A method for monitoring the operation of MVR evaporators used in industrial wastewater treatment

Abstract

This invention relates to the technical field of industrial equipment monitoring, specifically to a method for monitoring the operation of an MVR evaporator used in industrial wastewater treatment. This method is immune to amplitude interference under normal operating conditions and can dynamically fuse waveform patterns from multiple sensors. The method includes: S1, obtaining a normalized pattern sequence for each sensor; S2, calculating the pattern offset score between the normalized pattern sequence and its corresponding healthy pattern template; S3, weighted fusing the pattern offset scores of each sensor based on dynamic sensitivity weights to obtain an adaptive blockage index; and S4, analyzing the continuous offset trend of the adaptive blockage index relative to a preset healthy baseline using a cumulative sum control chart to monitor the operating status of the MVR evaporator. This invention improves the monitoring sensitivity and decision reliability for early-stage collaborative anomalies related to blockage.

Description

Technical Field

[0001] This invention relates to the technical field of industrial equipment monitoring, and specifically to a method for monitoring the operation of an MVR evaporator used in industrial wastewater treatment. Background Technology

[0002] In the fields of chemical engineering, pesticides, and new energy, industrial wastewater is typically characterized by high salinity and high chemical oxygen demand (COD). Mechanical vapor recompression evaporators (MVR evaporators) are the mainstream equipment for treating such industrial wastewater. In forced circulation MVR evaporators designed to handle high salinity or crystallization conditions, one of the core components is the forced circulation pump. The function of the forced circulation pump is to maintain high-speed circulation of industrial wastewater within the heat exchange tubes, which is crucial for improving heat transfer efficiency and inhibiting scaling on the tube walls.

[0003] However, during the evaporation process, industrial wastewater often contains suspended particulate matter or precipitates a large amount of crystals, leading to impeller wear or cavitation in the forced circulation pump. More seriously, these solids tend to accumulate in the pump body or pipelines, causing blockages. Once a blockage occurs, the flow rate of industrial wastewater in the pipes will decrease, the heat transfer coefficient will deteriorate, and instantaneous scaling will occur, forcing the entire MVR evaporator to shut down urgently, threatening the continuity and stability of production.

[0004] To achieve condition monitoring, existing technologies typically install sensors for current, vibration, and inlet / outlet pressure difference on forced circulation pumps and employ a dynamic time warping algorithm (DTW algorithm) to determine whether the equipment's operating status has deviated abnormally by calculating the warping distance between the real-time operating data sequence and the health status template sequence.

[0005] However, traditional DTW algorithms have two main technical drawbacks when applied to forced circulation pump monitoring: First, they are overly sensitive to fluctuations in normal operating conditions. When an MVR evaporator processes different batches and concentrations of industrial wastewater, the density and viscosity of the wastewater will change. This is normal production fluctuation, and such fluctuations will cause changes in the load on the forced circulation pump, resulting in a shift in the overall amplitude of sensor readings (such as current and differential pressure). Although the forced circulation pump is still operating in a healthy state at this time, the traditional DTW algorithm, based on Euclidean distance measurement, will judge this amplitude difference as a significant mode deviation, thus generating a large number of false alarms for blockage.

[0006] Secondly, it lacks the ability to synergistically integrate multi-dimensional features. Blockage of a forced circulation pump is a complex process, and its early signs often manifest as a specific combination of deviations in waveform patterns from multiple sensors (such as current, vibration, and differential pressure), rather than drastic changes from a single sensor. Traditional DTW algorithms typically calculate the normalized distances of each sensor independently and then perform a simple weighted sum, ignoring the synergistic changes between multi-sensor data. This results in insufficient monitoring sensitivity, making it difficult to effectively predict blockages in their early stages. Summary of the Invention

[0007] The present invention aims to solve the above problems and provide a method for monitoring the operation of MVR evaporators for industrial wastewater treatment that is immune to amplitude interference under normal operating conditions and can dynamically fuse waveform modes from multiple sensors.

[0008] A method for monitoring the operation of an MVR evaporator for industrial wastewater treatment includes:

[0009] S1. Collect multi-sensor time-series data of the forced circulation pump in the MVR evaporator, divide the multi-sensor time-series data into multiple analysis windows, and perform independent in-window normalization processing on the time-series data of each sensor in each analysis window to obtain the normalized mode sequence of each sensor.

[0010] S2, Based on the pre-established health pattern templates of each sensor, calculate the pattern offset score between the normalized pattern sequence and its corresponding health pattern template;

[0011] S3. Calculate the historical offset mean and historical offset instability of the mode offset score for each sensor. Based on the historical offset mean and historical offset instability, determine the dynamic sensitivity weight that characterizes the reliability of the corresponding sensor signal. Then, perform weighted fusion of the mode offset scores of each sensor based on the dynamic sensitivity weight to obtain the adaptive congestion index.

[0012] S4, based on the cumulative and control chart analysis, analyzes the continuous deviation trend of the adaptive blockage index relative to the preset health baseline. When the cumulative value of the continuous deviation trend exceeds the preset warning threshold, a blockage warning signal is generated to monitor the operating status of the MVR evaporator.

[0013] This invention eliminates sensor amplitude interference caused by fluctuations in operating conditions (such as changes in waste liquid density and viscosity) by performing independent in-window normalization on each analysis window. This allows monitoring to focus on the true waveform pattern distortion, solving the problem of false alarms easily generated by traditional DTW algorithms. Simultaneously, by constructing dynamic sensitivity weights to achieve the fusion of multi-dimensional features and combining cumulative sum control charts for trend early warning, it improves the monitoring sensitivity and decision reliability for early-stage blockage anomalies.

[0014] Furthermore, in S1, during the independent in-window normalization process, Z-Score normalization is performed on the time-series data of each sensor within the analysis window. The calculation formula is as follows:

[0015] ;

[0016] in, For sensor s within analysis window j Normalized pattern sequence at time step, Sensor s is within analysis window j The raw data at that moment Let be the mean of the time-series data of sensor s within the analysis window j. Let be the standard deviation of the time series data of sensor s within the analysis window j. For a unit and The same constant, whose value is 1, the denominator This is used to prevent division by zero errors or numerical divergence when the standard deviation approaches zero, thereby improving the stability of normalization calculations.

[0017] By adding a constant " to the denominator The robust design ensures that the normalization calculation will not result in a zero denominator in extreme cases (such as when the data within the analysis window is completely consistent, resulting in a standard deviation of 0), thus enhancing the robustness of the algorithm and ensuring the stable execution of the normalization process.

[0018] Furthermore, in S2, the creation of the health mode template includes the following steps:

[0019] When the MVR evaporator is in a healthy state, time-series data from each sensor in multiple analysis windows are collected;

[0020] Independent in-window normalization processing is performed on the time series data of each sensor within multiple analysis windows to obtain multiple sets of normalized health pattern sequences corresponding to each sensor.

[0021] Cluster analysis was performed on multiple sets of normalized health pattern sequences, and the centroid of the sequence cluster was selected as the template of the health pattern.

[0022] Furthermore, in S2, for each sensor's normalized pattern sequence, the dynamic time warping distance between it and the corresponding health pattern template is calculated using a dynamic time warping algorithm to obtain the pattern offset score.

[0023] The dynamic time warping algorithm is used to calculate the dynamic time warping distance, which can assess the morphological similarity between two sequences, even if they have slight differences in speed or phase shifts on the time axis. This makes the assessment of mode shift more accurate and is not limited by strict time alignment.

[0024] Furthermore, the formula for calculating the dynamic sensitivity weight is as follows:

[0025] ;

[0026] in, For sensors exist Dynamic sensitivity weights at any given time. For sensors The pattern offset score is the moving average over a set number of windows in the past. For sensors The mode offset score is the moving standard deviation within a set number of windows in the past. For a unit and The same constant, its value is 1.

[0027] The design of dynamic sensitivity weights can distinguish between signals and noise. When the sensor pattern deviation is continuous and stable, the weight increases, indicating that it is a reliable signal indicating a fault; when the deviation is brief and unstable, the weight decreases, treating it as random noise. This mechanism enables dynamic evaluation of the signal-to-noise ratio of sensor signals.

[0028] Furthermore, the formula for calculating the adaptive congestion index is as follows:

[0029] ;

[0030] in, for Adaptive congestion index at any time. For sensors exist Mode offset score at time This indicates that the process iterates through all monitored sensors and sums the results.

[0031] Further, in S4, the adaptive congestion index is analyzed based on the cumulative sum control chart, including the following steps:

[0032] Calculate the historical mean of the adaptive congestion index under healthy conditions. and historical standard deviation ;

[0033] Set the relaxation parameter to tolerate normal fluctuations. , The value is based on the historical standard deviation. Sure.

[0034] Furthermore, the cumulative value of the sustained offset trend The calculation formula is:

[0035] ;

[0036] in, This is the cumulative value from the previous moment. The adaptive congestion index for the current moment. The operation ensures that the cumulative value is reset when the adaptive congestion index falls back below the baseline.

[0037] The CUSUM cumulative control chart exhibits extremely high statistical sensitivity to small but persistent shifts in the mean. Through The implemented reset mechanism makes the indicator extremely sensitive to persistent congestion worsening trends, but insensitive to transient, recoverable disturbances, preventing historical disturbances from permanently contaminating monitoring results and making early warning decisions more reliable.

[0038] Furthermore, the multi-sensor time-series data includes at least two of the following: current data of the motor in the forced circulation pump, vibration data of the pump housing, and differential pressure data between the inlet and outlet of the forced circulation pump.

[0039] Furthermore, the MVR evaporator is a forced circulation evaporator, and the operation monitoring method is used to provide early warning of blockage of the forced circulation pump caused by suspended solids or precipitated crystals in industrial wastewater.

[0040] Beneficial effects: In the MVR evaporator operation monitoring method provided by the present invention, firstly, in-window normalization preprocessing is performed in step S1. By performing independent in-window normalization on each data window, the amplitude information and waveform pattern information in the sensor readings are separated. This makes the subsequent dynamic time warping algorithm only focus on the distortion of the waveform pattern, and completely immune to the amplitude shift caused by normal operating condition fluctuations such as changes in waste liquid density and viscosity. This solves the problem of a large number of false alarms caused by amplitude interference in the prior art.

[0041] Secondly, step S4 employs a dynamic feature fusion mechanism. By constructing dynamic sensitivity weights for the sensors, the signal-to-noise ratio (SNR) of each sensor's mode deviation is dynamically evaluated, i.e., the persistence and stability of the deviation. The adaptive congestion index obtained based on this weighted fusion amplifies fault indication signals with persistent and stable deviations while suppressing noise signals with instantaneous jumps or random fluctuations. This allows the monitoring system to more sensitively capture the collaborative abnormal patterns among multiple sensors in the early stages of congestion, improving monitoring sensitivity compared to traditional simple weighting methods. Attached Figure Description

[0042] Figure 1 A flowchart of an MVR evaporator operation monitoring method for industrial wastewater treatment;

[0043] Figure 2 A graph showing the change of mode offset scores for each sensor over time;

[0044] Figure 3 A graph showing the change of dynamic sensitivity weights of each sensor over time;

[0045] Figure 4 A graph showing the adaptive congestion index over time;

[0046] Figure 5 This is a graph showing the change and warning of the cumulative offset value of CUSUM over time. Detailed Implementation

[0047] An embodiment of the MVR evaporator operation monitoring method for industrial wastewater treatment provided by the present invention:

[0048] Reference Figure 1 The core of this invention lies in firstly removing amplitude influence and retaining only waveform mode features by performing in-window normalization on sensor data; then, independently calculating the mode offset score of each sensor; and finally, constructing dynamic sensitivity weights to weight and fuse the mode offset scores of each sensor to obtain the final adaptive congestion index, and combining it with the cumulative sum control chart to achieve high-sensitivity early warning.

[0049] S1. Collect multi-sensor time-series data of the forced circulation pump in the MVR evaporator, and divide the multi-sensor time-series data into multiple analysis windows. Perform independent in-window normalization processing on the time-series data of each sensor in each analysis window to obtain the normalized mode sequence of each sensor.

[0050] First, multi-sensor time-series data acquisition is performed: at key locations of the forced circulation pump in the MVR evaporator, time-series data from at least three key sensors are simultaneously acquired. In this embodiment, three key data points are specifically acquired: current data... Vibration signals and pressure difference Specifically, a high-frequency current transformer is installed at the motor inlet of the forced circulation pump to obtain current data. Accelerometers are installed at critical locations on the housing of the forced circulation pump (e.g., bearing housings) to acquire vibration signals. Pressure sensors are installed on the inlet and outlet pipes of the forced circulation pump to calculate the pressure difference. .

[0051] The key data acquisition frequency should be sufficient to capture the dynamic characteristics of the forced circulation pump operation. The acquired multi-sensor time-series data is divided into analysis windows of fixed duration. For example, the acquisition frequency is set to 10Hz, and every 10 seconds of data (i.e., 100 data points) constitutes an analysis window; all subsequent calculations are based on this analysis window.

[0052] Next, independent in-window normalization is performed on the time-series data of each sensor within each analysis window to solve the problem of false alarms caused by sensitivity to amplitude in the prior art. Specifically, this embodiment uses an improved Z-Score normalization method.

[0053] With current data For example, first calculate the mean of all current data within the current analysis window. and standard deviation .

[0054] Normalized current mode sequence within the current analysis window Calculated using the following formula:

[0055] ;

[0056] in, For analysis window The raw current data at time t. It is the analysis window Within the time point, For a unit and The same constant has a value of 1.

[0057] For example, suppose the analysis window j contains 3 current data points: .

[0058] Calculate the mean .

[0059] Calculate the standard deviation .

[0060] Calculate the normalized current mode sequence:

[0061] .

[0062] .

[0063] .

[0064] Vibration signals and pressure difference The same in-window normalization process is also performed to obtain... and .

[0065] The logic of step S1 is as follows: by performing independent in-window normalization on each analysis window, all data within the analysis window are converted into a sequence with a mean close to 0 and a standard deviation close to 1. No matter how high the actual load of the forced circulation pump is (for example, the average current increases from 50A to 80A due to the increase in industrial wastewater concentration), as long as its operating waveform pattern remains unchanged, the shape of the normalized sequence will be almost identical.

[0066] Specifically, the denominator in the formula uses This is a robust design; in extreme cases, such as when the data within the analysis window is completely identical (e.g., 50A, 50A, 50A), then... Traditional Z-Score normalization can lead to calculation errors where the denominator is zero.

[0067] In S1, by performing independent in-window normalization on each analysis window, the sensor amplitude changes caused by normal operating condition fluctuations such as changes in industrial wastewater density and viscosity can be removed, so that subsequent analysis only focuses on waveform distortion, thereby solving the defect of traditional methods that are easily affected by amplitude interference and produce false alarms.

[0068] S2, based on the pre-established health pattern templates of each sensor, calculate the pattern offset score between the normalized pattern sequence and its corresponding health pattern template.

[0069] First, a health mode template library is established: when the MVR evaporator is confirmed to be in a healthy state, multi-sensor time-series data are continuously collected for M analysis windows (e.g., M=50). The health state here can be a state corresponding to just completed chemical cleaning, stable operation of the circulating pump, or high heat exchange efficiency. The in-window normalization processing in step S1 is performed on each of the M analysis windows to obtain M sets of normalized health mode sequences.

[0070] To enhance the robustness of the health pattern template, density-based noise-applied spatial clustering (DBSCAN clustering) is employed to cluster M similar sequences (e.g., M sequences). The normalized pattern sequences are clustered, and the centroid of the sequence cluster with the most consistent morphology and the most representative core features among the M similar sequences is used as the final health pattern template. Thus, the health current template is obtained. Healthy vibration template Health pressure differential template These health pattern templates represent the health fingerprint of a forced circulation pump under optimal conditions.

[0071] Then, an independent mode offset calculation based on dynamic time warping is performed: for each sensor's normalized mode sequence obtained in step S1, the standard dynamic time warping algorithm is used to calculate the dynamic time warping distance between it and the corresponding health mode template established in step S2, such as the normalized current mode sequence in step S1. With step S2 correspond.

[0072] Dynamic time warping is a matching algorithm that allows for non-linear scaling of the time axis. It finds two sequences (e.g., ...) through dynamic programming. and The pattern shift score is the shortest regularized path between two sequences. The cumulative distance of the shortest regularized paths is the pattern shift score. Even if there is a slight phase shift between two sequences in time, the pattern shift score can still assess the morphological similarity between the two sequences.

[0073] This calculation yields three independent mode shift scores, namely the current mode shift score. Vibration mode shift score Differential pressure mode offset score .

[0074] Step S2 establishes a normalized health pattern template under healthy conditions and uses a dynamic time warping algorithm to calculate the dynamic time warping distance between the real-time pattern sequence and the health pattern template. This accurately assesses the distortion of the waveform pattern of the current forced circulation pump operation compared to the healthy pattern, providing independent feature inputs with amplitude interference eliminated for subsequent multidimensional fusion analysis.

[0075] S3. Calculate the historical mean and historical instability of the mode offset score for each sensor. Based on the historical mean and historical instability, determine the dynamic sensitivity weight that characterizes the reliability of the corresponding sensor signal. Then, perform weighted fusion of the mode offset scores of each sensor based on the dynamic sensitivity weight to obtain the adaptive congestion index.

[0076] To address the issue of multi-dimensional feature fusion, this embodiment no longer employs a fixed weighted summation but instead designs an adaptive fusion mechanism. This mechanism posits that the more consistently and stably a sensor deviates from its pattern, the stronger its indication of congestion at that moment, and its weight should be dynamically increased.

[0077] Specifically, the dynamic sensitivity weights of each sensor are calculated. Mode offset score for each sensor Maintain a sliding history, for example, recording the mode offset scores for the past N=10 analysis windows, where s represents the sensor. belong .

[0078] Calculate the moving average of this historical data. and moving standard deviation , This represents the sensor's recent average mode shift level, while This represents the instability of the sensor's recent mode shift.

[0079] Dynamic sensitivity weights The calculation formula is as follows:

[0080] ;

[0081] in, It is a sensor exist Dynamic sensitivity weights at any given time; It is a sensor Pattern offset score The moving average over the past N analysis windows; It is a sensor Pattern offset score Moving standard deviation over the past N analysis windows, For a unit and The same constant, its value is 1.

[0082] The logic of this formula is: when When it increases An increase, meaning a continuous and stable deviation from the pattern is highly likely caused by a physical fault, is a reliable signal with a high signal-to-noise ratio.

[0083] when When it increases A decrease, i.e. a brief and drastic pattern shift, is likely caused by random noise or transient disturbances and is an unreliable signal with a low signal-to-noise ratio.

[0084] Therefore, this formula tends to assign high weights to sensors that consistently and stably deviate from the healthy pattern template, while suppressing sensors that only exhibit momentary jumps or random fluctuations. The denominator also uses... The design ensures the stability of the computation.

[0085] Finally, calculate the adaptive congestion index: using the dynamic sensitivity weights obtained in the previous step. The mode shift scores of each sensor at the current time calculated in step S2. A weighted average is then performed to obtain the final adaptive congestion index. Adaptive congestion index The calculation formula is as follows:

[0086] ;

[0087] in, Is A real-time adaptive congestion risk index; This indicates that dynamic sensitivity weights are used to modulate the mode shift at the current time step.

[0088] For example, suppose in At time 1, the mode shift score of the current data is calculated. Dynamic sensitivity weights Vibration mode shift score Dynamic sensitivity weights Pattern offset score of pressure difference Dynamic sensitivity weights Among them, a dynamic sensitivity weight of 3.0 represents a continuous and stable deviation in current data, a dynamic sensitivity weight of 0.5 represents a sudden and severe fluctuation in the vibration signal, and a dynamic sensitivity weight of 1.0 represents a slight deviation in the differential pressure signal. The sum of the weights... .

[0089] Sum of weighted scores Calculated using the formula .

[0090] Despite the vibration mode shift score The current is the highest, but due to its low weight, its contribution to the adaptive congestion index is actually less than that of the current data; the calculated adaptive congestion index here mainly reflects the continuous deviation of the current.

[0091] Step S3 constructs dynamic sensitivity weights and uses them for weighted fusion to achieve intelligent fusion of multi-dimensional features. It can automatically amplify sensor signals that deviate continuously and stably, and suppress instantaneous noise signals. This allows the fused adaptive congestion index to capture early signs of congestion earlier and more sensitively, with higher sensitivity compared to traditional fixed-weight fusion methods.

[0092] S4, based on the cumulative and control chart analysis, analyzes the continuous deviation trend of the adaptive blockage index relative to the preset health baseline. When the cumulative value of the continuous deviation trend exceeds the preset warning threshold, a blockage warning signal is generated to monitor the operating status of the MVR evaporator.

[0093] Setting a fixed threshold directly on the adaptive congestion index generated in step S3 can easily lead to jitter and false alarms. Considering that congestion is usually a gradual accumulation process, this embodiment introduces a cumulative sum control chart, which has the highest statistical sensitivity to small but continuous shifts in the detection mean.

[0094] Specifically, first, a health baseline is established: based on the adaptive congestion index collected from multiple analysis windows during the healthy operation phase, the arithmetic mean of these indices is calculated to obtain the historical mean. And calculate its sample standard deviation to obtain the historical standard deviation. .

[0095] Then, define a sensitivity parameter. Also known as tolerance relaxation, it is used to prevent the accumulation of normal random noise triggers. For example, it can be set based on historical data analysis. .

[0096] Calculate the cumulative value of the persistent offset trend The calculation formula is as follows:

[0097] ;

[0098] in, yes The cumulative value at any given moment; It is the cumulative value from the previous moment; It is the current adaptive congestion index. Exceeding the historical average Add tolerance relaxation The part.

[0099] The logic of this formula is: as long as consistently above baseline , It will continue to accumulate; if Falling back to baseline The following times, It will begin to decrease if the cumulative value If reduced to below zero, then It will be immediately reset to 0. This makes the cumulative value of the continuous offset trend extremely sensitive to persistent blocking trends, but insensitive to short-term disturbances.

[0100] Finally, trigger an alert: set an alert decision threshold. Early warning decision threshold It is an absolute cumulative threshold set based on experience or historical data. When When the system determines that the forced circulation pump has entered a continuous and worsening blockage trend, it immediately issues a blockage warning to the operator.

[0101] Reference Figures 2 to 5 The effects of this invention can be demonstrated through a time series diagram of simulated data comprising four stages (health, early fluctuations, congestion development, and severe congestion). Figure 2The output of step S2, namely the mode offset score, is shown: the current mode offset score during the "blockage development" phase (times 200-300). and differential pressure mode offset score It exhibits a slow and steady linear rise, while the vibration mode shift score... Although the mean has increased, the fluctuations are significant. Figure 3 The core innovation of step S3 is demonstrated, namely the dynamic sensitivity weight: during the "congestion development" phase (times 200-300), the current weight... and pressure difference weight Its weight increases rapidly due to its stable offset; while its oscillation weight... Because its offset is unstable, its weight is continuously suppressed. Figure 4 The image shows the fusion result of step S3, namely the Adaptive Congestion Index (ABI): (The red line) remains relatively stable during the "early fluctuation" phase (times 100-200), almost unaffected by noise. From time t=200 onwards, it is affected by high-weighted... and drive, It has begun to show a clear and stable upward trend. Figure 5 The document illustrates the early warning mechanism for step S4, namely the CUSUM cumulative sum: The purple line remained at 0 throughout the "early fluctuation" phase (times 100-200), successfully avoiding false alarms. In the "congestion development" phase (starting from approximately time t=220), the line continued to move upwards. It begins to accumulate linearly and stably. Eventually, at t... At time 275, the purple line clearly broke through the red warning threshold H, achieving an effective early warning before the "severe congestion" stage (t=300) occurred.

[0102] By introducing the CUSUM cumulative sum control chart to analyze the fused adaptive blockage index, it is possible to sensitively detect the weak but continuous upward trend in the nascent stage of blockage failure, and provide early warning when the risk accumulates to the decision threshold, thus saving valuable time for on-site maintenance (such as arranging flushing and adjusting the process) and effectively avoiding catastrophic unplanned downtime.

[0103] In addition, in the description of this specification, "multiple" means at least two, such as two, three or more, etc., unless otherwise expressly and specifically defined.

Claims

1. A method for monitoring the operation of an MVR evaporator for industrial wastewater treatment, characterized in that, The method comprises the following steps: S1, collecting multi-element sensor time series data of a forced circulation pump in an MVR evaporator, dividing the multi-element sensor time series data into multiple analysis windows, performing independent window normalization processing on the time series data of each sensor in each analysis window, and obtaining normalized mode sequences of each sensor; S2, calculating the mode offset score between the normalized mode sequence and the corresponding health mode template according to the pre-established health mode template of each sensor; S3, calculating the historical offset mean and historical offset instability of the mode offset score of each sensor, determining the dynamic sensitivity weight representing the signal reliability of the corresponding sensor according to the historical offset mean and the historical offset instability, and performing weighted fusion on the mode offset scores of each sensor based on the dynamic sensitivity weight to obtain an adaptive blockage index; S4, analyzing the adaptive blockage index based on the cumulative sum control chart, and generating a blockage warning signal when the cumulative value of the continuous offset trend exceeds the preset warning threshold, so as to realize the monitoring of the running state of the MVR evaporator; The analysis of the adaptive blockage index based on the cumulative sum control chart comprises: calculating a historical mean of the adaptive clogging index in the healthy state and a historical standard deviation ; Setting a slack parameter for tolerating normal fluctuations , Values based on historical standard deviation Determination; cumulative value of the persistent drift trend is: ； is the accumulated value for the previous time instant, is the adaptive congestion index for the current time instant, The operation ensures that the accumulated value is reset when the adaptive congestion index falls below the baseline.

2. The method for monitoring the operation of an MVR evaporator for industrial wastewater treatment according to claim 1, characterized in that, In S1, when performing independent window normalization processing, Z-Score normalization processing is performed on the time series data of each sensor in the analysis window, and the calculation formula is: ； wherein, is the normalized pattern sequence of the sensor s at the time instant , is the raw data of the sensor s at the time instant , is the mean of the time series data of the sensor s within the analysis window j, is the standard deviation of the time series data of the sensor s within the analysis window j, is a unitless constant with the same dimension as , which has a value of 1, and the denominator of is used to prevent division by zero error or numerical divergence when the standard deviation approaches zero, to improve the stability of the normalization calculation.

3. The method for monitoring the operation of MVR evaporator for industrial wastewater treatment according to claim 1, characterized in that, In S2, the establishment of the health mode template comprises the following steps: When the MVR evaporator is in a healthy state, the time series data of each sensor in multiple analysis windows is collected; Perform independent window normalization processing on the time series data of each sensor in multiple analysis windows to obtain multiple sets of normalized health mode sequences corresponding to each sensor; Cluster analysis is performed on the multiple sets of normalized health mode sequences, and the centroid of the sequence cluster is selected as the health mode template.

4. The method for monitoring the operation of an MVR evaporator for industrial wastewater treatment according to claim 3, characterized in that, In S2, the dynamic time warping distance between the normalized mode sequence of each sensor and the corresponding health mode template is calculated by a dynamic time warping algorithm to obtain the mode offset score.

5. The method for monitoring the operation of MVR evaporator for industrial wastewater treatment according to claim 1, characterized in that, The calculation formula of the dynamic sensitivity weight is: ； wherein, is the sensor at the time of the dynamic sensitivity weight, is the sensor a moving average of the mode shift score of the sensor over a past set number of windows, is the moving standard deviation of the mode shift score of the sensor is a constant that is the same as and has a value of 1.

6. The method for monitoring the operation of an MVR evaporator for industrial wastewater treatment according to claim 5, characterized in that, The calculation formula of the adaptive blockage index is: ; wherein, is an adaptive clogging index at the time instant, is a sensor at a pattern offset score at the time instant denotes iterating over all monitored sensors and summing up.

7. The method for monitoring the operation of an MVR evaporator for industrial wastewater treatment according to any one of claims 1-6, characterized in that, The multi-element sensor time series data comprises at least two of the current data of the motor in the forced circulation pump, the vibration data of the pump body shell, and the pressure difference data of the inlet and outlet of the forced circulation pump.

8. The method for monitoring the operation of an MVR evaporator for industrial wastewater treatment according to any one of claims 1-6, characterized in that, The MVR evaporator is a forced circulation evaporator, and the running monitoring method is used to warn the blockage of the forced circulation pump caused by suspended solids or precipitated crystals in industrial wastewater.

Images