Method for predicting running state of water treatment equipment based on digital twinning

Abstract

The invention discloses a water treatment equipment operation state prediction method based on digital twinning, and the method comprises the following steps: collecting water treatment equipment data, carrying out the preprocessing, and generating an equipment data set; performing feature engineering on the equipment data set to generate a feature data set; based on the feature data set, constructing a digital twin model, and outputting a digital twin equipment operation model; an improved TimeSform model is used for modeling, and an equipment operation state prediction result is output; according to a prediction result, adjusting an equipment operation strategy by using an adaptive fuzzy genetic algorithm, and outputting an optimized equipment operation strategy; the equipment state is monitored in real time, and fault early warning information is sent out; and updating and optimizing the digital twinborn model and the improved TimeSform model. By combining the digital twinning and deep learning technology, the equipment state is accurately predicted, the operation strategy is optimized, and the method is widely applied to intelligent management and maintenance of the water treatment industry.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent manufacturing and water treatment equipment management, and particularly relates to a water treatment equipment operation state prediction method based on digital twinning. BACKGROUND

[0002] With the continuous development of intelligent water treatment systems and industrial equipment automation, real-time monitoring and operation state prediction technology for water treatment equipment has attracted more and more attention. The existing water treatment equipment operation monitoring method mainly relies on manual control and periodic maintenance based on traditional models. However, the existing technology generally has the following problems:

[0003] The existing water treatment equipment monitoring system mainly relies on the operation mode based on traditional models and empirical data, and cannot timely reflect the real-time changes of the equipment state, resulting in difficulty in predicting and accurately predicting the equipment fault diagnosis and maintenance. The traditional water treatment equipment fault warning system usually relies on single sensor data or rule-based prediction methods. These methods generally have the following defects when facing complex water treatment processes and dynamic environments: on the one hand, a single sensor cannot comprehensively capture all the operation data of the equipment, resulting in the omission of key fault signals; on the other hand, the rule-based method cannot accurately predict the dynamic state according to the historical data and real-time sensor data of the equipment, resulting in the inability to timely capture the abnormal fluctuations of the equipment operation state.

[0004] In the existing equipment state prediction method, most of them use traditional physical models or machine learning-based data analysis methods, but these methods generally have the following problems in the processing of large-scale equipment data and model updating: first, the traditional physical model lacks flexibility when facing complex operation states, and cannot effectively handle the real-time changes of the equipment state; second, although the machine learning model can provide certain prediction ability, it has the problems of long training time and high demand for computing resources when processing large-scale data, which is difficult to meet the real-time requirements of industrial equipment. In addition, the existing fault warning mechanism cannot effectively combine the real-time operation data and historical equipment operation data of the equipment, resulting in insufficient accuracy of the equipment fault warning and lack of adaptive adjustment ability of the equipment optimization strategy.

[0005] Therefore, how to provide a water treatment equipment operation state prediction method based on digital twinning is a problem that those skilled in the art need to solve. SUMMARY

[0006] One purpose of the present application is to propose a water treatment equipment operation state prediction method based on digital twinning. The present application fully combines digital twinning technology, deep learning and adaptive fuzzy genetic algorithm, and describes in detail how to predict the state by collecting equipment data in real time, constructing a digital twinning model, applying an improved TimeSformer model, and automatically adjusting the equipment operation strategy through an optimization algorithm. This method has the advantages of accurately predicting the equipment operation state, real-time monitoring of equipment failure, optimizing equipment operation efficiency and reducing energy consumption.

[0007] The water treatment equipment operation state prediction method based on digital twinning according to the embodiment of the present application comprises the following steps: Collecting flow data, pressure data, temperature data, chemical reaction data, sensor state data and equipment operation history data of the water treatment equipment, and preprocessing to generate an equipment data set; Performing feature selection, time series decomposition, LSTM feature selection, PCA dimension reduction and feature fusion on the equipment data set to generate a feature data set; Based on the feature data set, a digital twinning model is constructed to output a digital twinning equipment operation model; Based on the digital twinning equipment operation model and the feature data set, an improved TimeSformer model is used for modeling to output an equipment operation state prediction result; According to the equipment operation state prediction result, an adaptive fuzzy genetic algorithm is used to automatically adjust the operation strategy of the equipment, and an optimized equipment operation strategy is output; During equipment operation, the equipment state is monitored in real time through the equipment operation state prediction result, and a fault warning information is issued; Through real-time operation data and equipment operation state prediction results, the digital twinning model and the improved TimeSformer model are updated and optimized.

[0008] Optionally, the flow data, pressure data, temperature data, chemical reaction data, sensor state data and equipment operation history data of the water treatment equipment are collected and preprocessed to generate an equipment data set, which specifically comprises: The flow data includes inlet flow data and outlet flow data, the inlet flow data is the amount of water flowing into the water treatment equipment per unit time, and the outlet flow data is the amount of water flowing out of the water treatment equipment per unit time; The pressure data includes inlet pipe pressure data and outlet pipe pressure data, the inlet pipe pressure data is the pressure of water flowing into the water treatment equipment per unit time, and the outlet pipe pressure data is the pressure of water flowing out of the water treatment equipment per unit time; The temperature data includes water inlet temperature data and water outlet temperature data, the water inlet temperature data is the temperature of water flowing into the water treatment equipment per unit time, and the water outlet temperature data is the temperature of water flowing out of the water treatment equipment per unit time; The chemical reaction data includes pH value data, dissolved oxygen data, ammonia nitrogen concentration data and oxidation-reduction potential data; The sensor state data includes sensor working state data, sensor signal strength data and sensor fault state data; The equipment operation history data includes equipment start-stop time data, equipment operation mode data, equipment fault history data and equipment maintenance record data; The collected data is cleaned, repeated data is removed, and abnormal data is processed, the Z-score method is used to detect abnormal values, interpolation method is used to fill in missing data, and the values of different data sets are scaled to a unified range to generate an equipment data set.

[0009] Optionally, the equipment data set is subjected to feature selection, time series decomposition, LSTM feature selection, PCA dimension reduction and feature fusion to generate a feature data set, specifically including: The equipment data set is subjected to feature selection, the information gain method is used to calculate the information gain between each feature and the equipment operation state, the features with information gain values greater than a preset threshold are selected, and a feature selection data set is generated; The equipment data set is subjected to time series decomposition, the seasonal trend decomposition method is used to decompose the time series part of the equipment data set into long-term trend component, seasonal component and residual component, the long-term trend component representing the long-term change pattern of the equipment is extracted, the seasonal component representing the periodic fluctuation of the equipment is extracted, and the residual component representing the short-term instability of the equipment is extracted, each component is screened according to the variance explained rate, and the components with variance explained rates greater than a set threshold are retained to generate a time series decomposition data set; The equipment data set is subjected to LSTM feature selection, an LSTM model is used to model the equipment data set, the most important time-dependent features are automatically learned, the features with gradient importance greater than a set threshold are selected through gradient boosting, the selection is specifically performed by the weight normalization method, and the features with normalized weights greater than a set threshold are retained; The equipment data set is subjected to PCA dimension reduction, the PCA method is used to reduce the dimension of the features of the equipment data set, the principal components with cumulative variance contribution rates greater than a preset threshold are retained, and redundant features are removed to generate a PCA dimension reduction data set; The feature selection data set, the time series decomposition data set, the LSTM feature data set and the PCA dimension reduction data set are fused by using a weighted average method or a feature connection method to generate a feature data set.

[0010] Optionally, the feature data set is input into the digital twin model, and the digital twin model is used for dynamic simulation of the running state of the equipment and optimization prediction. Optionally, the feature data set is input into the digital twin model, and the digital twin model is used for dynamic simulation of the running state of the equipment and optimization prediction. The digital twin model is constructed by combining physical modeling and data-driven modeling, specifically including: The physical modeling uses quantum mechanics principles and uses a quantum computing model to model the behavior of microscopic particles in the equipment. Through the superposition state and entanglement characteristics of qubits, the dynamic behavior of molecules in the water treatment process is simulated, providing accurate microscopic simulation of the running state of the equipment. The quantum computing model uses quantum Monte Carlo simulation for calculation. The reaction kinetics model is based on a chemical reaction kinetics model and simulates the reaction process inside the water treatment equipment. The Arrhenius equation is used to describe the relationship between chemical reaction rate and temperature. The running state of the water treatment equipment is data-driven modeled using a graph neural network. The complex relationships between different components of the equipment are extracted through a graph convolution network, and the structure information of the graph is used to optimize the running state prediction of the water treatment equipment. The input graph data is the node features of the equipment, and feature propagation and aggregation are performed through the graph convolution network layer to output the state information of each component. The feature data set is combined with the physical model for simulation and prediction to obtain the digital twin equipment running model. Real-time sensor data is input into the digital twin equipment running model and dynamically adjusted. The running state of the equipment is real-time feedback and updated. By comparing the current state of the equipment with the prediction result of the digital twin equipment running model, the key parameters in the digital twin equipment running model are adjusted. The digital twin equipment running model includes the current running state of the equipment, future state prediction, equipment fault diagnosis, running efficiency, and load prediction.

[0011] Optionally, the improved TimeSformer model specifically includes: The digital twin equipment running model is applied to the improved TimeSformer model, and a physical constraint convolution layer is used for feature extraction. Based on the physical model of the equipment, convolution operation is performed on the time series data of the equipment to extract local features in the time series data. The design of the convolution kernel is based on the physical characteristics in the equipment running process, and the generated physical constraint convolution output data set is generated. The NSGA-II algorithm is used to optimize the convolution kernel in the physical constraint convolution output data set, the population is initialized, a plurality of convolution kernel configurations are randomly generated, the fitness evaluation method is used, the prediction accuracy and the calculation complexity of each convolution kernel configuration are evaluated to calculate the fitness, for the individuals with high fitness, the next round of crossover and mutation operations are preferentially selected, the individuals with better fitness are selected through non-dominated sorting, and the diversity of the population is maintained through congestion degree comparison; The space-time separation self-attention mechanism is set, the correlation between each time step is calculated through the time self-attention mechanism, the long-term trend in the device running process is identified, the mutual relationship between the components of the device is modeled through the space self-attention mechanism, the space self-attention mechanism captures the complex interaction relationship between the components of the device through the graph convolution network, and the corresponding state features of each component are generated; The topological data analysis is applied to the feature data set obtained through the space-time separation self-attention mechanism, the topological structure of the device time series data is analyzed using the persistent homology method, the key topological features in the data are identified, according to the analysis result, the most representative features for predicting the device running state are extracted, the redundant data is removed, and the optimized feature data set is generated; The time features and the space features in the optimized feature data set are weighted and synthesized to obtain comprehensive device running state features, the device running state is predicted through the prediction layer, and the device running state prediction result is output, including the load, the efficiency, the fault warning and the future running trend of the device.

[0012] Optionally, according to the device running state prediction result, an adaptive fuzzy genetic algorithm is used to automatically adjust the running strategy of the device, and an optimized device running strategy is output, specifically including: The device running state prediction result is input into the adaptive fuzzy genetic algorithm model; The population in the fuzzy genetic algorithm is initialized, a plurality of device running strategy configurations are randomly generated based on the current running state and the historical data of the device, and the effectiveness of each strategy is evaluated through the fitness function; The fuzzy rules based on the device running state are used to evaluate the running strategy of the device, the fuzzy rule base includes a plurality of control rules, the fuzzy inference mechanism is used for calculation, and the fuzzy fitness value of each strategy is generated; Based on the fuzzy fitness value, the initial population is optimized through the crossover, mutation and selection operations in the genetic algorithm, the fitness of each individual is calculated through the fitness evaluation function, and the individuals with fitness greater than a preset threshold are selected for the next generation of crossover and mutation operations; In the iteration process, the control parameters are dynamically adjusted using an adaptive adjustment mechanism; after multiple generations of iteration optimization, an optimized device operation strategy is output, which can automatically adjust the operation parameters of the device based on the real-time operation state and future prediction results of the device; According to the optimized device operation strategy, the water treatment device is adjusted in real time, and the operation parameters of the device are dynamically adjusted through real-time sensor data and device state prediction results.

[0013] Optionally, during the operation of the device, the device state is monitored in real time through the device state prediction results, and fault warning information is issued, specifically including: By receiving the sensor data and operation state data of the device in real time, the actual operation state of the device is monitored, the device state is compared with the prediction results output by the digital twin device operation model, the state deviation is identified, and when the parameters exceed the preset threshold, a fault warning is triggered; According to the operation history data and real-time sensor data of the device, the fault warning rules and thresholds are set, the operation of the device is monitored according to the fault mode, when the device state is abnormal and the deviation exceeds the threshold, the fault warning information is generated, and the alarm system is triggered; The fault warning information includes fault type, affected component and potential fault cause.

[0014] Optionally, the digital twin model and the improved TimeSformer model are updated and optimized through real-time operation data and device operation state prediction results, specifically including: The real-time operation data, including the flow, pressure, temperature, chemical reaction data and device operation state of the device, are collected in real time through sensors and device control systems, and are combined with the device operation state prediction results as input data, and are transmitted to the digital twin model and the improved TimeSformer model; Based on the real-time operation data, the key parameters in the digital twin model are dynamically adjusted; Through the device operation state prediction results, the improved TimeSformer model is incrementally learned, and through the model updating mechanism, the new operation data and prediction results are combined to further optimize the convolution kernel and self-attention mechanism in the TimeSformer model; The optimized digital twin model and improved TimeSformer model are used for continuous iteration training, and the time series processing ability and feature extraction ability in the structure of the digital twin model and the improved TimeSformer model are adjusted.

[0015] The beneficial effects of the present application are: The application is based on a device operation state prediction method based on digital twinning and an improved TimeSformer model. By combining physical modeling and data-driven modeling, the application effectively solves the problems of insufficient accuracy of device state monitoring and fault prediction, lack of adaptability of device operation strategy, and other issues in the prior art. First, the application uses a digital twinning model to perform dynamic simulation and optimized prediction of the device, which can reflect the running state of the device in real time, breaking through the limitations of traditional physical models that cannot cope with complex operating conditions. Second, through multi-level feature extraction and fusion, the application can accurately extract key features of the device state and use an improved TimeSformer model for efficient modeling to capture the temporal and spatial dependencies in the device operation data, thereby providing accurate device operation state prediction. Compared with the prior art, the improved model can better cope with real-time fluctuations during device operation, enhancing the sensitivity and accuracy of fault warning. Finally, the adaptive fuzzy genetic algorithm automatically adjusts the device operation strategy to ensure optimized control of the device during operation, further improving the efficiency and fault prevention capability of the device. In summary, the application can provide more accurate and real-time device state prediction and fault warning, improve the intelligent management level of the device, effectively reduce the probability of device failure, and improve the efficiency and reliability of the water treatment device. BRIEF DESCRIPTION OF DRAWINGS

[0016] The accompanying drawings are included to provide a further understanding of the application, and constitute a part of the specification, illustrate embodiments of the application, and are used to explain the application, and do not constitute a limitation on the application. In the drawings:

[0017] Fig. 1 Flowchart of the water treatment device operation state prediction method based on digital twinning proposed by the application; Fig. 2 Schematic diagram of the water treatment device operation state prediction method based on digital twinning proposed by the application; Fig. 3 Framework diagram of the improved TimeSformer model in the water treatment device operation state prediction method based on digital twinning proposed by the application. DETAILED DESCRIPTION

[0018] The application will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams that only illustrate the basic structure of the application in a schematic manner, and therefore only show the components related to the application.

[0019] Reference Figs. 1-3 The water treatment device operation state prediction method based on digital twinning includes the following steps: The flow data, pressure data, temperature data, chemical reaction data, sensor state data, and equipment operation history data of the water treatment equipment are collected and preprocessed to generate an equipment data set; The equipment data set is subjected to feature selection, time series decomposition, LSTM feature selection, PCA dimension reduction, and feature fusion to generate a feature data set; Based on the feature data set, a digital twin model is constructed, and a digital twin equipment operation model is output; Based on the digital twin equipment operation model and the feature data set, an improved TimeSformer model is used for modeling, and an equipment operation state prediction result is output; According to the equipment operation state prediction result, an adaptive fuzzy genetic algorithm is used to automatically adjust the operation strategy of the equipment, and an optimized equipment operation strategy is output; During the operation of the equipment, the equipment state is monitored in real time through the equipment operation state prediction result, and a fault warning information is issued; Through real-time operation data and equipment operation state prediction results, the digital twin model and the improved TimeSformer model are updated and optimized.

[0020] In the embodiment, the collection of the flow data, pressure data, temperature data, chemical reaction data, sensor state data, and equipment operation history data of the water treatment equipment, and the preprocessing to generate the equipment data set specifically include: The flow data includes inlet flow data and outlet flow data, the inlet flow data is the amount of water flowing into the water treatment equipment per unit time, and the outlet flow data is the amount of water discharged from the water treatment equipment per unit time; The pressure data includes inlet pipe pressure data and outlet pipe pressure data, the inlet pipe pressure data is the pressure of the water flowing into the water treatment equipment per unit time, and the outlet pipe pressure data is the pressure of the water flowing out of the water treatment equipment per unit time; The temperature data includes inlet temperature data and outlet temperature data, the inlet temperature data is the temperature of the water flowing into the water treatment equipment per unit time, and the outlet temperature data is the temperature of the water flowing out of the water treatment equipment per unit time; The chemical reaction data includes pH value data, dissolved oxygen data, ammonia nitrogen concentration data, and oxidation-reduction potential data; The sensor state data includes sensor working state data, sensor signal strength data, and sensor fault state data; The equipment operation history data includes equipment start-stop time data, equipment operation mode data, equipment fault history data, and equipment maintenance record data; The collected data is cleaned, repeated data is removed, and abnormal data is processed, the Z-score method is used to detect abnormal values, the interpolation method is used to fill in missing data, the values of different data sets are scaled to a unified range, and a device data set is generated.

[0021] In this embodiment, the feature selection, time series decomposition, LSTM feature selection, PCA dimension reduction, and feature fusion of the device data set are performed to generate a feature data set, specifically including: Feature selection is performed on the device data set, information gain method is used, information gain between each feature and device running state is calculated, features with information gain value greater than a preset threshold are selected, and a feature selection data set is generated; The device data set is decomposed into a time series, and the seasonal trend decomposition method is used to decompose the time series part of the device data set into long-term trend component, seasonal component and residual component, extract the long-term trend component representing the long-term change pattern of the device, extract the seasonal component representing the periodic fluctuation of the device, extract the residual component representing the short-term instability of the device, each component is screened according to the variance explained rate, and the components with variance explained rate greater than the set threshold are retained, and a time series decomposition data set is generated; The LSTM feature selection is performed on the device data set, the LSTM model is used to model the device data set, the most important time-dependent features are automatically learned, the gradient importance greater than the set threshold is selected through gradient boosting, the features are retained through the weight normalization method, and the features with normalized weight greater than the set threshold are retained; The PCA dimension reduction is performed on the device data set, the PCA method is used to reduce the dimension of the features of the device data set, the principal components with cumulative variance contribution rate greater than the preset threshold are retained, the redundant features are removed, and a PCA dimension reduction data set is generated; The feature selection data set, the time series decomposition data set, the LSTM feature data set and the PCA dimension reduction data set are fused using the weighted average method or the feature connection method to generate a feature data set.

[0022] In this embodiment, the feature data set is used to construct a digital twin model, and a digital twin device running model is output, specifically including: The feature data set is input into the digital twin model, and the digital twin model is used to dynamically simulate the running state of the device and perform optimization prediction; The digital twin model is constructed by combining physical modeling and data-driven modeling, specifically including: Physical modeling utilizes quantum mechanics principles to model the behavior of microscopic particles in the device using a quantum computing model, particularly the interaction of particles in fluid and chemical reaction processes, by utilizing the superposition and entanglement properties of qubits to simulate the molecular-level dynamics in water treatment processes, providing accurate microscopic simulations of the device's operating state. The quantum computing model uses quantum Monte Carlo simulation for calculations, as follows: ; where, is the wave function of the system, describing the state of particles in spatial position and time , is the quantum state coefficient, is the energy eigenvalue, is the eigenstate of the wave function, is the Planck constant; Reaction kinetics model simulates the reaction process inside the water treatment device based on chemical reaction kinetics model, using Arrhenius equation to describe the relationship between chemical reaction rate and temperature: ; where, is the reaction rate constant, is the pre-exponential factor, is the activation energy, is the gas constant, is the temperature, the reaction kinetics model is used to simulate the change of temperature and reaction rate in the reaction process, optimize the energy efficiency and reaction process of the water treatment device; Data-driven modeling of the water treatment device's operating state using graph neural networks, extracting complex relationships between different components of the device through graph convolution networks, and utilizing graph structure information to optimize the prediction of the water treatment device's operating state; The input graph data is the node features of the device, which are passed through the graph convolution network layer for feature transmission and aggregation, outputting the state information of each component. This method can effectively capture the complex relationships between the components of the water treatment device, improving the prediction accuracy; Simulate and predict by combining feature data sets with physical models to obtain a digital twin device operation model. Input real-time sensor data into the digital twin device operation model and make dynamic adjustments. Real-time feedback and update the device operating state. Compare the current state of the device with the prediction results of the digital twin device operation model to adjust the key parameters in the digital twin device operation model, so that the digital twin device operation model can always accurately reflect the actual operating state of the device; Output the digital twin device operation model, which includes the current operating state of the device, future state prediction, device fault diagnosis, operating efficiency, and load prediction.

[0023] In this embodiment, the improved TimeSformer model specifically includes: The digital twin device operation model is applied to the improved TimeSformer model, feature extraction is performed using a physical constraint convolution layer, convolution operation is performed on the device time series data based on the physical model of the device, local features in the time series data are extracted, and the design of the convolution kernel is based on the physical characteristics in the device operation process to ensure that the convolution operation can extract features that conform to the actual operation rules of the device, and a physical constraint convolution output data set is generated; The convolution kernel in the physical constraint convolution output data set is optimized by the NSGA-II algorithm, the population is initialized, a plurality of convolution kernel configurations are randomly generated, the fitness evaluation method is used, the prediction accuracy and computational complexity of each convolution kernel configuration are evaluated to calculate the fitness, for individuals with high fitness, the next round of crossover and mutation operations are preferentially selected, individuals with better fitness are selected by non-dominated sorting, and the diversity of the population is maintained by comparing the crowding degree, so that the optimization process of the convolution kernel can find the optimal solution, and the convolution kernel optimized by the NSGA-II algorithm improves the performance and adaptability of the convolution operation; The spatio-temporal separation self-attention mechanism is set, the correlation between each time step is calculated by the time self-attention mechanism, the long-term trend in the device operation process is identified, and the mutual relationship between the components of the device is modeled by the space self-attention mechanism. The space self-attention mechanism captures the complex interaction between the components of the device through the graph convolution network and generates corresponding state features for each component. The design of spatio-temporal separation can better handle the temporal dependence and spatial dependence in the time series data, and improve the prediction accuracy; Topological data analysis is applied to the feature data set obtained by the spatio-temporal separation self-attention mechanism, the topological structure of the device time series data is analyzed using the persistent homology method, the key topological features in the data are identified, and according to the analysis results, the most representative features for predicting the device operation state are extracted, redundant data is removed, and an optimized feature data set is generated. This step effectively reduces the dimension of the feature space while maintaining the key features for predicting the device operation state; The time features and space features in the optimized feature data set are weighted and synthesized to obtain comprehensive device operation state features, the device operation state is predicted through the prediction layer, and the device operation state prediction result is output, including the load, efficiency, fault warning and future operation trend of the device.

[0024] In this embodiment, the device operation strategy is automatically adjusted using an adaptive fuzzy genetic algorithm based on the device operation state prediction result, and the optimized device operation strategy is output, specifically including: The device running state prediction result is passed as an input parameter to the adaptive fuzzy genetic algorithm model; By initializing the population in the fuzzy genetic algorithm, based on the current running state and historical data of the device, a plurality of device running strategy configurations are randomly generated, and the effectiveness of each strategy is evaluated by the fitness function; The running strategy of the device is evaluated using fuzzy rules based on the device running state, and the fuzzy rule base includes a plurality of control rules, which are calculated by a fuzzy inference mechanism to generate a fuzzy fitness value for each strategy; Based on the fuzzy fitness value, the initial population is optimized using crossover, mutation and selection operations in the genetic algorithm, the fitness of each individual is calculated by the fitness evaluation function, and the individuals with fitness greater than a preset threshold are selected for the next generation of crossover and mutation operations, to ensure that the optimal strategy configuration that meets the actual running requirements of the device can be found; In the iteration process, the control parameters are dynamically adjusted using an adaptive adjustment mechanism to improve the optimization effect of the fuzzy genetic algorithm under different running states, so that the fuzzy genetic algorithm can adapt to the running characteristics and environmental changes of different devices; After multiple generations of iterative optimization, the optimized device running strategy is output, which can automatically adjust the operating parameters of the device based on the real-time running state and future prediction results of the device, to ensure that the device maintains the best running efficiency under different running states, reduces energy consumption and prolongs the service life of the device; According to the optimized device running strategy, the water treatment device is adjusted in real time, and the operating parameters of the device are dynamically adjusted through real-time sensor data and device state prediction results, so as to realize the optimized operation and fault warning of the device.

[0025] In this embodiment, the device running state prediction result is used to monitor the device state in real time during the device running process, and fault warning information is issued, which specifically includes: By receiving the sensor data and running state data of the device in real time, the actual running state of the device is monitored, the device state is compared with the prediction result output by the digital twin device running model, and the state deviation is identified, and when the parameter exceeds the preset threshold, the fault warning is triggered; According to the running historical data and real-time sensor data of the device, the fault warning rules and thresholds are set, and the device running is monitored according to the fault mode, when the device state is abnormal and the deviation exceeds the threshold, the fault warning information is generated, and the alarm system is triggered; The fault warning information includes fault type, affected component and potential fault cause, which can timely notify the device maintenance personnel or automatically dispatch the system to respond.

[0026] In this embodiment, the digital twin model and the improved TimeSformer model are updated and optimized by real-time operation data and equipment operation state prediction results, specifically including: Real-time operation data, including the flow, pressure, temperature, chemical reaction data of the equipment and the equipment operation state, are collected in real time through sensors and equipment control systems, and combined with the equipment operation state prediction results as input data, which are transmitted to the digital twin model and the improved TimeSformer model; Based on the real-time operation data, the key parameters in the digital twin model are dynamically adjusted to ensure that the digital twin model can always accurately simulate the current operation state of the equipment; Through the equipment operation state prediction results, incremental learning is performed on the improved TimeSformer model, and through the model updating mechanism, the new operation data and prediction results are combined to further optimize the convolution kernel and self-attention mechanism in the TimeSformer model, and improve the prediction accuracy of the model on the equipment operation state; Using the optimized digital twin model and improved TimeSformer model, continuous iterative training is performed to adjust the time series processing ability and feature extraction ability in the structure of the digital twin model and the improved TimeSformer model, so that the digital twin model and the improved TimeSformer model can more accurately predict the operation state, failure trend and equipment efficiency of the equipment according to the real-time operation data; Based on the continuous update of real-time operation data and equipment operation state prediction results, the optimized digital twin model and improved TimeSformer model can better adapt to the changes of the equipment under different working conditions, improve the stability and energy efficiency of the equipment operation, prolong the service life of the equipment, and reduce the occurrence of potential failures.

[0027] Example 1 In order to verify the feasibility of the application in implementation, the application is applied to the equipment operation state prediction and optimization management of a large water treatment plant in a certain city. The water treatment plant is mainly responsible for the supply of tap water in the city, and the equipment includes multiple water pumps, filters, pipelines, chemical reaction tanks and other equipment, with a processing capacity of 500,000 cubic meters of water per day. Due to the complexity of equipment operation and unpredictable changes in water quality, how to monitor the equipment state in real time and optimize the equipment operation efficiency is the main challenge faced by the water treatment plant.

[0028] In the practical application of this water treatment plant, first of all, through various sensors, real-time collection of equipment flow, pressure, temperature, chemical reaction data, equipment working state and operation history data. These data after pretreatment, constitute equipment data set, and input to feature selection and dimension reduction module. After feature selection, time series decomposition, LSTM feature selection and PCA dimension reduction, etc. Multi-step processing, finally a set of high-quality feature data set is obtained. These data sets are used to build digital twin model and improved TimeSformer model, and then carry out equipment running state prediction.

[0029] In specific applications, we combined the actual physical model of the equipment with the data-driven model for simulation, and dynamically updated the digital twin model and TimeSformer model through real-time monitoring data and prediction results. By combining real-time equipment data with prediction results, we can timely adjust the operating parameters of the equipment, such as flow, pressure and temperature, to cope with unexpected situations. Especially when the equipment load is too high or a fault warning occurs, the system will automatically generate a fault warning and adjust the equipment operation strategy through optimization algorithms to reduce the occurrence of faults and improve the efficiency of equipment operation.

[0030] In the implementation process, after several rounds of optimization, we have achieved significant results in equipment running state prediction and energy efficiency optimization. Taking the pressure data of a certain equipment as an example, when the system early detects that the pressure fluctuation is large, it timely adjusts the equipment operation strategy, reduces the equipment load, reduces the energy consumption, and avoids a potential equipment failure. Specifically, in the early stage of prediction, the efficiency of the equipment is improved by 15%, and the energy consumption is reduced by 10%.

[0031] In order to further verify the effectiveness of the invention, a month of equipment operation data is collected, and the relevant equipment operation data comparison is shown in Table 1: Table 1 Comparison of equipment running state and prediction results

[0032] As can be seen from Table 1, the device load is negatively correlated with the operation efficiency. For example, on September 1 and September 2, 2025, the device load is 80% and 85% respectively, but the corresponding operation efficiency is 92% and 90% respectively. On September 4, the device load is reduced to 70%, and the operation efficiency is increased to 96%. This phenomenon shows that when the device load is low, the operation efficiency can be significantly improved, otherwise, the efficiency will decrease. The energy consumption is positively correlated with the device load and efficiency. On September 5, the device load reaches 90%, but the operation efficiency decreases to 85%, and the energy consumption also increases to 5500 kWh. The number of device failures also increases significantly, reaching 4 times, while in the case of low load (such as September 4), the energy consumption of the device is only 4700 kWh, and the number of failures is 0, which shows that the device can effectively reduce energy consumption and reduce failures when running optimally. The increase in the number of fault warnings usually means that the device is running abnormally, and the prediction accuracy is directly related to the stability of the device running state. In the case of high device load (such as September 2 and September 5), the number of fault warnings is more, and the prediction accuracy decreases (96%-97%), however, in the case of low device load and stable running (such as September 3 and September 4), the prediction accuracy is close to 100%, and the number of fault warnings is significantly reduced, which shows that the present application can accurately predict the device state and effectively reduce the occurrence of failures.

[0033] From the examples, it can be concluded that the digital twin model and the improved TimeSformer model of the present application can effectively optimize the operation efficiency of the water treatment device, reduce the energy consumption and failure problems caused by high device load. The fault warning mechanism of the device can accurately predict potential failures and timely adjust the device operation strategy, effectively avoiding device failures. By updating and optimizing the digital twin model and the TimeSformer model in real time, the energy efficiency of the device is effectively improved, and the stability and reliability of the device are significantly improved. The method of the present application can greatly improve the operation efficiency of the water treatment device, reduce the failure rate, and reduce the energy consumption while ensuring stable operation of the device, which has strong practical application value.

[0034] The above describes only the preferred embodiments of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can make equivalent replacements or changes to the technical solutions and inventive concepts of the present application within the technical scope disclosed by the present application, which should be covered within the protection scope of the present application.

Claims

1. A method for predicting the operating status of water treatment equipment based on digital twins, characterized in that, Includes the following steps: Collect flow rate data, pressure data, temperature data, chemical reaction data, sensor status data, and historical equipment operation data from water treatment equipment, and preprocess them to generate equipment datasets; The device dataset is subjected to feature selection, time series decomposition, LSTM feature selection, PCA dimensionality reduction, and feature fusion to generate a feature dataset. Based on the feature dataset, a digital twin model is constructed, and a digital twin device operation model is output. Based on the digital twin device operation model and feature dataset, an improved TimeSformer model is used for modeling, and the device operation status prediction results are output. Based on the equipment operating status prediction results, an adaptive fuzzy genetic algorithm is used to automatically adjust the equipment operating strategy and output the optimized equipment operating strategy. During equipment operation, the equipment status is monitored in real time based on the equipment operation status prediction results, and fault warning information is issued. By using real-time operational data and equipment operating status prediction results, the digital twin model and the improved TimeSformer model are updated and optimized.

2. The method for predicting the operating status of water treatment equipment based on digital twins according to claim 1, characterized in that, The system collects flow rate data, pressure data, temperature data, chemical reaction data, sensor status data, and historical operating data of the water treatment equipment, and preprocesses them to generate an equipment dataset, specifically including: The flow data includes influent flow data and effluent flow data. The influent flow data is the amount of water flowing into the water treatment equipment per unit time, and the effluent flow data is the amount of water discharged from the water treatment equipment per unit time. The pressure data includes inlet pipe pressure data and outlet pipe pressure data. The inlet pipe pressure data is the pressure of water flowing into the water treatment equipment per unit time, and the outlet pipe pressure data is the pressure of water flowing out of the water treatment equipment per unit time. The temperature data includes inlet water temperature data and outlet water temperature data. The inlet water temperature data is the temperature of the water flowing into the water treatment equipment per unit time, and the outlet water temperature data is the temperature of the water flowing out of the water treatment equipment per unit time. The chemical reaction data includes pH data, dissolved oxygen data, ammonia nitrogen concentration data, and redox potential data; The sensor status data includes sensor operating status data, sensor signal strength data, and sensor fault status data; The equipment operation history data includes equipment start-up and shutdown time data, equipment operation mode data, equipment fault history data, and equipment maintenance record data; The collected data is cleaned to remove duplicate data and handle abnormal data. The Z-score method is used to detect outliers, and missing data is filled in by interpolation. The values ​​of different datasets are scaled to a uniform range to generate a device dataset.

3. The method for predicting the operating status of water treatment equipment based on digital twins according to claim 1, characterized in that, The process of performing feature selection, time series decomposition, LSTM feature selection, PCA dimensionality reduction, and feature fusion on the device dataset to generate a feature dataset specifically includes: Feature selection is performed on the device dataset using the information gain method. By calculating the information gain between each feature and the device operating state, features with information gain values ​​greater than a preset threshold are selected to generate a feature selection dataset. The equipment dataset is decomposed into time series components using a seasonal trend decomposition method. The time series component of the equipment dataset is decomposed into long-term trend components, seasonal components, and residual components. Long-term trend components representing the long-term change pattern of the equipment are extracted, seasonal components representing the periodic fluctuations of the equipment are extracted, and residual components representing the short-term instability of the equipment are extracted. Each component is filtered according to the variance explained rate, and components with a variance explained rate greater than a set threshold are retained to generate the time series decomposed dataset. LSTM feature selection is performed on the device dataset. The LSTM model is used to model the device dataset and automatically learn the most important time-dependent features. Through gradient boosting, features with gradient importance greater than a set threshold are selected to generate an LSTM feature dataset. Specifically, the selection is performed by weight normalization, retaining features with normalized weights greater than a set threshold. PCA dimensionality reduction is performed on the device dataset. The PCA method is used to reduce the dimensionality of the features of the device dataset, retaining the principal components whose cumulative variance contribution rate is greater than a preset threshold, removing redundant features, and generating a PCA dimensionality-reduced dataset. The feature selection dataset, time series decomposition dataset, LSTM feature dataset, and PCA dimensionality reduction dataset are fused using a weighted average method or a feature connection method to generate a feature dataset.

4. The method for predicting the operating status of water treatment equipment based on digital twins according to claim 1, characterized in that, The process of constructing a digital twin model based on the feature dataset and outputting a digital twin device operation model specifically includes: The feature dataset is input into the digital twin model, which is used to dynamically simulate the operating status of the equipment and perform optimization predictions. The construction of digital twin models combines physical modeling and data-driven modeling, specifically including: Physical modeling utilizes the principles of quantum mechanics and employs quantum computing models to model the behavior of microscopic particles in the equipment. By leveraging the superposition and entanglement characteristics of qubits, it simulates molecular-level dynamics during water treatment, providing an accurate microscopic simulation of the equipment's operating state. The quantum computing model is performed using quantum Monte Carlo simulation. The reaction kinetics model is based on the chemical reaction kinetics model and simulates the reaction process inside the water treatment equipment. The Arrhenius equation is used to describe the relationship between the chemical reaction rate and temperature. Graph neural networks are used to perform data-driven modeling of the operating status of water treatment equipment. Graph convolutional networks are used to extract the complex relationships between different components of the equipment, and the structural information of the graph is used to optimize the prediction of the operating status of water treatment equipment. The input graph data consists of the node features of the device. The features are transferred and aggregated through the graph convolutional network layer, and the output is the state information of each component. By combining feature datasets with physical models for simulation and prediction, a digital twin device operation model is obtained. Real-time sensor data is input into the digital twin device operation model and dynamically adjusted. The device operation status is updated in real time. By comparing the current device status with the prediction results of the digital twin device operation model, the key parameters in the digital twin device operation model are adjusted. Output a digital twin device operation model, which includes the current operating status of the device, future state prediction, device fault diagnosis, operating efficiency, and load prediction.

5. The method for predicting the operating status of water treatment equipment based on digital twins according to claim 1, characterized in that, The improved TimeSformer model specifically includes: The digital twin device operation model is applied to the improved TimeSformer model. Physically constrained convolutional layers are used for feature extraction. Based on the physical model of the device, convolution operations are performed on the device time series data to extract local features from the time series data. The design of the convolution kernel is based on the physical characteristics of the device during operation, and the generated physically constrained convolution output dataset is generated. The convolution kernels in the physically constrained convolution output dataset are optimized using the NSGA-II algorithm. The population is initialized, and multiple convolution kernel configurations are randomly generated. The fitness evaluation method is used to calculate the fitness by evaluating the prediction accuracy and computational complexity of each convolution kernel configuration. Individuals with high fitness are selected first and the next round of crossover and mutation operations are performed. Individuals with better fitness are selected by non-dominated sorting, and the diversity of the population is maintained by crowding comparison. A spatiotemporal separation self-attention mechanism is set up. The correlation between each time step is calculated through the temporal self-attention mechanism to identify the long-term trend in the operation of the device. The interrelationship between the various components of the device is modeled through the spatial self-attention mechanism. The spatial self-attention mechanism captures the complex interaction relationship between the various components of the device through a graph convolutional network and generates corresponding state features for each component. Topological data analysis is applied to the feature dataset obtained through the spatiotemporal separation self-attention mechanism. The persistent cohomology method is used to analyze the topological structure of the device time series data, identify key topological features in the data, extract the most representative features for predicting the device operating status based on the analysis results, remove redundant data, and generate an optimized feature dataset. The time and spatial features in the optimized feature dataset are weighted and synthesized to obtain comprehensive equipment operating status features. The prediction layer is then used to predict the equipment operating status, and the predicted equipment operating status results are output, including the equipment load, efficiency, fault warning and future operating trend.

6. The method for predicting the operating status of water treatment equipment based on digital twins according to claim 1, characterized in that, The step of automatically adjusting the equipment's operating strategy using an adaptive fuzzy genetic algorithm based on the equipment operating status prediction results, and outputting an optimized equipment operating strategy, specifically includes: The equipment operating status prediction results are passed as input parameters to the adaptive fuzzy genetic algorithm model. By initializing the population in the fuzzy genetic algorithm, multiple device operation strategy configurations are randomly generated based on the current operating status and historical data of the device, and the effectiveness of each strategy is evaluated through a fitness function. The device's operating strategy is evaluated using fuzzy rules based on the device's operating status. The fuzzy rule base includes multiple control rules, which are calculated through a fuzzy inference mechanism to generate a fuzzy fitness value for each strategy. Based on the fuzzy fitness value, the initial population is optimized using crossover, mutation, and selection operations in the genetic algorithm. The fitness of each individual is calculated through the fitness evaluation function, and individuals with fitness values ​​exceeding a preset threshold are selected for crossover and mutation operations in the next generation. During the iteration process, an adaptive adjustment mechanism is used to dynamically adjust the control parameters; after multiple generations of iterative optimization, an optimized equipment operation strategy is output, which can automatically adjust the equipment's operating parameters based on the equipment's real-time operating status and future prediction results. Based on the optimized equipment operation strategy, the water treatment equipment is adjusted in real time. The operating parameters of the equipment are dynamically adjusted by real-time sensor data and equipment status prediction results.

7. The method for predicting the operating status of water treatment equipment based on digital twins according to claim 1, characterized in that, The process of monitoring equipment status in real time based on equipment operating status prediction results and issuing fault warning information during equipment operation specifically includes: By receiving real-time sensor data and operating status data from the equipment, the actual operating status of the equipment is monitored. The equipment status is compared with the prediction results output by the digital twin equipment operating model to identify status deviations. When the parameters exceed the preset threshold, a fault warning is triggered. Based on the equipment's historical operating data and real-time sensor data, fault warning rules and thresholds are set, and the equipment's operation is monitored according to the fault mode. When the equipment status is abnormal and the deviation exceeds the threshold, fault warning information is generated and the alarm system is triggered. The fault warning information includes the fault type, affected components, and potential causes of the fault.

8. The method for predicting the operating status of water treatment equipment based on digital twins according to claim 1, characterized in that, The process of updating and optimizing the digital twin model and the improved TimeSformer model based on real-time operational data and equipment operating status prediction results specifically includes: Real-time operational data, including equipment flow, pressure, temperature, chemical reaction data, and equipment operating status, is collected in real time through sensors and equipment control systems. This data is then combined with equipment operating status prediction results and used as input data to transmit to the digital twin model and the improved TimeSformer model. Based on real-time operational data, key parameters in the digital twin model are dynamically adjusted; Based on the equipment operation status prediction results, the improved TimeSformer model is incrementally learned. Through the model update mechanism, combined with new operation data and prediction results, the convolution kernel and self-attention mechanism in the TimeSformer model are further optimized. We continuously iterate and train using the optimized digital twin model and the improved TimeSformer model, adjusting the time series processing and feature extraction capabilities in the structure of the digital twin model and the improved TimeSformer model.

Images