CN-122022034-A - Intelligent water quality early warning method based on multi-sensor data fusion

CN122022034ACN 122022034 ACN122022034 ACN 122022034ACN-122022034-A

Abstract

The invention relates to the technical field of water quality early warning, in particular to an intelligent water quality early warning method based on multi-sensor data fusion. The method includes the steps of acquiring and partitioning multi-sensor water quality monitoring data, and then performing linear coupling and residual non-linear processing on environmental factors in a historical sequence to extract transient interaction features. Generating prediction results of month, week, day and residual items through multi-scale decomposition and independent linear prediction channels, and fusing to obtain a future water quality prediction sequence. And then forming a prediction track by using repeated forward reasoning, calculating a prediction mean value and a confidence interval, finally judging by combining the mean value, the trend, the upper limit of the confidence interval and a multilevel threshold value, and outputting water quality early warning information of different levels. According to the intelligent water quality early warning method, the intelligent water quality early warning effects with higher precision, higher real-time performance, higher interpretability and stability are realized through the instantaneous coupling extraction of the environmental factors, the multi-scale trend and period prediction and the multi-level threshold probability early warning.

Inventors

HONG WENQING
JIN RUI
KONG JUNWEN
HUANG QUANPING
WANG NING
WU JUNXIANG
ZHANG YONG
LIU YINGYING

Assignees

北京三维天地科技股份有限公司
广东省水文局

Dates

Publication Date: 20260512
Application Date: 20260126

Claims (10)

1. An intelligent water quality early warning method based on multi-sensor data fusion is characterized by comprising the following steps: Step S1, acquiring multi-sensor monitoring data of a target water body, dividing the multi-sensor monitoring data into sample data containing a history sequence and a future target sequence; S2, carrying out linear coupling and residual nonlinear processing on environmental factors of sample data in each time step in a historical sequence, and extracting instantaneous interaction characteristics among the environmental factors to obtain a characteristic sequence; s3, decomposing the month scale, the week scale, the day scale and the residual items of the characteristic sequence, generating prediction results of corresponding time scales through independent linear prediction channels, and fusing the prediction results of all scales to obtain a prediction sequence of the future water quality factors; S4, performing forward reasoning on sample data of a future target sequence according to the predicted sequence, obtaining a plurality of predicted tracks, and obtaining a predicted mean value and a confidence interval based on statistics of the predicted tracks; and S5, judging based on the prediction mean value, the change trend of the prediction track, the upper limit of the confidence interval and a preset multilevel threshold value, and outputting water quality early warning information of corresponding grades.
2. The intelligent water quality early warning method based on multi-sensor data fusion according to claim 1, wherein the step S2 comprises the following steps: s21, extracting environmental factors of sample data in each time step in a history sequence; s22, taking the environmental factors as input vectors to perform dimension-lifting processing to obtain environment vectors after dimension lifting; S23, performing linear coupling and residual nonlinear processing on the environment vectors after the dimension rise in each time step, and extracting instantaneous interaction characteristics among environment factors to obtain updated characteristic vectors; and S24, vector combination is carried out on the updated feature vectors to obtain a feature sequence.
3. The intelligent water quality early warning method based on multi-sensor data fusion according to claim 2, wherein step S23 comprises: performing one-dimensional fast Fourier transform on the environment vector after the dimension rise in each time step to obtain an environment vector frequency domain; screening a low-frequency environment vector frequency domain based on a preset frequency domain threshold, and performing linear transformation on the corresponding environment vector according to the low-frequency environment vector frequency domain to obtain a linearly transformed environment vector frequency domain; Performing space domain conversion on the environment vector frequency domain after linear conversion through inverse Fourier transformation to obtain an updated environment vector; And obtaining an updated feature vector for representing the environment coupling characteristic of the t-th time step based on the nonlinear fusion of the updated environment vector and the environment vector residual error after the dimension rise.
4. The intelligent water quality early warning method based on multi-sensor data fusion according to claim 1, wherein the step S3 comprises the following steps: s31, decomposing the characteristic sequence according to a month scale, a week scale, a day scale and the rest items to obtain corresponding decomposition components; Step S32, performing linear prediction channel allocation through a multi-scale hierarchical DLinear network based on the corresponding decomposition components to obtain independent linear prediction channels; and step S33, carrying out hierarchical prediction on the corresponding decomposition components according to the independent linear prediction channels to obtain a hierarchical prediction result, and adding the hierarchical prediction result element by element to obtain a prediction sequence of the future water quality factor.
5. The intelligent water quality early warning method based on multi-sensor data fusion according to claim 1, wherein the step S4 comprises the following steps: step S41, repeatedly performing forward reasoning on sample data of a future target sequence according to the predicted sequence to obtain the future predicted sequence; step S42, summarizing future prediction sequences obtained by forward reasoning for multiple times according to time sequence to form a prediction track set with dimension of MxH x 1, wherein M is the reasoning times, and H is the time step length of future prediction; Step S43, carrying out time point-by-time statistics on the prediction track set along the forward reasoning dimension to obtain a prediction mean, a variance and a required confidence quantile of each time step in the future, wherein the quantiles comprise 5% quantiles and 95% quantiles; And S44, generating a confidence interval for representing future evolution uncertainty based on the mean, the variance and the confidence quantiles thereof.
6. The intelligent water quality early warning method based on multi-sensor data fusion according to claim 1, wherein the step S5 comprises the steps of: Step S51, extracting key dimension information from a probabilistic prediction result, wherein the key dimension information comprises a prediction mean value sequence, a mean value sequence change trend and a confidence interval upper limit; step S52, carrying out logic comparison on the predicted mean sequence, the change trend and the upper limit of the confidence interval with a preset multilevel concentration threshold value and a preset growth rate threshold value to obtain a logic judgment result, wherein the logic comparison comprises the following steps: When the predicted average value is more than or equal to 2.0mg/L and the average value sequence variation trend is more than or equal to 0.15 mg/L.h, judging that the emergency early warning is performed; when the prediction average value is less than 2.0mg/L and the upper limit of the confidence interval is more than or equal to 2.0mg/L, judging that the risk early warning is high; when the predicted mean value is 1.0-1.5 mg/L and the variation trend is less than 0.10 mg/L.h or the upper limit of the confidence interval is 1.5-2.0 mg/L, the warning is judged to be concerned; When the predicted average value is less than 1.0mg/L, the variation trend is less than 0.05 mg/(L.h), and the upper limit of the confidence interval is less than 1.5mg/L, judging that the safety state is the safety state; And step S53, generating water quality early warning information of corresponding grade based on the judging result.
7. The intelligent water quality early warning method based on multi-sensor data fusion according to claim 1, wherein obtaining multi-sensor monitoring data of a target water body further comprises: Arranging a plurality of groups of sensor acquisition nodes in a target water body, wherein each group of nodes comprises a dissolved oxygen electrode, a pH electrode, a thermometer, a nephelometer and a conductivity meter, and the nodes are fixed in the water body through buoys, pipes or underwater supports; The sensor is internally provided with a sampling pump or a water flow leading-in device for carrying out multichannel physical sampling, and the length and the flow speed of a sampling pipeline are adjusted to obtain vertical and horizontal multichannel measurement data; And adjusting the measurement time interval of each sensor based on the vertical and horizontal multichannel measurement data, and performing on-site physical calibration on the distributed sensors to acquire multi-sensor monitoring data.
8. The intelligent water quality early warning method based on multi-sensor data fusion according to claim 7, wherein performing on-site physical calibration on the laid out sensors comprises: Calibrating each sensor by using standard solutions, and sequentially introducing the standard solutions with different concentrations into a sensor sampling port to obtain corresponding output values so as to establish a sensor response curve; in the calibration process, the sampling flow rate and the circulating direction of the sensors are physically adjusted, and the adjacent sensors are calibrated in an interleaving sequence; performing on-site flushing treatment on a sensor sampling port, and acquiring a flushed sensor output result; And carrying out multi-node cross validation on the output result of the sensor to obtain multi-sensor monitoring data.
9. The intelligent water quality early warning method based on multi-sensor data fusion according to claim 8, wherein calibrating adjacent sensors in an interleaved order by physically adjusting the sampling flow rate and the circulation direction of the sensors comprises: acquiring a water inlet structure and a natural diversion path of the sensor; The method comprises the steps that based on a water inlet structure of a sensor, the sampling flow rate of the sensor is physically adjusted, and the circulation direction of the sensor is adjusted through a natural diversion path of the sensor, so that first adjustment data of the sensor are obtained; Stagger starting is carried out on the starting sequence of the adjacent sensors based on the first sensor adjustment data, so that second sensor adjustment data are obtained; And calibrating the sensor and the adjacent sensor according to the first sensor adjustment data and the second sensor adjustment data.
10. The intelligent water quality early warning method based on multi-sensor data fusion according to claim 9, wherein staggering the start sequence of adjacent sensors based on the first adjustment data of the sensors comprises: Starting the starting sequence of the adjacent sensors in an interleaving manner based on the first adjustment data of the sensors, so that the adjacent sensors do not enter a sampling state at the same time point at the same time, and confirming the first starting sensor and the second starting sensor; Confirming water flow trace sensing discrimination data of a second starting sensor according to the first starting sensor; Determining response sensitivity difference of the first starting sensor and the second starting sensor in a real water body flowing environment by analyzing response time of water flow trace perception discrimination data; Sensor second adjustment data capable of characterizing the activation characteristics of adjacent sensors is generated by responding to the sensitivity level differences.

Description

Intelligent water quality early warning method based on multi-sensor data fusion Technical Field The invention relates to the technical field of water quality early warning, in particular to an intelligent water quality early warning method based on multi-sensor data fusion. Background The water quality early warning is a key technology for monitoring the concentration and the variation trend of pollutants in a water body through a plurality of types of sensors and giving early warning when indexes are abnormal. The water quality state is influenced by various environmental factors such as water temperature, pH value, dissolved oxygen, chemical oxygen demand, total phosphorus, total nitrogen, ammonia nitrogen, permanganate index, sulfide, chlorophyll a and the like, the factors have nonlinear and dynamic coupling characteristics, and meanwhile, the water quality shows a change rule with trend and periodicity in the time dimension. However, the existing water quality early warning method has the following problems that firstly, the instantaneous interrelation and time sequence evolution among environmental factors are often mixed and modeled, the main stream method mostly adopts LSTM, transformer and other complex network structures, the model is huge, the calculation amount is high, and two types of dynamic information with different properties are difficult to process respectively and efficiently. And secondly, the factor-associated modeling based on the graph neural network regards environmental factors as discrete nodes, ignores the global smooth relation driven by continuous physical fields such as a temperature field, a concentration field and the like, and has insufficient generalization and expandability. And a complex coding-decoding structure is generally adopted in a time sequence evolution model, the real-time performance is poor, the training cost is high, the long-term change of water quality can be generally characterized by relatively simple trend items and period items in practice, and the inefficiency and the overfitting are easily caused by excessively complex modeling. Disclosure of Invention Based on this, it is necessary to provide an intelligent water quality early warning method based on multi-sensor data fusion, so as to solve at least one of the above technical problems. In order to achieve the purpose, the intelligent water quality early warning method based on multi-sensor data fusion comprises the following steps: Step S1, acquiring multi-sensor monitoring data of a target water body, dividing the multi-sensor monitoring data into sample data containing a history sequence and a future target sequence; S2, carrying out linear coupling and residual nonlinear processing on environmental factors of sample data in each time step in a historical sequence, and extracting instantaneous interaction characteristics among the environmental factors to obtain a characteristic sequence; s3, decomposing the month scale, the week scale, the day scale and the residual items of the characteristic sequence, generating prediction results of corresponding time scales through independent linear prediction channels, and fusing the prediction results of all scales to obtain a prediction sequence of the future water quality factors; S4, performing forward reasoning on sample data of a future target sequence according to the predicted sequence, obtaining a plurality of predicted tracks, and obtaining a predicted mean value and a confidence interval based on statistics of the predicted tracks; and S5, judging based on the prediction mean value, the change trend of the prediction track, the upper limit of the confidence interval and a preset multilevel threshold value, and outputting water quality early warning information of corresponding grades. The invention has the following beneficial effects: 1. Through the innovative design of multi-sensor data fusion and environmental factor instantaneous interaction modeling, the separation type treatment of 'spatial correlation' and 'time evolution' in the water quality state is realized. By means of linear coupling, residual nonlinear fusion and frequency domain screening mechanisms, the method can accurately extract instantaneous coupling characteristics among environmental factors under the condition of not depending on a complex depth network, and therefore the problems of structural bulkiness, training difficulty and high calculation cost caused by environmental factor mixed modeling in a traditional method are effectively solved. The structure not only improves the generalization capability and the robustness of the model, but also enables the water quality early warning system to have higher availability in a low-power consumption or edge computing scene. 2. The method adopts a mode of multi-scale decomposition and independent linear prediction channels to efficiently disassemble the long-term trend and periodic components of water quality, so that the pred