overlaying the sensor spatial coordinates with the pipeline topology, the spatial distance relationships between the sensors are determined. For example, the spatial distance between the temperature sensor at the outlet of the stenter flue and the temperature sensor at the inlet of the heat exchanger is calculated using the Euclidean distance between the two points. This distance reflects the actual path length of the hot flue gas flowing from the flue gas outlet to the heat exchanger inlet. Simultaneously, the heat transfer path relationships are determined based on the pipeline routing and fluid flow direction. These heat transfer path relationships describe the sequence and medium of heat transfer from high-temperature regions to low-temperature regions. For example, hot flue gas is first transported to the heat exchanger through the flue, where it transfers heat to the circulating water, which then transfers the heat to the heat-using equipment. This transfer path relationship constitutes the core content of the sensor spatial distribution characteristics.

[0075] After establishing the spatial distribution characteristics of the sensors, it is necessary to calculate the time delay compensation between sensor data from different spatial locations. Since heat transfer within the system takes time, the data collected by sensors installed at different locations have a temporal sequence. Taking a flue outlet temperature sensor and a heat exchanger inlet temperature sensor as examples, when the flue outlet temperature changes, there is a certain delay in the heat exchanger inlet sensing this change. This delay is mainly determined by the flow time of the hot flue gas within the flue. The calculation of the time delay compensation comprehensively considers spatial distance, fluid velocity, and heat transfer characteristics. For the transmission delay of the fluid medium, the transmission time is obtained by dividing the distance between sensors by the fluid velocity. For the delay caused by heat conduction, the thermal inertia characteristics of equipment such as heat exchangers are considered, and the time constant of the temperature response is estimated based on the equipment's heat capacity and heat transfer coefficient. In actual calculations, the hot flue gas velocity within the flue is set as... The distance between sensor A and sensor B along the flow direction is The time delay caused by the flow and transmission Calculated as For delays within equipment such as heat exchangers, the time constant is determined based on the characteristics of the first-order inertial element of the equipment. Based on the heat capacity of the equipment and overall heat transfer coefficient Determined. Taking into account both flow delay and thermal inertia delay, the total time delay compensation is the sum of the two.

[0076] After obtaining the time delay compensation, timestamp correction is performed on the multi-source heterogeneous sensor data. The goal of timestamp correction is to eliminate data time misalignment caused by differences in heat transfer time. The correction process uses the acquisition time of a certain reference sensor as a benchmark, and the timestamps of other sensors are adjusted according to the calculated delay compensation. For example, if the flue outlet temperature sensor is selected as the reference point, its timestamp remains unchanged, while the timestamp of the heat exchanger inlet temperature sensor is reduced by the corresponding delay compensation, so that the corrected timestamps reflect the true causal relationship of temperature changes at the two locations. For the entire system, timestamp correction is performed step by step according to the upstream and downstream relationships of the heat transfer path to ensure that all sensor data are aligned to the system state at the same physical moment in the time dimension.

[0077] After timestamp correction, spatial coordinate normalization is performed. Since the sensors are distributed at different locations in three-dimensional space, directly using the original coordinates for data fusion results in differences in dimensions and scale. Spatial coordinate normalization maps the sensor's three-dimensional coordinates to a unified dimensionless space. The normalization process uses a min-max normalization method, mapping the x, y, and z coordinates to the intervals of 0 to 1. Specifically, for the x-coordinate, the normalized coordinates... pass The calculation yielded, where and These are the minimum and maximum x-coordinates for all sensors, respectively; the y and z coordinates are processed using the same method. Normalized spatial coordinates eliminate the influence of physical dimensions, allowing spatial distances to be compared and calculated on a uniform scale.

[0078] After completing timestamp correction and spatial coordinate normalization, a spatiotemporally aligned multi-source sensor dataset is established. Each data record in this dataset includes normalized spatial coordinates, a corrected timestamp, sensor type identification, and measurement value. Through spatiotemporal alignment, the originally heterogeneous sensor data is transformed into homogeneous data with a unified spatiotemporal reference system, providing a foundation for subsequent data fusion calculations. Spatiotemporal alignment eliminates data inconsistencies caused by sensor physical distribution and thermal transfer delays, enabling data collected at different locations and times to accurately reflect the multidimensional characteristics of the same system state.

[0079] In practical applications, spatiotemporal alignment processing is dynamically adjusted based on the operating conditions of the waste heat recovery system. For example, when the production speed of the stenter changes, the flue gas velocity in the flue changes accordingly, causing changes in the flow delay between sensors. At this time, the time delay compensation is updated based on the real-time monitored flue gas velocity to ensure the accuracy of spatiotemporal alignment. For multiple parallel heat exchange channels, sensors in different channels establish independent spatiotemporal alignment relationships, and then these relationships are fused at a higher level. Through this hierarchical spatiotemporal alignment strategy, multi-source sensor data under complex pipeline topologies are processed, ensuring the monitoring system's accurate perception of the waste heat recovery process.

[0080] Figure 2 This is a schematic diagram of the process for determining the overall system status assessment result according to an embodiment of the present invention, such as... Figure 2 As shown, the spatiotemporally aligned multi-source sensor data is fused and calculated based on the physical correlation constraints between them to obtain a comprehensive system state assessment result, including:

[0081] Based on the energy transfer coupling relationship between the spatiotemporally aligned multi-source sensor data, a state evolution relationship describing the thermal energy flow process of the waste heat recovery system is determined. This state evolution relationship represents the constraint effect of physical correlation constraints on the system state changes.

[0082] The state evolution relationship is used to perform state estimation calculation on the spatiotemporally aligned multi-source sensing data to predict the theoretical state value of each sensor's monitoring position at the current moment, thus obtaining predicted state data;

[0083] The spatiotemporally aligned multi-source sensor data and the predicted state data are subjected to deviation analysis to calculate the deviation between the actual measured values ​​of each sensor and the predicted state data.

[0084] The measurement reliability of each sensor's measurement data is quantitatively evaluated based on the aforementioned deviation, resulting in a reliability evaluation index for each sensor.

[0085] A dynamic weight allocation strategy is determined based on the credibility assessment index, and the spatiotemporally aligned multi-source sensor data is fused and calculated according to the dynamic weight allocation strategy to obtain the system comprehensive state assessment result.

[0086] After obtaining spatiotemporally aligned multi-source sensor data, the data is deeply fused using the inherent physical laws of the stenter waste heat recovery system. The operation of the stenter waste heat recovery system is essentially a complex heat transfer and conversion process. After being discharged from the stenter, the flue gas passes through heat exchangers, condensers, and other equipment, with heat being recovered and utilized step by step. During this process, there are strict physical relationships between parameters such as temperature, pressure, and flow rate, which constitute the fundamental constraints for data fusion.

[0087] By analyzing the thermodynamic characteristics of the waste heat recovery system, a physical model describing the system's state evolution is established. Based on the law of conservation of energy and heat and mass transfer theory, this model characterizes the inherent laws governing temperature decay, heat transfer, and changes in the working fluid's state during flue gas flow. Specifically, the change in flue gas temperature along the flow direction is comprehensively affected by factors such as heat exchange area, convective heat transfer coefficient, and flue gas velocity. Inside the heat exchanger, heat exchange occurs between the flue gas side and the working fluid side, and the rate of flue gas temperature decrease matches the rate of heat absorption by the working fluid. Simultaneously, the working fluid's temperature rises after absorbing heat, and its heating rate is proportional to the absorbed heat flow rate and related to the working fluid's specific heat capacity and mass flow rate. These physical processes are precisely described by state evolution relationships, which link the state parameters such as temperature, pressure, and flow rate at each monitoring point, forming a closed physical constraint network.

[0088] The state evolution equation not only includes parameter correlations in the spatial dimension but also encompasses dynamic evolution patterns in the temporal dimension. Due to the thermal inertia of the waste heat recovery system, changes in state parameters at a certain location will be delayed before propagating to downstream locations. This spatiotemporal coupling characteristic requires the state evolution equation to consider both the current state of each measuring point and historical state information. By introducing a time delay term and a dynamic response coefficient, the state evolution equation can accurately predict the transient response of the system under specific operating conditions.

[0089] State estimation is performed on spatiotemporally aligned multi-source sensor data using a constructed state evolution relation. This process employs a dynamic recursive algorithm, using the system state from the previous moment as initial conditions and combining it with the current operating parameters. The theoretical state values ​​that each monitoring location should exhibit at the current moment are derived through the state evolution relation. For example, given the boundary conditions such as the inlet flue gas temperature, flue gas flow rate, and working fluid flow rate of the heat exchanger, the theoretical value of the outlet flue gas temperature can be calculated based on the heat balance equation and heat transfer equation. Similarly, for other key measuring points in the system, corresponding predicted state data can be obtained through the state evolution relation. These predicted values ​​represent the state that the system should exhibit under ideal physical constraints and serve as important reference benchmarks for subsequent deviation analysis.

[0090] After obtaining the predicted state data, it is compared one by one with the actual measurements collected by the sensors. For each physical quantity monitored by the sensor, the deviation between the measured value and the predicted value is calculated. This deviation reflects the difference between the actual system operating state and the expectations of the theoretical physical model. Ideally, if the sensor measurement accuracy is high enough and the system perfectly conforms to the assumptions of the physical model, the deviation should be close to zero. However, in actual operation, due to sensor measurement errors, environmental interference, and the non-ideal nature of the system's operating state, deviations are often unavoidable. Through in-depth analysis of these deviations, it is possible to identify which sensor measurement data are abnormal and which measurement values ​​are more reliable.

[0091] The magnitude of the deviation directly affects the reliability of sensor measurement data. When a sensor's measurement value is highly consistent with the theoretical value predicted based on physical constraints, it indicates that the sensor is working properly and its measurement data can accurately reflect the system state. Conversely, if the deviation is significantly large, it indicates that the sensor has malfunctioned, measurement drift has occurred, or the measurement point has been affected by local abnormal operating conditions. Based on this understanding, the measurement reliability of each sensor is quantitatively evaluated. The reliability evaluation adopts a function that is negatively correlated with the deviation; the smaller the deviation, the higher the reliability evaluation index; the larger the deviation, the lower the reliability evaluation index. Considering the differences in measurement difficulty and error tolerance for different physical quantities, different deviation tolerance thresholds are set for different types of sensors. For example, temperature sensors typically have high measurement accuracy, so their deviation tolerance threshold can be set relatively low; while flow sensors, due to limitations in measurement principles and field conditions, have a more relaxed deviation tolerance threshold.

[0092] After obtaining the reliability evaluation indicators for each sensor, a dynamic weight allocation strategy is constructed. This strategy dynamically adjusts the weight ratio of each sensor in the data fusion process based on its real-time reliability. Sensors with high reliability are assigned larger weights, and their measurement data dominates the fusion result; sensors with low reliability have their weights reduced accordingly to minimize their negative impact on the fusion result. Dynamic weight allocation avoids the shortcomings of traditional fixed-weight methods that cannot adapt to changes in sensor state, making the data fusion result more accurate and reliable.

[0093] The weighting process also needs to consider the spatial correlation and functional redundancy among sensors. For multiple redundant sensors monitoring the same physical quantity, when their reliability evaluation indicators are similar, an average weighting is used to fully utilize redundant information; when the reliability of one sensor is significantly higher than that of the others, the weight ratio of that sensor is increased. For sensors monitoring different physical quantities but with physical correlations, the mutual support relationship between them is determined by the constraint equations established through state evolution relationships. Cross-validation is performed by integrating information from multiple sensors to further improve the accuracy of the fusion results.

[0094] Based on a determined dynamic weighting strategy, weighted fusion calculations are performed on spatiotemporally aligned multi-source sensor data. The fusion process is not a simple numerical weighted average, but rather, under the premise of satisfying physical constraints, it determines an optimal system state estimate that minimizes the weighted deviation between this estimate and the measurements of all sensors, while also conforming to the constraints of the state evolution equation. This process is achieved by constructing a constrained optimization problem. The objective function is to minimize the sum of squares of the weighted deviations between the sensor measurements and the state estimate, and the constraints are the physical equations represented by the state evolution equation. By solving this optimization problem, a comprehensive system state assessment result that satisfies physical laws and is closest to the measured data is obtained.

[0095] The comprehensive system state assessment results are represented in vector form, containing optimal estimates of all key state parameters of the waste heat recovery system, such as flue gas inlet and outlet temperatures, working fluid inlet and outlet temperatures, pressure loss, and heat exchange efficiency of each stage of heat exchanger. These parameters comprehensively characterize the real-time thermal state of the system, providing a reliable data foundation for subsequent anomaly diagnosis and optimized control. Compared with single sensor data or simple average values, the comprehensive state assessment results obtained through physical constraint fusion calculations have higher accuracy and robustness, effectively suppressing interference from individual sensor anomalies on system monitoring and ensuring the stable and reliable operation of the monitoring system under complex operating conditions.

[0096] A dynamic weight allocation strategy is determined based on the reliability assessment index, and the spatiotemporally aligned multi-source sensor data is fused and calculated according to the dynamic weight allocation strategy to obtain a comprehensive system state assessment result, including:

[0097] Based on the credibility assessment index, it is determined that there are sensor data with measurement anomalies in the spatiotemporally aligned multi-source sensing data.

[0098] The reliability evaluation index corresponding to the sensor data with measurement anomalies is suppressed using a nonlinear decay function to obtain a corrected reliability evaluation index;

[0099] Based on the revised reliability assessment index, the fusion weight coefficient of each sensor data is calculated, and the dynamic weight allocation strategy is determined through the fusion weight coefficient.

[0100] According to the dynamic weight allocation strategy, the spatially adjacent sensor data in the spatiotemporally aligned multi-source sensing data are locally weighted and fused to obtain the regional state value. The fusion weight coefficient of the local weighted fusion is adjusted according to the spatial position relationship of the sensors.

[0101] A global weighted fusion of all regional state values ​​is performed to obtain the comprehensive system state assessment result. The fusion weight coefficient of the global weighted fusion is adjusted according to the data integrity of the regional state values.

[0102] In the waste heat recovery system of the stenter, the measurement accuracy of the sensors directly affects the reliability of the condition assessment. Sensors operate under high temperature and high humidity environments for extended periods, making them prone to drift, noise interference, or localized failure. To address the differences in sensor data quality reflected in the reliability assessment indicators, a dynamic weight allocation strategy is designed to rationally allocate the contribution of each sensor's data in the fusion calculation.

[0103] The credibility assessment index quantifies the evaluation of each sensor's data from three dimensions: data stability, temporal consistency, and physical rationality. When a sensor's credibility assessment index falls below a threshold... When this happens, it is determined that the sensor data has a measurement anomaly. Specifically, in the determination process, the reliability assessment index of sensor i is marked as... ,like If this happens, the sensor data is marked as abnormal. In practical applications... The value range is typically set between 0.6 and 0.75, with the specific value determined based on the system's tolerance for data quality. For waste heat recovery systems in stenters, considering the high requirements for process stability, a value is usually set... .

[0104] Directly removing sensor data marked as measurement anomalies leads to information loss, while complete retention introduces errors. A nonlinear decay function is used to suppress the reliability evaluation index of anomaly data, preserving some valid information while reducing the impact weight of anomalies. The nonlinear decay function is designed as follows: ,in The revised credibility assessment metric The attenuation intensity coefficient, It is a non-linear exponent. When The greater the deviation from 1, the more significant the attenuation effect. Parameter The value range is 2 to 5, parameter The value range is from 1.5 to 3. In the application scenario of the setting machine, after multiple sets of experimental calibrations, the value is... and It can effectively suppress interference from anomalous data. For normal sensor data where the reliability assessment index is higher than the threshold, the corrected reliability assessment index remains at its original value. .

[0105] The fusion weight coefficient is calculated based on the revised credibility evaluation index. The fusion weight coefficient of sensor i is... It is obtained through normalization calculation, that is Where M is the total number of sensors participating in the fusion. This normalization process ensures that the sum of all fusion weight coefficients is 1, satisfying the basic requirements of weighted fusion. The distribution of fusion weight coefficients reflects the importance of each sensor's data in the fusion process; sensors with higher reliability evaluation indicators receive larger weight coefficients. The dynamic weight allocation strategy adjusts in real time based on the fusion weight coefficients. When the measurement quality of a sensor changes, its corresponding fusion weight coefficient changes accordingly, thereby achieving adaptability in the fusion calculation.

[0106] The waste heat recovery system for the stenter exhibits distinct regional characteristics. The thermal parameters vary across different locations, such as the flue gas inlet area, heat exchanger area, and exhaust area. Local weighted fusion based on sensor spatial location allows for a more accurate reflection of the true state of each area. The system is divided into several monitoring zones, with multiple sensors deployed in each zone. When performing local weighted fusion of sensor data within zone k, the spatial distance between sensors must be considered. The local fusion weight coefficient for sensor i within zone k is defined. In global fusion weight coefficient Based on this, adjustments are made according to spatial location. The adjustment mechanism is designed as follows: ,in Represents the set of sensors within region k. The spatial distance between sensor i and sensor j This is the spatial correlation coefficient, ranging from 0.1 to 0.5. The closer the sensors are, the stronger the correlation between their measurements, and the greater the adjustment range of the local fusion weight coefficient.

[0107] Area status value It is obtained by locally weighted fusion of all sensor data within the region, and the calculation formula is as follows: ,in The measured value is from sensor i. This local weighted fusion process can eliminate local measurement fluctuations within the region and obtain a more stable regional state characterization. For temperature sensor data, the regional temperature value after local weighted fusion represents the average thermal state of the region; for pressure sensor data, the regional pressure value reflects the flow resistance characteristics of the region.

[0108] Global weighted fusion integrates the state values ​​of each region into a comprehensive system state assessment result. However, data integrity varies across different regions, with some regions experiencing reduced effective data due to sensor malfunctions. The data integrity index for region k is... Defined as the ratio of the number of normally functioning sensors to the planned number of sensors in the area, i.e. ,in The number of sensors operating normally within region k. The total number of sensors planned for region k. Regions with high data integrity indices have higher reliability of regional state values ​​and should be assigned a greater weight in global fusion.

[0109] Global fusion weight coefficients for region k The formula is calculated based on the data integrity index. Where K is the total number of regions divided by the system. This is an integrity sensitivity index, ranging from 1.2 to 2.0. This index controls the impact of data integrity on the global fusion weights; a higher value indicates stricter requirements for data integrity. In the waste heat recovery system of the stenter, to ensure the reliability of the evaluation results, a value of [value missing] is typically used. .

[0110] System integrated status assessment results The expression is obtained through global weighted fusion calculation. The assessment results comprehensively reflect the operational status of the entire waste heat recovery system, encompassing multi-dimensional information such as temperature field distribution, pressure field distribution, and flow distribution. For multi-parameter monitoring scenarios, the above fusion process is executed for each parameter separately, ultimately forming a comprehensive assessment vector containing multiple status parameters such as temperature, pressure, and flow rate.

[0111] The core of the dynamic weight allocation strategy lies in adjusting the fusion weights in real time based on the sensor data quality and spatial distribution characteristics. This avoids distortion of the fusion results caused by fixed weights. When a sensor experiences a short-term anomaly, its fusion weight is automatically reduced to minimize the impact of abnormal data. When the sensor returns to normal, the fusion weight is gradually increased to fully utilize effective information. The entire fusion process is divided into two levels: local fusion and global fusion. This preserves detailed information about regional characteristics while achieving accurate assessment of the overall system status, providing reliable data support for subsequent anomaly diagnosis and control decisions.

[0112] Based on the comprehensive system status assessment results and the preset normal system operating status boundaries, deviation calculations are performed to determine abnormal feature description information. Based on the abnormal feature description information and the comprehensive system status assessment results, multi-objective optimization is performed to generate a control execution scheme for the waste heat recovery system, including:

[0113] The system comprehensive status assessment result is compared with the preset system normal operation status boundary, the deviation value of the system comprehensive status assessment result from the system normal operation status boundary is calculated, and the abnormal feature description information is determined based on the deviation value;

[0114] Based on the description of the abnormal features, the key influencing factors causing state deviation in the waste heat recovery system are identified, and a state callback path planning strategy is determined for the key influencing factors.

[0115] Based on the state callback path planning strategy and the system comprehensive state evaluation results, a multi-objective optimization problem is determined, which includes a system energy efficiency optimization objective function and a state deviation minimization objective function. The system energy efficiency optimization objective function takes maximizing waste heat recovery efficiency as the optimization direction, and the state deviation minimization objective function takes minimizing the deviation value as the optimization direction.

[0116] The multi-objective optimization problem is solved to obtain the optimal values ​​of control variables that satisfy the comprehensive optimality of multiple objective functions. Based on these control variable values, a control execution scheme for the waste heat recovery system is generated.

[0117] After obtaining the comprehensive system status assessment results, in-depth analysis is required to identify whether the current operating state of the system deviates from the normal operating range. The preset normal operating state boundary of the system is obtained from the statistical analysis of historical operating data. This boundary characterizes the normal value range of each key parameter of the waste heat recovery system during stable and efficient operation. When comparing the comprehensive system status assessment results with this boundary, the degree of deviation from the corresponding boundary is calculated for each state parameter included in the assessment results.

[0118] The deviation value is calculated using a normalized distance metric. For a given state parameter, the deviation value is obtained by dividing the difference between the current actual value and the boundary center value by the allowable deviation range of the boundary. When the state parameter is a multi-dimensional vector, a weighted Euclidean distance is used to calculate the overall deviation. Specifically, assuming the system's overall state assessment results include multiple parameters such as flue gas temperature, waste heat recovery rate, hot water outlet temperature, and heat exchanger pressure drop, these parameters are combined into a state vector. The distance from this state vector to the boundary center of the normal operating state is calculated and divided by the characteristic radius of the boundary to obtain the dimensionless deviation value. When this value exceeds a preset threshold, it indicates that the system's operating state has deviated from the normal operating range.

[0119] Based on the calculated deviation values, further descriptive information about the anomalies is determined. This descriptive information includes not only the magnitude of the deviation but also the directional and temporal evolution characteristics of the deviation. The directional deviation is obtained by analyzing the contribution of each state parameter to the deviation, identifying the main direction of deviation in the multidimensional state space, thus clarifying which parameter anomalies caused the overall state deviation. The temporal evolution characteristics are obtained by trend analysis of the deviation values ​​over multiple consecutive sampling times, determining whether the deviation is abrupt or gradual, continuously deteriorating or fluctuating. These features together constitute a complete descriptive information about the anomalies.

[0120] Based on the description of abnormal characteristics, the key influencing factors leading to state deviations in the waste heat recovery system are further analyzed. The identification of these key influencing factors is based on causal correlation analysis. By establishing a causal network model between state parameters, the root causes of the deviations are traced. For example, when a decrease in waste heat recovery rate is detected, it is necessary to analyze whether it is due to reduced flue gas flow, decreased flue gas temperature, increased heat exchanger fouling, or insufficient hot water circulation flow. By calculating the sensitivity coefficients of each potential influencing factor to the state deviation, the factor with the highest sensitivity coefficient is identified as the key influencing factor. This sensitivity analysis can be performed analytically based on the system's thermodynamic model or through statistical regression analysis based on historical data.

[0121] For the identified key influencing factors, a state callback path planning strategy is determined. The goal of state callback path planning is to guide the currently deviating system state back to the normal operating state boundary, while avoiding the initiation of new anomalies and ensuring the smoothness of the adjustment process. The path planning strategy needs to consider the dynamic response characteristics of the system and the coupling relationship between multiple state parameters. For parameters with strong coupling relationships, a coordinated adjustment strategy is adopted to avoid drastic fluctuations in other parameters caused by large adjustments to a single parameter. For parameters with large differences in response time constants, a staged adjustment strategy is adopted, prioritizing the adjustment of parameters with fast responses to quickly alleviate deviations, and then adjusting parameters with slow responses to achieve steady-state optimization.

[0122] Based on the state callback path planning strategy and the system comprehensive state assessment results, a mathematical model for a multi-objective optimization problem is constructed. This optimization problem includes two core objective functions. The system energy efficiency optimization objective function aims to maximize waste heat recovery efficiency. This objective function uses the ratio of recovered waste heat to the total recoverable waste heat as the optimization target, and improves heat exchange efficiency by adjusting control variables such as the hot water flow rate of the heat exchanger and the flue gas velocity. The state deviation minimization objective function aims to minimize the deviation value obtained from the aforementioned calculation. This objective function drives the system state to return to the normal operating state boundary as quickly as possible.

[0123] In constructing a multi-objective optimization problem, in addition to the two objective functions, constraints also need to be set. These constraints include physical constraints and safety constraints on equipment operation. Physical constraints involve limitations on the value ranges of various control variables, such as the flow rate adjustment range of the hot water circulating pump and the speed adjustment range of the fan. Safety constraints involve the safety boundaries of system operation, such as the heat exchanger wall temperature not exceeding the upper limit of the material's temperature resistance and the flue gas pressure drop not exceeding the fan's head capacity. These constraints are added to the optimization problem in the form of inequality constraints.

[0124] When solving multi-objective optimization problems, either a weighted method with preference weights or a Pareto front search method is employed. The weighted method linearly combines two objective functions using weight coefficients to form a single comprehensive objective function. Adjusting these weight coefficients allows for a trade-off between energy efficiency optimization and state recovery. When the system deviation is large, the objective function minimizing state deviation is given higher weight to prioritize correcting abnormal states; conversely, when the system deviation is small, the energy efficiency optimization objective function is given higher weight to pursue higher waste heat recovery efficiency. The Pareto front search method seeks a set of non-dominated solutions where multiple objective functions are relatively optimal, and selects the most suitable solution based on the current operating scenario.

[0125] The solution process employs numerical optimization algorithms, commonly including sequential quadratic programming, particle swarm optimization, or genetic algorithms. For objective functions with explicit analytical gradient information, sequential quadratic programming can quickly converge to a local optimum. For objective functions with complex forms or non-convex characteristics, particle swarm optimization and genetic algorithms, through their swarm search mechanism, possess stronger global search capabilities. In practical applications, the advantages of both types of algorithms can be combined: first, a heuristic algorithm is used for global search to determine the optimization direction, and then gradient-based algorithms are used for local fine-tuning.

[0126] The optimization solution yields optimal values ​​for a set of control variables, including but not limited to the setpoints for hot water circulation flow, fan speed, heat exchanger bypass valve opening, and stenter exhaust temperature. Based on these optimal values, a control execution scheme for the waste heat recovery system is generated. This scheme includes not only the target setpoints for each control variable but also the timing of adjustments and adjustment rate limits. The timing ensures coordination between multiple control actions, preventing transient oscillations caused by improper adjustment sequence. The adjustment rate limits are set according to the characteristics of each actuator to prevent excessively rapid adjustments from causing equipment shocks and system instability. The generated control execution scheme is ultimately sent to the field controller of the waste heat recovery system in a standardized command format, driving each actuator to perform actions according to the scheme, achieving optimized system state adjustment and anomaly correction.

[0127] Based on the anomaly characteristic description information, the key influencing factors causing state deviation in the waste heat recovery system are identified, and a state callback path planning strategy is determined for the key influencing factors, including:

[0128] The abnormal feature description information is parsed to determine the feature vector describing the system state deviation characteristics. The feature vector includes the spatial distribution characteristics of the state deviation and the temporal evolution characteristics of the state deviation.

[0129] Based on the causal correlation analysis between the eigenvectors and the control variables of the waste heat recovery system, the key influencing factors that play a dominant role in state deviation are identified;

[0130] For the aforementioned key influencing factors, a dynamic transmission relationship between the control variable adjustment action and the system state change response is determined. This dynamic transmission relationship represents the process by which the system state evolves from the current deviation state to the boundary of the normal operating state after the control variable is adjusted.

[0131] Based on the dynamic transmission relationship, multiple candidate adjustment paths are determined to restore the system state from the current deviation state to the normal operating state boundary. The multiple candidate adjustment paths are evaluated, and the adjustment path with the shortest state recovery time and the smallest adjustment magnitude of the control variable is selected as the state callback path planning strategy.

[0132] During the operation of the waste heat recovery system in the stenter, when an anomaly is detected in the overall system status assessment results, a deep analysis of the anomaly characteristic description information is required. This anomaly characteristic description information contains multiple dimensions of state deviation representation. Through structured analysis of this information, feature vectors that can quantify the system's state deviation characteristics are extracted. These feature vectors have dual representation capabilities: firstly, they reflect the degree of state deviation at different physical locations in the waste heat recovery system through spatial distribution characteristics, such as abnormal temperature differences between the inlet and outlet of the heat exchanger and uneven flue gas flow distribution; secondly, they characterize the trend of state deviation over time through temporal evolution characteristics, including the growth rate of the deviation, the periodicity of fluctuations, and the stability of the evolution trajectory. When extracting spatial distribution characteristics, the waste heat recovery system is divided into multiple functional areas such as the flue gas side, water side, and heat exchange side. The deviation of sensor monitoring data from the normal operating state boundary is calculated for each area, forming a spatial distribution vector. The extraction of temporal evolution characteristics is based on historical data within a sliding time window. By calculating the rate of change, acceleration, and higher-order derivatives of state parameters, the dynamic characteristics of state evolution are captured.

[0133] After determining the eigenvector, a causal correlation analysis model is established between the eigenvector and the control variables of the waste heat recovery system. The control variables of the waste heat recovery system include multiple adjustable parameters such as the opening degree of the flue gas bypass valve, the frequency of the hot water circulating pump, the opening degree of the water-side flow regulating valve of the heat exchanger, and the speed of the flue gas-side induced draft fan. Causal correlation analysis needs to distinguish between correlation and causation to avoid misjudging accidental statistical correlations as substantial causal effects. The Granger causality test method is used to analyze whether changes in control variables in historical data precede changes in the state eigenvector in time, and whether this preceding change can significantly improve the prediction accuracy of changes in the state eigenvector. For each control variable, the causal influence strength on each component of the eigenvector is calculated, and this strength is quantified by constructing a vector autoregression model and performing a likelihood ratio test. When an adjustment to a control variable can statistically significantly affect multiple key components of the eigenvector, that control variable is marked as a potential key influencing factor. Furthermore, the interaction between control variables is considered; some control variables have limited impact when adjusted individually, but produce significant synergistic or antagonistic effects when combined with other control variables. By calculating the joint explanatory power of the combination of control variables on the state deviation, the set of key influencing factors that play a dominant role in the current state deviation is identified.

[0134] To address the identified key influencing factors, a dynamic transfer relationship model needs to be established between the control variable adjustment action and the system state change response. This dynamic transfer relationship differs from the static steady-state relationship; its core lies in describing the dynamic process from the application of the control adjustment action to the corresponding change in the system state, including characteristics such as response delay, dynamic gain, and time constant. In a waste heat recovery system, taking the adjustment of the flue gas bypass valve opening as an example, when the valve opening is increased, the amount of high-temperature flue gas entering the heat exchanger decreases. However, due to the heat capacity within the heat exchanger and the transport delay of the flue gas flow, the drop in the heat exchanger outlet water temperature does not occur instantaneously, but rather presents a dynamic response curve with time delay and a transient process. By conducting small-amplitude control variable excitation experiments under different operating conditions, system state response data is collected, and a transfer function model or state-space model from the control variable to the state variable is established using system identification methods. For transfer relationships with significant nonlinear characteristics, piecewise linearization methods or nonlinear mapping models based on neural networks are used for approximate characterization. Dynamic transfer relationship models not only include the direct transfer path from a single control variable to a single state variable, but also need to consider the indirect effects of multiple control variables on the state variable through the internal coupling mechanism of the system. For example, adjusting the frequency of the hot water circulation pump not only directly affects the water-side flow rate, but also indirectly affects the flue gas-side temperature distribution by changing the water-side heat transfer coefficient of the heat exchanger.

[0135] Based on the established dynamic transfer relationship model, the evolution trajectory of the system state after applying specific control adjustments can be predicted. The current system is in an abnormal state deviating from the normal operating state boundary. The goal is to adjust the control variables corresponding to key influencing factors to gradually evolve the system state along a certain trajectory and eventually restore it to within the normal operating state boundary. Multiple adjustment paths exist from the current abnormal state to the normal state boundary, with different paths corresponding to different control variable adjustment sequences and magnitudes. To generate candidate adjustment paths, forward simulation calculations are first performed based on the dynamic transfer relationship model, setting different combinations of control variable adjustments to predict the evolution trajectory of the system state over future time periods. Each candidate path can be represented as a time series, describing how the state parameters gradually approach and enter the normal operating state boundary after a series of control adjustments from the initial abnormal state. Various strategies can be used to generate candidate paths, including heuristic paths based on empirical rules, gradient paths based on the steepest descent principle, and optimization paths based on model predictive control.

[0136] When comprehensively evaluating multiple candidate adjustment paths, several performance indicators need to be considered. State recovery time is a key indicator, defined as the time required from the current moment until the system first enters the normal operating state boundary. A shorter recovery time means the system can quickly recover from abnormal states, reducing energy loss and equipment wear. The adjustment magnitude of control variables is another important indicator; excessively large adjustment magnitudes can lead to rapid wear of actuators, system oscillations, or trigger new anomalies. For each candidate path, the sum of the adjustment magnitudes of all control actions it contains, or the weighted sum of squares of the adjustment magnitudes, is calculated. In addition to these two main indicators, the smoothness of the path must also be considered to avoid drastic and repeated adjustments of control variables; the robustness of the path must be considered, i.e., whether the path can still guarantee state recovery effectiveness in the presence of model errors or external disturbances; and energy efficiency must be considered, as some adjustment paths, although having short recovery times, can lead to a surge in energy consumption in the short term.

[0137] Within a multi-objective evaluation framework, candidate paths are comprehensively ranked. Since the objectives of minimizing state recovery time and minimizing control variable adjustment magnitude often conflict, a Pareto front for the multi-objective optimization problem is constructed to filter out a set of non-dominated solutions. Within this set, decision preference weights are further introduced: higher weights are assigned to recovery time when system security requirements are high, and higher weights are assigned to adjustment magnitude when the lifespan of the actuator is limited. The optimal adjustment path is selected from the non-dominated solutions as the state callback path planning strategy using a weighted summation or ideal point distance minimization method. The selected state callback path planning strategy explicitly defines the adjustment time, direction, and magnitude of each control variable within a future time period, forming a time-seriesd sequence of control instructions.

[0138] This state callback path planning strategy dynamically corrects itself based on real-time feedback during execution. If the actual system response deviates from the model prediction after executing the planned path, path replanning is triggered. Based on the latest system state and the updated dynamic transmission relationship model, candidate paths are regenerated, and the optimal path is selected. This closed-loop feedback mechanism ensures that even with model uncertainties and external disturbances, the system can gradually recover to normal operation along a reasonable path, achieving stable and efficient operation of the waste heat recovery system.

[0139] After executing the control execution scheme, system response data is collected. The system response data is then compared and analyzed with the expected target of the control execution scheme to obtain control effect deviation information, including:

[0140] After executing the control execution scheme, system response data of the waste heat recovery system is collected within a preset response observation time window;

[0141] Temporal features are extracted from the system response data to obtain transient response features and steady-state response features.

[0142] Based on the transient response characteristics and the steady-state response characteristics, a system response integrity evaluation index is determined. The system response integrity evaluation index is used to quantify the completeness of the system response data in representing the control execution process.

[0143] The transient response characteristics are compared with the dynamic performance indicators in the expected target to calculate the transient process deviation, and the steady-state response characteristics are compared with the steady-state performance indicators in the expected target to calculate the steady-state result deviation.

[0144] The transient process deviation and the steady-state result deviation are fused and calculated based on the system response integrity evaluation index to obtain the control effect deviation information.

[0145] After the control execution plan is issued to the actuator of the waste heat recovery system, the actual execution effect needs to be tracked and verified. The setting of the response observation time window comprehensively considers the system's thermal inertia, fluid transport delay, and the time constant of the heat exchange process. For the stenter waste heat recovery system, the typical response observation time window is 120 to 300 seconds after the execution command is issued. The start time of this time window is not the instant the control command is issued, but after the system begins to generate an observable response, and the end time is dynamically determined according to whether the system reaches a new steady state. Within this time window, various sensor data are continuously collected, including but not limited to hot air temperature, fan speed, heat exchanger outlet temperature, cooling water flow rate, and system pressure difference. The sampling frequency is set differently according to the rate of change of different physical quantities. Rapidly changing quantities such as temperature are sampled at a high frequency of 5 times per second, while slowly changing quantities such as pressure are sampled at a conventional frequency of 1 time per second.

[0146] The extraction of time-series features from system response data employs a segmented processing strategy. Transient response features primarily focus on the dynamic behavior of the system during its transition from the initial state to the target state. By differentiating the time series of key parameters such as temperature and flow rate, rate-of-change curves are obtained, from which typical transient indicators such as rise time, peak time, overshoot, and settling time are extracted. For example, when adjusting hot air temperature, the time required from the issuance of the control command to the temperature first reaching 95% of the target value is recorded as the rise time, and the ratio of the actual peak temperature to the target temperature is recorded as the overshoot. These parameters directly reflect the system's response speed and control stability. Steady-state response features focus on the system's ability to maintain a stable state after entering stable operation. This is described by calculating the statistical characteristics of the response data in the steady-state phase, such as mean, variance, and fluctuation amplitude. Steady-state judgment uses a sliding window variance test method. When the data variance is less than a preset threshold for 30 consecutive seconds, the system is considered to have entered a steady state, and the calculated mean at this time is the steady-state response value.

[0147] The system response integrity evaluation index is established based on the combination of information entropy theory and physical process coverage. Integrity evaluation needs to be carried out from two dimensions: in the time dimension, the response data must cover the entire process from the start of the control action to system stability; in the spatial dimension, each key measuring point must have effective data feedback. Temporal integrity is quantified by calculating the ratio of the observation time window to the theoretical response period; when this ratio is greater than 0.9, the time coverage is considered sufficient. Spatial integrity is calculated by statistically analyzing the data effectiveness rate of each sensor within the observation window, removing outliers and missing values, and then calculating the percentage of valid data points. A comprehensive integrity evaluation index is then established. Through weighted fusion time integrity Spatial integrity The weighting coefficients are adjusted according to the emphasis of different control scenarios. Typically, the time integrity weight is set to 0.6, and the spatial integrity weight is set to 0.4. When the value is below 0.75, it indicates that the response data is insufficient to support accurate deviation calculation. In this case, the observation window should be extended or the sensor self-test process should be triggered.

[0148] The calculation of transient process deviation focuses on dynamic performance indicators, including expected rise time, allowable overshoot range, and settling time limit. The measured transient response characteristics are compared with these indicators one by one to calculate the relative deviation. For example, if the expected rise time is 45 seconds and the measured rise time is 52 seconds, the rise time deviation is 15.6%. For overshoot, when the measured overshoot is 8% and the allowable range is 0-5%, the overshoot deviation is 3 percentage points. Different weights are assigned to the deviations of different dynamic indicators according to their severity of impact on system performance, with higher weights for response speed indicators and lower weights for oscillation indicators, thus constructing a comprehensive transient process deviation. This deviation not only reflects numerical differences, but also distinguishes between positive and negative deviations by introducing a deviation direction coefficient. For example, an overly fast response can lead to system shock, while an overly slow response can affect control timeliness. The two have different mechanisms of influence, which are reflected in the deviation calculation.

[0149] The calculation of steady-state deviation focuses on the degree of conformity between the final operating state and the target state. Steady-state performance indicators include the target setpoints and allowable fluctuation ranges of each key parameter. The difference between the steady-state values ​​of each parameter extracted from the steady-state response characteristics and their corresponding target setpoints is calculated to obtain the absolute deviation. This absolute deviation is then normalized by dividing by the allowable fluctuation range to obtain the relative deviation. When the steady-state value of a parameter is exactly within the allowable range boundary, the relative deviation is 1; if it exceeds the allowable range, the relative deviation is greater than 1. For the steady-state deviations of multiple parameters, a weighted summation method is used to calculate the comprehensive steady-state deviation. The weighting is determined based on the contribution of each parameter to the waste heat recovery efficiency. The weight of the steady-state deviation of the hot air temperature is usually set to 0.4, the weight of the heat exchanger efficiency deviation is 0.3, and the sum of the weights of the deviations of the other auxiliary parameters is 0.3. In addition, volatility evaluation is introduced into the calculation of steady-state deviation. The steady-state maintenance quality is quantified by calculating the standard deviation of the steady-state data. Excessive volatility will lead to an increase in steady-state deviation even if the mean meets the standard.

[0150] The fusion calculation process uses the system response integrity evaluation index to adjust the credibility of transient process deviations and steady-state result deviations. When the integrity of the response data is low, the credibility of the deviations calculated based on incomplete data decreases, and their weighting needs to be reduced during fusion. Specifically, the fusion method involves adjusting the transient process deviation... Completeness indicators Multiplying them yields the corrected transient bias contribution, and the steady-state result bias is then calculated. and The corrected steady-state deviation contribution is obtained by multiplying the two values, and then a weighted sum is calculated according to the importance ratio of transient and steady-state performance in the control evaluation. In waste heat recovery scenarios, steady-state performance is usually more important because the system operates in steady state for most of the time, and steady-state efficiency directly determines the total energy recovered. Therefore, the weight of steady-state deviation is set to 0.65, and the weight of transient deviation is set to 0.35. The final control effect deviation information is obtained. It is a comprehensive indicator; the smaller the value, the closer the actual effect of the control implementation plan is to the expected goal.

[0151] Control performance deviation information also needs to be analyzed historically, statistically comparing the current deviation with the deviations from recent control executions to calculate the deviation trend. If the deviation shows a continuous increasing trend, it indicates that the system characteristics have drifted or the matching degree between the control model and the actual operating conditions has decreased, triggering the parameter adaptive adjustment mechanism. The deviation information also records the specific stage at which the deviation occurred: whether it is a transient process deviation or a steady-state result deviation, whether a deviation of a specific parameter is dominant or a combined deviation of multiple parameters. This detailed information provides precise guidance for subsequent parameter adjustments. By establishing a mapping relationship between deviation information and adjustment strategies, a closed-loop feedback from deviation identification to parameter optimization is achieved, enabling the monitoring system to have the ability to continuously learn and self-optimize.

[0152] A second aspect of the present invention provides a multi-sensor fusion monitoring system for a waste heat recovery system of a stenter, comprising:

[0153] The data alignment unit is used to acquire multi-source heterogeneous sensing data characterizing the thermal state of the system through multiple sensors distributed in different spatial locations of the waste heat recovery system of the stenter. Based on the spatial location information and data acquisition time information of each sensor, the multi-source heterogeneous sensing data is spatiotemporally aligned to obtain spatiotemporally aligned multi-source sensing data.

[0154] The data fusion unit is used to perform fusion calculations on the spatiotemporally aligned multi-source sensor data according to the physical association constraints between the spatiotemporally aligned multi-source sensor data, and obtain the system comprehensive state evaluation result.

[0155] The optimization control unit is used to calculate the deviation between the system comprehensive state assessment result and the preset system normal operation state boundary to determine the abnormal feature description information, and to perform multi-objective optimization solution based on the abnormal feature description information and the system comprehensive state assessment result to generate a control execution scheme for the waste heat recovery system.

[0156] The feedback adjustment unit is used to collect system response data after executing the control execution scheme, compare and analyze the system response data with the expected target of the control execution scheme, and obtain control effect deviation information. The control effect deviation information is used to adaptively adjust the parameters in the fusion calculation and the multi-objective optimization solution.

[0157] A third aspect of the present invention provides an electronic device, comprising:

[0158] processor;

[0159] Memory used to store processor-executable instructions;

[0160] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0161] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0162] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0163] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multi-sensor fusion monitoring method for a waste heat recovery system in a stenter, characterized in that, include: Multi-source heterogeneous sensing data characterizing the thermal state of the waste heat recovery system of the stenter is acquired by multiple sensors distributed at different spatial locations. Based on the spatial location information and data acquisition time information of each sensor, the multi-source heterogeneous sensing data is spatiotemporally aligned to obtain spatiotemporally aligned multi-source sensing data. Based on the physical correlation constraints between the spatiotemporally aligned multi-source sensor data, the spatiotemporally aligned multi-source sensor data are fused and calculated to obtain the system comprehensive state evaluation result; Based on the comprehensive system status assessment results and the preset system normal operation status boundary, deviation is calculated to determine abnormal feature description information. Based on the abnormal feature description information and the comprehensive system status assessment results, multi-objective optimization is performed to generate a control execution scheme for the waste heat recovery system. After executing the control execution scheme, system response data is collected. The system response data is compared and analyzed with the expected target of the control execution scheme to obtain control effect deviation information. The control effect deviation information is used to adaptively adjust the parameters in the fusion calculation and the multi-objective optimization solution.

2. The method according to claim 1, characterized in that, Multi-source heterogeneous sensing data characterizing the thermal state of the waste heat recovery system of the stenter is acquired by multiple sensors distributed at different spatial locations. Based on the spatial location information and data acquisition time information of each sensor, the multi-source heterogeneous sensing data is spatiotemporally aligned to obtain spatiotemporally aligned multi-source sensing data, including: When acquiring multi-source heterogeneous sensing data characterizing the thermal state of the system using multiple sensors distributed at different spatial locations in the waste heat recovery system of the stenter, the spatial coordinate position information and data acquisition time information of each sensor are recorded simultaneously. Based on the spatial coordinate location information of each sensor and the pipeline topology information of the waste heat recovery system, the spatial distribution relationship characteristics of the sensors are determined. These characteristics describe the spatial distance relationships and heat transfer path relationships between the sensors. Based on the spatial distribution characteristics of the sensors and the data acquisition time information, the time delay compensation between sensor data at different spatial locations is calculated; The timestamp correction process is performed on the multi-source heterogeneous sensing data based on the time delay compensation amount, and the spatial coordinate normalization process is performed on the timestamp-corrected multi-source heterogeneous sensing data according to the spatial distribution relationship characteristics of the sensors, so as to obtain spatiotemporally aligned multi-source sensing data.

3. The method according to claim 1, characterized in that, Based on the physical correlation constraints between the spatiotemporally aligned multi-source sensor data, the spatiotemporally aligned multi-source sensor data are fused and calculated to obtain the system comprehensive state evaluation result, including: Based on the energy transfer coupling relationship between the spatiotemporally aligned multi-source sensor data, a state evolution relationship describing the thermal energy flow process of the waste heat recovery system is determined. This state evolution relationship represents the constraint effect of physical correlation constraints on the system state changes. The state evolution relationship is used to perform state estimation calculation on the spatiotemporally aligned multi-source sensing data to predict the theoretical state value of each sensor's monitoring position at the current moment, thus obtaining predicted state data; The spatiotemporally aligned multi-source sensor data and the predicted state data are subjected to deviation analysis to calculate the deviation between the actual measured values ​​of each sensor and the predicted state data. The measurement reliability of each sensor's measurement data is quantitatively evaluated based on the aforementioned deviation, resulting in a reliability evaluation index for each sensor. A dynamic weight allocation strategy is determined based on the credibility assessment index, and the spatiotemporally aligned multi-source sensor data is fused and calculated according to the dynamic weight allocation strategy to obtain the system comprehensive state assessment result.

4. The method according to claim 3, characterized in that, A dynamic weight allocation strategy is determined based on the reliability assessment index, and the spatiotemporally aligned multi-source sensor data is fused and calculated according to the dynamic weight allocation strategy to obtain a comprehensive system state assessment result, including: Based on the credibility assessment index, it is determined that there are sensor data with measurement anomalies in the spatiotemporally aligned multi-source sensing data. The reliability evaluation index corresponding to the sensor data with measurement anomalies is suppressed using a nonlinear decay function to obtain a corrected reliability evaluation index; Based on the revised reliability assessment index, the fusion weight coefficient of each sensor data is calculated, and the dynamic weight allocation strategy is determined through the fusion weight coefficient. According to the dynamic weight allocation strategy, the spatially adjacent sensor data in the spatiotemporally aligned multi-source sensing data are locally weighted and fused to obtain the regional state value. The fusion weight coefficient of the local weighted fusion is adjusted according to the spatial position relationship of the sensors. A global weighted fusion of all regional state values ​​is performed to obtain the comprehensive system state assessment result. The fusion weight coefficient of the global weighted fusion is adjusted according to the data integrity of the regional state values.

5. The method according to claim 1, characterized in that, Based on the comprehensive system status assessment results and the preset normal system operating status boundaries, deviation calculations are performed to determine abnormal feature description information. Based on the abnormal feature description information and the comprehensive system status assessment results, multi-objective optimization is performed to generate a control execution scheme for the waste heat recovery system, including: The system comprehensive status assessment result is compared with the preset system normal operation status boundary, the deviation value of the system comprehensive status assessment result from the system normal operation status boundary is calculated, and the abnormal feature description information is determined based on the deviation value; Based on the description of the abnormal features, the key influencing factors causing state deviation in the waste heat recovery system are identified, and a state callback path planning strategy is determined for the key influencing factors. Based on the state callback path planning strategy and the system comprehensive state evaluation results, a multi-objective optimization problem is determined, which includes a system energy efficiency optimization objective function and a state deviation minimization objective function. The system energy efficiency optimization objective function takes maximizing waste heat recovery efficiency as the optimization direction, and the state deviation minimization objective function takes minimizing the deviation value as the optimization direction. The multi-objective optimization problem is solved to obtain the optimal values ​​of control variables that satisfy the comprehensive optimality of multiple objective functions. Based on these control variable values, a control execution scheme for the waste heat recovery system is generated.

6. The method according to claim 5, characterized in that, Based on the anomaly characteristic description information, the key influencing factors causing state deviation in the waste heat recovery system are identified, and a state callback path planning strategy is determined for the key influencing factors, including: The abnormal feature description information is parsed to determine the feature vector describing the system state deviation characteristics. The feature vector includes the spatial distribution characteristics of the state deviation and the temporal evolution characteristics of the state deviation. Based on the causal correlation analysis between the eigenvectors and the control variables of the waste heat recovery system, the key influencing factors that play a dominant role in state deviation are identified; For the aforementioned key influencing factors, a dynamic transmission relationship between the control variable adjustment action and the system state change response is determined. This dynamic transmission relationship represents the process by which the system state evolves from the current deviation state to the boundary of the normal operating state after the control variable is adjusted. Based on the dynamic transmission relationship, multiple candidate adjustment paths are determined to restore the system state from the current deviation state to the normal operating state boundary. The multiple candidate adjustment paths are evaluated, and the adjustment path with the shortest state recovery time and the smallest adjustment magnitude of the control variable is selected as the state callback path planning strategy.

7. The method according to claim 1, characterized in that, After executing the control execution scheme, system response data is collected. The system response data is then compared and analyzed with the expected target of the control execution scheme to obtain control effect deviation information, including: After executing the control execution scheme, system response data of the waste heat recovery system is collected within a preset response observation time window; The system response data is subjected to time-series feature extraction to obtain transient response features and steady-state response features. Based on the transient response characteristics and the steady-state response characteristics, a system response integrity evaluation index is determined. The system response integrity evaluation index is used to quantify the completeness of the system response data in representing the control execution process. The transient response characteristics are compared with the dynamic performance indicators in the expected target to calculate the transient process deviation, and the steady-state response characteristics are compared with the steady-state performance indicators in the expected target to calculate the steady-state result deviation. The transient process deviation and the steady-state result deviation are fused and calculated based on the system response integrity evaluation index to obtain the control effect deviation information.

8. A multi-sensor fusion monitoring system for waste heat recovery in a stenter machine, used to implement the method as described in any one of claims 1-7, characterized in that, include: The data alignment unit is used to acquire multi-source heterogeneous sensing data characterizing the thermal state of the system through multiple sensors distributed in different spatial locations of the waste heat recovery system of the stenter. Based on the spatial location information and data acquisition time information of each sensor, the multi-source heterogeneous sensing data is spatiotemporally aligned to obtain spatiotemporally aligned multi-source sensing data. The data fusion unit is used to perform fusion calculations on the spatiotemporally aligned multi-source sensor data according to the physical association constraints between the spatiotemporally aligned multi-source sensor data, and obtain the system comprehensive state evaluation result. The optimization control unit is used to calculate the deviation between the system comprehensive state assessment result and the preset system normal operation state boundary to determine the abnormal feature description information, and to perform multi-objective optimization solution based on the abnormal feature description information and the system comprehensive state assessment result to generate a control execution scheme for the waste heat recovery system. The feedback adjustment unit is used to collect system response data after executing the control execution scheme, compare and analyze the system response data with the expected target of the control execution scheme, and obtain control effect deviation information. The control effect deviation information is used to adaptively adjust the parameters in the fusion calculation and the multi-objective optimization solution.

9. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 7.

Images