A multi-modal liquid level monitoring method and system based on time sequence feature analysis

Abstract

This invention relates to the field of water purification equipment monitoring technology, and particularly to a multimodal liquid level monitoring method and system based on time-series feature analysis. The method includes: collecting and preprocessing multi-dimensional data through sensors to obtain the original pressure value, real-time water temperature value, real-time TDS value, and current operating condition indicator after interference removal; inputting the original pressure value and water density basic parameters into the hydrostatic pressure formula to obtain the initial liquid level value; combining the real-time water temperature value, real-time liquid level fluctuation frequency, and reference fluctuation frequency to obtain the final high-precision liquid level value; based on the final high-precision liquid level value, real-time TDS value, inlet water pressure data, and current operating condition indicator, calculating the flow rate attenuation score, TDS stability score, and correlation score according to weighted and threshold rules, and combining the three scores to obtain the overall health score; calculating the liquid level change rate and comparing it with a standard range, determining and outputting the anomaly type, and generating control commands. This solution improves the accuracy of liquid level measurement through multi-dimensional liquid level compensation.

Description

Technical Field

[0001] This invention relates to the field of water purification equipment monitoring technology, and in particular to a multimodal liquid level monitoring method and system based on time-series feature analysis. Background Technology

[0002] In the field of equipment involving liquid treatment and storage, such as water purification and storage, accurate liquid level monitoring, health management of core components, and efficient handling of anomalies are crucial to ensuring stable equipment operation, water quality compliance, and controllable operation and maintenance costs. However, several common problems still exist in the current application of these technologies and urgently need to be addressed.

[0003] In terms of basic parameter monitoring, key data such as liquid level, TDS value, and water temperature are easily affected by interference. Changes in ambient temperature cause fluctuations in water density, air bubbles in the liquid cause pressure signal distortion, and long-term use of sensors leads to performance degradation, all of which reduce measurement accuracy and affect subsequent steps such as water production flow calculation and water storage status judgment, resulting in misjudgment of equipment operating parameters.

[0004] In terms of core component management, core components such as RO membranes lack systematic evaluation methods. Existing methods mostly rely on manual periodic inspections or experience-based judgments, which cannot dynamically quantify the degree of wear and tear, making it difficult to predict membrane blockage, damage, and other faults in advance. This often leads to over-maintenance, wasting costs, or delayed maintenance, affecting water quality. Furthermore, it cannot accurately correlate health status with remaining lifespan, lacking a scientific basis for maintenance. Regarding anomaly handling, when faced with issues such as abnormal liquid levels or delayed valve responses, the equipment can often only issue alarms without autonomous handling capabilities, requiring manual on-site investigation and treatment. Delayed responses can easily lead to overflows, leaks, and equipment damage. Frequent manual intervention also significantly increases labor costs and reduces equipment operational stability and management efficiency. The current technical shortcomings in basic monitoring, core component management, and anomaly handling of water purification and storage equipment urgently require a more comprehensive solution to improve equipment operation quality and management efficiency. Summary of the Invention

[0005] This invention improves the accuracy of liquid level measurement through multi-dimensional liquid level compensation, avoids misjudgment of equipment operating status due to liquid level deviation, and ensures the accuracy of decision-making in core functional modules.

[0006] The technical solution proposed in this invention is: a multimodal liquid level monitoring method based on time-series feature analysis, the method comprising:

[0007] Multi-dimensional data is collected by sensors, and the collected multi-dimensional data is preprocessed to obtain the original pressure value, real-time water temperature value, real-time TDS value and current operating condition indicator after interference removal.

[0008] The original pressure value after interference removal and the basic parameters of water density are input into the hydrostatic pressure formula to obtain the initial liquid level value. Combined with the real-time water temperature value, the liquid level value after water temperature compensation is obtained. Combined with the real-time liquid level fluctuation frequency and the reference fluctuation frequency, the liquid level value after bubble compensation is obtained. Drift compensation is performed according to the periodic zero drift amount and drift conversion coefficient to obtain the final high-precision liquid level value.

[0009] Based on the final high-precision liquid level value, real-time TDS value, inlet water pressure data, and current operating condition indicator, the water production flow rate is calculated by liquid level changes to obtain the water production flow rate for a given period. The daily TDS fluctuation value is obtained based on the extreme fluctuation of TDS. The flow stability is analyzed under a fixed pressure to obtain the flow fluctuation value. Based on the water production flow rate for a given period, the daily TDS fluctuation value, and the flow fluctuation value, the flow attenuation score, TDS stability score, and correlation score are calculated according to weight and threshold rules. The overall health score is obtained by combining the three scores.

[0010] By combining multiple consecutive high-precision liquid level values, current operating condition indicators, and a normal operating condition feature library, the liquid level change rate is calculated and compared with the standard range. The abnormality type is determined and output. Based on the abnormality type, control commands are generated.

[0011] Preferably, the specific process for preprocessing the collected multivariate data is as follows:

[0012] Data from three types of sensors and operating condition signals are collected synchronously at fixed short intervals. The average value of each type of data is calculated using a sliding window containing a fixed number of continuous sampled values. If the difference between the average value and the previous result exceeds the upper limit of the normal fluctuation of the corresponding data, it is judged as interference and the current result is discarded. After preprocessing, the data retains the decimal places according to the corresponding format, and the operating condition signals maintain the enumerated text identifier. The data are then integrated into a valid data set for output.

[0013] Preferably, the process for obtaining the final high-precision liquid level value is as follows:

[0014] The pre-processed, interference-free original pressure value is combined with the basic parameters of water density and the standard value of gravitational acceleration to obtain the initial liquid level value through the principle of hydrostatic pressure. Based on the difference between the real-time water temperature value and the standard temperature, a compensation coefficient is calculated using the temperature influence coefficient to correct the initial liquid level value, resulting in a water temperature-compensated liquid level value. The real-time liquid level fluctuation frequency is calculated based on the water temperature-compensated liquid level value, and a bubble compensation coefficient is calculated by combining the reference fluctuation frequency and the bubble influence coefficient to obtain a bubble-compensated liquid level value. Based on the periodic zero-point drift amount statistically obtained under empty tank conditions, the liquid level offset is calculated using the drift conversion coefficient. The final high-precision liquid level value is obtained by subtracting this offset from the bubble-compensated liquid level value.

[0015] Preferably, the process for obtaining the overall health score is as follows:

[0016] Based on the final high-precision liquid level value, real-time TDS value, inlet water pressure data, and operating condition indicators, three types of feature data are extracted: time-period water production flow rate, daily TDS fluctuation value, and flow rate fluctuation value. According to the time-period water production flow rate, daily TDS fluctuation value, and flow rate fluctuation value over a recent period, flow rate attenuation score, TDS stability score, and correlation score are calculated according to preset weight and threshold rules. The three scores are added together to obtain the total health of the RO membrane.

[0017] Preferably, the specific process for obtaining the exception type is as follows:

[0018] Acquire multiple consecutive high-precision liquid level values, current operating condition identifier, and a normal operating condition feature library containing the normal range of liquid level change rates for each operating condition; calculate the current liquid level change rate based on the continuous liquid level data; compare the calculated rate with the normal rate range of the corresponding operating condition in the feature library; if it exceeds the range, determine and output the specific abnormality type; if it is within the normal range, output no abnormality.

[0019] Preferably, the specific content of the control command is as follows:

[0020] For different abnormal scenarios, corresponding instructions are output. If it is sensor interference, an instruction to adjust the sampling interval to a low interference interval is generated. If it is valve response lag, a valve opening and closing instruction to send the average historical valve response time is generated in advance. If it is overflow risk, an instruction to adjust the alarm threshold to the highest water level preset ratio and switch the sampling interval to a high-frequency monitoring interval is generated. At the same time, when the liquid level reaches the new threshold, an emergency valve closing instruction will also be generated.

[0021] The present invention also provides a multimodal liquid level monitoring method based on time-series feature analysis, wherein the system is used to execute the aforementioned multimodal liquid level monitoring method based on time-series feature analysis.

[0022] The present invention also provides a computer-readable storage medium storing a computer program, which is executed by a processor to implement the multimodal liquid level monitoring method based on time-series feature analysis.

[0023] The beneficial effects of this invention are:

[0024] 1. A four-level progressive compensation mechanism—initial level conversion, water temperature compensation, bubble compensation, and long-term drift compensation—is implemented to address various interference factors such as water temperature changes, liquid bubbles, and sensor aging. Compared to the traditional method of converting liquid level to pressure, this mechanism effectively offsets the effects of water temperature on water density, the interference of bubbles on pressure signals, and the drift error of sensors over long-term use, resulting in a final high-precision liquid level value that more closely reflects the actual liquid level. This mechanism provides a precise data foundation for subsequent core processes such as water flow rate calculation and RO membrane health assessment, avoiding misjudgments of health status and misidentification of anomalies caused by liquid level measurement deviations, thus ensuring the accuracy of equipment operation status assessment. The pressure-to-liquid-level conversion method effectively offsets the effects of water temperature on water density, the interference of bubbles on pressure signals, and the drift error of sensors over long-term use, resulting in a final high-precision liquid level value that more closely reflects the actual liquid level. This mechanism provides a precise data foundation for subsequent core processes such as water flow rate calculation and RO membrane health assessment, avoiding misjudgments of health status and misidentification of anomalies caused by liquid level measurement deviations, thus ensuring the accuracy of equipment operation status assessment.

[0025] 2. First, extract three core characteristic data points: water production flow rate, TDS fluctuation, and flow rate fluctuation. Then, according to preset weights, break down the health status into three sub-scores: flow rate decline, TDS stability, and pressure-flow rate correlation. Finally, summarize the overall health status and determine the fault type based on data characteristics. This hierarchical evaluation mode not only quantifies the RO membrane health status (0-100 points), allowing users to intuitively understand the degree of membrane wear, but also identifies membrane blockage, membrane damage, and other faults in advance through correlation analysis between a sudden drop in health status and key data characteristics. Compared to the traditional method of relying solely on experience to determine replacement, it can accurately predict the remaining lifespan, guide users to maintain as needed, and reduce the cost waste of premature replacement or the risk of substandard water quality due to delayed replacement. The three sub-scores (pressure-flow rate correlation, etc.) are summarized to obtain the overall health status, and the fault type is determined based on data characteristics. This hierarchical evaluation mode not only quantifies the RO membrane health status (0-100 points), allowing users to intuitively understand the degree of membrane wear, but also identifies membrane blockage, membrane damage, and other faults in advance through correlation analysis between a sudden drop in health status and key data characteristics.

[0026] 3. First, by comparing operating conditions and liquid level characteristics, and combining continuous liquid level data with the matching of operating conditions, the anomaly type is determined. Then, corresponding control commands are automatically generated for different anomalies (such as initiating filtering and adjusting the sampling interval when there is sensor interference, and triggering high-frequency monitoring and emergency valve closure when there is a risk of overflow). This "identification-self-healing" linkage mode can quickly alleviate most common anomalies without manual intervention. For example, by sending valve commands in advance to compensate for response lag and adjusting the sampling interval to reduce the impact of interference, it effectively reduces the downtime of equipment due to anomalies, reduces the frequency and cost of manual inspection and maintenance, and avoids more serious problems such as equipment damage and overflow leakage caused by the failure to handle anomalies in a timely manner, thereby improving the stability and reliability of equipment operation. Attached Figure Description

[0027] Figure 1 This is a flowchart of a multimodal liquid level monitoring method based on time-series feature analysis according to the present invention;

[0028] Figure 2 This is a flowchart illustrating the monitoring process of a multimodal liquid level monitoring method based on time-series feature analysis according to the present invention. Detailed Implementation

[0029] The following description is intended to disclose the present invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art. The basic principles of the invention defined in the following description can be applied to other embodiments, modifications, improvements, equivalents, and other technical solutions that do not depart from the spirit and scope of the invention.

[0030] It is understood that the term "a" should be understood as "at least one" or "one or more," that is, in one embodiment, the number of an element can be one, while in another embodiment, the number of the element can be multiple, and the term "a" should not be understood as a limitation on the number.

[0031] like Figure 1 and Figure 2 As shown, the pressure sensor in this scheme is a high-precision piezoresistive type, with a range of 0-100kPa and an accuracy of ±0.1%FS. It is installed at the bottom of the water storage container and connected to the main control system via a four-wire connection, powered by 5VDC, and can operate stably in an environment of -10℃ to 60℃. After the device starts up, it converts the liquid pressure into an electrical signal at a fixed sampling interval of 10 milliseconds. After A / D conversion, it outputs a digital signal in integer form, with kPa as the unit. The water temperature sensor is an NTC thermistor type, with a temperature measurement range of 0℃-100℃ and an accuracy of ±0.5℃. The probe is completely immersed in the liquid and is connected to the main control system via a two-wire connection. When the device is running and liquid is present, it amplifies, filters, and converts the temperature signal according to the resistance-temperature characteristics of the thermistor at 10-millisecond intervals, and then transmits it in floating-point form, with ℃ as the unit. The conductivity-type TDS sensor has a measurement range of 0-5000. With an accuracy of ±2%, the electrodes are immersed in the liquid and connected via a dedicated signal processing circuit. When the device is running and the liquid's conductivity is stable, conductivity data is collected every 10 milliseconds. This data is then converted to a TDS value using an internal algorithm (TDS value = conductivity × conversion factor, which is preset based on the liquid type; for example, it is typically 0.55-0.7 in water scenarios), and presented as a floating-point number. The output is in units; the equipment operating condition signal is represented by the level of the internal control circuit. 5V high level is for water replenishment, 0V low level is for water discharge, and 2.5V is for static state. The digital input port monitors the data in real time, and records the data of the enumeration type "water replenishment", "water discharge", and "static state" when the operating condition changes.

[0032] After data acquisition, a preprocessing stage is initiated to eliminate interference and ensure data validity. First, synchronous calibration is performed. The main control system simultaneously sends acquisition commands to three types of sensors and reads operating condition signals at 10-millisecond intervals to ensure that pressure, water temperature, and TDS data correspond perfectly with operating condition indicators in the time dimension, avoiding data misalignment due to acquisition time differences. Then, a mean filtering method is used to process various data types: For pressure data, a sliding window containing five consecutive sampled values ​​is constructed. After each new data acquisition, the oldest data in the window is removed and the new data is added. The filtering result is calculated using the formula: "Average pressure within the window = (Pressure value 1 + Pressure value 2 + Pressure value 3 + Pressure value 4 + Pressure value 5) ÷ 5". If the difference between this average value and the previous filtering result exceeds ±0.5 kPa (the upper limit of normal pressure fluctuation, set based on sensor accuracy and actual water scenario fluctuations), it is determined to be pulse interference, and the current average value is discarded. The average value is used and the previous valid result is retained. The water temperature data processing logic is the same, and the sliding window also contains 5 sampled values. The average water temperature is calculated as "(water temperature value 1 + water temperature value 2 + water temperature value 3 + water temperature value 4 + water temperature value 5) ÷ 5". If the difference from the previous result exceeds ±1℃ (the upper limit of normal water temperature fluctuation), a new set of data is collected and filtered again. When filtering TDS data, the 5 sampled values ​​in the window are calculated as "TDS average value = (TDS value 1 + TDS value 2 + TDS value 3 + TDS value 4 + TDS value 5) ÷ 5". If the difference exceeds ±50℃, the average value is retained. When the TDS fluctuation reaches the upper limit, discard the current result and retain the previous valid data. After preprocessing, pressure data should be kept in integer form (retaining one decimal place, e.g., 30.2 kPa) in kPa, while water temperature and TDS data should be kept in floating-point form (retaining two decimal places, e.g., 20.14℃, 100.30℃). The units are ℃ and , respectively. The operating condition signals maintain the enumeration of text identifiers, which are then integrated into a complete data set and output to the subsequent algorithm module.

[0033] Taking a water purification device in actual operation as an example, in the initial stage, the device is in a static state. The pressure sensor continuously collects five sets of data: 30.2 kPa, 30.1 kPa, 30.3 kPa, 30.2 kPa, and 30.1 kPa. Using the average value formula, the calculated value is (30.2 + 30.1 + 30.3 + 30.2 + 30.1) ÷ 5 = 30.18 kPa. This is compared to the previous interference-free reference value (30.2 kPa calibrated during initial startup). The difference between the readings (0 kPa and 0 kPa) and the readings is 0.02 kPa, which is within ±0.5 kPa and is therefore considered valid data. The water temperature sensor collected data at 20.1℃, 20.2℃, 20.0℃, 20.3℃, and 20.1℃. The calculated value (20.1 + 20.2 + 20.0 + 20.3 + 20.1) ÷ 5 = 20.14℃, which is 0.01℃ different from the previous reading of 20.15℃ and is within ±1℃, therefore valid. The TDS sensor collected 100.2 kPa. 100.5 100.3 100.4 100.1 The calculation yields (100.2 + 100.5 + 100.3 + 100.4 + 100.1) ÷ 5 = 100.30 Compared to the previous 100.28 Difference 0.02 Not exceeding ±50 The readings are valid, and the operating condition is marked as "static". When the equipment switches to water replenishment mode, in the next sampling cycle, the pressure sensor collects readings of 30.5 kPa, 30.6 kPa, 30.7 kPa, 30.8 kPa, and 30.6 kPa. The calculated value is (30.5 + 30.6 + 30.7 + 30.8 + 30.6) ÷ 5 = 30.64 kPa, which is 0.46 kPa less than the previous reading of 30.18 kPa, and is therefore valid. The water temperature readings are 20.2℃, 20.3℃, 20.4℃, 20.3℃, and 20.2℃, which is calculated as (20.2 + 20.3 + 20.4 + 20.3 + 20.2) ÷ 5 = 20.28℃, a difference of 0.14℃, and is therefore valid. The TDS reading is 100.6 kPa. 100.7 100.8 100.7 100.6 The calculation yields (100.6 + 100.7 + 100.8 + 100.7 + 100.6) ÷ 5 = 100.70 The difference is 0.40. It is effective, and the working condition label has been updated to "water replenishment", fully presenting the entire process from data acquisition, calculation and processing to result output.

[0034] After data preprocessing, the pressure signal needs to be accurately converted into liquid level height. This step, the basic liquid level conversion, is the core input source for subsequent multi-factor compensation. The input includes the original pressure value after interference removal (denoted as...). The unit is kilopascal (kPa). ), is the first after preprocessing Pressure data at each sampling time, for example, 30.18 30.64 (etc.), basic parameters of water density (the constant density of pure water at standard atmospheric pressure at a standard temperature such as 25℃). Determined by the physical properties of water, it is a key medium parameter for pressure-level conversion, as well as the standard value of gravitational acceleration (using the universal gravitational constant of the Earth's surface). (used to characterize the effect of gravity on liquid pressure).

[0035] The processing is based on the physical principle of hydrostatic pressure, the hydrostatic pressure of a liquid. With liquid level height Liquid density Gravitational acceleration Satisfying Relationships Therefore, the formula for calculating the liquid level by inferring the liquid level from the pressure is: In the formula, Multiplying by 1000 is to change the pressure unit from Convert to (because ), making the unit system similar to the SI unit of density and gravitational acceleration ( , The final liquid level unit is meters (m). ).

[0036] The output is the initial liquid level value (denoted as ). The unit is meters ( ), is the first The initial liquid level height corresponding to the pressure signal at each sampling time is retained to two decimal places to ensure accuracy, and serves as the core input for subsequent algorithms such as water temperature compensation and bubble compensation.

[0037] Taking actual data as an example, when the first The original pressure value after removing interference at each sampling time. Substituting this into the formula, we get: When the first The original pressure value after removing interference at each sampling time. Substituting this into the formula, we get: .

[0038] The initial liquid level value strictly follows the principles of fluid mechanics, ensuring the accuracy of the linear conversion between pressure and liquid level, and providing reliable basic data support for subsequent multi-factor dynamic compensation.

[0039] Water temperature compensation calculation is a crucial step in correcting the impact of temperature on liquid level measurement. Its inputs include real-time water temperature values ​​(denoted as...). The unit is ℃, which is the temperature after pretreatment. Water temperature data at each sampling time (e.g., 20.14℃, 20.28℃, etc.) and initial liquid level value (denoted as...). The unit is That is, the first output of the basic liquid level conversion process. The initial liquid level at each sampling time, such as 3.08. 3.13 (etc.), while introducing a standard temperature (set to 25℃, consistent with the temperature corresponding to the basic parameters of water density) and a temperature influence coefficient (a fixed value set based on the law of water density change with temperature, taken as 0.0002 / ℃, this coefficient was measured experimentally: for every 1℃ change in water temperature, the water density changes by about 0.02%, corresponding to the same liquid level compensation ratio).

[0040] During the process, it is necessary to first determine the relationship between the real-time water temperature and the standard temperature: when the real-time water temperature is higher than the standard temperature, the water density decreases as the temperature increases, and the actual liquid level under the same pressure is higher than the initial liquid level conversion result. Therefore, the "water temperature compensation coefficient" is applied. "Calculation: When the real-time water temperature is lower than the standard temperature, the water density increases as the temperature decreases. Under the same pressure, the actual liquid level is lower than the initial liquid level conversion result. Therefore, according to the 'water temperature compensation coefficient'..." "Calculate. After obtaining the compensation coefficient, multiply the initial liquid level value by this coefficient, which is the "liquid level value after water temperature compensation". This completes the temperature correction of the liquid level.

[0041] The output is the water level value after temperature compensation (denoted as ). The unit is (Retain two decimal places as input for subsequent bubble compensation steps).

[0042] Taking actual data as an example, when the first Real-time water temperature at each sampling time =20.14℃ (below 25℃), initial liquid level First, calculate the water temperature compensation coefficient. Then obtain the liquid level value after water temperature compensation. When the first Real-time water temperature at each sampling time (Below 25℃), initial liquid level At that time, water temperature compensation coefficient Liquid level after water temperature compensation .

[0043] This compensation process quantifies the effect of temperature on water density, corrects the liquid level measurement error caused by temperature deviation from the standard value, and makes the liquid level data more consistent with the actual working conditions.

[0044] Bubble compensation calculation is used to correct the interference of air bubbles in the liquid on the liquid level measurement. Its input includes the liquid level value after water temperature compensation (denoted as...). The unit is That is, the first output of the water temperature compensation stage Liquid level data at each sampling time, such as 3.08 3.13 etc.), real-time liquid level fluctuation frequency (denoted as The unit is times / second, calculated from N consecutive liquid level data (N is set to 10 times, based on a 10ms sampling interval, which can cover a fluctuation period of 0.1 seconds), specifically the number of times the liquid level change direction reverses per unit time, and the reference fluctuation frequency (denoted as ). The unit is times / second. The stable fluctuation value obtained through calibration under bubble-free conditions is set to 0.2 times / second, reflecting the normal fluctuation level of natural liquid disturbance. At the same time, a bubble influence coefficient is introduced (a fixed value of 0.05 is set based on the interference law of bubbles on pressure signals. This coefficient is determined experimentally: when the real-time fluctuation frequency is twice the reference value, the liquid level measurement deviation caused by bubbles is about 1%).

[0045] During the processing, the real-time liquid level fluctuation frequency is first calculated by measuring the liquid level values ​​after 10 consecutive water temperature compensations: the sign (positive / negative) of the difference between two adjacent liquid level data points is recorded, and a count is made when the sign changes. Ten data points will generate nine differences, converting the count per unit time into a frequency (times / second). Then, the formula "bubble compensation coefficient" is used. The calculation involves using the ratio of the real-time fluctuation frequency to the reference frequency to reflect the relative intensity of bubble interference. A larger ratio indicates more bubbles and a smaller compensation coefficient. Finally, the liquid level value after water temperature compensation is multiplied by this coefficient to obtain the "liquid level value after bubble compensation". "This completes the correction of the liquid level by the bubbles."

[0046] The output is the liquid level value after bubble compensation (denoted as...). (The unit is meters, rounded to two decimal places, and used as input for subsequent long-term drift compensation).

[0047] Taking actual data as an example, when the first Liquid level value after temperature compensation at each sampling time Real-time liquid level fluctuation frequency When calculating the bubble compensation coefficient (calculated from 10 data points, with 0.03 sign inversions per millisecond out of 9 differences, converted to 0.3 times per second), first calculate the bubble compensation coefficient. Then obtain the liquid level value after bubble compensation. When the first Liquid level value after temperature compensation at each sampling time Real-time liquid level fluctuation frequency bubble compensation coefficient at times / second Liquid level after bubble compensation .

[0048] This compensation process corrects the pressure signal distortion caused by bubble interference by quantifying the intensity of liquid level fluctuations caused by bubbles, making the liquid level data closer to the true liquid height.

[0049] Long-term drift compensation calculation aims to eliminate measurement deviations caused by factors such as sensor aging and scale buildup, ensuring long-term stable and reliable liquid level data. Its input parameters include the liquid level value after bubble compensation (denoted as...). The unit is meters (m), which is the output of the bubble compensation stage. Liquid level data at each sampling time, such as 2.85m, 2.82m, etc., and periodic zero-point drift (denoted as...). The unit is kilopascal (kPa). Under empty tank conditions, the offset of the pressure sensor zero point is statistically analyzed at a preset period (set to 7 days, taking into account sensor aging and scale accumulation cycles), along with the drift conversion coefficient (denoted as...). The unit is meters per kilopascal (m / kPa), and the value is 0.102 m / kPa. This coefficient is determined based on the physical relationship between the pressure sensor installation position and the container height, that is, every 1 kPa pressure change corresponds to a 0.102 m liquid level height change.

[0050] The process begins with zero-point calibration of the pressure sensor according to a preset cycle. With the tank empty, the pressure sensor output value is recorded at the same time each day (e.g., 2 AM, to ensure system stability and no interference), forming a zero-point pressure data sequence over seven consecutive days. Calculate the zero-point drift during the calculation period. Next, the formula "long-term drift level offset" is used. "Calculate the liquid level offset caused by zero-pressure drift, and finally subtract this offset from the liquid level value after bubble compensation, which is the "final high-precision liquid level value". This eliminates measurement errors caused by long-term drift.

[0051] The output is the final high-precision liquid level value (denoted as ). The unit is meters (m), rounded to two decimal places, and used as the core input parameter for subsequent algorithms such as RO membrane health assessment and liquid level anomaly identification.

[0052] Taking actual data as an example, let's assume the first... Liquid level after bubble compensation at each sampling time =2.85m. In the past 7 days under empty tank conditions, the pressure sensor zero point drifted from an initial 2.1kPa to 2.3kPa. What is the periodic zero point drift amount? kPa, long-term drift level offset Final high-precision liquid level value If the first Liquid level after bubble compensation at each sampling time =2.82m, the zero-pressure drift is 0.3kPa, then the long-term drift level offset is... Ultimately, high-precision liquid level value .

[0053] This compensation process, through periodic calibration and quantitative calculation, effectively corrects the offset error of the sensor during long-term use, significantly improving the long-term stability and accuracy of liquid level measurement.

[0054] The feature data extraction stage aims to extract key features for RO membrane health assessment from raw data such as the final high-precision liquid level value and real-time TDS value. Its input data includes: the final high-precision liquid level value (denoted as...). The unit is meters (m), rounded to two decimal places, such as 2.83m or 2.79m, output from the long-term drift compensation calculation stage; real-time TDS value (denoted as...). The data includes: ppm (unit: ppm, rounded to one decimal place, e.g., 100.3ppm, 100.7ppm, obtained after preprocessing); inlet pressure data (denoted as pin[n], unit: kilopascals (kPa), collected and preprocessed by pressure sensor); and current operating status indicator (enumerated type, values: "water replenishment", "water discharge", "static", "water production", reflecting the real-time operating status of the equipment).

[0055] The specific processing procedure is as follows: When the operating condition indicator displays "Water Production," the system records liquid level data at three fixed time points daily: 8:00, 14:00, and 20:00. Assume the... The liquid level at each time point is... Change in liquid level at adjacent nodes The cross-sectional area of ​​the equipment is known. (Determined based on equipment specifications, such as 0.01m²), and the corresponding water production duration is obtained by combining the system operation logs. (Unit: seconds), obtained through the formula "time period water production flow rate" Calculate the flow rate, in cubic meters per second (m³ / s). (), rounded to four decimal places.

[0056] Meanwhile, the system monitors the daily TDS value in real time and records the maximum value. and minimum value The formula "Daily TDS fluctuation value" is used to determine the daily TDS fluctuation value. "Calculate the fluctuation range, and keep the result to one decimal place. The unit is ppm."

[0057] In addition, when the inlet water pressure stabilizes at the standard value (Set at 0.3 MPa, i.e., 300 kPa) Within a range of ±5 kPa, record the water production flow rate 5 times consecutively. First calculate the average value. Then, through the formula "flow fluctuation value" "Calculate the degree of flow fluctuation, in cubic meters per second (m³ / s)" (), rounded to four decimal places.

[0058] This step ultimately outputs three key characteristic data: water production flow rate over time period. This reflects the average flow rate during the water production period; daily TDS fluctuation value. This reflects the daily TDS change range; traffic fluctuation value. This characterizes the stability of the water production flow rate under a fixed inlet pressure. For example, the liquid level at 8:00 AM on a certain day... 14:00 liquid level equipment cross-sectional area Water production time The calculated water flow rate for the time period The maximum TDS value on that day minimum value Daily TDS fluctuation value Flow rate was recorded five times under a pressure of 300±5 kPa, and the flow rate fluctuation value was calculated. These data will serve as core inputs for subsequent RO membrane health assessments.

[0059] The health assessment component uses quantitative analysis of three dimensions—flow rate decline, TDS stability, and pressure-flow correlation—to provide a basic score for RO membrane health evaluation. Its inputs include: water production flow rate over a given period (preset to be continuous data for the past 30 days, denoted as pressure-flow correlation). The unit is ), daily TDS fluctuation value (denoted as (unit: ppm), flow fluctuation value (denoted as...) The unit is The system introduces a weighted distribution for each score (flow attenuation 40 points, TDS stability 30 points, stress-flow correlation 30 points, total 100 points) and corresponding threshold standards (preset normal attenuation threshold is 15%, meaning natural flow attenuation within 30 days does not exceed 15% of the initial value; preset TDS stability threshold is 5ppm, meaning daily TDS fluctuation does not exceed 5ppm; preset flow stability threshold is 30 points, total 100 points). That is, the flow rate fluctuation should not exceed this value under a fixed inlet pressure.

[0060] The specific processing involves three steps: When calculating the flow attenuation score, the initial flow is first determined. This is the absolute value of the water production flow rate on the first day within a 30-day period (the absolute value is used because a negative flow rate only indicates the direction of the liquid level drop), and the current flow rate. The absolute value of the water flow rate during the 30th day is calculated using the formula "flow rate decay rate". Calculate the attenuation ratio. If ≤15%, get 40 points; if For every 1% exceeding the limit, 1 point will be deducted (until all points are deducted); if 0 points.

[0061] When calculating the TDS stability score, if ≤5 30 points; if For every 1 Deduct 2 points; if 0 points.

[0062] When calculating the pressure-flow correlation score, if 30 points; if Each exceeding Deduct 3 points; if 0 points.

[0063] The output consists of three scores: flow attenuation score (denoted as...). (0-40 points), TDS stability score (recorded as...) (0-30 points), pressure-flow correlation score (recorded as...) (0-30 points), all rounded to the nearest integer.

[0064] Taking actual data as an example, if the initial traffic within 30 days Current traffic The attenuation rate was calculated. Less than 15%, flow attenuation score Points; if the daily TDS fluctuation value Exceeding the threshold Deduct 2.4 points (rounded to the nearest 2 points), TDS stability score Points; if the flow fluctuation value Exceeding the threshold Deduct 4.5 points (rounded to the nearest 5 points), pressure-flow correlation score These three scores will be used in the subsequent overall health assessment.

[0065] The overall health assessment and lifespan prediction process quantifies the overall condition of the RO membrane and estimates its remaining service life. Its inputs include a flow rate decay score (denoted as...). (0-40 points), TDS stability score (recorded as...) (0-30 points), pressure-flow correlation score (recorded as...) The system includes a health score range of 0-30 points, and a health-lifetime mapping rule (preset health scores of 90-100 points correspond to a remaining lifetime of 12-18 months, 70-89 points correspond to 8-12 months, 50-69 points correspond to 4-8 months, 30-49 points correspond to 1-4 months, and 0-29 points correspond to 0-1 month, with a maximum design life of 18 months for a full score of 100 points, and a critical warning lifetime for scores below 30 points). It also introduces a health score drop threshold (health score drops by more than 15 points in a single day or within 24 consecutive hours) and a combination of fault judgment features ("sudden drop in flow + sudden increase in pressure" refers to a flow rate drop of more than 30% and an inlet water pressure increase of more than 20% within 1 hour; "sudden increase in TDS + slow drop in flow" refers to a TDS value increase of more than 50% and a flow rate drop of 10%-20% within 1 hour).

[0066] The process consists of two steps: First, the total health score is calculated using the formula "Total Health Score". "A quantitative result of 0-100 points is obtained; then, according to the health-lifetime mapping rule, the corresponding remaining lifetime range is matched. Next, the fault type is determined. If the total health score does not drop sharply (the daily drop is ≤15 points), the fault type is marked as "no fault"; if the health score drops sharply and meets the characteristics of "sudden drop in flow + sudden increase in pressure", it is judged as "membrane blockage"; if the health score drops sharply and meets the characteristics of "sudden increase in TDS + slow decrease in flow", it is judged as "membrane rupture".

[0067] The output consists of three results: Overall RO Membrane Health (0-100 points, rounded to the nearest integer), remaining lifetime range (presented in the form of "XY months"), fault type identifier (enumerated type, values ​​"no fault", "membrane blockage", "membrane rupture").

[0068] Taking actual data as an example, if the traffic attenuation score =40 points, TDS stability score =28 points, pressure-flow correlation score =25 points, the total health score H is calculated to be 40 + 28 + 25 = 93 points. According to the mapping rule, the remaining lifespan is 12-18 months, and the health score does not drop suddenly. The fault type is marked as "no fault". If the scores of the three items at another moment are 30 points, 15 points, and 10 points respectively, the total health score is... =55 points, corresponding to a remaining lifespan of 4-8 months. If simultaneously there is a 35% decrease in flow rate and a 22% increase in inlet pressure within 1 hour, and the health score decreases by 20 points in a single day, it is judged as "membrane clogging"; if the overall health score is... A membrane is considered "damaged" if it scores 45 points (corresponding to 1-4 months), and if the TDS value increases by 55% and the flow rate decreases by 15% within 1 hour, and the health score decreases by 18 points within 24 hours. These results will provide a direct basis for decision-making regarding RO membrane maintenance and replacement.

[0069] The operating condition-liquid level characteristic comparison step is used to monitor the matching between liquid level changes and equipment operating conditions in real time, and to promptly detect anomalies. Its input data includes: the final high-precision liquid level value (denoted as...). The unit is meters (m), rounded to two decimal places, and is the continuous output of the long-term drift compensation calculation stage. Secondary sampling data, for example, selecting =10 times, i.e., liquid level data at 10 consecutive sampling times), current operating condition identifier (enumerated type, with values ​​of "water replenishment", "water discharge", "static", and "water production", reflecting the real-time operating status of the equipment), and normal operating condition feature library (containing the normal range of liquid level change rate under different operating conditions: the normal range of rise rate under water replenishment condition is 0.05-0.15m / min, the normal range of fall rate under water discharge condition is 0.08-0.2m / min, and the normal range of fluctuation amplitude under static condition is ±0.02m. These thresholds are determined through experiments and statistical analysis based on equipment specifications and historical operating data).

[0070] The process first calculates the current rate of change of the liquid level. This is done according to the formula "current rate of change of liquid level". ",in The preset sampling interval is set to 10 seconds, which is equivalent to 6010 minutes. For example, when... When the value is 10, the liquid level change rate in meters per minute (m / min) is obtained by dividing the difference between the liquid level at the current time and the liquid level 10 times before the sampling by the product of the number of samplings and the sampling interval.

[0071] Then the calculated rate of change of liquid level Compare the speed range with the corresponding operating condition in the normal operating condition feature library:

[0072] If the current operating condition is "water replenishment", then When the flow rate is m / min, it is determined that there may be an inlet valve malfunction (such as the valve not being fully open, resulting in excessive flow); when When the flow rate is m / min, it may be due to a blocked inlet valve or insufficient water supply pressure.

[0073] If the current operating condition is "water discharge", then At a flow rate of m / min, there may be an abnormality in the drain valve (such as valve malfunction causing excessively rapid drainage); when When the flow rate is m / min, it may be due to a blockage in the drainage pipe.

[0074] If the current operating condition is "stationary", when When the liquid level is m, it may be due to sensor interference generating incorrect data, or there may be leakage in the container. When the liquid level continues to rise and approaches 90% of the container height (overflow risk threshold, set based on container specifications), and the rate of change exceeds the normal range, it is judged as an overflow risk.

[0075] Based on the comparison results, output an anomaly type identifier (enumerated type, with values ​​of "inlet valve anomaly", "outlet valve anomaly", "sensor interference", and "overflow risk"). If the liquid level change rate is within the normal range of the corresponding operating condition, output "no anomaly".

[0076] Taking actual data as an example, assuming that under water replenishment conditions, the liquid level values ​​of 10 consecutive samples are 2.00m, 2.01m, 2.03m, 2.05m, 2.07m, 2.09m, 2.12m, 2.15m, 2.18m, and 2.21m, with a sampling interval of 10 seconds, the rate of change of liquid level is calculated using a formula. The value of m / min is greater than the upper limit of the normal rise rate under water replenishment conditions (0.15 m / min), therefore the anomaly type is identified as "inlet valve anomaly". These output results will serve as an important basis for the subsequent software self-healing execution phase, enabling the system to take timely measures to deal with abnormal situations.

[0077] The software self-healing execution phase aims to identify and address anomalies promptly by dynamically adjusting parameters and control strategies to achieve system self-repair. Its inputs include anomaly type identifiers (enumerated types, such as "sensor interference," "inlet valve malfunction," "overflow risk," etc.) and the final high-precision liquid level value (denoted as...). (Unit: m, rounded to two decimal places) Average historical valve response time (denoted as...) The unit is seconds. It calculates the average response time from the issuance of valve control commands to a significant change in liquid level in the last 50 times (e.g., 1.2 seconds). It also introduces preset parameters: sliding window length (using 20 sampled data for filtering), low interference sampling interval (adjusted to 50ms when there is sensor interference, which is more stable than conventional data), high frequency monitoring interval (set to 5ms when there is a risk of overflow, which improves the real-time monitoring), safety distance (liquid level less than 0.1m from the highest threshold is considered high risk), risk rate (the rate of rise of water exceeding 0.3m / min during water replenishment is considered a risk rate), and the preset ratio of the highest water level (the alarm threshold is adjusted to 90% of the highest water level, such as when the highest water level in the container is 3m, the alarm threshold is set to 2.7m).

[0078] The processing procedure takes targeted measures based on the type of anomaly: When the anomaly type is "sensor interference" (such as air bubbles causing fluctuations in liquid level data), the system activates sliding window filtering and takes the average of the last 20 high-precision liquid level values ​​as the current valid liquid level value (formula: self-healing liquid level value). Meanwhile, the sampling interval was adjusted from the default 10ms to 50ms to reduce the impact of high-frequency interference; when "valve response lag" is determined (e.g., after the inlet / outlet valve command is issued, the liquid level change is later than the historical average), the system advances the originally planned valve opening / closing time. Sending instructions, for example, if the inlet valve was originally scheduled to close at 10:00, the instruction is sent 1.2 seconds earlier at 09:59:58.8 to compensate for lag errors; when the "overflow risk" is triggered (liquid level is below the highest threshold)... 0.1m and rising rate (0.3 m / min), immediately lower the alarm threshold to 90% of the highest water level, and switch the sampling interval to 5 ms to monitor the liquid level change at high frequency. If the liquid level touches the new threshold, automatically trigger the emergency valve closing command.

[0079] The output results include: post-healing liquid level value ( The unit is m, and two decimal places are retained to replace the original data that has been disturbed. Valve advance control command (including command type "open valve / close valve" and advance time, such as "close valve command, advance 1.2 seconds"), and adjusted sampling interval (unit is ms, such as 50ms or 5ms).

[0080] Taking a real-world scenario as an example, if the liquid level values ​​for the last 20 times during sensor interference are 2.83m, 2.85m, 2.82m…2.84m, the calculated self-healing liquid level value is approximately 2.83m, and the sampling interval is adjusted to 50ms. If the historical average valve response time is 1.2 seconds, the original plan to open the drain valve at 14:00 is moved forward to 13:59:58.8. If the container's highest water level is 3m, the current liquid level is 2.92m (0.08m from the threshold), and the water replenishment rate is 0.35m / min, triggering an overflow risk, the alarm threshold is set to 2.7m, the sampling interval is changed to 5ms, and continuous monitoring continues until the liquid level drops to a safe range. These measures can quickly mitigate anomalies and ensure stable system operation.

[0081] The core results integration stage is a crucial step in transforming multi-dimensional data processing outcomes into usable information for the equipment. Its input data includes several key indicators: the final high-precision liquid level value (denoted as...). The unit is meters (m), rounded to two decimal places. If an anomaly triggers self-healing, the level value after self-healing will be used. Alternative); cumulative water flow rate over a specific time period (referred to as The unit is cubic meters (m3), which is obtained by summing the water flow rates at three different times of the day; the total health of the RO membrane (denoted as...) (0-100 points, rounded to the nearest integer); Remaining lifespan range (presented as "XY months" in text format); Fault type identifier (enumerated type, values ​​are "no fault", "membrane blockage", "membrane rupture"); Anomaly type identifier (enumerated type, values ​​are "inlet valve anomaly", "outlet valve anomaly", "sensor interference", "overflow risk", "no anomaly").

[0082] The processing begins by structuring all data according to the equipment communication protocol (using Modbus RTU protocol). Invalid data is then cleaned: if a self-healing operation exists, the original abnormal liquid level data is deleted, retaining only the self-healed liquid level value; if the fault type or abnormality type is "no abnormality," the corresponding identifier field is deleted to avoid redundant information.

[0083] In specific integration, the real-time liquid level information adopts the final valid liquid level value ( or ), stored in floating-point form at the address specified in the data frame as stipulated in the protocol; cumulative water production Retain three decimal places and store in the corresponding address; the RO membrane health status will be the overall health score. The remaining lifespan range is merged with the data into a text string (e.g., "Health: 85 points, Remaining Lifespan: 10-12 months") and stored. Safety alarm information is generated based on fault type and anomaly type identifiers. If both are "no anomaly," then "Equipment is operating normally" is generated. If an anomaly exists, information is integrated according to priority (e.g., "Membrane blockage, overflow risk"). Finally, a CRC checksum is calculated based on the data content to complete the data frame encapsulation. Safety alarm information is generated based on fault type and anomaly type identifiers. If both are "no anomaly," then "Equipment is operating normally" is generated. If an anomaly exists, information is integrated according to priority (e.g., "Membrane blockage, overflow risk").

[0084] The output is a final data set that the device can directly display or upload. For example, if the final high-precision liquid level value... =2.83m, cumulative water flow rate over a given period =0.568m3, RO membrane overall health =85 points, remaining lifespan range is "10-12 months", fault type is identified as "no fault", anomaly type is identified as "no anomaly", then the integrated output includes: real-time liquid level 2.83m; cumulative water production 0.568 The RO membrane health status is "Health score: 85 points, remaining lifespan: 10-12 months"; the safety alarm information is "Equipment is operating normally". This result can be displayed in real time on the device's screen or uploaded to the cloud management platform via the IoT module, providing users with an intuitive understanding of the equipment's operating status and a basis for maintenance decisions.

[0085] The processes described above with reference to the flowcharts in the embodiments disclosed in this invention can be implemented as computer software programs. The embodiments disclosed in this invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication component, and / or installed from a removable medium. When the computer program is executed by a central processing unit (CPU), it performs the functions defined in the methods of this application. It should be noted that the computer-readable medium described above in this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wire segments, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in connection with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on a computer-readable medium may be transmitted using any suitable medium, including but not limited to: wireless segments, wire segments, optical fibers, RF, etc., or any suitable combination thereof.

[0086] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0087] Those skilled in the art should understand that the embodiments of the present invention described above and shown in the accompanying drawings are merely examples and do not limit the present invention. The purpose of the present invention has been fully and effectively achieved, and the functions and structural principles of the present invention have been shown and explained in the embodiments. Without departing from the stated principles, the implementation of the present invention may have any variations or modifications.

Claims

1. A multimodal liquid level monitoring method based on time-series feature analysis, characterized in that, The method includes: Multi-dimensional data is collected by sensors, and the collected multi-dimensional data is preprocessed to obtain the original pressure value, real-time water temperature value, real-time TDS value and current operating condition indicator after interference removal. The original pressure value after interference removal and the basic parameters of water density are input into the hydrostatic pressure formula to obtain the initial liquid level value. Combined with the real-time water temperature value, the liquid level value after water temperature compensation is obtained. Combined with the real-time liquid level fluctuation frequency and the reference fluctuation frequency, the liquid level value after bubble compensation is obtained. Drift compensation is performed according to the periodic zero drift amount and drift conversion coefficient to obtain the final high-precision liquid level value. Based on the final high-precision liquid level value, real-time TDS value, inlet water pressure data, and current operating condition indicator, the water production flow rate is calculated by liquid level changes to obtain the water production flow rate for a given period. The daily TDS fluctuation value is obtained based on the extreme fluctuation of TDS. The flow stability is analyzed under a fixed pressure to obtain the flow fluctuation value. Based on the water production flow rate for a given period, the daily TDS fluctuation value, and the flow fluctuation value, the flow attenuation score, TDS stability score, and correlation score are calculated according to weight and threshold rules. The overall health score is obtained by combining the three scores. By combining multiple consecutive high-precision liquid level values, current operating condition indicators, and a normal operating condition feature library, the liquid level change rate is calculated and compared with the standard range. The abnormality type is determined and output. Based on the abnormality type, control commands are generated.

2. The multimodal liquid level monitoring method based on time-series feature analysis according to claim 1, characterized in that, The specific process for preprocessing the collected multivariate data is as follows: Data from three types of sensors and operating condition signals are collected synchronously at fixed short intervals. The average value of each type of data is calculated using a sliding window containing a fixed number of continuous sampled values. If the difference between the average value and the previous result exceeds the upper limit of the normal fluctuation of the corresponding data, it is judged as interference and the current result is discarded. After preprocessing, the data retains the decimal places according to the corresponding format, and the operating condition signals maintain the enumerated text identifier. The data are then integrated into a valid data set for output.

3. The multimodal liquid level monitoring method based on time-series feature analysis according to claim 2, characterized in that, The process for obtaining the final high-precision liquid level value is as follows: The original pressure value after pretreatment and removal of interference is combined with the basic parameters of water density and the standard value of gravitational acceleration to obtain the initial liquid level value through the principle of hydrostatic pressure. Based on the difference between the real-time water temperature value and the standard temperature, the compensation coefficient is calculated using the temperature influence coefficient to correct the initial liquid level value and obtain the liquid level value after water temperature compensation. The real-time liquid level fluctuation frequency is calculated based on the liquid level value after water temperature compensation. The bubble compensation coefficient is calculated by combining the reference fluctuation frequency and the bubble influence coefficient. The liquid level value after bubble compensation is obtained after correction. Based on the periodic zero-point drift amount statistically obtained under empty tank conditions, the liquid level offset is calculated using the drift conversion coefficient. After bubble compensation, the liquid level value is subtracted from the offset to obtain the final high-precision liquid level value.

4. The multimodal liquid level monitoring method based on time-series feature analysis according to claim 3, characterized in that, The specific process for obtaining the overall health score is as follows: Based on the final high-precision liquid level value, real-time TDS value, inlet water pressure data and operating condition label, three types of feature data are extracted: time period water production flow rate, daily TDS fluctuation value, and flow rate fluctuation value. Based on the water production flow rate, daily TDS fluctuation value, and flow fluctuation value over a recent period, the flow rate decay score, TDS stability score, and correlation score are calculated according to preset weights and threshold rules. The three scores are added together to obtain the overall health of the RO membrane.

5. The multimodal liquid level monitoring method based on time-series feature analysis according to claim 4, characterized in that, The specific process for obtaining the exception type is as follows: Acquire multiple consecutive high-precision liquid level values, current operating condition identifier, and a normal operating condition feature library containing the normal range of liquid level change rates for each operating condition; calculate the current liquid level change rate based on the continuous liquid level data; compare the calculated rate with the normal rate range of the corresponding operating condition in the feature library; if it exceeds the range, determine and output the specific abnormality type; if it is within the normal range, output no abnormality.

6. The multimodal liquid level monitoring method based on time-series feature analysis according to claim 5, characterized in that, The specific content of the control command is as follows: For different abnormal scenarios, corresponding instructions are output. If it is sensor interference, an instruction to adjust the sampling interval to a low interference interval is generated. If it is valve response lag, a valve opening and closing instruction to send the average historical valve response time is generated in advance. If it is overflow risk, an instruction to adjust the alarm threshold to the highest water level preset ratio and switch the sampling interval to a high-frequency monitoring interval is generated. At the same time, when the liquid level reaches the new threshold, an emergency valve closing instruction will also be generated.

7. A multimodal liquid level monitoring system based on time-series feature analysis, characterized in that, The system is used to execute the multimodal liquid level monitoring method based on time-series feature analysis as described in any one of claims 1-6.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that is executed by a processor to implement the multimodal liquid level monitoring method based on time-series feature analysis as described in any one of claims 1-6.

Images