Intelligent water dispenser with real-time water quality monitoring and filter life warning

Abstract

The application relates to the technical field of intelligent water purification monitoring, and particularly discloses an intelligent water dispenser with real-time water quality monitoring and filter core life warning, which comprises the following steps: collecting relevant data of instantaneous flow of water supplement, start-stop frequency and TDS removal rate of the intelligent water dispenser, building a two-threshold system of primary strong disturbance and secondary weak disturbance to identify water disturbance monitoring points, screening water quality abnormal monitoring points, determining water quality monitoring distortion risk through time sequence coincidence matching analysis, dividing distortion periods, calculating filter core life damage factors only for the distortion periods corresponding to the secondary weak disturbance, constructing a TDS removal rate prediction model, outputting theoretical TDS removal rate of the prediction period, correcting the theoretical TDS removal rate, calculating the remaining life of the filter core and realizing intelligent warning; the application can accurately distinguish between false water quality detection abnormalities caused by water flow disturbance and real performance degradation of the filter core, avoid invalid warning, and significantly improve the real-time water quality monitoring accuracy and the reliability of filter core life prediction and warning.

Description

Technical Field

[0001] This invention relates to the field of intelligent water purification monitoring technology, specifically to an intelligent water dispenser with real-time water quality monitoring and filter life warning. Background Technology

[0002] With the rapid development of the smart home water purification equipment industry, smart water dispensers have become widely used in homes, offices and public drinking water scenarios due to their advantages such as real-time water quality visualization, automated water supply and intelligent filter management. As existing smart water dispensers rely on basic technologies such as TDS detection and flow collection, they can achieve simple water quality monitoring and fixed-cycle filter reminders. The intelligent water purification system is gradually improving and has the technological space and market demand for continuous iteration and upgrading.

[0003] However, in the actual working conditions of smart water dispensers, where water is replenished and dispensed alternately, frequent water flow fluctuations and disturbances easily occur inside the water tank. At this point, existing conventional technologies have significant shortcomings: Firstly, traditional water quality monitoring mechanisms cannot distinguish between false anomalies in TDS detection caused by water disturbances and the actual performance degradation of the filter cartridge itself. Relying solely on a single TDS removal rate threshold to determine the filter cartridge's condition easily distorts the monitoring data caused by water replenishment and dispensing disturbances, misjudging it as a filter cartridge failure. This leads to frequent triggering of invalid filter cartridge warnings, resulting in wasted filter cartridges and masking the true aging problem of the filter cartridge. The problems are twofold: firstly, they create hidden dangers to drinking water quality safety; secondly, existing technologies have not established a hierarchical identification logic for water flow disturbances, making it impossible to distinguish between strong water body disturbances that only interfere with detection and weak water body disturbances that simultaneously cause filter element fatigue wear during start-up and shutdown. In the filter element life calculation process, all disturbance conditions are uniformly superimposed with loss corrections, which will not only incorrectly deduct the filter element life for strong disturbances that have no actual physical damage, but also fail to accurately quantify the filter element start-up and shutdown impact damage caused by weak disturbances. Ultimately, this leads to large errors in filter element life calculation and low accuracy in early warning, making it difficult to achieve high-precision real-time water quality monitoring and reliable dynamic early warning of filter element life.

[0004] Therefore, the present invention provides an intelligent water dispenser with real-time water quality monitoring and filter life warning. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent water dispenser with real-time water quality monitoring and filter life warning to solve the aforementioned background problems.

[0006] The objective of this invention can be achieved through the following technical solutions:

[0007] A smart water dispenser with real-time water quality monitoring and filter life warning features includes:

[0008] Disturbance risk assessment module: Collects instantaneous flow rate, frequency and historical TDS removal rate data of water dispenser replenishment, calculates the instantaneous flow rate difference of replenishment water, identifies water disturbance monitoring points through standard comparison, determines water quality anomaly monitoring points based on abnormal TDS removal rate data, and judges the risk of water quality monitoring result distortion through monitoring point time sequence overlap matching analysis.

[0009] Distortion period segmentation module: If there is a risk, extract the primary and secondary overlapping risk monitoring points, and use cluster analysis to segment the distortion period that is only data correction type, the distortion period that is related to filter element wear type, and the non-distortion period;

[0010] Damage Cause TDS Prediction Module: Extracts data related to filter cartridge wear-related distortion periods to calculate filter cartridge life damage factors, and simultaneously screens effective water quality monitoring points based on non-distortion periods, and outputs the theoretical TDS removal rate for the predicted period based on the TDS removal rate prediction model;

[0011] Lifetime warning and judgment module: The theoretical TDS removal rate is corrected according to the first-level and second-level disturbance differentiation correction rules to obtain the effective TDS removal rate. Then, the filter life is predicted based on the effective TDS removal rate, and the filter life warning trigger judgment is made.

[0012] Furthermore, the process of calculating the instantaneous flow difference of the replenished water is as follows:

[0013] Set a fixed sampling period, preferably 100ms;

[0014] The instantaneous flow rates of water replenishment and water discharge with a unified timestamp are collected synchronously at the sampling period. The absolute value of the difference between the instantaneous flow rates of water replenishment and water discharge at the same sampling time is calculated to obtain the difference between the instantaneous flow rates of water replenishment and water discharge at that sampling time.

[0015] Furthermore, the process of identifying water disturbance monitoring points through standard comparison and determining water quality anomaly monitoring points based on abnormal TDS removal rate data is as follows:

[0016] Set instantaneous flow difference standards and frequency standards for Level 2 weak disturbances and Level 1 strong disturbances;

[0017] Based on any acquisition time:

[0018] If the instantaneous flow difference between the replenishment and the replenishment frequency are both greater than or equal to the standard corresponding to the second-level weak disturbance, then the collection time will be marked as the second-level weak disturbance monitoring point.

[0019] If the instantaneous flow difference of the replenished water and the frequency of the replenished water are both greater than or equal to the standard corresponding to the first level of strong disturbance, then the collection time will be marked as the first level of strong disturbance monitoring point.

[0020] If any historical TDS removal rate is not within the reasonable range of TDS removal rates, the sampling point corresponding to the historical TDS removal rate will be marked as a water quality anomaly monitoring point.

[0021] Furthermore, the process of determining the risk of distortion in water quality monitoring results through time-series overlap matching analysis of monitoring points is as follows:

[0022] Set the overlap determination time window;

[0023] For all water body disturbance monitoring points and water quality anomaly monitoring points within the same unit time, timestamp matching is performed. If the timestamp of a water quality anomaly monitoring point falls within the overlap judgment time window of a certain water body disturbance monitoring point, the monitoring point is judged as an overlap risk monitoring point. Secondary weak disturbance matching generates secondary overlap risk monitoring points, and primary strong disturbance matching generates primary overlap risk monitoring points. The overlap degree is obtained by calculating the proportion of overlap risk monitoring points to the total number of water quality anomaly monitoring points within a unit time.

[0024] Based on any unit of time, if the overlap is greater than or equal to the overlap judgment standard, it is determined that there is a risk of distortion in the water quality monitoring results within that unit of time.

[0025] Furthermore, the process of dividing the data-correction-only distortion period, the filter wear-related distortion period, and the non-distortion period through cluster analysis is as follows:

[0026] The first-level overlapping risk monitoring points and the second-level overlapping risk monitoring points are arranged in chronological order according to their corresponding timestamps to obtain two independent time series sequences of overlapping risk monitoring points.

[0027] The DBSCAN density clustering algorithm was used to perform clustering on the time series of overlapping risk monitoring points to obtain first-level overlapping clusters and second-level overlapping clusters of overlapping risk monitoring points.

[0028] Based on any overlapping risk monitoring point cluster, the sampling time with the smallest and largest timestamps within the cluster is extracted as the start and end times of the distortion period, respectively, and correspondingly, a data-corrected distortion period or a filter wear-related distortion period is generated.

[0029] Merge two adjacent distortion time intervals of the same type;

[0030] The sampling time range outside the coverage of all distortion time intervals is divided into non-distortion time intervals.

[0031] Furthermore, the process of extracting data from periods of filter cartridge wear-related distortion to calculate filter cartridge lifespan damage factors, and simultaneously selecting effective water quality monitoring points based on non-distortion periods, is as follows:

[0032] Extract the instantaneous flow difference of the replenished water at all sampling times within the secondary correlation loss distortion period;

[0033] Based on the instantaneous flow difference of the replenished water at any sampling time, the ratio of the instantaneous flow difference of the replenished water to the standard instantaneous flow difference of the secondary weak disturbance is calculated to obtain the flow difference exceedance ratio at the sampling time.

[0034] Calculate the average of all exceeding ratios to obtain the filter life damage factor for excessive flow difference;

[0035] Extract the water replenishment frequency F within a unit of time corresponding to the distortion period, and the frequency standard for secondary weak disturbances. ;

[0036] Through formula The filter life damage factor due to excessive frequency was calculated. ;

[0037] in, The filter element start-stop impact loss coefficient;

[0038] Valid water quality monitoring points are obtained by extracting sampling times when no primary or secondary water disturbances are triggered and the corresponding TDS removal rates are within a reasonable range.

[0039] Furthermore, the process of outputting the theoretical TDS removal rate for the prediction period based on the TDS removal rate prediction model is as follows:

[0040] Construct a TDS removal rate prediction model:

[0041] The TDS removal rate and the corresponding sampling time in the historical TDS removal rate data during the non-distortion period are combined to obtain the model input group of the TDS removal rate prediction model.

[0042] The 1D-MobileNetV3 network architecture is adopted, which includes an input layer, an inverse residual structure layer, an attention mechanism module layer, a fully connected layer, and an output layer.

[0043] The error between the predicted and measured values ​​is calculated using the mean squared error loss function and backpropagation is completed. The AdamW algorithm is used to iteratively optimize the network parameters until the model training loss converges and the training ends.

[0044] The TDS removal rate time series of continuous undistorted time periods is extracted as input to the TDS removal rate time series deep prediction model, and the theoretical TDS removal rate of the prediction period is output.

[0045] Furthermore, the theoretical TDS removal rate is corrected according to the differential correction rules for first- and second-level perturbations to obtain the effective TDS removal rate. The process is as follows:

[0046] Based on the timestamp intervals of the first-level distortion-only period and the non-distortion period, the prediction period is divided into continuous non-distortion sub-periods and second-level associated loss distortion sub-periods according to the sampling time sequence;

[0047] The theoretical TDS removal rate corresponding to each sampling time within the non-distortion sub-period is extracted and effectively verified. The theoretical TDS removal rate that passes the effective verification can be directly used as the effective TDS removal rate of the non-distortion sub-period. The theoretical TDS removal rate that fails the effective verification is removed and filled using the two-point neighborhood mean interpolation method to obtain the effective TDS removal rate of the non-distortion sub-period.

[0048] The difference between the instantaneous flow rate difference of the replenished water at each collection point within the distortion sub-period and the standard difference between the secondary weak disturbance flow rate, and the difference between the replenished water frequency and the standard secondary weak disturbance frequency, are multiplied with the corresponding filter life damage factor to obtain the flow difference correction value and the frequency correction value. The difference between the theoretical TDS removal rate and the sum of the two correction values ​​is then calculated to obtain the effective TDS removal rate of the distortion sub-period.

[0049] The effective TDS removal rate is integrated over time to obtain the effective TDS removal rate for the predicted time period.

[0050] Furthermore, the effective verification process is as follows:

[0051] If the theoretical TDS removal rate is within a reasonable range and there is no reverse rebound, then the effective verification has passed.

[0052] If the theoretical TDS removal rate is not within the reasonable range of TDS removal rate, or if it rises in the opposite direction, it means that the effective verification has failed.

[0053] The term "reverse rebound" refers to the theoretical TDS removal rate at the next sampling time being higher than the theoretical TDS removal rate at the previous time.

[0054] Furthermore, the process of calculating and predicting filter lifespan based on the effective TDS removal rate, and then triggering a filter lifespan warning, is as follows:

[0055] The sampling moment when the effective TDS removal rate first fails to meet the reasonable range of the removal rate during the prediction period is extracted and marked as the filter failure sampling moment.

[0056] Using the baseline real-time moment when the TDS removal rate prediction model starts prediction as the sole starting point, the time period between the current sampling moment and the filter failure sampling moment is counted to obtain the predicted filter life.

[0057] When the predicted filter lifespan is less than 30 days, a filter lifespan warning is triggered.

[0058] The beneficial effects of this invention are:

[0059] (1) Distinguish between false water quality abnormalities caused by water disturbance and true performance degradation of filter cartridges, effectively avoid false warnings of ineffective filter cartridges. By collecting flow frequency and TDS water quality data, and combining time sequence overlap matching, identify monitoring distortion caused by water flow disturbance. After dividing the data distortion period, build a prediction model based on undisturbed effective water quality data, and perform correction processing on the interference data to remove false monitoring deviations.

[0060] (2) Realize the classification and identification of water disturbance, quantify the actual loss of filter element, and improve the accuracy of filter element life calculation and early warning. By setting two levels of water disturbance judgment rules, we can distinguish between strong disturbances that only interfere with the detection and weak disturbances that are accompanied by filter element start-up and shutdown losses. Then, by dividing different distortion periods, we can calculate the filter element damage factor only for weak disturbances that cause actual losses, and use a differentiated method to correct the TDS value. We can accurately count the hidden aging problem of filter element caused by frequent water flow impact, so that the filter element life calculation fits the actual use conditions. Attached Figure Description

[0061] The invention will now be further described with reference to the accompanying drawings.

[0062] Figure 1 This is a system block diagram of an intelligent water dispenser with real-time water quality monitoring and filter life warning.

[0063] Figure 2 This is a logic diagram for an intelligent water dispenser that features real-time water quality monitoring and filter life warning. Detailed Implementation

[0064] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0065] Please see Figure 1 - Figure 2 As shown, this invention is an intelligent water dispenser with real-time water quality monitoring and filter life warning, comprising the following steps:

[0066] Disturbance Risk Assessment Module: Acquires instantaneous flow rate and frequency data of water replenished by the smart water dispenser, as well as historical TDS removal rate data. Calculates the instantaneous flow rate difference of the replenished water based on the instantaneous flow rate, compares the instantaneous flow rate difference and the frequency of water replenishment with standards to identify water disturbance monitoring points. Simultaneously, identifies abnormal water quality monitoring points based on abnormal TDS removal rate data. Finally, through overlap matching analysis of water disturbance monitoring points and abnormal water quality monitoring points, it determines whether there is a risk of distorted water quality monitoring results.

[0067] In the disturbance risk assessment module, the instantaneous flow rate, frequency data, and historical TDS removal rate data of the makeup water are obtained as follows:

[0068] Hall effect flow sensors are installed at the inlet and outlet of the automatic water replenishment pipeline of the smart water dispenser to collect the instantaneous flow rate of water replenishment (corresponding to the inlet) and the instantaneous flow rate of water outlet (corresponding to the outlet) in real time. Based on the flow trigger signal (the Hall effect flow sensor outputs a pulse electrical signal indicating the water flow interruption state), the frequency of water replenishment and the frequency of water outlet are counted within a unit time (1 minute).

[0069] It should be noted that the instantaneous flow rate of the replenished water includes the instantaneous flow rate of the replenished water and the instantaneous flow rate of the discharged water. The frequency of replenishment and discharge refers to the independently counted frequency of replenishment and frequency of discharge, and they are not summed together.

[0070] TDS detection probes are installed at the inlet and outlet of the water dispenser filter cartridge. The TDS values ​​of raw water (corresponding to the inlet) and purified water (corresponding to the outlet) are collected simultaneously by the dual TDS probes. The TDS removal rate data is calculated according to the formula TDS removal rate = (raw water TDS value - purified water TDS value) / raw water TDS value × 100%. Historical TDS removal rate data is extracted through the data storage system of the smart water dispenser.

[0071] It should be noted that TDS removal rate is the core monitoring object for water quality inspection in smart water dispensers. It can reflect the filtration effect of the filter cartridge in the smart water dispenser on the raw water, and is also the core object for evaluating the life of the filter cartridge. When the TDS removal rate is lower than a certain value, it means that the filtration effect of the filter cartridge no longer meets the standard, that is, the life of the filter cartridge is 0.

[0072] In the disturbance risk assessment module, the process of calculating the instantaneous flow difference of the replenishment water is as follows:

[0073] Set a fixed sampling period, preferably 100ms;

[0074] The instantaneous flow rates of water replenishment and water discharge with a unified timestamp are collected synchronously at the sampling period. The absolute value of the difference between the instantaneous flow rates of water replenishment and water discharge at the same sampling time is calculated to obtain the difference between the instantaneous flow rates of water replenishment and water discharge at that sampling time.

[0075] In the disturbance risk assessment module, the process of identifying primary and secondary water body disturbance monitoring points by comparing the instantaneous flow difference of replenishment water and the frequency of replenishment water with standards is as follows:

[0076] The criteria for distinguishing between secondary weak disturbances and primary strong disturbances are set, and a two-level threshold classification and comparison are adopted to distinguish them. The monitoring points for secondary weak disturbances and primary strong disturbances are marked.

[0077] The criteria for identifying secondary weak disturbances are: the preferred instantaneous flow difference standard is 1.5 L / min and the frequency standard is 8 times / min; the criteria for identifying primary strong disturbances are: the preferred instantaneous flow difference standard is 2.5 L / min and the frequency standard is 12 times / min.

[0078] The process of comparing and judging the two-level threshold is as follows:

[0079] Secondary weak disturbance comparison: Sampling times with an instantaneous flow difference of ≥1.5L / min between replenishment and outflow and a replenishment frequency and outflow frequency of ≥8 times / min within the unit time are marked as secondary weak disturbance monitoring points. These points correspond to frequent small fluctuations in water flow, which not only interfere with TDS detection but also cause filter cartridge start-stop impact loss.

[0080] Level 1 Strong Disturbance Comparison: Sampling times with an instantaneous flow difference of ≥2.5L / min between replenishment and outflow and a replenishment frequency and outflow frequency of ≥12 times / min within a 1-minute unit time are marked as Level 1 Strong Disturbance Monitoring Points. These points correspond to violent turbulence of the water in the tank, which only causes TDS detection distortion and does not cause physical wear and tear on the filter element.

[0081] It should be noted that sampling times where the instantaneous flow difference and frequency do not reach any of the above-mentioned threshold levels will not be marked as any disturbance monitoring points. If a sampling time meets both disturbance threshold levels, it will be preferentially classified as a level one strong disturbance monitoring point.

[0082] It should be noted that: Level 2 weak disturbance is a low-frequency, small-amplitude continuous water flow oscillation, which is the core operating condition that causes start-stop fatigue wear of filter membrane components and pipeline joints. Level 1 strong disturbance is an instantaneous large flow of water surging, which only changes the uniformity of water mixing in the cavity and does not have continuous filter start-stop loading, so it does not cause substantial lifespan loss.

[0083] In the disturbance risk assessment module, abnormal data of TDS removal rate data refers to: based on the reasonable range of TDS removal rate, the historical TDS removal rate data is filtered for outliers, and TDS removal rate data that exceeds the reasonable range of TDS removal rate is marked as abnormal TDS removal rate data.

[0084] It should be noted that the reasonable range of TDS removal rate is set based on the desalination rate (TDS removal rate) requirements in QB / T 4144-2019 "Household and Similar Use Water Purifiers Only", and the reasonable range of TDS removal rate is set as 90%-99%.

[0085] In the disturbance risk assessment module, the process of identifying abnormal water quality monitoring points is as follows: extract the sampling time corresponding to the abnormal TDS removal rate and mark it as an abnormal water quality monitoring point;

[0086] In the disturbance risk assessment module, the process of determining whether there is a risk of distorted water quality monitoring results through overlap and matching analysis of water disturbance monitoring points and water quality anomaly monitoring points is as follows:

[0087] Set the overlap determination time window, preferably ±200ms;

[0088] For all water body disturbance monitoring points and water quality anomaly monitoring points within the same unit time, timestamp matching is performed. If the timestamp of a water quality anomaly monitoring point falls within the overlap judgment time window of a certain water body disturbance monitoring point, the monitoring points in this group are determined to be overlap risk monitoring points, and are split according to the disturbance level: Level II weak disturbance matching generates Level II overlap risk monitoring points, and Level I strong disturbance matching generates Level I overlap risk monitoring points. The overlap degree is obtained by the proportion of overlap risk monitoring points to the total number of water quality anomaly monitoring points within the statistical unit time.

[0089] It should be noted that only the secondary overlap risk monitoring points can be used as the basis for calculating the subsequent filter life damage factor.

[0090] Based on any unit of time, the overlap rate is compared with the overlap rate determination criteria:

[0091] When the overlap rate is greater than or equal to the overlap rate judgment standard (preferred value is 85%), it is determined that there is a risk of water quality monitoring results being distorted within that unit of time.

[0092] When the overlap is less than the overlap judgment standard, it is determined that there is no risk of water quality monitoring results being distorted within that unit of time, and the normal calculation and replacement of the remaining filter life is performed.

[0093] It should be noted that the reason for determining whether there is a risk of water quality monitoring results being distorted is that the timing of water flow disturbances and the timing of TDS water quality anomalies highly overlap, that is, the overlap is greater than or equal to the overlap judgment standard, which proves that the water quality deviation is false data generated by water body fluctuation interference detection, and thus it is determined that there is a risk of water quality monitoring distortion.

[0094] Distortion period segmentation module: If there is a risk of distortion in water quality monitoring results, based on the overlap matching analysis results of water disturbance monitoring points and water quality anomaly monitoring points, the data is split and extracted into two categories: primary overlap risk monitoring points and secondary overlap risk monitoring points. Through cluster analysis of the two types of overlap risk monitoring points, the data is divided into data correction-type distortion periods, filter cartridge wear-related distortion periods, and non-distortion periods.

[0095] In the distortion period segmentation module, the clustering analysis process for overlapping risk monitoring points is as follows:

[0096] The first-level overlapping risk monitoring points and the second-level overlapping risk monitoring points are arranged in chronological order according to their corresponding timestamps to obtain two independent time series sequences of overlapping risk monitoring points.

[0097] The time threshold P of the time-series neighborhood is preset (preferably 500ms), the minimum number of clustered samples is preset (preferably 3), and the DBSCAN density clustering algorithm is used to perform clustering:

[0098] Traverse each overlapping risk monitoring point in the time series of overlapping risk monitoring points, count the number of overlapping risk monitoring points in the neighborhood P before and after the timestamp of the point. If the number is greater than or equal to the minimum number of cluster samples, mark the point as the core point, and classify all overlapping risk monitoring points in its neighborhood into the same cluster, thus obtaining several groups of overlapping risk monitoring point clusters, including first-level overlapping clusters and second-level overlapping clusters.

[0099] In the distortion period segmentation module, the process of segmenting distortion periods into data-correction-only distortion periods, filter wear-related distortion periods, and non-distortion periods is as follows:

[0100] For any cluster of overlapping risk monitoring points obtained by cluster analysis, the sampling time with the smallest timestamp in the cluster is extracted as the start time of the distortion period, and the sampling time with the largest timestamp in the cluster is extracted as the end time of the distortion period. Correspondingly, a first-level distortion-only period and a second-level related loss distortion period are generated. Two adjacent distortion period intervals of the same type are merged, and the sampling time range outside the coverage of all distortion period intervals is divided into non-distortion periods.

[0101] It should be noted that if the time interval between adjacent distorted time periods is less than the preset time period merging threshold (preferably 1 second), the two adjacent distorted time periods will also be merged into one distorted time period.

[0102] Damage Cause TDS Prediction Module: Extracts data on instantaneous flow difference, frequency of replenishment, and TDS removal rate of water quality monitoring during the secondary correlation loss distortion period. Calculates the filter life damage factors corresponding to exceeding limits for instantaneous flow difference and frequency of replenishment. Simultaneously, based on the non-distortion period of water quality monitoring, it screens effective water quality monitoring points and extracts TDS removal rate data from these points. Through the TDS removal rate prediction model, it obtains the theoretical TDS removal rate for the prediction period.

[0103] In the TDS prediction module for damage causes, the process of calculating the filter cartridge life damage factors corresponding to excessive instantaneous flow difference of makeup water and excessive makeup water frequency is as follows:

[0104] Extract the instantaneous flow difference of the replenished water at all sampling times within the secondary correlation loss distortion period;

[0105] Based on the instantaneous flow difference of the replenished water at any sampling time, and taking the instantaneous flow difference standard of the secondary weak disturbance as the unified flow difference standard, the ratio of the instantaneous flow difference of the replenished water to the instantaneous flow difference standard is calculated to obtain the flow difference exceedance ratio at the sampling time (when the instantaneous flow difference of the replenished water is less than the instantaneous flow difference standard, the flow difference exceedance ratio is taken as 1).

[0106] The arithmetic mean of all the over-limit ratios during the distortion period is used to obtain the filter cartridge life damage factor for the excessive flow difference (which represents the amplification factor of the filter cartridge attenuation rate due to the portion of the instantaneous flow difference of the makeup water exceeding the standard).

[0107] Extract the water replenishment frequency F within a unit time period of the distorted period, using 8 times / min of secondary weak disturbance as the unified frequency standard. Calculate the ratio of the frequency of water replenishment to the frequency standard to obtain the frequency exceedance ratio (when the frequency of water replenishment is less than the frequency standard, the frequency exceedance ratio is taken as 1).

[0108] Through formula The filter life damage factor due to excessive frequency was calculated. (This indicates the amplification factor for the additional wear and tear on the filter element caused by the frequency of water replenishment exceeding the standard).

[0109] Among them, the filter element start-stop impact loss coefficient This information can be obtained from the material manual provided by the filter manufacturer.

[0110] In the TDS prediction module for damage causes, the process of selecting effective water quality monitoring points is as follows:

[0111] Valid water quality monitoring points are obtained by extracting sampling times when no primary or secondary water body disturbance is triggered and the corresponding TDS removal rate is within a reasonable range.

[0112] In the TDS prediction module for causes of damage, the process of constructing the TDS removal rate prediction model is as follows:

[0113] The TDS removal rate and the corresponding sampling time in the historical TDS removal rate data of the non-distortion period are combined to obtain the model input group of the TDS removal rate prediction model, and then integrated to obtain the model input group set.

[0114] The model input set is divided into time series: the first 70% of the time series data is the training set, the middle 20% is the validation set, and the last 10% is the test set.

[0115] The 1D-MobileNetV3 network architecture is adopted, which includes an input layer, an inverse residual structure layer, an attention mechanism module layer, a fully connected layer, and an output layer.

[0116] The time-series input group of the training set is fed into the input layer to complete the dimensional alignment and min-max normalization preprocessing of the time-series data, and outputs the dimensionally matched normalized time-series feature sequence. The sequence is then fed into the inverse residual structure layer to mine the nonlinear decay features of the TDS removal rate time-series data and output a deep time-series feature map. The deep time-series feature map is then fed into the attention mechanism module layer, where global max pooling and global average pooling are performed on the deep time-series feature map respectively. The channel weights are adaptively learned through the bottleneck fully connected layer, and the weighted optimized deep time-series features are output. The deep time-series features are then fed into the fully connected layer, where nonlinear activation is performed using the GELU activation function with a negative half-axis slope of 0.02. Finally, the linear activation layer of the output layer performs regression prediction and outputs the theoretical TDS removal rate time-series data for the corresponding sampling time within the prediction period.

[0117] It should be noted that the inverse residual structure layer consists of 4 sets of 1D inverse residual modules connected in series. Each set of modules is provided with an expanded convolutional layer, a depth-separable convolutional layer and a linear projection convolutional layer in sequence. The expansion factor is preferably 6, the kernel size is preferably 3, and the stride is preferably 1.

[0118] It should be noted that the attention mechanism module layer is a 1D SE channel attention layer, and the dimensions of the fully connected layer are preferably 128→64→32 in sequence.

[0119] The model is trained based on the training set. The error between the predicted and measured values ​​is calculated using the mean squared error loss function, and backpropagation is completed. The network parameters are iteratively optimized using the AdamW algorithm, where the weight decay coefficient is preferably 1e-4 and the initial learning rate is preferably 1e-3. During the training process, the generalization ability of the model is verified in real time using the validation set. Training is terminated when the model training loss converges, the validation set loss decreases by ≤0.001 for 5 consecutive rounds without any decrease, or the preset maximum number of training rounds (preferably 200 rounds) is reached, thus obtaining the TDS removal rate time series deep prediction model.

[0120] In the TDS prediction module for the cause of loss, the process of obtaining the theoretical TDS removal rate for the prediction period is as follows:

[0121] The TDS removal rate time series of continuous undistorted time periods is extracted as the input of the TDS removal rate time series deep prediction model, and the theoretical TDS removal rate of the prediction period is output.

[0122] Lifetime warning and judgment module: Based on the differentiated TDS correction rules of the first and second level disturbances, the theoretical TDS removal rate of the prediction period is corrected to obtain the effective TDS removal rate of the prediction period. Then, based on the effective TDS removal rate, the predicted filter life is calculated and the filter life warning trigger judgment is made.

[0123] In the lifespan early warning and judgment module, the process of correcting the theoretical TDS removal rate for the prediction period to obtain the effective TDS removal rate for the prediction period is as follows:

[0124] Based on the timestamp intervals of the first-level distortion-only period, the second-level associated loss-distortion period, and the non-distortion period, the prediction period is divided into continuous non-distortion sub-periods, first-level distortion-only period, and second-level associated loss-distortion sub-periods according to the sampling time sequence. The effective TDS removal rate of the three types of sub-periods is calculated respectively, and the time sequence is integrated to obtain the effective TDS removal rate of the prediction period.

[0125] The calculation process for the effective TDS removal rate in the non-distortion sub-period is as follows:

[0126] The theoretical TDS removal rate corresponding to each sampling time within the non-distortion sub-period is extracted and effectively verified. The theoretical TDS removal rate that passes the effective verification can be directly used as the effective TDS removal rate of the sampling time corresponding to the non-distortion sub-period. The theoretical TDS removal rate that fails the effective verification is removed, and the removal points are filled by two-point neighborhood mean interpolation method to finally obtain the effective TDS removal rate of the non-distortion sub-period.

[0127] The process of valid verification is as follows:

[0128] If the theoretical TDS removal rate is within a reasonable range and there is no reverse rebound (the value at the next sampling time must not be higher than the previous time), then the valid verification is passed.

[0129] If the theoretical TDS removal rate is not within the reasonable range of TDS removal rate, or if it rises in the opposite direction (the value at the next sampling time must not be higher than the value at the previous time), it means that the valid verification has failed (including both of these).

[0130] The two-point neighborhood mean interpolation method refers to: first, removing outlier data points that fail verification in the time series, and then filling the gap with the arithmetic mean of the two valid data points that pass verification immediately before and after the outlier point.

[0131] The calculation process for the effective TDS removal rate of the first-level correction of only the distortion sub-period is as follows: only the two-point neighborhood mean interpolation method of the non-distortion sub-period is used to complete the data smoothing correction, without superimposing any filter life damage factor or deducting the theoretical TDS removal rate value.

[0132] The calculation process for the effective TDS removal rate of the distorted sub-period is as follows:

[0133] The calculation is only performed for the secondary correlation loss distortion sub-period: the difference between the instantaneous flow rate difference of the replenished water at each collection point within the distortion sub-period and the standard difference between the secondary weak disturbance flow rate, and the difference between the replenished water frequency and the standard secondary weak disturbance frequency, are multiplied with the corresponding filter life damage factor to obtain the flow difference correction value and frequency correction value. The difference between the theoretical TDS removal rate and the sum of the two correction values ​​is then calculated to obtain the effective TDS removal rate of the distortion sub-period.

[0134] It should be noted that since the instantaneous flow difference and frequency of water replenishment at all sampling points during the secondary distortion sub-period exceed the standard, the difference between these two values ​​and the standard must be non-negative. The filter life damage factor must be a positive number greater than 1. Therefore, both the flow difference correction value and the frequency correction value are positive. The greater the damage to the filter element, the lower the corresponding TDS removal rate. Therefore, the difference between the theoretical TDS removal rate and the sum of the two correction values ​​is calculated. There is no physical loss of the filter element during the primary distortion sub-period, so this deduction calculation logic is not executed.

[0135] In the lifespan early warning and judgment module, the process of calculating and predicting the filter element's lifespan is as follows:

[0136] Based on the effective TDS removal rate during the prediction period, since the TDS removal rate theoretically tends to decrease, the sampling time when the effective TDS removal rate first fails to meet the reasonable range during the prediction period is extracted and marked as the filter failure sampling time. The baseline real-time time when the TDS removal rate prediction model starts prediction is used as the only timing starting point. The time period between the current sampling time and the filter failure sampling time is counted to obtain the predicted filter life.

[0137] In the lifespan warning assessment module, the process for triggering a filter lifespan warning is as follows:

[0138] Preferably, when the predicted filter life is less than 30 days, a filter life warning is triggered, and the filter flushing operation is performed.

[0139] If the predicted filter lifespan is greater than or equal to 30 days, the filter lifespan warning will not be triggered, and water quality monitoring will continue.

[0140] The working principle of this invention is as follows: By collecting real-time water flow rate, water flow frequency, and raw water / purified water TD data, the instantaneous flow difference of the replenished water is first calculated. Combined with two-level thresholds, monitoring points for disturbances in strong and weak water bodies are identified. Then, TDS anomaly points are screened. Time-series window matching is used to determine whether water quality detection data is distorted due to water flow disturbance. Subsequently, the DBSCAN clustering algorithm is used to divide the data overlapping with disturbances and water quality anomalies into three time periods: data correction only, associated filter cartridge wear, and normal monitoring. For the wear period, the filter cartridge damage factor caused by excessive water flow is quantified. Simultaneously, based on non-distorted effective data... By training a 1D-MobileNetV3 deep learning model, the theoretical TDS removal rate is accurately predicted. Finally, differentiated TDS data correction is performed based on different distortion periods, and filter wear correction is superimposed to obtain the effective TDS removal rate. The filter failure node is accurately determined and the remaining service life is calculated, triggering a low life warning. It can effectively identify false anomalies in TDS detection caused by dynamic water replenishment and water turbulence. It can not only eliminate interference to avoid false warnings of the filter, but also accurately quantify the hidden wear of the filter due to water flow start and stop, greatly improving the accuracy of water quality detection and the reliability of filter life assessment.

[0141] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the present invention should still fall within the scope of the present invention.

Claims

1. A smart water dispenser with real-time water quality monitoring and filter life warning, characterized in that: include: Disturbance risk assessment module: Collects instantaneous flow rate, frequency and historical TDS removal rate data of water dispenser replenishment, calculates the instantaneous flow rate difference of replenishment water, identifies water disturbance monitoring points through standard comparison, determines water quality anomaly monitoring points based on abnormal TDS removal rate data, and judges the risk of water quality monitoring result distortion through monitoring point time sequence overlap matching analysis. Distortion period segmentation module: If there is a risk, extract the primary and secondary overlapping risk monitoring points, and use cluster analysis to segment the distortion period that is only data correction type, the distortion period that is related to filter element wear type, and the non-distortion period; Damage Cause TDS Prediction Module: Extracts data related to filter cartridge wear-related distortion periods to calculate filter cartridge life damage factors, and simultaneously screens effective water quality monitoring points based on non-distortion periods, and outputs the theoretical TDS removal rate for the predicted period based on the TDS removal rate prediction model; Lifetime warning and judgment module: The theoretical TDS removal rate is corrected according to the first-level and second-level disturbance differentiation correction rules to obtain the effective TDS removal rate. Then, the filter life is predicted based on the effective TDS removal rate, and the filter life warning trigger judgment is made.

2. The intelligent water dispenser with real-time water quality monitoring and filter life warning as described in claim 1, characterized in that: The process of calculating the instantaneous flow difference of the replenished water is as follows: Set a fixed sampling period, preferably 100ms; The instantaneous flow rates of water replenishment and water discharge with a unified timestamp are collected synchronously at the sampling period. The absolute value of the difference between the instantaneous flow rates of water replenishment and water discharge at the same sampling time is calculated to obtain the difference between the instantaneous flow rates of water replenishment and water discharge at that sampling time.

3. The intelligent water dispenser with real-time water quality monitoring and filter life warning as described in claim 1, characterized in that: The process of identifying water disturbance monitoring points through standard comparison and determining water quality anomaly monitoring points based on abnormal TDS removal rate data is as follows: Set instantaneous flow difference standards and frequency standards for Level 2 weak disturbances and Level 1 strong disturbances; Based on any acquisition time: If the instantaneous flow difference between the replenishment and the replenishment frequency are both greater than or equal to the standard corresponding to the second-level weak disturbance, then the collection time will be marked as the second-level weak disturbance monitoring point. If the instantaneous flow difference of the replenished water and the frequency of the replenished water are both greater than or equal to the standard corresponding to the first level of strong disturbance, then the collection time will be marked as the first level of strong disturbance monitoring point. If any historical TDS removal rate is not within the reasonable range of TDS removal rates, the sampling point corresponding to the historical TDS removal rate will be marked as a water quality anomaly monitoring point.

4. The intelligent water dispenser with real-time water quality monitoring and filter life warning as described in claim 1, characterized in that: The process of determining the risk of distortion in water quality monitoring results through time-series overlap matching analysis of monitoring points is as follows: Set the overlap determination time window; For all water body disturbance monitoring points and water quality anomaly monitoring points within the same unit time, timestamp matching is performed. If the timestamp of a water quality anomaly monitoring point falls within the overlap judgment time window of a certain water body disturbance monitoring point, the monitoring point is judged as an overlap risk monitoring point. Secondary weak disturbance matching generates secondary overlap risk monitoring points, and primary strong disturbance matching generates primary overlap risk monitoring points. The overlap degree is obtained by calculating the proportion of overlap risk monitoring points to the total number of water quality anomaly monitoring points within a unit time. Based on any unit of time, if the overlap is greater than or equal to the overlap judgment standard, it is determined that there is a risk of distortion in the water quality monitoring results within that unit of time.

5. The intelligent water dispenser with real-time water quality monitoring and filter life warning as described in claim 1, characterized in that: The process of dividing the data-correction-only distortion period, the filter wear-related distortion period, and the non-distortion period through cluster analysis is as follows: The first-level overlapping risk monitoring points and the second-level overlapping risk monitoring points are arranged in chronological order according to their corresponding timestamps to obtain two independent time series sequences of overlapping risk monitoring points. The DBSCAN density clustering algorithm was used to perform clustering on the time series of overlapping risk monitoring points to obtain first-level overlapping clusters and second-level overlapping clusters of overlapping risk monitoring points. Based on any overlapping risk monitoring point cluster, the sampling time with the smallest and largest timestamps within the cluster is extracted as the start and end times of the distortion period, respectively, and correspondingly, a data-corrected distortion period or a filter wear-related distortion period is generated. Merge two adjacent distortion time intervals of the same type; The sampling time range outside the coverage of all distortion time intervals is divided into non-distortion time intervals.

6. The intelligent water dispenser with real-time water quality monitoring and filter life warning as described in claim 3, characterized in that: The process of extracting data from periods of filter cartridge wear-related distortion to calculate filter cartridge lifespan damage factors, and simultaneously selecting effective water quality monitoring points based on non-distortion periods, is as follows: Extract the instantaneous flow difference of the replenished water at all sampling times within the secondary correlation loss distortion period; Based on the instantaneous flow difference of the replenished water at any sampling time, the ratio of the instantaneous flow difference of the replenished water to the standard instantaneous flow difference of the secondary weak disturbance is calculated to obtain the flow difference exceedance ratio at the sampling time. Calculate the average of all exceeding ratios to obtain the filter life damage factor for excessive flow difference; Extract the water replenishment frequency F within a unit of time of the distorted period, and the frequency standard for secondary weak disturbances. ; Through formula The filter life damage factor due to excessive frequency was calculated. ; in, The filter element start-stop impact loss coefficient; Valid water quality monitoring points are obtained by extracting sampling times when no primary or secondary water disturbances are triggered and the corresponding TDS removal rates are within a reasonable range.

7. The intelligent water dispenser with real-time water quality monitoring and filter life warning as described in claim 1, characterized in that: The process of outputting the theoretical TDS removal rate for the prediction period based on the TDS removal rate prediction model is as follows: Construct a TDS removal rate prediction model: The TDS removal rate and the corresponding sampling time in the historical TDS removal rate data during the non-distortion period are combined to obtain the model input group of the TDS removal rate prediction model. The 1D-MobileNetV3 network architecture is adopted, which includes an input layer, an inverse residual structure layer, an attention mechanism module layer, a fully connected layer, and an output layer. The error between the predicted and measured values ​​is calculated using the mean squared error loss function and backpropagation is completed. The AdamW algorithm is used to iteratively optimize the network parameters until the model training loss converges and the training ends. The TDS removal rate time series of continuous undistorted time periods is extracted as input to the TDS removal rate time series deep prediction model, and the theoretical TDS removal rate of the prediction period is output.

8. The intelligent water dispenser with real-time water quality monitoring and filter life warning as described in claim 1, characterized in that: The process of correcting the theoretical TDS removal rate according to the first- and second-level perturbation differentiation correction rules to obtain the effective TDS removal rate is as follows: Based on the timestamp intervals of the first-level distortion-only period and the non-distortion period, the prediction period is divided into continuous non-distortion sub-periods and second-level associated loss distortion sub-periods according to the sampling time sequence; The theoretical TDS removal rate corresponding to each sampling time within the non-distortion sub-period is extracted and effectively verified. The theoretical TDS removal rate that passes the effective verification can be directly used as the effective TDS removal rate of the non-distortion sub-period. The theoretical TDS removal rate that fails the effective verification is removed and filled using the two-point neighborhood mean interpolation method to obtain the effective TDS removal rate of the non-distortion sub-period. The difference between the instantaneous flow rate difference of the replenished water at each collection point within the distortion sub-period and the standard difference between the secondary weak disturbance flow rate, and the difference between the replenished water frequency and the standard secondary weak disturbance frequency, are multiplied with the corresponding filter life damage factor to obtain the flow difference correction value and the frequency correction value. The difference between the theoretical TDS removal rate and the sum of the two correction values ​​is then calculated to obtain the effective TDS removal rate of the distortion sub-period. The effective TDS removal rate is integrated over time to obtain the effective TDS removal rate for the predicted time period.

9. A smart water dispenser with real-time water quality monitoring and filter life warning as described in claim 8, characterized in that: The effective verification process is as follows: If the theoretical TDS removal rate is within a reasonable range and there is no reverse rebound, then the effective verification has passed. If the theoretical TDS removal rate is not within the reasonable range of TDS removal rate, or if it rises in the opposite direction, it means that the effective verification has failed. The term "reverse rebound" refers to the theoretical TDS removal rate at the next sampling time being higher than the theoretical TDS removal rate at the previous time.

10. A smart water dispenser with real-time water quality monitoring and filter life warning as described in claim 1, characterized in that: The process of calculating and predicting filter cartridge lifespan based on effective TDS removal rate, and triggering filter cartridge lifespan warning decisions, is as follows: The sampling moment when the effective TDS removal rate first fails to meet the reasonable range of the removal rate during the prediction period is extracted and marked as the filter failure sampling moment. Using the baseline real-time moment when the TDS removal rate prediction model starts prediction as the sole starting point, the time period between the current sampling moment and the filter failure sampling moment is counted to obtain the predicted filter life. When the predicted filter lifespan is less than 30 days, a filter lifespan warning is triggered.

Images