Intelligent evaluation system for health monitoring of reverse osmosis (RO) membrane

Abstract

The invention relates to the technical field of reverse osmosis (RO) membrane health monitoring, and discloses an intelligent evaluation system for reverse osmosis (RO) membrane health monitoring. The system comprises an operation data acquisition module used for acquiring historical operation data of a plurality of groups of reverse osmosis (RO) membranes; the feature association modeling module is used for constructing a health feature map according to historical operation data association health influence parameters and initializing association weights; the state trend classification module is used for classifying the historical operation data to generate a health state trend and calling a corresponding map; the aging feature extraction module and the abnormal feature extraction module are respectively used for extracting core features and optimizing atlas weights through a health parameter analysis model; and the health state evaluation module performs evaluation according to the atlas calling parameters. The system integrates technologies such as a moving average algorithm, an Euclidean distance algorithm, an auto-encoder, a GRU and an LSTM to realize deep analysis of running data of the RO membrane and accurate evaluation of a health state, can early warn membrane performance attenuation and abnormity in advance, and reduces maintenance cost.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of reverse osmosis (RO) membrane health monitoring, and particularly to an intelligent evaluation system for RO membrane health monitoring. BACKGROUND

[0002] As a core component in water treatment processes, the performance of reverse osmosis (RO) membranes directly affects water quality and system operation efficiency. In actual applications, RO membranes may exhibit aging, pollution, and other problems due to long-term operation, resulting in decreased water production and reduced desalination rates. If not promptly monitored and addressed, this will seriously affect the normal operation of water treatment systems.

[0003] There are many deficiencies in the technical means for RO membrane health monitoring. Traditional monitoring methods rely heavily on manual analysis of operation data, which is not only time-consuming and labor-intensive, but also difficult to capture subtle changes in membrane performance in real time, making it difficult to meet the demand for rapid evaluation of RO membrane health status in actual production. Existing evaluation systems often use fixed evaluation models and weight settings, which cannot be dynamically adjusted according to different operating stages and maintenance frequencies of RO membranes, resulting in insufficient accuracy in identifying membrane aging and abnormal states.

[0004] Existing technologies lack deep mining of data features when processing RO membrane operation data, making it difficult to effectively extract core features reflecting the health status of the membrane, resulting in inaccurate health status trend analysis and inability to predict membrane performance degradation and potential failures in advance. Meanwhile, most evaluation systems do not consider the impact of maintenance time and frequency on membrane performance, and the data classification method is simple, resulting in an unscientific construction of health feature maps, which further affects the reliability of health status evaluation. These problems make the existing RO membrane health monitoring technology have the disadvantages of evaluation lag, high maintenance cost, and short service life of the membrane in actual applications, and there is an urgent need for a more intelligent and accurate health monitoring system to solve the above problems. SUMMARY

[0005] The present application aims to provide an intelligent evaluation system for reverse osmosis (RO) membrane health monitoring to solve the problems raised in the background.

[0006] To achieve the above-mentioned purpose, the present application provides an intelligent evaluation system for reverse osmosis (RO) membrane health monitoring, which comprises: an operation data acquisition module for acquiring historical operation data of multiple groups of RO membranes; a feature correlation modeling module for correlating first and second health influence parameters based on the historical operation data, constructing a health feature map, and initializing the correlation weights of the health feature map; a state trend classification module, which classifies a plurality of sets of the historical operation data of the reverse osmosis RO membranes within an S period, generates health state trends, compares and labels the health state trends, and calls corresponding health feature maps; an aging feature extraction module, which calls the first health impact parameter according to the health feature maps of the same category, extracts first and second core features of the first health impact parameter through a health parameter analysis model, and optimizes a first weight of the health feature map; an abnormal feature extraction module, which calls the second health impact parameter according to the health feature maps of the same category, extracts third and fourth core features of the second health impact parameter through the health parameter analysis model, and optimizes a second weight of the health feature map; a health state evaluation module, which, when evaluating, calls the first and second health impact parameters according to the health feature map to evaluate.

[0007] Preferably, classifying a plurality of sets of the historical operation data of the reverse osmosis RO membranes within an S period comprises: dividing the historical operation data according to the maintenance time S as a dividing point, and classifying the data according to the total number of maintenance times as a maintenance frequency.

[0008] Preferably, comparing the health state trends and calling corresponding health feature maps comprises: extracting the health state trends of the divided historical operation data through a sliding average algorithm, calculating the mean of the health state trends of the same maintenance frequency as a standard health trend; calculating the deviation of a plurality of the health state trends of the same maintenance frequency from the standard health trend, labeling a plurality of trend categories according to the deviation, and calling the health feature map of the corresponding category; comparing the health state trends of the same reverse osmosis RO membrane at the mth maintenance frequency and the m+1th maintenance frequency according to the Euclidean distance algorithm, and if the comparison result is less than a critical value of the health state trend, calling the health feature map of the reverse osmosis RO membrane corresponding to the m+1th maintenance frequency.

[0009] Preferably, calling the first health impact parameter according to the health feature maps of the same category, extracting first and second core features of the first health impact parameter through a health parameter analysis model, and optimizing a first weight of the health feature map comprises: The first core feature and the second core feature are obtained according to a performance attenuation analysis model of the health parameter analysis model; the performance attenuation analysis model comprises an influence parameter preprocessing unit, an influence parameter feature analysis unit, a first weight optimization unit and a first weight storage unit; The influence parameter preprocessing unit performs data segmentation and data normalization on the first health influence parameter to generate input data; the influence parameter feature analysis unit predicts next segmented time data through a 1-layer self-encoder and a 5-layer GRU, and identifies the first core feature of the input data and the second core feature of the next segmented time data through a 3-layer deep convolution layer, a 1-layer hollow convolution layer and a 4-layer GRU; The first weight optimization unit optimizes the correlation weight of the health feature map according to the first core feature and the second core feature; and the first weight storage unit stores the correlation weight of the health feature map.

[0010] Preferably, the first weight optimization unit optimizing the correlation weight of the health feature map according to the first core feature and the second core feature comprises: linearly adjusting the initial correlation weight of the health feature map according to the feature importance degree output by the performance attenuation analysis model, combining the stability of the first core feature and the predictability of the second core feature.

[0011] Preferably, the calculation of the feature importance degree comprises: grouping the input core features through a hierarchical clustering algorithm, calculating the similarity between the features by combining a dynamic time warping algorithm, and determining the importance level label of each group of features according to the feature similarity distribution and the within-group dispersion.

[0012] Preferably, according to the second health influence parameter of the health feature map of the same category, the third core feature and the fourth core feature of the second health influence parameter are extracted by the health parameter analysis model and the second weight of the health feature map is optimized. The third core feature and the fourth core feature are obtained according to an abnormal diagnosis analysis model of the health parameter analysis model; the abnormal diagnosis analysis model comprises an abnormal parameter preprocessing unit, an abnormal parameter feature extraction unit, a second weight optimization unit and a second weight storage unit; The abnormal parameter preprocessing unit performs data segmentation and data normalization on the second health influence parameter to generate to-be-identified data; the abnormal parameter feature extraction unit extracts the third core feature of the to-be-identified data according to a 1-layer attention mechanism, a 1-layer time convolution layer and a 4-layer bidirectional LSTM unit, and extracts the fourth core feature of the to-be-identified data through a 2-layer residual convolution layer, a 1-layer hollow convolution layer and a 4-layer LSTM; The second weight optimization unit optimizes the correlation weight of the second health influence parameter based on the third core feature and the fourth core feature; and the second weight storage unit stores the optimized correlation weight.

[0013] Preferably, the second weight optimization unit optimizing the correlation weight of the second health influence parameter based on the third core feature and the fourth core feature comprises: according to the feature abnormality degree output by the abnormality diagnosis analysis model, combining the significance of the third core feature and the persistence of the fourth core feature, performing exponential adjustment on the initial correlation weight of the second health influence parameter.

[0014] Preferably, when the total number of maintenance times S is taken as the maintenance frequency for data classification, if the maintenance frequency is less than the set frequency threshold, it is classified into the initial maintenance category, and if the maintenance frequency is greater than or equal to the set frequency threshold, it is classified into the stable maintenance category.

[0015] Preferably, the way of calling the corresponding first health influence parameter and second health influence parameter according to the health feature map when the health state evaluation module evaluates comprises: extracting the first health influence parameter and the second health influence parameter in the current running data, respectively matching the optimized first weight and the second weight in the health feature map, and calculating a comprehensive health index as an evaluation basis through weighted summation.

[0016] Compared with the prior art, the present application has the following beneficial effects: The system acquires multiple groups of historical running data through the running data acquisition module, providing a rich data basis for subsequent feature analysis and state evaluation, ensuring the reliability of the evaluation results. The feature correlation modeling module constructs a health feature map and initializes the correlation weight, establishing a scientific framework for the evaluation of the health state of the RO membrane, making the evaluation process have rules to follow. The state trend classification module classifies the historical running data according to the maintenance time and the maintenance frequency, generates a health state trend, and through the comparison and analysis of the sliding average algorithm and the Euclidean distance algorithm, can accurately identify the membrane state trend at different maintenance stages, improving the pertinence and accuracy of the evaluation.

[0017] The aging feature extraction module and the abnormal feature extraction module respectively utilize the performance decay analysis model and the abnormality diagnosis analysis model, combining various machine learning algorithms such as autoencoder, GRU, attention mechanism, and bidirectional LSTM, to deeply mine the core features of the first health influence parameter and the second health influence parameter, and dynamically optimize the correlation weight of the health feature map according to the feature importance and abnormality, so that the system can adaptively adjust the evaluation model, better adapt to the performance changes of the RO membrane at different stages, and improve the recognition ability of the membrane aging and abnormal state.

[0018] The health state evaluation module extracts the impact parameters in the current running data, matches the optimized weights for weighted summation, calculates a comprehensive health index, and realizes accurate evaluation of the health state of the RO membrane. This dynamic weight adjustment mechanism and multi-model collaborative working mode not only improve the processing capacity of the system for complex data and the evaluation accuracy of the membrane health state, but also can detect the performance degradation trend and abnormal signs of the membrane in advance, provide a scientific basis for maintenance decision-making, reduce unnecessary maintenance costs, prolong the service life of the RO membrane, and improve the operation efficiency and stability of the water treatment system. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 A working principle diagram of the intelligent evaluation system for reverse osmosis (RO) membrane health monitoring is provided. Figure 2 A flowchart for health state trend comparison and atlas calling is provided. Figure 3 A flowchart for first health impact parameter feature extraction is provided. Figure 4 A flowchart for second health impact parameter feature extraction is provided. DETAILED DESCRIPTION

[0020] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0021] Please refer to Figures 1-4 The present application provides an intelligent evaluation system for reverse osmosis (RO) membrane health monitoring, which comprises.

[0022] A running data acquisition module acquires historical running data of multiple groups of reverse osmosis (RO) membranes; A feature correlation modeling module correlates a first health impact parameter and a second health impact parameter according to the historical running data, constructs a health feature atlas, and initializes the correlation weight of the health feature atlas; A state trend classification module classifies the historical running data of multiple groups of reverse osmosis (RO) membranes in an S period, generates a health state trend, compares and labels the health state trend, and calls the corresponding health feature atlas; An aging feature extraction module calls the first health impact parameter according to the health feature atlas of the same category, extracts first and second core features of the first health impact parameter through a health parameter analysis model, and optimizes the first weight of the health feature atlas; The abnormal feature extraction module extracts third core features and fourth core features of the second health influence parameter through the health parameter analysis model and optimizes the second weight of the health feature graph according to the same category of the health feature graph and the second health influence parameter. The health state evaluation module evaluates the first health influence parameter and the second health influence parameter according to the health feature graph. Embodiments

[0023] The historical running data of multiple groups of reverse osmosis (RO) membranes in an S period is classified, and the specific implementation is as follows: The system first acquires the historical running data of multiple groups of reverse osmosis (RO) membranes, which contains the running state information of the RO membranes at different time points. Then, according to the historical running data of all RO membranes, the historical running data is divided into two parts by taking maintenance time S as the dividing point. The maintenance time S here refers to the specific time point of the maintenance operation of the RO membrane, and the system needs to accurately identify and determine the specific time of each maintenance time S.

[0024] After the maintenance time S is determined, the system classifies the data according to the total number of maintenance times S as the maintenance frequency. In the specific operation, for the historical running data of each group of RO membranes, the system will count the number of maintenance times S contained in the data, which is the maintenance frequency of the RO membrane. For example, if the historical running data of a group of RO membranes records 3 maintenance times S, then the maintenance frequency of the data is 3.

[0025] After the maintenance frequency is counted, the system needs to divide the data into different categories according to the size of the maintenance frequency. Here, a frequency threshold is set. If the maintenance frequency is less than the set frequency threshold, the corresponding historical running data is divided into the initial maintenance category; if the maintenance frequency is greater than or equal to the set frequency threshold, it is divided into the stable maintenance category. The determination of the set frequency threshold needs to consider the characteristics of the RO membrane and the maintenance strategy in the actual application scenario and other factors. For example, in some application scenarios, the frequency threshold can be set to 5 times. When the maintenance frequency is less than 5 times, it is considered that the RO membrane is in the initial maintenance stage, and its running state and performance may still be in the process of gradual stabilization; when the maintenance frequency is greater than or equal to 5 times, it is considered that the RO membrane enters the stable maintenance stage, and its running state is relatively stable.

[0026] In the implementation process, the system needs to efficiently process and analyze historical operation data. First, the information of maintenance time S needs to be accurately extracted from a large amount of historical operation data, which may involve data parsing and screening. For example, historical operation data may be stored in the form of log files or database records, and the system needs to be able to identify the corresponding timestamp or related records of maintenance operations from them. Then, the extracted maintenance time S is counted, and the maintenance frequency of each reverse osmosis RO membrane is calculated. This step needs to ensure the accuracy of the statistics to avoid errors caused by data omission or errors.

[0027] When dividing the maintenance category, the system needs to compare the maintenance frequency with the set frequency threshold. This comparison process needs to strictly follow the set logic to ensure the accuracy of data classification. For example, when the maintenance frequency of a reverse osmosis RO membrane is 4 times, and the frequency threshold is set to 5 times, the data should be correctly classified into the initial maintenance category; when the maintenance frequency is 5 times, it should be classified into the stable maintenance category.

[0028] In addition, the system also needs to have the ability to process different data formats and storage methods. Since historical operation data may come from different data sources, their formats and storage methods may differ, and the system needs to be able to uniformly process and convert these data to ensure the extraction of maintenance time S and the statistics of maintenance frequency. For example, for structured data, maintenance time S can be extracted through database query statements; for unstructured data, text parsing technology may be used to identify maintenance time information.

[0029] In actual application, some special situations may be encountered, such as incomplete or erroneous records of maintenance time S. In response to such situations, the system needs to have certain fault tolerance and processing capabilities. For example, a certain time tolerance range can be set, and maintenance time records within the tolerance range are considered valid; for missing maintenance time records, data interpolation or other methods can be used to complete them to ensure the accuracy of maintenance frequency statistics.

[0030] In addition, the frequency threshold is not fixed and the system needs to allow it to be adjusted according to the actual operation and maintenance effect. For example, after the system has been running for a period of time, through the analysis of the health status evaluation results and maintenance effect of the reverse osmosis RO membrane, it is found that the current frequency threshold setting is unreasonable, resulting in inaccurate data classification, and the frequency threshold can be re-set and adjusted to improve the adaptability and accuracy of the system.

[0031] Through the above series of operations and processing, the system can accurately divide the historical running data of multiple groups of reverse osmosis RO membranes in the S period into initial maintenance categories and stable maintenance categories according to the maintenance frequency. This classification method provides an important basis for subsequent health status trend analysis and health feature spectrum calling, enabling the system to perform more targeted analysis and evaluation on reverse osmosis RO membranes in different maintenance stages, thereby improving the accuracy and effectiveness of the entire intelligent evaluation system. Embodiments

[0032] Compare the health status trends and call the corresponding health feature spectrum. The specific implementation is as follows: After completing the classification of the historical running data of multiple groups of reverse osmosis RO membranes, the system needs to compare the health status trends and call the corresponding health feature spectrum. The health status trends of the classified historical running data are extracted through the sliding average algorithm. In the application process of the sliding average algorithm, the system sets a suitable sliding window size, which needs to consider factors such as the time span and data density of the historical running data. For example, if the time span of the historical running data is large and the data points are dense, a smaller sliding window may be selected to capture the data trend more carefully; if the data points are sparse, a larger sliding window may be selected to smooth the data fluctuations.

[0033] In the application of the sliding average algorithm, the system moves the sliding window on the classified historical running data in turn, and calculates the average of the data in each window to obtain the corresponding health status trend value. In this way, random noise in the data can be effectively reduced, and the health trend that better reflects the actual running state of the reverse osmosis RO membrane can be extracted.

[0034] The mean of the health status trends of the same maintenance frequency is calculated as the standard health trend. The system first groups the historical running data according to the maintenance frequency. For all health status trend data in each maintenance frequency group, the arithmetic mean is calculated. For example, there are 10 health status trend data in a certain maintenance frequency group, and the system adds these 10 data and divides by 10 to obtain the standard health trend under this maintenance frequency. This standard health trend represents the typical health status trend of the reverse osmosis RO membrane under normal operation at this maintenance frequency.

[0035] The deviation of multiple health status trends of the same maintenance frequency from the standard health trend is calculated, and the multiple trend categories are labeled and the corresponding health feature spectrum is called according to the deviation. The deviation can be calculated using the Euclidean distance method. Specifically, for each health status trend data, the system calculates the Euclidean distance between it and the standard health trend under this maintenance frequency, and this distance value is the deviation.

[0036] After obtaining the deviation degree, the system needs to set different deviation degree intervals to divide the trend categories. For example, the deviation degree less than a certain threshold A can be set as the normal trend category, the deviation degree between the threshold A and the threshold B as the slight deviation trend category, and the deviation degree greater than the threshold B as the serious deviation trend category. The setting of these thresholds needs to be determined in combination with the actual operation of the reverse osmosis RO membrane and engineering experience to ensure that the division of the trend category can accurately reflect the health status of the reverse osmosis RO membrane.

[0037] For each health status trend data, the system determines the trend category to which it belongs according to its deviation degree and labels it with the corresponding category identifier. After completing the labeling, the system calls the corresponding health feature map according to different trend categories, and different trend categories may correspond to different health feature maps to more accurately analyze the health status of the reverse osmosis RO membrane.

[0038] According to the Euclidean distance algorithm, the health status trends of the same reverse osmosis RO membrane at the mth maintenance frequency and the m+1th maintenance frequency are compared. If the comparison result is less than the critical value of the health status trend, the health feature map of the reverse osmosis RO membrane corresponding to the m+1th maintenance frequency is called. In this process, the system first obtains the health status trend data of the same reverse osmosis RO membrane at the mth and m+1th maintenance frequencies.

[0039] The Euclidean distance algorithm is used to calculate the distance between the two health status trends. In the calculation process of the Euclidean distance algorithm, the two health status trend data are regarded as vectors in a multi-dimensional space, and the straight-line distance between them is calculated. The critical value of the health status trend is a preset standard value used to determine whether the health status trends at two maintenance frequencies are similar enough.

[0040] If the calculated Euclidean distance is less than the critical value, it means that the health status trends at the mth and m+1th maintenance frequencies are similar. In this case, the system calls the health feature map corresponding to the m+1th maintenance frequency. This is because in this case, the health status trend after the m+1th maintenance is less different from the trend after the previous maintenance, and calling the health feature map of the m+1th maintenance can more accurately reflect the current health status.

[0041] During the entire implementation process, the system needs to ensure that the parameter settings of the sliding average algorithm are reasonable to accurately extract the health status trend; the calculation of the standard health trend should be based on enough valid data to ensure its representativeness; the calculation of the deviation degree and the division of the trend category should be scientific and reasonable to truly reflect the health status difference of the reverse osmosis RO membrane; the calculation of the Euclidean distance and the setting of the critical value should be combined with the actual situation to ensure the accuracy of the health feature map calling.

[0042] In addition, the system also needs to have the ability to process large amounts of data, and can efficiently complete operations such as moving average calculation, mean calculation, deviation calculation, and Euclidean distance calculation. When facing data missing or abnormal situations, the system needs to have corresponding processing mechanisms, such as interpolation processing for missing data, identification and filtering for abnormal data, to ensure the accuracy of the calculation results.

[0043] At the same time, the system should allow relevant parameters to be adjusted according to actual operation conditions, such as sliding window size, deviation threshold, critical value, etc., to adapt to different reverse osmosis RO membrane operation scenarios and maintenance needs, and improve the flexibility and adaptability of the system. Embodiments

[0044] According to the same category of health feature map, the first health impact parameter is called, the first core feature and the second core feature are extracted through the health parameter analysis model, and the first weight of the health feature map is optimized. The specific implementation is as follows: Taking a group of reverse osmosis RO membranes running for half a year as an example, the system first divides its historical operation data into stable maintenance categories through the state trend classification module. At this time, the feature correlation modeling module has constructed the health feature map of this category, and the first health impact parameter includes the running data such as water inlet pressure, temperature, and pH value. The aging feature extraction module calls the first health impact parameter in the map of this category, for example, obtains the water inlet pressure data, temperature data, and pH value data of this RO membrane every day in the past 3 months. These data are stored in the form of time series, including every minute of sampling points.

[0045] The influence parameter preprocessing unit in the performance degradation analysis model processes these first health impact parameters. Taking the water inlet pressure data as an example, the preprocessing unit divides the 3-month data into 12 data segments according to weeks, and each data segment corresponds to a week of pressure data. For each data segment, normalization processing is performed to map the pressure value from 0.8MPa-1.2MPa to the 0-1 interval, for example, the actual pressure value 1.0MPa is converted to 0.5. Similarly, the temperature data (range 20℃-30℃) and pH value data (range 6.5-8.5) are segmented and normalized to generate a data matrix to be input. Each row in the matrix represents the normalized parameter value of a day, and each column corresponds to a different parameter.

[0046] After the impact parameter feature analysis unit receives the input data, it performs dimensionality reduction processing on the data through a 1-layer autoencoder. The encoding layer of the autoencoder contains 10 neurons, which compresses the 7-day influent pressure data (1440 sampling points per day) into a 50-dimensional feature vector. The decoding layer then attempts to reconstruct the original data to extract the core feature representation of the data. At the same time, a 5-layer GRU network processes the time series, for example, the first layer of GRU receives the normalized pressure data of each day, the second layer of GRU combines the temperature data, the third layer of GRU adds the pH value data, the fourth layer of GRU predicts the pressure trend of the next week, and the fifth layer of GRU outputs the predicted pressure data sequence. In this way, the GRU network captures the time series rule of the influent pressure changing with temperature and pH value, and predicts the influent pressure data of the next segment of time (i.e. the next week).

[0047] In identifying the core features, a 3-layer deep convolutional layer processes the normalized pressure, temperature, and pH data. The first convolutional layer uses a 3x3 convolution kernel with a stride of 1 to extract local features of the data, such as the fluctuation pattern of pressure in a certain time period. The second convolutional layer uses a larger convolution kernel (5x5) to capture features with a longer time span. The third convolutional layer combines the features of the previous two layers to generate more abstract feature representations. A 1-layer dilated convolution layer processes the data with a dilation rate of 2 to fill the data interval and capture more sparse feature patterns, such as the periodic pressure changes every week. A 4-layer GRU network further processes the features output by the convolutional layer, the first layer of GRU identifies the short-term features of daily parameter changes, the second layer of GRU captures the medium-term features of each week, the third layer of GRU analyzes the long-term features of each month, and the fourth layer of GRU integrates all features to identify the first core feature of the input data (such as the sensitivity parameter of influent pressure changing with temperature) and the second core feature of the next segment of time data (such as the amplitude parameter of pressure fluctuation in the next week).

[0048] The first weight optimization unit linearly adjusts the initial association weight of the health feature map according to the feature importance output by the performance decay analysis model, combined with the stability of the first core feature and the predictability of the second core feature. For example, in the feature importance calculation, the hierarchical clustering algorithm divides the first core feature (such as the pressure sensitivity parameter) and the second core feature (such as the fluctuation amplitude of the next week's pressure) into two groups, and the dynamic time warping algorithm calculates the similarity of the pressure sensitivity parameter with historical failure data as 0.75, and the similarity of the fluctuation amplitude parameter as 0.6. According to the feature similarity distribution (set similarity above 0.7 as high importance) and the within-group dispersion (the dispersion of the pressure sensitivity parameter is 0.15, and the dispersion of the fluctuation amplitude parameter is 0.2), the importance level label of the pressure sensitivity parameter is determined as "high", and the fluctuation amplitude parameter is "medium".

[0049] Therefore, the weight of the initial correlation weight of the water inlet pressure is 0.3, the weight of the temperature is 0.25, and the weight of the pH value is 0.25. Since the pressure sensitivity parameter belongs to high importance, the stability index is 0.8 (full score is 1), and the predictability index is 0.7, the weight of the water inlet pressure is linearly adjusted: 0.3 + 0.3 x (0.8 x 0.75 + 0.7 x 0.25) x 0.1 (adjustment coefficient), and the weight is changed to 0.33 after calculation. The weights of the temperature and the pH value are also adjusted according to the importance of the respective core features and the indexes, and the adjusted weights are stored in the health feature map through the first weight storage unit, and the original correlation weight is updated.

[0050] During the entire implementation process, the system needs to ensure the rationality of data segmentation, such as considering the maintenance period of the RO membrane when segmenting by week; the normalization processing needs to maintain the physical meaning of the parameters, and avoid feature distortion caused by excessive mapping. The parameter settings of the autoencoder and the GRU network need to be adjusted according to the data volume, for example, increasing the number of neurons when the data volume is large. The number of categories of hierarchical clustering needs to be determined according to the feature distribution to avoid too fine or too coarse clustering. The distance measurement method of the dynamic time warping algorithm needs to match the feature type, such as using Euclidean distance for continuous parameters. The setting of the adjustment coefficient needs to refer to engineering experience, for example, a coefficient of 0.1 can avoid excessive weight adjustment. When abnormal data (such as sudden drop of water inlet pressure) appears in a certain period of time, the system needs to filter the data through the outlier detection algorithm first, and then extract the features, to prevent abnormal data from affecting the core feature recognition. If the maintenance frequency changes, the system will re-call the health feature map of the corresponding category, re-extract the core features and optimize the weight, to ensure that the weight always matches the current running state. Embodiment

[0051] According to the same category of health feature map, the second health influence parameter is called, the third core feature and the fourth core feature are extracted through the health parameter analysis model, and the second weight of the health feature map is optimized, and the specific implementation is as follows: Taking a certain reverse osmosis RO membrane system as an example, the system has been running for 8 months, and has been classified into a stable maintenance category by the state trend classification module, and the second health influence parameter of the health feature map includes water production, desalination rate, and water inlet flow data. The abnormal feature extraction module calls the second health influence parameter in the category map, for example, obtains the daily water production data, desalination rate data and water inlet flow data of the RO membrane in the last 2 months, and these data are recorded in hours, with 24 sampling points per day.

[0052] The abnormal parameter preprocessing unit in the anomaly diagnosis analysis model processes these second health impact parameters. Taking the water production data as an example, the preprocessing unit divides the 2-month data into 8 data segments according to every 5 days, and each data segment corresponds to 5-day water production data. For each data segment, normalization processing is performed to map the water production from 80 m³ / h-120 m³ / h to the 0-1 interval, for example, the actual water production of 100 m³ / h is converted to 0.5. Similarly, the desalination rate data (range 98%-99.5%) and the water inflow data (range 150 m³ / h-180 m³ / h) are segmented and normalized to generate a to-be-identified data matrix, where each row represents the normalized parameter value of a certain hour, and each column corresponds to a different parameter.

[0053] The abnormal parameter feature extraction unit receives the to-be-identified data and processes the data through a 1-layer attention mechanism. The attention mechanism assigns different weights to different parameters such as water production, desalination rate, and water inflow. For example, when the water production shows abnormal fluctuations, the system automatically increases the attention weight of the water production parameter, making it occupy a more important position in the feature extraction process. Then, a 1-layer time convolution layer processes the time series data. The time convolution layer uses a causal convolution method to ensure that it does not rely on future data when processing current data, thereby more accurately capturing the time dependence of the data. For example, the time convolution layer can identify a continuous downward trend in water production over the past few hours.

[0054] A 4-layer bidirectional LSTM unit further extracts features from the data. The first layer of bidirectional LSTM processes daily water production data from both forward and reverse directions, capturing the change characteristics of water production within the day. The second layer of bidirectional LSTM combines desalination rate data to analyze the correlation characteristics between water production and desalination rate. The third layer of bidirectional LSTM adds water inflow data to explore the comprehensive influence characteristics among the three. The fourth layer of bidirectional LSTM integrates all features to extract the third core feature of the to-be-identified data, such as the abnormal decline rate parameter of water production within a specific time period.

[0055] Meanwhile, the data is processed by two residual convolution layers. The first residual convolution layer preserves part of the features of the original data through a skip connection, avoiding the loss of important information in the convolution process, such as preserving the basic trend of water production. The second residual convolution layer further extracts more complex features, such as the pattern characteristics of abnormal fluctuations in water production. The 1st layer of the hollow convolution layer processes the data with an expansion rate of 3, expands the receptive field, and captures longer-term feature dependencies, such as the periodic variation characteristics of weekly water production. The 4th layer of the LSTM unit processes the features output by the residual convolution layer and the hollow convolution layer. The first layer of LSTM identifies short-term abnormal features of water production, the second layer of LSTM captures medium-term desalination rate change features, the third layer of LSTM analyzes long-term water inflow fluctuation features, and the fourth layer of LSTM integrates all features to extract the fourth core feature of the data to be identified, such as the persistence parameter of the abnormal association between desalination rate and water inflow.

[0056] The second weight optimization unit optimizes the association weight of the second health impact parameter based on the third core feature and the fourth core feature. Specifically, according to the feature abnormality degree output by the abnormal diagnosis analysis model, combined with the significance of the third core feature and the persistence of the fourth core feature, the initial association weight of the second health impact parameter is exponentially adjusted. For example, in the feature abnormality degree calculation, by comparing the difference between the current feature and the historical normal feature, the abnormality degree of the water production abnormal decline rate is 0.8, and the abnormality degree of the abnormal association between the desalination rate and the water inflow is 0.6. The significance index of the third core feature is 0.9 (full score is 1), indicating that the water production abnormal decline rate feature performs outstandingly in the abnormal diagnosis; the persistence index of the fourth core feature is 0.7, indicating that the abnormal association between the desalination rate and the water inflow has lasted for a period of time.

[0057] In the initial association weight, the weight of water production is 0.35, the desalination rate is 0.3, and the water inflow is 0.25. Due to the high abnormality degree and significance of the water production abnormal decline rate, the system exponentially adjusts the weight of water production: 0.35 x exp(0.8 x 0.9 x 0.2) (the adjustment coefficient is 0.2), and the weight after calculation is about 0.35 x 1.15 = 0.4025. The weights of the desalination rate and the water inflow are also adjusted according to the abnormality degree and index of their respective core features. The adjusted weight is stored in the health feature map through the second weight storage unit, updating the original second weight association.

[0058] During the entire implementation process, the system needs to ensure that the time interval of data segmentation matches the abnormal monitoring period of the RO membrane, such as 5-day segmentation, which can effectively capture the abnormal pattern every week. The physical dimension corresponding relationship of the parameter needs to be preserved during the normalization process to avoid distortion of abnormal characteristics caused by normalization. The weight distribution of the attention mechanism needs to be trained according to historical abnormal data, for example, by learning from past failure cases to determine the attention priority of different parameters. The convolution kernel size and step of the time convolution layer need to be adjusted according to the data sampling frequency, such as 3x1 convolution kernel for hourly data. The number of hidden layer neurons of bidirectional LSTM needs to be set according to the complexity of data features, and the number of neurons can be increased to 128 when the features are complex.

[0059] The skip connection design of the residual convolution layer needs to ensure the effective transmission of the original features, avoiding the gradient vanishing problem in the deep network. The dilation rate setting of the dilated convolution layer needs to consider the time span of the data, such as dilation rate 3 can capture the feature dependence of 3 time steps. The forget gate, input gate and output gate parameters of the LSTM unit need to be trained through the back propagation algorithm to optimize the feature extraction effect. The calculation of feature abnormality needs to establish a reasonable normal feature baseline, for example, based on the normal operation data of the past 3 months to establish a baseline model. The adjustment coefficient in the exponential adjustment needs to refer to the industry standard, for example, a coefficient of 0.2 can control the weight adjustment range within a reasonable range.

[0060] When the system detects that there is missing data in a certain period of time (such as missing 2 hours of water production data due to sensor failure), linear interpolation method is used to complete the missing data first, and then feature extraction is performed to prevent data missing from affecting the accuracy of core features. If the maintenance operation of the RO membrane causes a significant change in its operating state, the system will reclassify the data and call the corresponding health feature map according to the new category, re-extract the core features and optimize the weights to ensure that the weights can reflect the current operating state of the RO membrane in real time. Embodiment

[0061] The health status evaluation module evaluates the corresponding first health impact parameter and second health impact parameter according to the health feature map, and the specific implementation method is as follows: Taking a set of RO membranes in a water treatment plant as an example, the set of RO membranes has been running for 10 months and is classified into the stable maintenance category by the state trend classification module, the first health influence parameter in the health feature map of the set of RO membranes includes the inlet water pressure, temperature, and pH value, and the corresponding optimized first weights are 0.35, 0.25, and 0.2 respectively; the second health influence parameter includes the water production, desalination rate, and inlet water flow, and the corresponding optimized second weights are 0.3, 0.3, and 0.25 respectively. The system collects the current running data of the set of RO membranes in real time, for example, the first health influence parameter obtained at a certain time is that the inlet water pressure is 1.1 MPa, the temperature is 25°C, and the pH value is 7.2; the second health influence parameter is that the water production is 110 m³ / h, the desalination rate is 98.8%, and the inlet water flow is 165 m³ / h.

[0062] After extracting the first health influence parameter and the second health influence parameter in the current running data, the optimized first weight and the second weight in the health feature map need to be matched respectively. For the first health influence parameter, the inlet water pressure 1.1 MPa needs to be normalized first, and the normalized range of the inlet water pressure recorded in the health feature map is 0.8 MPa-1.2 MPa, so the calculation method of mapping 1.1 MPa to the 0-1 interval is (1.1-0.8) / (1.2-0.8)=0.75. The normalized range of the temperature 25°C is 20°C-30°C, and the normalized value is (25-20) / (30-20)=0.5. The normalized range of the pH value 7.2 is 6.5-8.5, and the normalized value is (7.2-6.5) / (8.5-6.5)=0.35.

[0063] For the second health influence parameter, the normalized range of the water production 110 m³ / h is 80 m³ / h-120 m³ / h, and the normalized value is (110-80) / (120-80)=0.75. The normalized range of the desalination rate 98.8% is 98%-99.5%, and the normalized value is (98.8-98) / (99.5-98)≈0.533. The normalized range of the inlet water flow 165 m³ / h is 150 m³ / h-180 m³ / h, and the normalized value is (165-150) / (180-150)=0.5.

[0064] After the normalization is completed, the system starts to calculate the comprehensive health index by weighted summation. The weighted calculation of the first health influence parameter is: the inlet water pressure normalized value 0.75×weight 0.35=0.2625, the temperature normalized value 0.5×weight 0.25=0.125, and the pH value normalized value 0.35×weight 0.2=0.07, and the sum of the three is 0.2625+0.125+0.07=0.4575.

[0065] The weighted calculation of the second health impact parameter is: the normalized value of water production 0.75 x weight 0.3 = 0.225, the normalized value of desalination rate 0.533 x weight 0.3 ≈ 0.1599, the normalized value of influent flow 0.5 x weight 0.25 = 0.125, and the sum of the three is the weighted sum of the second health impact parameter 0.225 + 0.1599 + 0.125 ≈ 0.5099.

[0066] The weighted sum of the first health impact parameter and the weighted sum of the second health impact parameter are added to obtain the comprehensive health index 0.4575 + 0.5099 ≈ 0.9674. The system preset health evaluation threshold is 0.8, when the comprehensive health index is greater than or equal to 0.8, it is determined that the current health status of the RO membrane is good; when the comprehensive health index is between 0.5 and 0.8, it is determined to be in a sub-health state and needs attention; when the comprehensive health index is less than 0.5, it is determined to be in an abnormal state and needs maintenance. Since the comprehensive health index of the RO membrane is about 0.9674, which is greater than 0.8, the system evaluates that its current health status is good.

[0067] In actual application, the weight in the health feature spectrum is not fixed, but is updated in real time with the optimization of the aging feature extraction module and the abnormal feature extraction module. For example, if the water production of the RO membrane group continues to decrease in subsequent operation, the abnormal feature extraction module will extract the core features of the second health impact parameter and optimize the second weight. Assuming that the weight of water production is adjusted from 0.3 to 0.35 after optimization. When the water production 110m³ / h (0.75 after normalization) is collected again, its weighted value will be 0.75 x 0.35 = 0.2625, which is higher than the previous 0.225, so that the calculation result of the comprehensive health index can better reflect the influence of water production on the health status of the RO membrane.

[0068] When the system collects the current running data, it will verify the validity of the data. For example, if the water inlet pressure data at a certain time suddenly jumps and exceeds the normal range (such as suddenly rising from 1.1 MPa to 1.8 MPa), the system will first process the abnormal data through a data filtering algorithm to determine whether it is an abnormality caused by sensor failure or a real change in running state. If it is confirmed as a sensor failure, the previous valid data is used for interpolation replacement to ensure the accuracy of the input first health impact parameter and second health impact parameter.

[0069] In addition, the health status evaluation module will call the corresponding health feature map according to different maintenance frequency categories. For example, when the set of RO membranes is maintained for the 6th time, the maintenance frequency increases from 5 times to 6 times, the system will automatically call the health feature map corresponding to the 6th maintenance frequency from the stable maintenance category. The weights in this map may be different from the map at the 5th maintenance due to previous optimization. At this time, even if the current running data collected is the same, the comprehensive health index calculated will be different due to the change of the weight, so as to more accurately reflect the health status of the RO membrane after maintenance.

[0070] When calculating the comprehensive health index, the system will also consider the influence of different running time periods. For example, during the high temperature period in summer, the influence of temperature on the performance of RO membranes may be greater, and the temperature weight in the health feature map may be increased according to the optimization of historical data. Assuming that the temperature weight is adjusted from 0.25 to 0.3, when the temperature of 25℃ (0.5 after normalization) is collected, the weighted value will be 0.5x0.3=0.15, which is increased from the previous 0.125, so that the comprehensive health index can better reflect the influence of temperature on the health of RO membranes during the high temperature period.

[0071] Through the above specific embodiments, the health status evaluation module can accurately weight and calculate the first health influence parameter and the second health influence parameter in the current running data according to the optimized weights in the health feature map, obtain the comprehensive health index, and thus realize scientific evaluation of the health status of the reverse osmosis RO membrane. In this process, the system updates the weights in real time, verifies the validity of the data, matches the maps of different maintenance frequencies, and considers the influence of the running period, so as to ensure the accuracy and reliability of the evaluation results and provide a basis for the maintenance decision of the RO membrane.

[0072] It should be noted that, in this text, relationship terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between the entities or operations. Moreover, the terms "include", "contain" or any other variant thereof are intended to cover non-exclusive inclusion, so that the process, method, article or equipment including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or equipment.

[0073] Although the embodiments of the present application have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, replacements and variations of these embodiments can be made without departing from the principles and spirits of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. An intelligent assessment system for reverse osmosis (RO) membrane health monitoring, characterized in that: include: Run the data acquisition module to obtain historical operating data of multiple sets of reverse osmosis RO membranes; a feature association modeling module, which associates the first health impact parameter and the second health impact parameter according to the historical operation data, constructs a health feature map, and initializes an association weight of the health feature map; a status trend classification module, classifying the historical operating data of multiple groups of reverse osmosis RO membranes within the S cycle, generating health status trends, comparing and annotating the health status trends, and calling the corresponding health feature maps; an aging feature extraction module, which calls the first health impact parameter based on the health feature map of the same category, extracts the first core feature and the second core feature of the first health impact parameter through a health parameter analysis model, and optimizes the first weight of the health feature map; an abnormal feature extraction module, calling the second health impact parameter based on the health feature map of the same category, extracting the third core feature and the fourth core feature of the second health impact parameter through the health parameter analysis model, and optimizing the second weight of the health feature map; The health status assessment module calls the corresponding first health impact parameter and the second health impact parameter for assessment according to the health characteristic map during assessment.

2. The intelligent assessment system for reverse osmosis (RO) membrane health monitoring according to claim 1, characterized in that: Classifying the historical operation data of multiple groups of the reverse osmosis RO membranes within the S cycle includes: dividing the historical operation data based on the historical operation data of all the reverse osmosis RO membranes with the maintenance time S as a dividing point, and classifying the data according to the total number of the maintenance time S as the maintenance frequency.

3. The intelligent assessment system for reverse osmosis (RO) membrane health monitoring according to claim 1, characterized in that: Comparing the health status trend and calling the corresponding health characteristic map includes: Extracting the health status trend of the divided historical operation data by a sliding average algorithm, and calculating the average of the health status trends with the same maintenance frequency as a standard health trend; Calculating the deviations of the plurality of health status trends with the same maintenance frequency from the standard health trend, dividing the trend categories into a plurality of categories according to the deviations, marking the trends, and calling the health characteristic maps of the corresponding categories; The health status trends of the same reverse osmosis RO membrane at the mth maintenance frequency and the m+1th maintenance frequency are compared according to the Euclidean distance algorithm. If the comparison result is less than the critical value of the health status trend, the health characteristic map of the reverse osmosis RO membrane corresponding to the m+1th maintenance frequency is called.

4. The intelligent assessment system for reverse osmosis (RO) membrane health monitoring according to claim 1, characterized in that: Calling the first health impact parameter according to the health feature map of the same category, extracting the first core feature and the second core feature of the first health impact parameter through a health parameter analysis model, and optimizing the first weight of the health feature map includes: The first core feature and the second core feature are obtained according to a performance decay analysis model of the health parameter analysis model; the performance decay analysis model includes an influencing parameter preprocessing unit, an influencing parameter feature parsing unit, a first weight optimization unit and a first weight storage unit; The influencing parameter preprocessing unit performs data segmentation and data normalization on the first health influencing parameter to generate data to be input; the influencing parameter feature parsing unit predicts the next segmented time data through a 1-layer autoencoder and a 5-layer GRU, and identifies the first core feature of the data to be input and the second core feature of the next segmented time data through a 3-layer deep convolution layer, a 1-layer dilated convolution layer, and a 4-layer GRU; The first weight optimization unit optimizes the associated weight of the health feature map according to the first core feature and the second core feature; the first weight storage unit stores the associated weight of the health feature map.

5. The intelligent assessment system for reverse osmosis (RO) membrane health monitoring according to claim 4, characterized in that: The first weight optimization unit optimizes the association weight of the health feature map based on the first core feature and the second core feature, including: linearly adjusting the initial association weight of the health feature map according to the feature importance output by the performance decay analysis model, combined with the stability of the first core feature and the predictability of the second core feature.

6. The intelligent assessment system for reverse osmosis (RO) membrane health monitoring according to claim 5, characterized in that: The calculation of the feature importance includes: grouping the input core features through a hierarchical clustering algorithm, calculating the similarity between features in combination with a dynamic time warping algorithm, and determining the importance level label of each group of features based on the feature similarity distribution and the intra-group dispersion.

7. The intelligent assessment system for reverse osmosis (RO) membrane health monitoring according to claim 1, characterized in that: Calling the second health impact parameter according to the health feature map of the same category, extracting the third core feature and the fourth core feature of the second health impact parameter through the health parameter analysis model, and optimizing the second weight of the health feature map includes: The third core feature and the fourth core feature are obtained according to the abnormal diagnosis analysis model of the health parameter analysis model; the abnormal diagnosis analysis model includes an abnormal parameter preprocessing unit, an abnormal parameter feature extraction unit, a second weight optimization unit and a second weight storage unit; The abnormal parameter preprocessing unit performs data segmentation and data normalization on the second health impact parameter to generate data to be identified; the abnormal parameter feature extraction unit extracts the third core feature of the data to be identified based on 1 layer of attention mechanism, 1 layer of temporal convolution layer and 4 layers of bidirectional LSTM units, and extracts the fourth core feature of the data to be identified through 2 layers of residual convolution layers, 1 layer of void convolution layer and 4 layers of LSTM; The second weight optimization unit optimizes the associated weight of the second health impact parameter based on the third core feature and the fourth core feature; and the second weight storage unit stores the optimized associated weight.

8. The intelligent assessment system for reverse osmosis (RO) membrane health monitoring according to claim 7, characterized in that: The second weight optimization unit optimizes the association weight of the second health impact parameter based on the third core feature and the fourth core feature, including: exponentially adjusting the initial association weight of the second health impact parameter according to the feature abnormality output by the abnormal diagnosis analysis model, combined with the significance of the third core feature and the persistence of the fourth core feature.

9. The intelligent assessment system for reverse osmosis (RO) membrane health monitoring according to claim 2, characterized in that: When data is classified according to the total number of maintenance times S as maintenance frequency, if the maintenance frequency is less than the set frequency threshold, it is classified as the initial maintenance category; if the maintenance frequency is greater than or equal to the set frequency threshold, it is classified as the stable maintenance category.

10. The intelligent assessment system for reverse osmosis (RO) membrane health monitoring according to claim 1, characterized in that: The health status assessment module calls the corresponding first health impact parameter and the second health impact parameter according to the health characteristic map during assessment, including: extracting the first health impact parameter and the second health impact parameter in the current operating data, matching the optimized first weight and the second weight in the health characteristic map respectively, and calculating the comprehensive health index through weighted summation as the basis for assessment.

Images