Filtering device and control method for sludge filtration in gear oil

Abstract

The present application relates to the technical field of device control, and more particularly to a filter device and control method for filtering sludge in gear oil, wherein a data acquisition module acquires and pre-processes pressure, flow and temperature data to generate a multidimensional data stream; a digital twin module drives a hybrid virtual model based on the data stream to output system state estimates; a parameter optimization module performs multi-objective optimization calculation to generate pressure and flow set values, and iteratively corrects them through virtual verification; a learning adaptation module constructs a knowledge base and adapts to new working conditions using a meta-learning algorithm; and an execution control module converts the optimized set values into control instructions and feeds back execution deviations. The modules constitute a multi-level feedback loop and update path through data quality evaluation results, optimized set value virtual verification, execution deviations and adaptation rules, realize accurate collaborative control of filtering pressure and flow, and effectively improve the thoroughness of sludge removal and the filtering efficiency.

Description

Technical Field

[0001] This invention relates to the field of equipment control technology, and in particular to a filtration device and control method for filtering sludge in gear oil. Background Technology

[0002] During operation, gear oil produces sludge particles due to oxidation and contaminant intrusion. The accumulation of sludge particles may reduce lubrication efficiency and accelerate equipment wear. Therefore, the filtration device is integrated into the oil circuit system. The pump drives the oil to flow through multiple layers of filter media. Larger particles are intercepted by surface filtration, while fine particles are adsorbed by deep filtration, realizing continuous separation of sludge. The purified oil is returned to the circulation to maintain lubrication performance and extend component life.

[0003] Existing filtration technology for filtering sludge in gear oil suffers from the following technical pain points: inaccurate control of filtration pressure and flow rate leads to incomplete sludge removal and low filtration efficiency. This is because existing devices lack an automated adjustment mechanism and cannot dynamically optimize parameters based on the oil condition. For example, in industrial gear transmission systems, sludge easily accumulates and forms viscous impurities. If the filtration pressure is insufficient, the impurities cannot be effectively trapped, while flow fluctuations can cause filter element blockage or bypass, ultimately resulting in gear wear and system downtime, highlighting the deficiency of insufficient control precision. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a filtration device and control method for filtering sludge in gear oil, solving the technical problems of incomplete sludge removal and low filtration efficiency caused by inaccurate control of filtration pressure and flow rate.

[0005] To solve the above-mentioned technical problems, the specific contents of the present invention are as follows: In a first aspect, the filtration device for filtering sludge in gear oil provided by the present invention includes a physical device and a control device communicatively connected to the physical device. The physical device includes a gear oil circulation pipeline, a multi-layer composite filter installed in the pipeline, a pressure sensor, a flow sensor, an intelligent electric regulating valve, and a variable frequency drive pump. The control device includes: The data acquisition module is used to acquire pressure data and flow data during the gear oil filtration process through the pressure sensor and flow sensor, preprocess the pressure data and flow data to eliminate measurement noise, and output the preprocessed sensor data to the digital twin module. The digital twin module is used to receive the preprocessed sensor data, drive the virtual model to run using the preprocessed sensor data as input, and output system state estimates and trend prediction data; the digital twin module also feeds back the data quality assessment results to the data acquisition module; The parameter optimization module is used to receive the system state estimate output by the digital twin module, perform multi-objective optimization calculations on the system state estimate, and generate optimized setpoints for pressure and flow. The parameter optimization module sends the optimized setpoints to the digital twin module for virtual verification, and iteratively corrects the optimized setpoints based on the virtual verification results. The learning and adaptation module is used to receive the system state estimate and trend prediction data output by the digital twin module, update the control strategy mapping relationship based on historical operating data and current operating data, and generate adaptation rules when a new operating mode is identified and send them to the digital twin module. The execution control module is used to receive the optimized setpoint determined by the parameter optimization module after iterative correction, and generate control commands for the variable frequency drive pump and the intelligent electric regulating valve according to the optimized setpoint; the execution control module also feeds back the deviation between the actual system response and the predicted response after the command is executed to the digital twin module; Specifically, the data acquisition module and the digital twin module form a first feedback loop through the data quality assessment results; the parameter optimization module and the digital twin module form a second feedback loop through virtual verification of the optimized set value; the execution control module and the digital twin module form a third feedback loop through the deviation; and the learning adaptation module and the digital twin module form an update path through the adaptation rules.

[0006] Furthermore, in the filtration device for filtering sludge in gear oil according to the present invention, the data acquisition module is further configured as follows: Acquire temperature data measured by a temperature sensor; Based on the temperature data, a preset oil viscosity-temperature characteristic curve is queried, and the dynamic viscosity value of the oil is output. The dynamic viscosity value of the oil is spatiotemporally aligned and fused with the preprocessed pressure data and flow data to generate a multidimensional data stream with timestamps. The multidimensional data stream is used as the input data for the digital twin module.

[0007] Furthermore, in the filtration device for filtering sludge in gear oil according to the present invention, the digital twin module is further configured as follows: A mechanistic model based on fluid mechanics and porous media filtration theory is used to simulate the oil flow and particulate matter deposition process; Run a long short-term memory network data-driven model to make time-series predictions of dynamic changes in differential pressure. The mechanism model is mixed with the data-driven model, and the hybrid model is driven by the multidimensional data flow to adjust the internal state variables of the virtual model. The system state estimate and trend prediction data output by the calibrated virtual model are sent to the parameter optimization module and the learning adaptation module.

[0008] Furthermore, in the filtration device for filtering sludge in gear oil according to the present invention, the parameter optimization module is further configured as follows: Invoke a multi-objective particle swarm optimization algorithm with filtration efficiency, energy consumption index and equipment life as optimization objectives; The pressure limit of the filter element and the minimum stable flow rate of the system are loaded into the multi-objective particle swarm optimization algorithm as constraints. The candidate optimization settings obtained by the algorithm are sent to the digital twin module to request the execution of virtual tests on stress shock and flow stability in the virtual model; The system receives the virtual test results returned by the digital twin module. If the results do not meet the robustness requirements, the system triggers the multi-objective particle swarm optimization algorithm to solve the problem again.

[0009] Furthermore, in the filtration device for filtering sludge in gear oil according to the present invention, the learning and adaptation module is further configured as follows: Record the control parameters and corresponding filtration effect evaluation indicators under different sludge characteristics, ambient temperature and load conditions, and build an experience knowledge base; Analyze the characteristics of real-time operational data and output the operational mode recognition results; When the operating mode recognition result indicates a new mode, the meta-learning algorithm is invoked to extract prior knowledge from the experience knowledge base and to fine-tune the control strategy using new operating condition samples. The fine-tuned adaptation rules are sent to the digital twin module to update the parameters of the data-driven model.

[0010] Furthermore, in the filtration device for filtering sludge in gear oil according to the present invention, the execution control module is further configured as follows: Receive the final optimized setting value sent by the parameter optimization module; The final optimized setpoint is input into the embedded system dynamic response model to predict the response of the variable frequency drive pump and the intelligent electric regulating valve in the future time domain. Based on the predicted response, the optimal speed curve of the variable frequency drive pump and the optimal opening sequence of the intelligent electric regulating valve are calculated using a rolling optimization method. The control command for the current moment is extracted from the optimal speed curve and the optimal opening sequence and sent to the variable frequency drive pump and the intelligent electric regulating valve.

[0011] Furthermore, in the filtration device for filtering sludge in gear oil according to the present invention, the data acquisition module and the digital twin module form a first feedback closed loop through the data quality assessment results, including: The data acquisition module transmits the multidimensional data stream to the digital twin module; The digital twin module uses the multidimensional data stream to drive the virtual model to run, performs data consistency analysis, and generates data quality assessment results. The digital twin module sends the data quality assessment results to the data acquisition module; The data acquisition module dynamically adjusts the signal sampling frequency of the pressure sensor and flow sensor and the noise covariance parameter of the Kalman filter algorithm based on the received data quality assessment results.

[0012] Furthermore, in the filtration device for filtering sludge in gear oil according to the present invention, the parameter optimization module and the digital twin module form a second feedback closed loop through virtual verification of the optimized set value, including: The parameter optimization module sends the candidate optimization settings to the digital twin module; The digital twin module performs virtual tests on the candidate optimization settings and traffic stability in a virtual model, generating robustness assessment results. The digital twin module returns the robustness assessment results to the parameter optimization module; The parameter optimization module determines whether the requirements are met based on the robustness assessment results. If not, it adjusts the search weights or particle initialization range of the multi-objective particle swarm optimization algorithm and recalculates the candidate optimization settings.

[0013] Furthermore, in the filtration device for filtering sludge in gear oil according to the present invention, the execution control module and the digital twin module form a third feedback closed loop through the deviation, including: After the execution control module sends the control command, it collects the actual speed of the variable frequency drive pump, the actual opening degree of the intelligent electric regulating valve, and the actual system pressure and flow values ​​measured by the pressure sensor and flow sensor. The execution control module compares the actual system pressure and flow values ​​with the predicted response data made by the system dynamic response model, and calculates the prediction deviation. The execution control module sends the prediction deviation to the digital twin module; The digital twin module uses the prediction bias to correct the parameters describing the dynamic characteristics of the system in the virtual model.

[0014] Secondly, the filtration device control method for filtering sludge in gear oil provided by the present invention, applied to the filtration device for filtering sludge in gear oil, includes: Step 1: Acquire pressure and flow data during the gear oil filtration process using the pressure sensor and flow sensor, preprocess the pressure and flow data to eliminate measurement noise, and output the preprocessed sensor data. Step 2: Receive the preprocessed sensor data, use the preprocessed sensor data as input to drive the virtual model to run, and output system state estimates and trend prediction data; at the same time, perform data consistency analysis to generate data quality assessment results, and feed back the data quality assessment results to adjust the data acquisition strategy; Step 3: Receive the system state estimate, perform multi-objective optimization calculation on the system state estimate to generate optimized setpoints for pressure and flow; send the optimized setpoints for virtual verification, and iteratively correct the optimized setpoints based on the virtual verification results; Step 4: Receive the system state estimate and trend prediction data, update the control strategy mapping relationship based on historical operating data and current operating data; when a new operating mode is identified, generate adaptation rules and send them to update the virtual model; Step 5: Receive the optimized setpoints determined after iterative correction, generate control commands for the variable frequency drive pump and the intelligent electric regulating valve based on the optimized setpoints, and use the deviation between the actual system response and the predicted response after the command is executed as feedback for model correction; In this process, the adjustment of the data acquisition strategy in step 1, based on the data quality assessment results from step 2, constitutes the first feedback loop; the virtual verification and iterative correction of the optimized setpoints in step 3 constitute the second feedback loop; the model correction in step 5, based on the feedback deviation, constitutes the third feedback loop; and the adaptation rules generated in step 4 are used to update the virtual model, forming the update path.

[0015] Beneficial effects of this invention; This invention constructs a multi-level feedback network, including a first feedback loop, a second feedback loop, a third feedback loop, and an update path, through the coordinated operation of a data acquisition module, a digital twin module, a parameter optimization module, a learning and adaptation module, and an execution control module. The data acquisition module collects pressure, flow, and temperature data in real time, preprocesses them to generate a high-quality multi-dimensional data stream, providing reliable input to the digital twin module. The digital twin module uses a hybrid model to output system state estimates and trend prediction data, and optimizes the data acquisition strategy by providing feedback data quality assessment results through data consistency analysis. The parameter optimization module performs multi-objective operations based on the state estimates. The system optimizes calculations to generate optimal pressure and flow setpoints, which are then iteratively corrected through virtual verification to ensure robustness. The learning and adaptation module, through the construction of an experience knowledge base and meta-learning algorithms, quickly adapts control strategies and updates model parameters when new operating modes are identified. The execution control module transforms the optimized setpoints into control commands, driving the variable frequency drive pump and intelligent electric regulating valve, and achieves online model correction through predictive deviation feedback. This integrated feedback mechanism enables the system to dynamically maintain precise and coordinated control of filtration pressure and flow, effectively improving the thoroughness of sludge removal and filtration efficiency, while also enhancing the system's adaptability to unknown operating conditions and extending equipment lifespan. Attached Figure Description

[0016] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on the drawings without creative effort.

[0017] Figure 1 This is a flowchart of the control method for the filtration device used for filtering sludge in gear oil according to the present invention. Detailed Implementation

[0018] To make the technical solution of the present invention clearer, the present invention will be clearly and completely described below with reference to specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. 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. The present invention provided by various embodiments will be described in detail below with reference to the accompanying drawings. To better understand the purpose of the present invention, the present invention will be described in further detail below.

[0019] In a first aspect, the filtration device for filtering sludge in gear oil provided by the present invention includes a physical device and a control device communicatively connected to the physical device. The physical device includes a gear oil circulation pipeline, a multi-layer composite filter installed in the pipeline, a pressure sensor, a flow sensor, an intelligent electric regulating valve, and a variable frequency drive pump. The control device includes: The data acquisition module is used to acquire pressure data and flow data during the gear oil filtration process through the pressure sensor and flow sensor, preprocess the pressure data and flow data to eliminate measurement noise, and output the preprocessed sensor data to the digital twin module. The digital twin module is used to receive the preprocessed sensor data, drive the virtual model to run using the preprocessed sensor data as input, and output system state estimates and trend prediction data; the digital twin module also feeds back the data quality assessment results to the data acquisition module; The parameter optimization module is used to receive the system state estimate output by the digital twin module, perform multi-objective optimization calculations on the system state estimate, and generate optimized setpoints for pressure and flow. The parameter optimization module sends the optimized setpoints to the digital twin module for virtual verification, and iteratively corrects the optimized setpoints based on the virtual verification results. The learning and adaptation module is used to receive the system state estimate and trend prediction data output by the digital twin module, update the control strategy mapping relationship based on historical operating data and current operating data, and generate adaptation rules when a new operating mode is identified and send them to the digital twin module. The execution control module is used to receive the optimized setpoint determined by the parameter optimization module after iterative correction, and generate control commands for the variable frequency drive pump and the intelligent electric regulating valve according to the optimized setpoint; the execution control module also feeds back the deviation between the actual system response and the predicted response after the command is executed to the digital twin module; Specifically, the data acquisition module and the digital twin module form a first feedback loop through the data quality assessment results; the parameter optimization module and the digital twin module form a second feedback loop through virtual verification of the optimized set value; the execution control module and the digital twin module form a third feedback loop through the deviation; and the learning adaptation module and the digital twin module form an update path through the adaptation rules.

[0020] The control unit of the filtration device achieves precise control of the filtration process through the coordinated operation of multiple modules. The data acquisition module collects pressure and flow data in real time during the gear oil filtration process using pressure and flow sensors. The raw data is amplified and preliminarily filtered by a signal conditioning circuit, and then converted into digital signals by a high-precision analog-to-digital converter. The digital signals are recursively processed using a Kalman filter algorithm, continuously correcting the state estimate through prediction and correction stages to effectively suppress measurement noise, ultimately outputting preprocessed sensor data. This preprocessing provides a high-quality data foundation for subsequent modules.

[0021] The digital twin module receives preprocessed sensor data from the data acquisition module, using this as input to drive the virtual model. This virtual model employs a hybrid modeling approach, combining a mechanistic model based on fluid dynamics and porous media filtration theory with a long short-term memory (LSTM) network-driven model. The mechanistic model simulates the flow behavior of oil in the filter medium and the particulate matter deposition process based on fluid dynamics principles, while the data-driven model focuses on time-series prediction of dynamic changes in system differential pressure. The virtual model dynamically adjusts its internal state variables using the real-time input data stream to achieve model calibration and outputs system state estimates and trend predictions. Simultaneously, the digital twin module performs data consistency analysis during model calibration, generating data quality assessment results characterizing data reliability, and feeding these results back to the data acquisition module.

[0022] The parameter optimization module receives the system state estimate output by the digital twin module and invokes a multi-objective particle swarm optimization algorithm with filtration efficiency, energy consumption, and equipment lifespan as optimization objectives. During the algorithm's solution process, the pressure limit of the filter element and the minimum stable flow rate of the system are treated as constraints. The candidate optimization setpoints obtained from the algorithm are sent to the digital twin module, requesting virtual testing of pressure shock and flow stability in a virtual model environment. The robustness assessment results generated by the digital twin module after executing the virtual tests are returned to the parameter optimization module. Based on these assessment results, the parameter optimization module determines whether the candidate optimization setpoints meet the robustness requirements. If not, the search weights or particle initialization range of the multi-objective particle swarm optimization algorithm are adjusted, and the optimization calculation is re-performed, forming an iterative correction loop until a satisfactory optimization setpoint is obtained.

[0023] The learning and adaptation module continuously monitors the long-term operating performance of the system, recording the control parameters and corresponding filtration effect evaluation indicators used under different sludge characteristics, ambient temperatures, and load conditions, and constructs and updates the experience knowledge base. The module analyzes the characteristics of real-time operating data and outputs the operating mode recognition results. When a new operating mode is identified, the learning and adaptation module calls a meta-learning algorithm to extract prior knowledge from the experience knowledge base and quickly fine-tunes the control strategy using a small number of new operating condition samples. The resulting adaptation rules are sent to the digital twin module to update the parameters of its data-driven model, thereby enhancing the virtual model's adaptability to new operating conditions and its predictive accuracy.

[0024] The execution control module receives the final optimized setpoint determined by the parameter optimization module after iterative correction, and uses this as the control target. An embedded system dynamic response model is used to predict the response of the variable frequency drive pump and the intelligent electric regulating valve to control commands over a future period. Using a rolling optimization method, the optimal speed curve of the variable frequency drive pump and the optimal opening sequence of the intelligent electric regulating valve are calculated. The execution control module sends the optimal control command for the current moment in the sequence to the variable frequency drive pump and the intelligent electric regulating valve. After the command is executed, the module collects the actual speed of the variable frequency drive pump, the actual opening of the intelligent electric regulating valve, and the actual system pressure and flow values ​​measured by pressure and flow sensors. It compares the actual response data with the predicted response data made by the system dynamic response model and calculates the prediction deviation. This prediction deviation is fed back to the digital twin module for online correction of the parameters describing the system's dynamic characteristics in the virtual model, thereby reducing the behavioral differences between the virtual model and the physical entity.

[0025] Through the collaborative operation and data interaction among the aforementioned modules, the data acquisition module and the digital twin module form a first feedback loop by transmitting and responding to data quality assessment results, aiming to improve the quality of source data. The parameter optimization module and the digital twin module form a second feedback loop by virtual verification and iterative correction of optimized setpoints, aiming to ensure the robustness of optimization results. The execution control module and the digital twin module form a third feedback loop by feedback on prediction deviations and model correction, aiming to maintain the consistency between the virtual model and the physical entity. The learning and adaptation module and the digital twin module form a knowledge update path by transmitting adaptation rules and updating the model, aiming to improve the system's adaptability to unknown operating conditions. This multi-layered, bidirectional feedback network enables the system to achieve precise and coordinated control of filtration pressure and flow, effectively solving the technical problems of incomplete sludge removal and low filtration efficiency.

[0026] Specifically, in the filtration device for filtering sludge in gear oil according to the present invention, the data acquisition module is further configured as follows: Acquire temperature data measured by a temperature sensor; Based on the temperature data, a preset oil viscosity-temperature characteristic curve is queried, and the dynamic viscosity value of the oil is output. The dynamic viscosity value of the oil is spatiotemporally aligned and fused with the preprocessed pressure data and flow data to generate a multidimensional data stream with timestamps. The multidimensional data stream is used as the input data for the digital twin module.

[0027] The data acquisition module collects temperature data in real time during the gear oil filtration process using a temperature sensor installed in the gear oil circulation pipeline. This sensor continuously monitors oil temperature changes, providing raw temperature information for subsequent data processing. After acquiring the temperature data, the module queries a preset oil viscosity-temperature characteristic curve. This curve is pre-stored in the system based on the physical properties of the gear oil. The module outputs the corresponding dynamic viscosity value of the oil through interpolation or a lookup table, thus accurately reflecting the influence of temperature on oil viscosity and enhancing the system's ability to perceive oil state parameters.

[0028] Next, the data acquisition module performs spatiotemporal alignment and fusion of the calculated oil dynamic viscosity value with the preprocessed pressure and flow data. The preprocessing includes amplifying and initially filtering the raw sensor signals using a signal conditioning circuit, converting them into digital signals via a high-precision analog-to-digital converter, and applying a Kalman filter algorithm to eliminate measurement noise. Spatiotemporal alignment ensures synchronization of pressure, flow, and viscosity data across time by adding a uniform timestamp to each data point, generating a timestamped multidimensional data stream. This data stream integrates multiple source parameters, forming a structured dataset reflecting the overall state of the system.

[0029] Finally, the data acquisition module uses the generated multidimensional data stream as input data for the digital twin module. The multidimensional data stream is transmitted to the digital twin module through the communication interface, providing the virtual model with real-time, multi-dimensional, high-quality driving data. Each step is sequentially linked: the introduction of temperature data expands the data dimensions; the calculation of oil dynamic viscosity values ​​enhances parameter correlation; and spatiotemporal alignment and fusion ensure data consistency and timeliness. This lays the foundation for accurate state estimation and trend prediction by the digital twin module, supporting the coordinated control of the filtration device.

[0030] Specifically, in the filtration device for filtering sludge in gear oil according to the present invention, the digital twin module is further configured as follows: A mechanistic model based on fluid mechanics and porous media filtration theory is used to simulate the oil flow and particulate matter deposition process; Run a long short-term memory network data-driven model to make time-series predictions of dynamic changes in differential pressure. The mechanism model is mixed with the data-driven model, and the hybrid model is driven by the multidimensional data flow to adjust the internal state variables of the virtual model. The system state estimate and trend prediction data output by the calibrated virtual model are sent to the parameter optimization module and the learning adaptation module.

[0031] The digital twin module operates based on a mechanistic model derived from fluid mechanics and porous media filtration theory. This model describes oil flow behavior using the Navier-Stokes equations from fluid mechanics and simulates particulate matter deposition in the filter media using Darcy's law from porous media filtration theory. The mechanistic model divides the gear oil circulation pipeline and multi-layer composite filter into discrete computational units, sets boundary conditions such as inlet pressure and outlet flow rate, and simulates the dynamics of oil velocity distribution, pressure field changes, and the retention and accumulation of sludge particles on the filter surface through numerical solutions. This establishes a physical representation of the oil filtration process, providing a theoretical basis for system state estimation.

[0032] The digital twin module simultaneously runs a Long Short-Term Memory (LSTM) network data-driven model. This model uses historical differential pressure data as training samples to construct a multi-layer recurrent neural network structure, learning the sequential characteristics of differential pressure changes over time. The data-driven model receives differential pressure time-series information from real-time multi-dimensional data streams, captures long-term dependencies through a gating mechanism, and outputs predicted differential pressure values ​​for future time steps. This enables time-series prediction of system differential pressure dynamics, compensating for the limitations of mechanistic models under complex nonlinear conditions and enhancing the model's adaptability to actual operational fluctuations.

[0033] The digital twin module blends mechanistic and data-driven models, employing data fusion strategies such as the Kalman filter algorithm based on error covariance to collaboratively correct the physical simulation output of the mechanistic model with the time-series prediction results of the data-driven model. The hybrid model is driven by a multi-dimensional data stream, including synchronously collected parameters such as pressure, flow rate, and oil dynamic viscosity. This data stream serves as input to update the internal weights of the hybrid model in real time, dynamically adjusting the internal state variables of the virtual model, including filter porosity, oil viscosity coefficient, and particulate deposition rate, ensuring that the state variables of the virtual model remain consistent with the actual operating state of the physical entity.

[0034] The calibrated virtual model outputs system state estimates and trend prediction data. The system state estimates cover current system pressure, flow rate, and sludge filtration efficiency, while the trend prediction data includes future pressure differential change trends and filter clogging risk predictions. The digital twin module sends the output data to the parameter optimization module and the learning and adaptation module via a data bus or communication protocol. The parameter optimization module uses the state estimates to perform multi-objective optimization calculations, and the learning and adaptation module updates its experience knowledge base based on the trend prediction data, forming a data collaboration and feedback pathway between modules.

[0035] Mechanistic models provide the physical basis for the filtering process, data-driven models enhance the accuracy of time series predictions, hybrid models improve the overall representation capability through data fusion, drive adjustments to ensure that the model and the entity are synchronized, and output data supports subsequent optimization and learning, ultimately enabling the digital twin module to accurately estimate and predict the system state.

[0036] Specifically, in the filtration device for filtering sludge in gear oil according to the present invention, the parameter optimization module is further configured as follows: Invoke a multi-objective particle swarm optimization algorithm with filtration efficiency, energy consumption index and equipment life as optimization objectives; The pressure limit of the filter element and the minimum stable flow rate of the system are loaded into the multi-objective particle swarm optimization algorithm as constraints. The candidate optimization settings obtained by the algorithm are sent to the digital twin module to request the execution of virtual tests on stress shock and flow stability in the virtual model; The system receives the virtual test results returned by the digital twin module. If the results do not meet the robustness requirements, the system triggers the multi-objective particle swarm optimization algorithm to solve the problem again.

[0037] The parameter optimization module invokes a multi-objective particle swarm optimization algorithm (MPS) with filtration efficiency, energy consumption, and equipment lifespan as optimization objectives. MPS is a swarm intelligence-based optimization method that initializes particle swarm position and velocity parameters, sets the particle swarm size and iteration count, and defines a multi-objective function to transform filtration efficiency, energy consumption, and equipment lifespan into quantifiable optimization objectives. The filtration efficiency objective focuses on improving sludge removal rate, the energy consumption objective aims to reduce the power consumption of the variable frequency drive pump and intelligent electric regulating valve, and the equipment lifespan objective focuses on extending the service life of the multi-layer composite filter and system components. The algorithm evaluates the quality of solutions in the multi-objective search space by calculating the fitness value of each particle and uses Pareto dominance to guide particles towards the non-dominated solution set, gradually approaching the optimal front.

[0038] In the multi-objective particle swarm optimization algorithm, the pressure limit of the filter element and the minimum stable flow rate of the system are loaded as constraints. The pressure limit of the filter element refers to the maximum working pressure that the multi-layer composite filter structure can withstand, preventing excessive pressure from causing filter damage or failure. The minimum stable flow rate of the system is the minimum flow rate threshold required to maintain continuous flow of oil in the gear oil circulation pipeline, preventing system instability or filtration interruption caused by excessively low flow rates. The constraints are integrated into the optimization process in the form of boundary restrictions. For example, constraint processing techniques such as the penalty function method are used to automatically adjust the fitness value when the particle position exceeds the constraint range, ensuring that the algorithm search always takes place within the feasible solution space. The loading of constraints enhances the practicality and safety of the optimization results.

[0039] The parameter optimization module sends the candidate optimization setpoints obtained from the algorithm to the digital twin module. These candidate setpoints include pressure and flow setpoints, transmitted to the digital twin module via communication interfaces such as Ethernet or fieldbus. The parameter optimization module requests the digital twin module to perform virtual tests of pressure surge and flow stability in the virtual model. The pressure surge test simulates sudden increases or decreases in system pressure to evaluate the response characteristics of the candidate setpoints to pressure changes and detects the presence of pressure overshoot or oscillations. The flow stability test applies flow disturbances in the virtual environment and observes the speed and magnitude of flow recovery to stable conditions under the setpoints, verifying the robustness of flow control. The digital twin module uses a hybrid virtual model to simulate the test process in real time, generating quantitative evaluation data.

[0040] The parameter optimization module receives virtual test results returned by the digital twin module. These results include robustness assessment metrics, such as maximum overshoot and settling time in stress shock tests, and variance and recovery time in flow stability tests. The parameter optimization module compares these metrics with preset robustness requirements, such as requiring stress overshoot to not exceed a safety margin and flow volatility to be below an allowable threshold. When the virtual test results do not meet robustness requirements, the parameter optimization module triggers a re-solution using a multi-objective particle swarm optimization algorithm. This re-solution process may adjust algorithm parameters, such as modifying the inertia weights, learning factors, or particle initialization range of the particle swarm optimization algorithm, to expand the search space or focus on potential optimal solution regions. This iterative correction loop continues until the virtual test results fully meet robustness requirements, at which point the optimized setpoints are finally sent to the execution control module.

[0041] A multi-objective particle swarm optimization algorithm generates candidate setpoints, constraints ensure that the setpoints meet physical limitations, virtual testing verifies the performance of the setpoints under dynamic operating conditions, and a feedback mechanism drives iterative optimization of the algorithm. In this way, the parameter optimization module achieves robust optimization of pressure and flow setpoints, providing a reliable parameter basis for the control of the filtration device.

[0042] Specifically, in the filtration device for filtering sludge in gear oil according to the present invention, the learning and adaptation module is further configured as follows: Record the control parameters and corresponding filtration effect evaluation indicators under different sludge characteristics, ambient temperature and load conditions, and build an experience knowledge base; Analyze the characteristics of real-time operational data and output the operational mode recognition results; When the operating mode recognition result indicates a new mode, the meta-learning algorithm is invoked to extract prior knowledge from the experience knowledge base and to fine-tune the control strategy using new operating condition samples. The fine-tuned adaptation rules are sent to the digital twin module to update the parameters of the data-driven model.

[0043] The learning and adaptation module continuously monitors the system's operational status, recording control parameters and corresponding filtration performance evaluation indicators under different sludge characteristics, ambient temperatures, and load conditions to build an experience knowledge base. Sludge characteristics include particle size distribution and concentration parameters; ambient temperature covers the gear oil's operating temperature range; and load conditions involve load variation data from the gear transmission system. Control parameters include pressure setpoints, flow setpoints, and actuator control commands; filtration performance evaluation indicators include filtration efficiency, differential pressure change rate, and sludge removal rate. The experience knowledge base uses a database management system for structured storage, with data records accompanied by timestamps and operating condition tags, forming a historical operational dataset that provides a data foundation for subsequent pattern recognition and strategy optimization.

[0044] The learning and adaptation module analyzes real-time operational data characteristics and outputs operational mode recognition results. The real-time operational data feature extraction process includes calculating statistical features of sensor data such as pressure, flow rate, and temperature, including sliding window mean, variance, and trend slope, and integrating derived parameters such as oil dynamic viscosity. The module applies clustering algorithms such as K-means clustering or classification algorithms such as support vector machines to match the real-time data features with historical patterns in an experience knowledge base, identifying whether the current operational mode belongs to a known mode or a new mode. The operational mode recognition results are output in the form of mode labels, such as normal operating mode, high sludge load mode, or a new, unrecorded operating mode.

[0045] When the operation mode recognition result indicates a new mode, the learning and adaptation module invokes the meta-learning algorithm to extract prior knowledge from the experience knowledge base and fine-tune the control strategy using new operating condition samples. The meta-learning algorithm employs a model-independent meta-learning framework, selecting historical operating condition data with similar characteristics to the new mode from the experience knowledge base as the meta-training set to learn the initialization parameters or basic model of the control strategy. New operating condition samples refer to a small amount of real-time collected operating data. The module uses these samples to quickly fine-tune the control strategy using gradient descent, adjusting the weight parameters in the strategy network to enable the control strategy to adapt to the characteristics of the new operating conditions while retaining historical experience.

[0046] The learning and adaptation module sends the fine-tuned adaptation rules to the digital twin module to update the parameters of the data-driven model. The adaptation rules include the fine-tuned control strategy parameters, operating condition characteristic boundaries, and strategy mapping relationships, transmitted to the digital twin module in data packet form via a communication protocol. Upon receiving the adaptation rules, the digital twin module applies them to update the internal weights of the Long Short-Term Memory network data-driven model, adjusting the model's prediction logic for new operating conditions and enhancing the accuracy of the virtual model's system state estimation. The adaptation rule update cycle is synchronized with the operating mode recognition frequency, achieving dynamic collaboration between the knowledge base and the model.

[0047] An experience knowledge base accumulates historical operational experience, providing a reference benchmark for pattern recognition; operational pattern recognition monitors system state changes in real time and triggers adaptation mechanisms promptly; meta-learning algorithms utilize prior knowledge to quickly adapt to new operating conditions, reducing learning costs; and adaptation rule updates ensure consistency between the virtual model and the actual system. In this way, the learning adaptation module achieves online optimization of control strategies and knowledge transfer, improving the system's adaptability to unknown operating conditions.

[0048] Specifically, in the filtration device for filtering sludge in gear oil according to the present invention, the execution control module is further configured as follows: Receive the final optimized setting value sent by the parameter optimization module; The final optimized setpoint is input into the embedded system dynamic response model to predict the response of the variable frequency drive pump and the intelligent electric regulating valve in the future time domain. Based on the predicted response, the optimal speed curve of the variable frequency drive pump and the optimal opening sequence of the intelligent electric regulating valve are calculated using a rolling optimization method. The control command for the current moment is extracted from the optimal speed curve and the optimal opening sequence and sent to the variable frequency drive pump and the intelligent electric regulating valve.

[0049] The execution control module receives the final optimized setpoints from the parameter optimization module. These final optimized setpoints include pressure and flow rate setpoints, which are transmitted to the execution control module via a communication interface such as Ethernet or fieldbus. The execution control module performs format verification and range checks on the received setpoints to ensure data integrity and provide accurate input for subsequent control command generation.

[0050] The execution control module inputs the final optimized setpoint into the embedded system dynamic response model. This model is constructed based on the physical characteristics of the gear oil circulation pipeline and the multi-layer composite filter, and uses state-space equations to describe the system's dynamic behavior. Taking the pressure and flow setpoints as input, the model predicts the responses of the variable frequency drive pump and the intelligent electric regulating valve in the future time domain using numerical integration methods, including the pump's speed variation trend and the valve's opening dynamic characteristics. The prediction process considers oil viscosity changes and filter clogging factors, outputting predicted system state values ​​for future time steps, providing forward-looking information for rolling optimization.

[0051] Based on the predicted response, the execution control module employs a rolling optimization method to calculate the optimal speed curve of the variable frequency drive pump and the optimal opening sequence of the intelligent electric regulating valve. The rolling optimization method, based on model predictive control principles, solves an optimization problem within a finite time domain in each control cycle, with the objective function being the minimization of system tracking error and control energy, combined with physical constraints of the actuator such as speed limits and opening ranges. The optimization algorithm generates a smooth speed curve and opening sequence, ensuring a stable transition in pressure and flow, and avoiding system oscillations caused by abrupt changes.

[0052] The control command for the current moment is extracted from the optimal speed curve and optimal opening sequence. The execution control module obtains the instantaneous control value from the sequence through an interpolation algorithm, converting it into a frequency signal acceptable to the variable frequency drive pump and an opening command recognizable by the intelligent electric regulating valve. The control command is output via a digital-to-analog converter or pulse width modulation, driving the variable frequency drive pump to adjust the motor speed and the intelligent electric regulating valve to change the throttling area, thereby achieving precise regulation of filtration pressure and flow rate. After the command is sent, the execution control module enters the next control cycle, forming a continuous closed-loop control.

[0053] The final optimized setpoint provides the control objective, the system's dynamic response model predicts behavior, the rolling optimization method generates an optimization sequence, and instruction extraction and execution complete real-time control. In this way, the execution control module transforms the optimized setpoint into executable actions, maintaining the filtering process stably and efficiently.

[0054] Specifically, the filtration device for filtering sludge in gear oil according to the present invention includes a first feedback loop between the data acquisition module and the digital twin module based on the data quality assessment results, comprising: The data acquisition module transmits the multidimensional data stream to the digital twin module; The digital twin module uses the multidimensional data stream to drive the virtual model to run, performs data consistency analysis, and generates data quality assessment results. The digital twin module sends the data quality assessment results to the data acquisition module; The data acquisition module dynamically adjusts the signal sampling frequency of the pressure sensor and flow sensor and the noise covariance parameter of the Kalman filter algorithm based on the received data quality assessment results.

[0055] The data acquisition module transmits the pre-processed multidimensional data stream to the digital twin module via a communication interface such as Ethernet or fieldbus. The multidimensional data stream includes timestamped pressure data, flow data, and oil dynamic viscosity values. The data acquisition module uses a data packetization protocol to ensure data integrity and real-time performance, providing a synchronous input data source for the digital twin module.

[0056] After receiving the multidimensional data stream, the digital twin module performs data consistency analysis during the operation of the virtual model. This analysis compares the residuals between the real-time data stream and the virtual model's predictions, calculating statistical indicators such as mean shift, variance anomaly, and trend deviation coefficient. The analysis process uses a sliding window technique to process time-series data, identifying noise spikes, sensor drift, or communication delays in data acquisition, and generating quantitative data quality assessment results, including data reliability scores and anomaly markers.

[0057] The digital twin module encapsulates the data quality assessment results into feedback data packets and sends them to the data acquisition module via a bidirectional communication link. The feedback transmission employs a priority queue mechanism to ensure the high timeliness of the assessment results. The receiving unit of the data acquisition module parses the data packet content and extracts the data quality score and anomaly type information.

[0058] The data acquisition module dynamically adjusts the signal sampling frequency of the pressure and flow sensors and the noise covariance parameter of the Kalman filter algorithm based on the received data quality assessment results. When the data quality assessment results indicate a high noise level, the data acquisition module increases the signal sampling frequency to capture more details, while simultaneously increasing the noise covariance parameter of the Kalman filter algorithm to enhance the filtering strength. If the assessment results indicate stable data, the sampling frequency is reduced to decrease resource consumption, and the noise covariance parameter is decreased to improve filtering accuracy. The adjustment process achieves adaptive optimization through embedded control logic, forming a real-time interaction between data quality and acquisition strategy.

[0059] Multidimensional data stream delivery provides the foundation for consistency analysis. Data consistency analysis generates quantitative evaluation indicators, and the feedback of evaluation results triggers dynamic adjustments to the acquisition parameters. The adjusted acquisition strategy then affects the quality of subsequent data streams. In this way, the first feedback loop achieves collaborative optimization between data acquisition and model-driven approaches, improving the reliability and accuracy of the system's data source.

[0060] Specifically, in the filtration device for filtering sludge in gear oil described in this invention, the parameter optimization module and the digital twin module form a second feedback closed loop through virtual verification of the optimized set value, including: The parameter optimization module sends the candidate optimization settings to the digital twin module; The digital twin module performs virtual tests on the candidate optimization settings and traffic stability in a virtual model, generating robustness assessment results. The digital twin module returns the robustness assessment results to the parameter optimization module; The parameter optimization module determines whether the requirements are met based on the robustness assessment results. If not, it adjusts the search weights or particle initialization range of the multi-objective particle swarm optimization algorithm and recalculates the candidate optimization settings.

[0061] The parameter optimization module sends candidate optimization settings obtained from the multi-objective particle swarm optimization algorithm to the digital twin module via a communication interface such as Ethernet or fieldbus. These candidate optimization settings include pressure and flow settings. Data transmission uses standard protocol encapsulation to ensure data integrity and real-time performance. The parameter optimization module includes virtual test instructions with the request, specifying the virtual tests to be performed on pressure surges and flow stability.

[0062] After receiving candidate optimization settings, the digital twin module performs a pressure shock virtual test in the virtual model. The pressure shock test simulates sudden increases or decreases in system pressure by applying step or ramp pressure change signals, observing the response characteristics of the filter elements in the virtual model, and recording indicators such as pressure overshoot, settling time, and oscillation frequency. The flow stability test introduces flow disturbances, such as sine waves or random fluctuations, into the virtual environment to evaluate the response speed and fluctuation amplitude of flow recovery to stability. The virtual test utilizes the mechanistic part of the hybrid model to simulate physical dynamics and the data-driven part to predict time-series behavior, generating robustness assessment results including quantitative indicators.

[0063] The digital twin module encapsulates the robustness assessment results into data packets and returns them to the parameter optimization module via a bidirectional communication link. The robustness assessment results include parameters such as maximum overshoot rate and settling time in the stress shock test, and fluctuation variance and recovery time in the flow stability test, and are comprehensively characterized by a score to represent the robustness level of the setpoints. The return process employs a priority transmission mechanism to ensure the timeliness of the feedback information.

[0064] The parameter optimization module analyzes the robustness assessment results and compares each indicator with preset robustness requirements. These requirements include standards such as pressure overshoot not exceeding a safe threshold and flow fluctuation rate being below the allowable range. If the assessment results meet all requirements, the parameter optimization module confirms the candidate optimization settings as the final optimization settings; otherwise, it triggers a re-solution using the multi-objective particle swarm optimization algorithm. The re-solution process adjusts the algorithm parameters, such as modifying the inertia weights of the particle swarm optimization algorithm to change the search inertia, adjusting the learning factor to affect the particle learning speed, or expanding the particle initialization range to explore new solution regions. This iterative correction loop continues until the robustness assessment results fully meet the requirements.

[0065] Sending candidate optimized setpoints initiates a virtual testing process. The virtual test generates quantitative evaluation data, and the evaluation results drive decision-making. If requirements are not met, algorithm parameters are adjusted and recalculated. In this way, the second feedback loop achieves virtual verification and iterative correction of the optimized setpoints, ensuring the feasibility and robustness of the setpoints under dynamic operating conditions and providing reliable input for execution control.

[0066] Specifically, the filtration device for filtering sludge in gear oil according to the present invention, wherein the execution control module and the digital twin module form a third feedback closed loop through the deviation, including: After the execution control module sends the control command, it collects the actual speed of the variable frequency drive pump, the actual opening degree of the intelligent electric regulating valve, and the actual system pressure and flow values ​​measured by the pressure sensor and flow sensor. The execution control module compares the actual system pressure and flow values ​​with the predicted response data made by the system dynamic response model, and calculates the prediction deviation. The execution control module sends the prediction deviation to the digital twin module; The digital twin module uses the prediction bias to correct the parameters describing the dynamic characteristics of the system in the virtual model.

[0067] After sending control commands to the variable frequency drive pump and the intelligent electric regulating valve, the execution control module uses an integrated data acquisition unit to collect in real time the actual speed signal of the variable frequency drive pump, the actual opening signal of the intelligent electric regulating valve, and the actual system pressure and flow values ​​measured by pressure and flow sensors. The acquisition process uses a high-precision analog-to-digital converter to convert the analog signals output by the sensors into digital signals, and applies a time synchronization protocol to add a unified timestamp to each data point, ensuring that the multi-source data are aligned in time sequence, forming a complete actual operating dataset, and providing accurate input for subsequent deviation calculations.

[0068] The execution control module inputs the collected actual system pressure and flow values ​​into the embedded system dynamic response model and compares them with the predicted response data made by the model based on optimized setpoints. The comparison process uses a residual calculation method to calculate the difference sequence between the actual and predicted values, generating a prediction deviation index. The prediction deviation quantifies the degree of deviation between the actual dynamic behavior of the system and the model prediction, reflecting the accuracy level of the virtual model under current operating conditions and providing a quantitative basis for model calibration.

[0069] The execution control module encapsulates the calculated predicted deviation into a standard data packet and sends it to the digital twin module via industrial Ethernet or fieldbus communication protocols. Cyclic redundancy check (CRC) codes are used to ensure data integrity during data transmission, and a priority queue is set up to ensure real-time transmission of deviation information. The predicted deviation data packet includes the deviation value, timestamp, and data quality identifier, providing the digital twin module with structured correction input.

[0070] After receiving prediction bias, the digital twin module uses the bias value to correct the parameters describing the system's dynamic characteristics in the virtual model through a parameter adaptation algorithm. The correction process adjusts physical parameters in the mechanistic model, such as the filter flow resistance coefficient and oil viscosity coefficient, while simultaneously updating the weight coefficients of the long short-term memory network in the data-driven model. The parameter adaptation algorithm, based on the gradient descent principle, optimizes model parameters with the goal of reducing prediction bias, making the virtual model's output closer to the actual behavior of the physical entity and improving the model's prediction accuracy.

[0071] Actual data acquisition provides the real-world state of the system's operation; deviation calculation reveals the differences between the model and the physical entity; deviation transmission enables information exchange between the control and model modules; and model calibration completes parameter optimization and model updates. In this way, the third feedback loop achieves continuous collaboration between the virtual model and the physical entity, ensuring the adaptability and accuracy of the system's control strategy.

[0072] Secondly, please refer to Figure 1 The filter control method for filtering sludge in gear oil provided by the present invention is applied to the filter device for filtering sludge in gear oil, and includes: Step 1: Acquire pressure and flow data during the gear oil filtration process using the pressure sensor and flow sensor, preprocess the pressure and flow data to eliminate measurement noise, and output the preprocessed sensor data. Step 2: Receive the preprocessed sensor data, use the preprocessed sensor data as input to drive the virtual model to run, and output system state estimates and trend prediction data; at the same time, perform data consistency analysis to generate data quality assessment results, and feed back the data quality assessment results to adjust the data acquisition strategy; Step 3: Receive the system state estimate, perform multi-objective optimization calculation on the system state estimate to generate optimized setpoints for pressure and flow; send the optimized setpoints for virtual verification, and iteratively correct the optimized setpoints based on the virtual verification results; Step 4: Receive the system state estimate and trend prediction data, update the control strategy mapping relationship based on historical operating data and current operating data; when a new operating mode is identified, generate adaptation rules and send them to update the virtual model; Step 5: Receive the optimized setpoints determined after iterative correction, generate control commands for the variable frequency drive pump and the intelligent electric regulating valve based on the optimized setpoints, and use the deviation between the actual system response and the predicted response after the command is executed as feedback for model correction; In this process, the adjustment of the data acquisition strategy in step 1, based on the data quality assessment results from step 2, constitutes the first feedback loop; the virtual verification and iterative correction of the optimized setpoints in step 3 constitute the second feedback loop; the model correction in step 5, based on the feedback deviation, constitutes the third feedback loop; and the adaptation rules generated in step 4 are used to update the virtual model, forming the update path.

[0073] The present invention provides a control method for a filtration device used for filtering sludge in gear oil, which achieves precise coordinated control of filtration pressure and flow rate through the coordinated execution of multiple steps. In step 1, pressure and flow sensors collect pressure and flow data in real time during the gear oil filtration process. The raw data is amplified and preliminarily filtered by a signal conditioning circuit, then converted into digital signals by a high-precision analog-to-digital converter, and recursively processed using a Kalman filter algorithm to eliminate measurement noise, outputting preprocessed sensor data. This preprocessing provides high-quality input for subsequent modules and reduces noise interference.

[0074] Step 2 receives the preprocessed sensor data and uses it as input to drive the virtual model. The virtual model employs a hybrid modeling approach, combining a mechanistic model based on fluid mechanics and porous media filtration theory with a data-driven model using long short-term memory networks. The mechanistic model simulates the flow behavior of oil in a multi-layer composite filter and the particulate matter deposition process, while the data-driven model performs time-series predictions of dynamic pressure differential changes. The virtual model outputs system state estimates and trend predictions, and simultaneously performs data consistency analysis, generating data quality assessment results by comparing the residuals between real-time data and model predictions. The data quality assessment results are fed back to the data acquisition strategy, dynamically adjusting the sampling frequency and filtering parameters of the pressure and flow sensors, forming the first feedback loop to improve the reliability and consistency of the data source.

[0075] Step 3 receives the system state estimate and invokes a multi-objective particle swarm optimization algorithm to perform multi-objective optimization calculations. The optimization algorithm targets filtration efficiency, energy consumption, and equipment lifespan, loading the pressure limit of the filter element and the minimum stable flow rate of the system as constraints to generate optimized setpoints for pressure and flow. These optimized setpoints are sent to a virtual model for virtual testing of pressure shocks and flow stability, generating robustness assessment results. If the results do not meet the requirements, the optimization algorithm parameters are adjusted and recalculated, forming an iterative correction loop, constituting a second feedback loop to ensure the feasibility and robustness of the optimized setpoints under dynamic operating conditions.

[0076] Step 4 receives system state estimates and trend prediction data. The learning and adaptation module records historical operating data and builds an experience knowledge base. The module analyzes real-time operating characteristics, identifies operating modes, and when a new mode appears, it calls a meta-learning algorithm to extract prior knowledge from the knowledge base and uses new operating condition samples to fine-tune the control strategy. The generated adaptation rules are sent to the virtual model to update data-driven model parameters, forming an update path and enhancing the system's adaptability to unknown operating conditions.

[0077] Step 5 receives the optimized setpoints determined after iterative correction. The execution control module inputs these setpoints into the system dynamic response model to predict the responses of the variable frequency drive pump and the intelligent electric regulating valve. A rolling optimization method is used to calculate the optimal speed curve and opening sequence, generating control commands to drive the actuators. After command execution, actual speed, opening, and pressure / flow values ​​are collected and compared with the predicted values ​​to calculate the prediction deviation. The prediction deviation is fed back to the virtual model to correct model parameters, forming a third feedback loop to achieve online model calibration and consistency with the physical entity.

[0078] The data acquisition in step 1 provides input for step 2, the model output in step 2 supports the optimization calculation in step 3, the optimized setpoints in step 3 are transformed into control actions in step 5, and the adaptation rules in step 4 optimize the model parameters in step 2. The first feedback loop optimizes the acquisition strategy through data quality assessment, improving the data quality in step 1; the second feedback loop ensures the robustness of the setpoints in step 3 through virtual verification; the third feedback loop corrects the model parameters in step 2 through deviation feedback; and the update path enhances the model's adaptability through the adaptation rules in step 4. This multi-level feedback network enables the system to dynamically maintain precise control of pressure and flow, improving sludge removal efficiency and filtration thoroughness.

[0079] This invention addresses the technical problem of inaccurate pressure and flow control in filtration by constructing an intelligent control device comprised of a data acquisition module, a digital twin module, a parameter optimization module, a learning and adaptation module, and an execution control module. The data acquisition module collects pressure, flow, and temperature data in real time, preprocesses them to generate a high-quality multidimensional data stream, and provides reliable input for subsequent modules. The digital twin module uses this data stream to drive a hybrid virtual model, combining a mechanistic model and a data-driven model to output system state estimates and trend predictions. Simultaneously, it generates data quality assessment results through data consistency analysis and feeds them back to the data acquisition module, forming a first feedback loop and achieving adaptive optimization at the data source.

[0080] The parameter optimization module calculates optimized setpoints for pressure and flow based on system state estimates using a multi-objective particle swarm optimization algorithm, and sends candidate setpoints to the digital twin module for virtual verification. The robustness assessment results generated from the virtual test are returned to the parameter optimization module, driving iterative correction and forming a second feedback loop to ensure the feasibility and robustness of the optimized setpoints in the virtual environment. The learning and adaptation module continuously records historical operating data, building an experience knowledge base. When a new operating mode is identified, it quickly fine-tunes the control strategy using a meta-learning algorithm and sends the adaptation rules to the digital twin module to update model parameters, enhancing the system's adaptability to unknown operating conditions.

[0081] The execution control module translates the final optimized setpoints into control commands, driving the variable frequency drive pump and the intelligent electric regulating valve. Simultaneously, it collects actual response data and compares it with predicted values, generating prediction deviations that are fed back to the digital twin module, forming a third feedback loop for online correction of model parameters. This multi-level feedback network dynamically adjusts control parameters through inter-module collaborative data interaction, ensuring that filtration pressure and flow rate remain within the optimal range, thereby improving the thoroughness of sludge removal and filtration efficiency.

[0082] Embodiment 1 of this invention: In the actual operation of industrial gear transmission systems, multi-layer composite filters in gear oil circulation pipelines face the challenge of sludge accumulation. In this embodiment, the data acquisition module collects pressure and flow data in real time during the filtration process using pressure and flow sensors, while simultaneously integrating a temperature sensor to acquire temperature data. After amplification and preliminary filtering by a signal conditioning circuit, the raw data is converted into digital signals by a high-precision analog-to-digital converter and recursively processed using a Kalman filter algorithm to effectively suppress measurement noise. Based on the temperature data, the data acquisition module queries a preset oil viscosity-temperature characteristic curve to calculate the dynamic viscosity value of the oil. Then, it performs spatiotemporal alignment and fusion of the viscosity value with the pre-processed pressure and flow data to generate a timestamped multi-dimensional data stream. This data stream serves as the input to the digital twin module, driving the virtual model. The digital twin module uses a mechanism model based on fluid mechanics and porous media filtration theory to simulate the oil flow and particulate matter deposition process, and combines a long short-term memory network data-driven model to perform time-series prediction of dynamic pressure changes. The hybrid model dynamically adjusts internal state variables using multidimensional data streams, outputting system state estimates and trend predictions to the parameter optimization and learning / adaptation modules. The parameter optimization module invokes a multi-objective particle swarm optimization algorithm, using filtration efficiency, energy consumption, and equipment lifespan as optimization objectives, and loading the filter element's pressure limit and the system's minimum stable flow rate as constraints to generate candidate optimization setpoints. These setpoints are sent to the digital twin module for virtual testing of pressure shocks and flow stability. If the robustness assessment results do not meet the requirements, the algorithm parameters are adjusted and re-solved, forming an iterative correction loop. The execution control module receives the final optimization setpoints, predicts the responses of the variable frequency drive pump and intelligent electric regulating valve using its internally embedded system dynamic response model, calculates the optimal speed curve and opening sequence using a rolling optimization method, and sends control commands. After command execution, the module collects actual speed, opening, and pressure / flow values, compares them with the predicted values ​​to generate prediction deviations, and feeds them back to the digital twin module to correct model parameters. Through this process, the system maintains a dynamic balance of pressure and flow rate when sludge load changes, preventing filter element clogging or bypass.

[0083] Embodiment 2 of this invention: When the gear transmission system enters a new, unverified operating condition due to fluctuations in ambient temperature or changes in sludge characteristics, the learning and adaptation module activates an adaptive mechanism. The module continuously records control parameters and filtration effect evaluation indicators under different sludge characteristics, ambient temperatures, and load conditions, constructing an experience knowledge base. After real-time data feature analysis identifies a new pattern, it calls a meta-learning algorithm to extract prior knowledge from the knowledge base and fine-tunes the control strategy using a small number of new operating condition samples. The generated adaptation rules are sent to the digital twin module to update the parameters of the data-driven model, enhancing the accuracy of the virtual model's predictions for unknown operating conditions. Simultaneously, the data acquisition module dynamically adjusts the sampling frequency and filtering parameters of the pressure and flow sensors based on the data quality evaluation results fed back by the digital twin module, improving data reliability. The parameter optimization module tests the adaptability of candidate setpoints to new operating conditions in virtual verification, iteratively correcting them through a multi-objective optimization algorithm to ensure the robustness of the optimized setpoints in terms of pressure shocks and flow stability. The execution control module converts the corrected setpoints into control commands, driving the actuators to adjust their operating status. Through a third feedback loop, it feeds back the actual response deviation to the digital twin module, enabling online correction of model parameters. This series of coordinated operations allows the system to quickly converge to the optimal control point during sudden changes in operating conditions, effectively preventing efficiency decline caused by incomplete sludge removal due to insufficient pressure or flow fluctuations, thus extending equipment lifespan.

[0084] This invention relates to a filtration device and control method for filtering sludge in gear oil. Addressing the technical problems of inaccurate filtration pressure and flow control leading to incomplete sludge removal and low filtration efficiency in existing technologies, the invention is implemented as follows: The filtration device includes a physical device and a control device communicatively connected to the physical device. The physical device consists of a gear oil circulation pipeline, a multi-layer composite filter installed in the pipeline, a pressure sensor, a flow sensor, an intelligent electric regulating valve, and a variable frequency drive pump. The control device integrates a data acquisition module, a digital twin module, a parameter optimization module, a learning and adaptation module, and an execution control module. These modules interact through data to form a multi-level feedback network.

[0085] During gear oil filtration, the data acquisition module collects pressure and flow data in real time using pressure and flow sensors, while simultaneously acquiring temperature data measured by a temperature sensor. The module preprocesses the raw data, amplifying and initially filtering it using a signal conditioning circuit, then converting it into a digital signal via a high-precision analog-to-digital converter and applying a Kalman filter algorithm to eliminate measurement noise. The preprocessed sensor data is combined with the temperature data, and a preset oil viscosity-temperature characteristic curve is retrieved to output the oil's dynamic viscosity value. This viscosity value is then spatiotemporally aligned and fused with the pressure and flow data to generate a timestamped multidimensional data stream. This multidimensional data stream serves as the input data for the digital twin module.

[0086] After receiving the multidimensional data stream, the digital twin module drives the virtual model. The virtual model employs a hybrid modeling approach, combining a mechanistic model based on fluid mechanics and porous media filtration theory with a data-driven model using long short-term memory networks. The mechanistic model simulates the flow behavior of oil in a multi-layer composite filter and the particulate matter deposition process, while the data-driven model performs time-series predictions of dynamic pressure differential changes. The digital twin module uses the multidimensional data stream to adjust the internal state variables of the virtual model, outputting system state estimates and trend prediction data to the parameter optimization and learning / adaptation modules. Simultaneously, the digital twin module performs data consistency analysis, generates data quality assessment results, and feeds these results back to the data acquisition module. Based on the data quality assessment results, the data acquisition module dynamically adjusts the signal sampling frequencies of the pressure and flow sensors and the noise covariance parameters of the Kalman filter algorithm, forming a first feedback loop to improve the accuracy and reliability of data acquisition.

[0087] The parameter optimization module receives the system state estimate output by the digital twin module and invokes a multi-objective particle swarm optimization algorithm with filtration efficiency, energy consumption, and equipment lifespan as optimization objectives. The algorithm loads the pressure limit of the filter element and the minimum stable flow rate of the system as constraints to calculate candidate optimized setpoints for pressure and flow. The parameter optimization module sends the candidate optimized setpoints to the digital twin module, requesting virtual testing of pressure shock and flow stability in the virtual model. After executing the virtual test, the digital twin module generates a robustness assessment result and returns it to the parameter optimization module. The parameter optimization module determines whether the candidate optimized setpoints meet the requirements based on the robustness assessment result. If not, it adjusts the search weights or particle initialization range of the multi-objective particle swarm optimization algorithm, recalculates the candidate optimized setpoints, forming a second feedback loop. Through iterative correction, the robustness and feasibility of the optimized setpoints are ensured.

[0088] The learning and adaptation module receives system state estimates and trend prediction data from the digital twin module, records control parameters and corresponding filtration performance evaluation indicators under different sludge characteristics, ambient temperatures, and load conditions, and constructs an experience knowledge base. The learning and adaptation module analyzes real-time operating data characteristics and outputs operating mode recognition results. When a new operating mode is identified, a meta-learning algorithm is invoked to extract prior knowledge from the experience knowledge base, and the control strategy is fine-tuned using new operating condition samples to generate adaptation rules. These adaptation rules are sent to the digital twin module to update the parameters of the data-driven model, forming an update path and enhancing the system's adaptability to unknown operating conditions.

[0089] The execution control module receives the final optimized setpoints determined by the parameter optimization module after iterative correction. These setpoints are then input into the embedded system dynamic response model to predict the responses of the variable frequency drive pump and the intelligent electric regulating valve in the future time domain. A rolling optimization method is used to calculate the optimal speed curve of the variable frequency drive pump and the optimal opening sequence of the intelligent electric regulating valve. The control command for the current moment is extracted from these parameters and sent to the actuator. After the control command is executed, the execution control module collects the actual speed of the variable frequency drive pump, the actual opening of the intelligent electric regulating valve, and the actual system pressure and flow values ​​measured by pressure and flow sensors. The actual response data is compared with the predicted response data from the system dynamic response model to calculate the prediction deviation. This prediction deviation is fed back to the digital twin module to correct the parameters describing the system's dynamic characteristics in the virtual model, forming a third feedback loop. This achieves online correction of model parameters and consistency between the virtual model and the physical entity.

[0090] Through the above implementation methods, the collaborative operation of the data acquisition module, digital twin module, parameter optimization module, learning and adaptation module, and execution control module constructs a multi-level feedback closed loop and update path. The system dynamically maintains precise coordinated control of filtration pressure and flow, effectively improving the thoroughness of sludge removal and filtration efficiency, while enhancing the system's adaptability to changes in operating conditions and extending equipment lifespan.

[0091] Long Short-Term Memory (LSTM) network data-driven models are special recurrent neural networks (RNNs) capable of handling long-term dependencies in time-series data. This model employs three gating mechanisms—input gate, forget gate, and output gate—to selectively memorize or forget information, effectively overcoming the vanishing or exploding gradient problems encountered by existing RNNs during training. In this invention, the model's construction first requires collecting historical operational data, particularly time-varying sequence data of system pressure differentials, as training samples. By cleaning, normalizing, and serializing the samples, a dataset suitable for supervised learning is constructed. Subsequently, the network structure is designed, determining the number of hidden layers and neurons. Backpropagation is used to train the model over time, optimizing network weight parameters so that the model can learn complex patterns of dynamic pressure differential changes from historical data.

[0092] The mechanistic model based on fluid mechanics and porous media filtration theory starts from physical laws and describes the system's intrinsic physicochemical processes through mathematical equations. The model is constructed based on fundamental principles such as the conservation of mass and momentum in fluid mechanics, and Darcy's law in porous media. In this invention, the gear oil circulation pipeline and multi-layer composite filter are physically abstracted and divided into several discrete computational units. For each unit, the Navier-Stokes equations describing oil flow and the convection-diffusion equations describing the transport and deposition of particulate matter in the porous media are established. Then, appropriate boundary and initial conditions are set, such as inlet pressure, outlet flow rate, and the initial porosity of the filter media. Finally, numerical methods, such as the finite volume method or the finite element method, are used to discretize and solve the partial differential equations, thereby constructing a mechanistic model capable of simulating the physical behavior of the system.

[0093] In the model's data processing path, the Long Short-Term Memory (LSTM) network data-driven model receives preprocessed sensor data from the data acquisition module, particularly time-series data of differential pressure. The model uses the differential pressure data at each time step as input, selectively memorizing and updating historical information through an internal gating mechanism, ultimately outputting a predicted value for differential pressure changes over a future period. Essentially, its data processing involves mining temporal patterns in historical data to achieve black-box prediction of system behavior.

[0094] The data processing for the mechanistic model is a white-box simulation process. The model receives a wider range of input data, including real-time collected system pressure, flow rate, oil viscosity, and filter structural parameters. These data are substituted into pre-established governing equations, and the numerical solver performs calculations to simulate the oil velocity distribution, pressure field changes, and particulate matter deposition process within the filter. By solving the physical equations, the model visually reproduces the dynamic behavior within the system.

[0095] The two models work collaboratively in a hybrid manner. The Long Short-Term Memory (LSTM) network data-driven model is responsible for capturing and predicting complex nonlinear dynamics that are difficult to describe with precise physical equations, such as abrupt changes or oscillatory behavior of pressure differentials under certain operating conditions. The mechanistic model provides a solid physical foundation, ensuring that the predictions conform to basic physical laws and simulating physical processes such as the effect of changes in filter porosity on flow resistance. The digital twin module fuses the outputs of both models, for example, by using a Kalman filter algorithm to perform a weighted average or error correction on the simulation results from the mechanistic model and the prediction results from the data-driven model, generating more accurate and reliable system state estimates.

[0096] The model's input data primarily consists of a preprocessed and spatiotemporally aligned multidimensional data stream provided by the data acquisition module. This includes specific data related to pressure, flow rate, temperature-derived dynamic viscosity values ​​of the oil, and structural parameters reflecting the filter's condition. All these data collectively drive the hybrid model's operation.

[0097] In the data processing stage, the hybrid model dynamically adjusts its internal state variables using the input data stream. The mechanistic model updates boundary conditions based on real-time data, resolves the governing equations, and refreshes the estimate of the system's physical state. The data-driven model inputs the latest time-series data into a trained Long Short-Term Memory network for forward propagation computation, obtaining predictions based on data patterns. The outputs of both models are combined in a fusion algorithm to ultimately generate a calibrated system state estimate.

[0098] The model's output data primarily includes system state estimates and trend predictions. System state estimates cover key system parameters at the current moment, such as filter inlet and outlet pressures, real-time flow rate, current filtration efficiency, and sludge deposition. Trend predictions provide forward-looking assessments of system behavior over a future period, such as pressure differential trends, early warnings of filter clogging risks, and system stability evaluations. Each output data point is sent to the parameter optimization module and the learning and adaptation module, respectively, to provide a basis for optimization decisions and knowledge updates. Through this hybrid modeling and collaborative processing approach, the digital twin module achieves accurate perception of the filtration system's state and reliable prediction of future trends. The Kalman filter algorithm is a highly efficient recursive estimation algorithm widely used in signal processing and control systems. It estimates the state of a dynamic system through two steps: prediction and correction, effectively reducing the impact of measurement noise. In the technical solution of this invention, the Kalman filter algorithm is built into the data acquisition module to preprocess the raw data collected by the pressure sensor and flow sensor. The specific data processing route is as follows: the raw sensor signal is first amplified and pre-filtered by the signal conditioning circuit, and then converted into a digital signal by a high-precision analog-to-digital converter. The Kalman filter algorithm then runs recursively on this basis, estimating the current pressure and flow values ​​through the state prediction model, and then correcting them by combining the real-time measurement values. It continuously corrects the state estimation to suppress noise, and finally outputs smooth and reliable preprocessed sensor data, providing high-quality input for the digital twin module.

[0099] Multi-objective particle swarm optimization (PSO) is an optimization method based on swarm intelligence. It simulates the social behavior of flocks of birds or schools of fish, and finds the optimal set of solutions for multiple objective functions, i.e., the Pareto front, by moving particles in the solution space. In this invention, the algorithm is integrated into the parameter optimization module. During construction, filtration efficiency, energy consumption index, and equipment lifespan are used as optimization objectives, while the pressure limit of the filter element and the minimum stable flow rate of the system are loaded as constraints. The data processing route is as follows: the algorithm receives the system state estimate output by the digital twin module as input, initializes the particle swarm position and velocity parameters, iteratively calculates the fitness value of each particle, and updates the particle position using the Pareto dominance relation. Finally, candidate optimization settings for pressure and flow rate are generated. Each setting value is then sent to the digital twin module for virtual verification. If the robustness assessment does not meet the requirements, the search weight or particle initialization range is adjusted and the solution is re-solved, forming an iterative correction loop to ensure the feasibility and robustness of the optimization results.

[0100] Meta-learning algorithms are advanced machine learning paradigms designed to enable models to quickly adapt to new tasks by extracting prior knowledge from multiple related tasks, thus reducing the need for new data. In this invention, meta-learning algorithms are applied to the learning adaptation module. During construction, an experience knowledge base is built based on historical operating data, recording control parameters and filtration performance indicators under different sludge characteristics, ambient temperatures, and load conditions. The data processing route is as follows: the module analyzes the characteristics of operating data in real time, identifies operating modes, and when a new mode is detected, the meta-learning algorithm is invoked to extract prior knowledge from the knowledge base. Using a small number of new operating condition samples, the control strategy parameters are fine-tuned using gradient descent, generating adaptation rules which are then sent to the digital twin module to update the weights of the data-driven model. This enhances the system's adaptability to unknown operating conditions and enables rapid optimization of the control strategy. Through the synergistic application of the above algorithms, this invention achieves seamless integration of data acquisition, parameter optimization, and adaptive learning, improving the accuracy and efficiency of filtering control.

Claims

1. A filtration device for filtering sludge from gear oil, characterized in that, Includes a physical device and a control device communicatively connected to the physical device; The physical device includes a gear oil circulation pipeline, a multi-layer composite filter installed in the pipeline, a pressure sensor, a flow sensor, an intelligent electric regulating valve, and a variable frequency drive pump. The control device includes: The data acquisition module is used to acquire pressure data and flow data during the gear oil filtration process through the pressure sensor and flow sensor, preprocess the pressure data and flow data to eliminate measurement noise, and output the preprocessed sensor data to the digital twin module. The digital twin module is used to receive the preprocessed sensor data, drive the virtual model to run using the preprocessed sensor data as input, and output system state estimates and trend prediction data; the digital twin module also feeds back the data quality assessment results to the data acquisition module; The parameter optimization module is used to receive the system state estimate output by the digital twin module, perform multi-objective optimization calculations on the system state estimate, and generate optimized setpoints for pressure and flow. The parameter optimization module sends the optimized setpoints to the digital twin module for virtual verification, and iteratively corrects the optimized setpoints based on the virtual verification results. The learning and adaptation module is used to receive the system state estimate and trend prediction data output by the digital twin module, update the control strategy mapping relationship based on historical operating data and current operating data, and generate adaptation rules when a new operating mode is identified and send them to the digital twin module. The execution control module is used to receive the optimized setpoint determined by the parameter optimization module after iterative correction, and generate control commands for the variable frequency drive pump and the intelligent electric regulating valve according to the optimized setpoint; the execution control module also feeds back the deviation between the actual system response and the predicted response after the command is executed to the digital twin module; Specifically, the data acquisition module and the digital twin module form a first feedback loop through the data quality assessment results; the parameter optimization module and the digital twin module form a second feedback loop through virtual verification of the optimized set value; the execution control module and the digital twin module form a third feedback loop through the deviation; and the learning adaptation module and the digital twin module form an update path through the adaptation rules. The data acquisition module is also configured to: Acquire temperature data measured by a temperature sensor; Based on the temperature data, a preset oil viscosity-temperature characteristic curve is queried, and the dynamic viscosity value of the oil is output. The dynamic viscosity value of the oil is spatiotemporally aligned and fused with the preprocessed pressure data and flow data to generate a multidimensional data stream with timestamps. The multidimensional data stream is used as the input data for the digital twin module; The digital twin module is also configured to: A mechanistic model based on fluid mechanics and porous media filtration theory is used to simulate the oil flow and particulate matter deposition process; Run a long short-term memory network data-driven model to make time-series predictions of dynamic changes in differential pressure. The mechanism model is mixed with the data-driven model, and the hybrid model is driven by the multidimensional data flow to adjust the internal state variables of the virtual model. The system state estimate and trend prediction data output by the calibrated virtual model are sent to the parameter optimization module and the learning adaptation module. The parameter optimization module is also configured to: Invoke a multi-objective particle swarm optimization algorithm with filtration efficiency, energy consumption index and equipment life as optimization objectives; The pressure limit of the filter element and the minimum stable flow rate of the system are loaded into the multi-objective particle swarm optimization algorithm as constraints. The candidate optimization settings obtained by the algorithm are sent to the digital twin module to request the execution of virtual tests on stress shock and flow stability in the virtual model; The system receives the virtual test results returned by the digital twin module. If the results do not meet the robustness requirements, the system triggers the multi-objective particle swarm optimization algorithm to solve the problem again.

2. The filtration device for filtering sludge in gear oil according to claim 1, characterized in that, The learning adaptation module is also configured as follows: Record the control parameters and corresponding filtration effect evaluation indicators under different sludge characteristics, ambient temperature and load conditions, and build an experience knowledge base; Analyze the characteristics of real-time operational data and output the operational mode recognition results; When the operating mode recognition result indicates a new mode, the meta-learning algorithm is invoked to extract prior knowledge from the experience knowledge base and to fine-tune the control strategy using new operating condition samples. The fine-tuned adaptation rules are sent to the digital twin module to update the parameters of the data-driven model.

3. The filtration device for filtering sludge in gear oil according to claim 2, characterized in that, The execution control module is also configured to: Receive the final optimized setting value sent by the parameter optimization module; The final optimized setpoint is input into the embedded system dynamic response model to predict the response of the variable frequency drive pump and the intelligent electric regulating valve in the future time domain. Based on the predicted response, the optimal speed curve of the variable frequency drive pump and the optimal opening sequence of the intelligent electric regulating valve are calculated using a rolling optimization method. The control command for the current moment is extracted from the optimal speed curve and the optimal opening sequence and sent to the variable frequency drive pump and the intelligent electric regulating valve.

4. The filtration device for filtering sludge in gear oil according to claim 3, characterized in that, The data acquisition module and the digital twin module form a first feedback loop based on the data quality assessment results, including: The data acquisition module transmits the multidimensional data stream to the digital twin module; The digital twin module uses the multidimensional data stream to drive the virtual model to run, performs data consistency analysis, and generates data quality assessment results. The digital twin module sends the data quality assessment results to the data acquisition module; The data acquisition module dynamically adjusts the signal sampling frequency of the pressure sensor and flow sensor and the noise covariance parameter of the Kalman filter algorithm based on the received data quality assessment results.

5. The filtration device for filtering sludge in gear oil according to claim 4, characterized in that, The parameter optimization module and the digital twin module form a second feedback loop through virtual verification of the optimized set value, including: The parameter optimization module sends the candidate optimization settings to the digital twin module; The digital twin module performs virtual tests on the candidate optimization settings and traffic stability in a virtual model, generating robustness assessment results. The digital twin module returns the robustness assessment results to the parameter optimization module; The parameter optimization module determines whether the requirements are met based on the robustness assessment results. If not, it adjusts the search weights or particle initialization range of the multi-objective particle swarm optimization algorithm and recalculates the candidate optimization settings.

6. The filtration device for filtering sludge in gear oil according to claim 5, characterized in that, The execution control module and the digital twin module form a third feedback closed loop through the deviation, including: After the execution control module sends a control command, it collects the actual speed of the variable frequency drive pump, the actual opening degree of the intelligent electric regulating valve, and the actual system pressure and flow values ​​measured by the pressure sensor and flow sensor. The execution control module compares the actual system pressure and flow values ​​with the predicted response data made by the system dynamic response model, and calculates the prediction deviation. The execution control module sends the prediction deviation to the digital twin module; The digital twin module uses the prediction bias to correct the parameters describing the dynamic characteristics of the system in the virtual model.

7. A control method for a filtration device for filtering sludge in gear oil, applicable to the filtration device for filtering sludge in gear oil as described in any one of claims 1 to 6, characterized in that, include: Step 1: Acquire pressure and flow data during the gear oil filtration process using the pressure sensor and flow sensor, preprocess the pressure and flow data to eliminate measurement noise, and output the preprocessed sensor data. Step 2: Receive the preprocessed sensor data, use the preprocessed sensor data as input to drive the virtual model to run, and output system state estimates and trend prediction data; at the same time, perform data consistency analysis to generate data quality assessment results, and feed back the data quality assessment results to adjust the data acquisition strategy; Step 3: Receive the system state estimate, perform multi-objective optimization calculation on the system state estimate to generate optimized setpoints for pressure and flow; send the optimized setpoints for virtual verification, and iteratively correct the optimized setpoints based on the virtual verification results; Step 4: Receive the system state estimate and trend prediction data, update the control strategy mapping relationship based on historical operating data and current operating data; when a new operating mode is identified, generate adaptation rules and send them to update the virtual model; Step 5: Receive the optimized setpoints determined after iterative correction, generate control commands for the variable frequency drive pump and the intelligent electric regulating valve based on the optimized setpoints, and use the deviation between the actual system response and the predicted response after the command is executed as feedback for model correction; In this process, the adjustment of the data acquisition strategy in step 1, based on the data quality assessment results from step 2, constitutes the first feedback loop; the virtual verification and iterative correction of the optimized setpoints in step 3 constitute the second feedback loop; the model correction in step 5, based on the feedback deviation, constitutes the third feedback loop; and the adaptation rules generated in step 4 are used to update the virtual model, forming the update path.

Images