Intelligent monitoring method for drinking water source based on artificial intelligence

Abstract

The invention relates to the technical field of water quality detection, and discloses a drinking water source intelligent monitoring method based on artificial intelligence, and the method comprises the steps: S1, a sensor arranged in a drinking water source region collects water quality parameters and hydrological and hydraulic information in real time; s2, on the basis of the collected hydrological and hydraulic information, establishing a pollution diffusion simulation model, when a sensor detects that water quality parameters of any monitoring point are abnormal, carrying out pollution diffusion simulation by taking the current monitoring point as an initial condition, and predicting a future diffusion trajectory and an influence area of pollution; and S3, the generated pollution diffusion trajectory and the influence area are acquired. According to the technical scheme of fusion of the pollution diffusion trend simulation module and the artificial intelligence identification process, the two-dimensional pollution diffusion model is constructed to dynamically simulate the space-time propagation trajectory of the pollutants, and the simulation result is used as a time sequence to assist in inputting the AI model, so that the technical effect of improving the prediction timeliness of the pollution event and the space coverage precision is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of water quality detection, and in particular to an intelligent monitoring method for drinking water sources based on artificial intelligence. Background Art

[0002] Currently, the typical method for regulating drinking water sources is to collect water samples from the water area under supervision and analyze them based on traditional water quality models to determine the water quality of the water area. However, due to the fluidity of water in the water area and the large area, it has been found that simply collecting water samples and analyzing the water quality based on traditional water quality models cannot accurately analyze the water quality of the water area.

[0003] For example, Chinese invention application publication number CN106560711B discloses a drinking water source monitoring and early warning system, comprising a housing located at the water source, a processor, a wireless transmitter, a memory, and a GPS locator within the housing, a computer, and a wireless receiver electrically connected to the computer. The housing is equipped with an upper sensor group, a middle sensor group, and a bottom sensor group, each of which includes a dissolved oxygen sensor, a temperature sensor, a pH sensor, an ammonia nitrogen sensor, a phosphate sensor, and a turbidity sensor. This system has high detection sensitivity and good accuracy.

[0004] For example, the Chinese invention application with publication number CN115236007B discloses an intelligent monitoring method and device for drinking water sources. The method divides the water area into sub-water areas according to the type and location of the water area, thereby improving the accuracy of the water area division, and generates equipment control parameters of the intelligent flight equipment based on the accurately divided water area, thereby improving the accuracy of the generation of the equipment control parameters, and controls the intelligent flight equipment based on the accurately generated equipment control parameters to collect high-spectral information of each sub-water area after division, thereby improving the accuracy of the collection of high-spectral information of each sub-water area, and performs hyperspectral inversion analysis on it to obtain the analysis of the pollutant type and concentration of each sub-water area. Finally, the pollutant situation of the entire water area is analyzed based on the analysis of the pollutants in each sub-water area, thereby improving the analysis accuracy and efficiency of the pollutants in the entire water area, realizing intelligent monitoring of the water area, improving the analysis accuracy of the water area data, and providing an accurate basis for water source management decision-making.

[0005] The shortcomings of the above patents are:

[0006] On the one hand, current intelligent drinking water source monitoring systems mostly rely on sensors deployed in fixed locations within water bodies to collect water quality data, which is then analyzed using artificial intelligence algorithms. However, when water sources are affected by events such as rainfall erosion and sudden sewage discharges, the spread of pollutants in the water exhibits distinct spatiotemporal dynamics, such as uncertain diffusion paths and complex variations in contaminated areas. This results in localized and delayed information obtained from fixed monitoring points, making it impossible for artificial intelligence systems to fully and timely grasp the pollution diffusion process. This leads to delays or errors in prediction results, making it impossible to meet the needs of rapid response and accurate early warning of pollution incidents.

[0007] On the other hand, in complex natural water environments, water quality sensors often face physical interference such as entanglement with aquatic plants, obstruction by sediment, and coverage by floating debris, resulting in abnormal fluctuations in collected water quality data. Existing AI systems can detect anomalies in water quality indicators, but lack effective mechanisms for distinguishing the source of these anomalies. They cannot distinguish between actual water pollution, sensor failure, or environmental interference. This can lead to false alarms or missed alarms, compromising the stability and accuracy of water quality sensors in real-world applications.

[0008] To this end, the present invention proposes an intelligent monitoring method for drinking water sources based on artificial intelligence to solve the above-mentioned problems. Summary of the Invention

[0009] In view of the deficiencies of the existing technology, the present invention provides an intelligent monitoring method for drinking water sources based on artificial intelligence to solve the problems raised in the above background technology.

[0010] To achieve the above objectives, the present invention is implemented through the following technical solutions: an intelligent monitoring method for drinking water sources based on artificial intelligence, comprising:

[0011] S1, sensors deployed in drinking water source areas collect water quality parameters and hydrological and hydraulic information in real time;

[0012] S2, based on the collected hydrological and hydraulic information, establishes a pollution diffusion simulation model. When the sensor detects abnormal water quality parameters at any monitoring point, the pollution diffusion simulation is carried out with the current monitoring point as the initial condition to predict the future diffusion trajectory of the pollution and the affected area;

[0013] S3, the generated pollution diffusion trajectory and impact area are input into the artificial intelligence identification process as time series data, and the artificial intelligence identification process is integrated with the collected water quality parameters to generate pollution identification results;

[0014] S4: Once the AI ​​recognition process identifies a potential pollution incident, shore-based fixed cameras or drones are used to capture images of the affected area.

[0015] S5, if no contamination features are found in the image acquisition, the sensor abnormality marking program will be triggered to record the status information of the relevant sensors for subsequent re-inspection;

[0016] S6, for sensor locations marked as abnormal, dispatching backup sensors in the adjacent sensor network to perform coverage supplementary monitoring and collect supplementary water quality data;

[0017] S7, fuses and analyzes the pollution diffusion trajectory with the supplementary collected water quality data, and updates the input data structure of the artificial intelligence recognition process based on the fusion analysis, so that the recognition process can adapt to the current changes in water quality and diffusion environment.

[0018] Preferably, in said S1, collecting water quality parameters and hydrological and hydraulic information of the drinking water source further includes:

[0019] S1.1, using multiple water quality sensors deployed in the drinking water source area to collect basic water quality parameters in real time, including: water temperature T, pH, conductivity σ, turbidity NTU, and dissolved oxygen concentration DO;

[0020] The collected data is recorded as vector Q i =[T i ,pH i ,σ i ,NTU i ,DO i ];

[0021] Set the initial threshold vector Q ref =[T0,pH0,σ0,NTU0,DO0] and the allowable fluctuation coefficient δ = [δ T ,δ pH ,δ σ ,δ NTU ,δ DO ], conduct preliminary screening and judgment:

[0022] If present: |Q ij -Q ref,j |>δ j Q ref,j ,

[0023] Among them, Q ij is the jth water quality parameter value at the i-th monitoring point, δ j is the allowable fluctuation coefficient of the jth water quality parameter, Q ref,j is the reference threshold value of j water quality parameter;

[0024] Mark the data as initially suspected abnormal data for further analysis by the subsequent pollution judgment module;

[0025] S1.2. Collect hydrological and hydraulic information through flowmeters and water level gauges installed at river sections and nodes. The information includes:

[0026] Velocity vector v = [u(X, Y, t), v(X, Y, t)], where u and v are the instantaneous water velocity in the X and Y directions, and (X, Y, t) are the two-dimensional spatial coordinates and time variables;

[0027] Water depth h(X,Y,t), which represents the height of the water column at each coordinate point;

[0028] Boundary morphology Γ(X,Y), is the distribution of the water area boundary curve on the two-dimensional plane (X,Y);

[0029] Inflow and outflow flow Q in With Q out ;

[0030] Based on the above information, the instantaneous flow field equation is constructed:

[0031]

[0032] Among them, V(X,Y,t) is the divergence of the water flow vector field, is a two-dimensional gradient operator, u(X,Y,t) is the velocity component of the water flow in the X direction, and v(X,Y,t) is the velocity component of the water flow in the Y direction;

[0033] S1.3, the water quality parameter Q obtained in S1.1 i The local water flow vector V calculated in S1.2 i Conduct joint judgment to determine whether there is real driving force behind the suspected pollution;

[0034] Define the following joint decision function:

[0035] Ψ i =α·Norm(Q i -Q ref )+β·||V i ||,

[0036] Among them, i is the joint anomaly determination function of monitoring point i, Norm(·) is the normalized distance function, α and β are the empirical weight coefficients, ||V i || is the local hydrodynamic strength;

[0037] When i >Θ, where Θ is the empirically set joint anomaly threshold, and the monitoring point i is marked as a pollution candidate point.

[0038] Preferably, in S2, establishing a pollution diffusion simulation model further includes:

[0039] S2.1, according to the pollution candidate point i marked in S1.3, extract the spatial coordinates (X, Y) and time t, and define the initial pollution concentration C0 = Q ij , where Q ij is the j-th water quality parameter value at the i-th monitoring point;

[0040] Set the simulation starting point to (X i ,Y i ), the starting time is t i , providing clear initial conditions for subsequent concentration diffusion calculations;

[0041] S2.2, according to S2.1 (X i ,Y i ,t i , C0), combined with the flow velocity u and flow rate Q of S1.2 in , background concentration C p and attenuation coefficient K, a one-dimensional steady-state diffusion model is established:

[0042]

[0043] Among them, C(x) is the pollution concentration at x meters along the flow direction from the initial point, u is the flow velocity, K is the comprehensive attenuation coefficient of the pollutant, C p is the background concentration;

[0044] According to the length of the river section L, N = L / Δx groups of discrete nodes are divided according to the step size Δx, and C(x k ) Generate concentration array Provide data basis for subsequent trajectory input;

[0045] S2.3, using the concentration array generated in S2.2, set the concentration threshold C th , when node k has C(x k )>C th When the corresponding position x k and upstream and downstream adjacent nodes are included in the impact area;

[0046] By judging max k C(x k ) and C th Ratio, generating the diffusion range index R:

[0047] R={x|C(x)>C th}, and output the final pollution trajectory {(x k ,C(x k )),R}, providing well-defined diffusion path data for S3 input.

[0048] Preferably, said S3 further includes:

[0049] S3.1, the predicted value of pollution concentration C pred (t i ) to perform normalization processing and calculate the normalized concentration value C nrm (t i ), execute as follows:

[0050]

[0051] Among them, C pred (t i ) is the time t i The predicted concentration, C min is the historical minimum predicted concentration, C max is the maximum predicted concentration in history, t i is the prediction time series;

[0052] If any C nrm >0.8, marked as high risk;

[0053] S3.2, combined with real-time observation of concentration C obs (t i ), calculate the relative error E at each moment i And judge the consistency, and calculate according to the following formula:

[0054]

[0055] Among them, C obs (t i ) is the time t i Observed concentration of

[0056] If there is E i >0.5, marking the current sequence as inconsistent in prediction;

[0057] If for all i, E i <0.2, marked as reliable prediction;

[0058] S3.3, normalized concentration value vector C in S3.1 nrm =[C nrm (t1),…,C nrm (t n )], and the error index vector E=[E1,…,E n ] is concatenated into a feature vector x = [C nrm ,E], input the multi-layer perceptron model to perform classification, and the internal operation of the model is as follows:

[0059] z=Wx+b,

[0060]

[0061] Among them, W is the weight matrix, b is the bias vector, z is the unnormalized output vector, y j is the probability output of the jth class;

[0062] When the pollution category probability y pollution When it is greater than 0.7, it is determined to be a pollution event and an early warning is output.

[0063] Preferably, in said S4, when the artificial intelligence recognition process determines that a potential pollution event has occurred, calling a shore-based fixed camera device or a drone to collect images of the involved area further includes:

[0064] S4.1, the device parameter setting module calculates the ground sampling distance GSD according to the following formula and generates raw image data:

[0065]

[0066] Where h is the flight altitude, p is the camera sensor pixel size, and f is the lens focal length;

[0067] When the calculated GSD is greater than the preset maximum allowable value GSD max , adjust the flight altitude to:

[0068] To meet the resolution requirements and complete the original image acquisition;

[0069] S4.2, the image preprocessing module performs grayscale normalization on the original image data generated in S4.1 according to the following formula:

[0070]

[0071] Among them, I(x,y) is the pixel gray value, I min with I max is the minimum and maximum grayscale value of the image;

[0072] When the image grayscale standard deviation σ is lower than the threshold σ th When , calculate the Gamma correction index:

[0073]

[0074] Where μ is the mean grayscale value of the image, and applies:

[0075] I′(x,y)=255×[I n (x,y)] γ , generate pre-processed images for subsequent analysis;

[0076] S4.3, the feature extraction module converts the pre-processed image generated in S4.2 into HSV space and generates a binary mask according to the following formula:

[0077]

[0078] Among them, H(x,y) and S(x,y) are the hue and saturation of the pixel, H th With S th is the judgment threshold;

[0079] Calculate the proportion of polluted pixels:

[0080]

[0081] Among them, N total is the total number of pixels in the image;

[0082] When M is greater than the preset ratio threshold M th When , it is determined that there are pollution features in the area and the recognition result is output.

[0083] Preferably, the S5 further includes:

[0084] S5.1, based on the proportion of contaminated pixels calculated in step S4.3:

[0085]

[0086] When the ratio values ​​at consecutive sampling moments all satisfy M<M th When , it is determined that no pollution features are found in the current image acquisition and the abnormal marking process is entered;

[0087] S5.2, the marking process generates a state variable according to the determination result:

[0088]

[0089] Among them, the flag value 1 indicates that the sensor data is abnormal, and the recording module is called later;

[0090] S5.3, the recording module records the state variable S generated in step S5.2 status The sensor identifier ID s 、Position coordinates (x s ,y s ) and timestamp t mark Assemble into an abnormal record vector as follows:

[0091] V rec =[ID s ,x s ,y s ,t mark ,S status ], written into the log storage for subsequent review and scheduling.

[0092] Preferably, in S6, for the sensor location marked as abnormal, dispatching a backup sensor in the adjacent sensor network to perform coverage supplementary monitoring and collect supplementary water quality data, further includes:

[0093] S6.1, the scheduling module calculates the position P of each abnormal main sensor s =(x s ,y s ) in the neighborhood radius D max Retrieve the list of backup sensors and calculate the distance between each backup sensor j and the abnormal location:

[0094]

[0095] Among them, (x j ,y j ) is the position coordinate of the backup sensor j;

[0096] When there is d j ≤D max When , the spare sensor is marked as a candidate. If the number of candidates is greater than zero, the sensor with the smallest distance is selected for call monitoring;

[0097] S6.2, the supplementary monitoring module starts the sampling process for the selected backup sensor and calculates the supplementary collection concentration C sup (t k ), the sampling time is t k ;

[0098] At the same time, record the sensor temperature T at that moment s (t k ) and electrical conductivity EC s (t k );

[0099] If the measured concentration exceeds the set abnormal threshold C th , generate a new abnormal event report and return to step S6 for subsequent processing; otherwise, continue to S6.3;

[0100] S6.3, the data fusion module converts the pollution diffusion trajectory concentration prediction sequence C generated by S2 pred (t k ) and the supplementary concentration data C collected in step S6.2 sup (t k ) are merged into fusion data pairs {t k ,C pred (t k ),C sup (t k )}, calculate the fusion error:

[0101]

[0102] When there exists any moment that satisfies E sup (t k )>0.3, a fusion quality warning will be triggered and transmitted to the fusion analysis module of S7, otherwise the supplementary data will be added to the training samples for subsequent recognition process updates.

[0103] Preferably, the pollution diffusion trend simulation model is a diffusion prediction model constructed based on a two-dimensional hydrodynamic equation, wherein the simulation input parameters include the location of the pollution source, the water flow velocity field and the water body boundary conditions.

[0104] Preferably, the diffusion prediction result and the sensor sampling data are fused using a weighted fusion mechanism, and the weight is dynamically adjusted according to the distance from the pollution source, the data credibility and the sampling timeliness.

[0105] Preferably, the image acquisition device in S4 includes a fixed shore high-definition camera and a low-altitude flying drone system equipped with a visual positioning module.

[0106] The present invention provides an intelligent monitoring method for drinking water sources based on artificial intelligence. It has the following beneficial effects:

[0107] 1. The present invention adopts a technical solution that integrates a pollution diffusion trend simulation module with an artificial intelligence identification process. By constructing a two-dimensional pollution diffusion model to dynamically simulate the spatiotemporal propagation trajectory of pollutants, the simulation results are used as a time series auxiliary input into the AI ​​model, thereby achieving the technical effect of improving the timeliness and spatial coverage accuracy of pollution event prediction. Compared with the existing analysis method that relies on single-point data from fixed sensors, it solves the problems of prediction delay and error caused by the locality and lag of monitoring information, and significantly enhances the ability to respond quickly to sudden pollution events.

[0108] 2. The present invention adopts the technical solution of multi-source anomaly inversion verification and intelligent scheduling of redundant sensors. It verifies the authenticity of pollution characteristics through image recognition and combines the backup sensors to supplement the monitoring data for cross-validation, so as to achieve the technical effect of accurately distinguishing between real pollution and sensor anomalies. Compared with the existing technology that simply relies on the mechanism of sensor data anomaly judgment, it solves the problems of false alarms and missed alarms caused by physical interference or equipment failure, and effectively improves the stability of the monitoring system and data reliability. BRIEF DESCRIPTION OF THE DRAWINGS

[0109] Figure 1 Flowchart of the present invention. DETAILED DESCRIPTION

[0110] To help those skilled in the art understand the present invention, the following will provide a clear and complete description of the technical solutions in the embodiments of the present invention, in conjunction with the accompanying drawings. Obviously, the described embodiments are only partial embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0111] The present invention is described in detail below with reference to the accompanying drawings:

[0112] Example:

[0113] Please see the attached Figure 1 , an embodiment of the present invention provides an intelligent monitoring method for drinking water sources based on artificial intelligence, comprising:

[0114] S1, sensors deployed in drinking water source areas collect water quality parameters and hydrological and hydraulic information in real time;

[0115] S1.1, using multiple water quality sensors deployed in drinking water source areas to collect basic water quality parameters in real time, including: water temperature T, pH, conductivity σ, turbidity NTU, and dissolved oxygen concentration DO;

[0116] The collected data is recorded as vector Q i =[T i ,pH i ,σ i ,NTU i ,DO i ];

[0117] Set the initial threshold vector Q ref =[T0,pH0,σ0,NTU0,DO0] and the allowable fluctuation coefficient δ = [δ T ,δ pH ,δ σ ,δ NTU ,δ DO ], conduct preliminary screening and judgment:

[0118] If present: |Q ij -Q ref,j |>δ j Q ref,j ,

[0119] Among them, Q ij is the jth water quality parameter value at the i-th monitoring point, δ j is the allowable fluctuation coefficient of the jth water quality parameter, Q ref,j is the reference threshold value of j water quality parameter;

[0120] Mark the data as initially suspected abnormal data for further analysis by the subsequent pollution judgment module;

[0121] S1.2. Collect hydrological and hydraulic information through flowmeters and water level gauges installed at river sections and nodes. The information includes:

[0122] Velocity vector v = [u(X, Y, t), v(X, Y, t)], where u and v are the instantaneous water velocity in the X and Y directions, and (X, Y, t) are the two-dimensional spatial coordinates and time variables;

[0123] Water depth h(X,Y,t), which represents the height of the water column at each coordinate point;

[0124] Boundary morphology Γ(X,Y), is the distribution of the water area boundary curve on the two-dimensional plane (X,Y);

[0125] Inflow and outflow flow Q in With Q out ;

[0126] Based on the above information, the instantaneous flow field equation is constructed:

[0127]

[0128] Among them, V(X,Y,t) is the divergence of the water flow vector field, is a two-dimensional gradient operator, u(X,Y,t) is the velocity component of the water flow in the X direction, and v(X,Y,t) is the velocity component of the water flow in the Y direction;

[0129] S1.3, the water quality parameter Q obtained in S1.1 i The local water flow vector V calculated in S1.2 i Conduct joint judgment to determine whether there is real driving force behind the suspected pollution;

[0130] Define the following joint decision function:

[0131] Ψ i =α·Norm(Q i -Q ref )+β·||V i ||,

[0132] Among them, i is the joint anomaly determination function of monitoring point i, Norm(·) is the normalized distance function, α and β are the empirical weight coefficients, ||V i || is the local hydrodynamic strength;

[0133] When i >Θ, where Θ is the empirically set joint anomaly threshold, marking the monitoring point i as a pollution candidate point;

[0134] S2, based on the collected hydrological and hydraulic information, establishes a pollution diffusion simulation model. When the sensor detects abnormal water quality parameters at any monitoring point, the pollution diffusion simulation is carried out with the current monitoring point as the initial condition to predict the future diffusion trajectory of the pollution and the affected area;

[0135] S2.1, according to the pollution candidate point i marked in S1.3, extract the spatial coordinates (X, Y) and time t, and define the initial pollution concentration C0 = Q ij , where Q ij is the j-th water quality parameter value at the i-th monitoring point;

[0136] Set the simulation starting point to (X i ,Y i ), the starting time is t i , providing clear initial conditions for subsequent concentration diffusion calculations;

[0137] S2.2, according to S2.1 (X i ,Y i ,t i , C0), combined with the flow velocity u and flow rate Q of S1.2 in , background concentration C p and attenuation coefficient K, a one-dimensional steady-state diffusion model is established:

[0138]

[0139] Among them, C(x) is the pollution concentration at x meters along the flow direction from the initial point, u is the flow velocity, K is the comprehensive attenuation coefficient of the pollutant, C p is the background concentration;

[0140] According to the length of the river section L, N = L / Δx groups of discrete nodes are divided according to the step size Δx, and C(x k ) Generate concentration array Provide data basis for subsequent trajectory input;

[0141] S2.3, using the concentration array generated in S2.2, set the concentration threshold C th , when node k has C(x k )>C th When the corresponding position x k and upstream and downstream adjacent nodes are included in the impact area;

[0142] By judging max k C(x k ) and C th Ratio, generating the diffusion range index R:

[0143] R={x|C(x)>C th}, and output the final pollution trajectory {(xk ,C(x k )),R}, providing well-defined diffusion path data for S3 input;

[0144] S3, the generated pollution diffusion trajectory and impact area are input into the artificial intelligence identification process as time series data, and the artificial intelligence identification process is integrated with the collected water quality parameters to generate pollution identification results;

[0145] S3.1, the predicted value of pollution concentration C pred (t i ) to perform normalization processing and calculate the normalized concentration value C nrm (t i ), execute as follows:

[0146]

[0147] Among them, C pred (t i ) is the time t i The predicted concentration, C min is the historical minimum predicted concentration, C max is the maximum predicted concentration in history, t i is the prediction time series;

[0148] If any C nrm >0.8, marked as high risk;

[0149] S3.2, combined with real-time observation of concentration C obs (t i ), calculate the relative error E at each moment i And judge the consistency, and calculate according to the following formula:

[0150]

[0151] Among them, C obs (t i ) is the time t i Observed concentration of

[0152] If there is E i >0.5, marking the current sequence as inconsistent in prediction;

[0153] If for all i, E i <0.2, marked as reliable prediction;

[0154] S3.3, normalized concentration value vector C in S3.1 nrm =[C nrm (t1),…,C nrm (t n )], and the error index vector E=[E1,…,En ] is concatenated into a feature vector x = [C nrm ,E], input the multi-layer perceptron model to perform classification, and the internal operation of the model is as follows:

[0155] z=Wx+b,

[0156]

[0157] Among them, W is the weight matrix, b is the bias vector, z is the unnormalized output vector, y j is the probability output of the jth class;

[0158] When the pollution category probability y pollution When it is greater than 0.7, it is determined to be a pollution event and an early warning is output;

[0159] S4: Once the AI ​​recognition process identifies a potential pollution incident, shore-based fixed cameras or drones are used to capture images of the affected area.

[0160] S4.1, the device parameter setting module calculates the ground sampling distance GSD according to the following formula and generates raw image data:

[0161]

[0162] Where h is the flight altitude, p is the camera sensor pixel size, and f is the lens focal length;

[0163] When the calculated GSD is greater than the preset maximum allowable value GSD max , adjust the flight altitude to:

[0164] To meet the resolution requirements and complete the original image acquisition;

[0165] S4.2, the image preprocessing module performs grayscale normalization on the original image data generated in S4.1 according to the following formula:

[0166]

[0167] Among them, I(x,y) is the pixel gray value, I min with I max is the minimum and maximum grayscale value of the image;

[0168] When the image grayscale standard deviation σ is lower than the threshold σ th When , calculate the Gamma correction index:

[0169]

[0170] Where μ is the mean grayscale value of the image, and applies:

[0171] I′(x,y)=255×[I n (x,y)] γ , generate pre-processed images for subsequent analysis;

[0172] S4.3, the feature extraction module converts the pre-processed image generated in S4.2 into HSV space and generates a binary mask according to the following formula:

[0173]

[0174] Among them, H(x,y) and S(x,y) are the hue and saturation of the pixel, H th With S th is the judgment threshold;

[0175] Calculate the proportion of polluted pixels:

[0176]

[0177] Among them, N total is the total number of pixels in the image;

[0178] When M is greater than the preset ratio threshold M th When , it is determined that there are pollution features in the area and the recognition result is output;

[0179] S5, if no contamination features are found in the image acquisition, the sensor abnormality marking program will be triggered to record the status information of the relevant sensors for subsequent re-inspection;

[0180] S5.1, based on the proportion of contaminated pixels calculated in step S4.3:

[0181]

[0182] When the ratio values ​​at consecutive sampling moments all satisfy M<M th When , it is determined that no pollution features are found in the current image acquisition and the abnormal marking process is entered;

[0183] S5.2, the marking process generates state variables based on the judgment results:

[0184]

[0185] Among them, the flag value 1 indicates that the sensor data is abnormal, and the recording module is called later;

[0186] S5.3, the recording module records the state variable S generated in step S5.2 status The sensor identifier ID s 、Position coordinates (x s ,y s ) and timestamp t mark Assemble into an abnormal record vector as follows:

[0187] V rec =[ID s ,x s ,y s ,t mark ,S status ], written into the log storage for subsequent review and scheduling;

[0188] S6, for sensor locations marked as abnormal, dispatching backup sensors in the adjacent sensor network to perform coverage supplementary monitoring and collect supplementary water quality data;

[0189] S6.1, the scheduling module calculates the position P of each abnormal main sensor s =(x s ,y s ) in the neighborhood radius D max Retrieve the list of backup sensors and calculate the distance between each backup sensor j and the abnormal location:

[0190]

[0191] Among them, (x j ,y j ) is the position coordinate of the backup sensor j;

[0192] When there is d j ≤D max When , the spare sensor is marked as a candidate. If the number of candidates is greater than zero, the sensor with the smallest distance is selected for call monitoring;

[0193] S6.2, the supplementary monitoring module starts the sampling process for the selected backup sensor and calculates the supplementary collection concentration C sup (t k ), the sampling time is t k ;

[0194] At the same time, record the sensor temperature T at that moment s (t k ) and electrical conductivity EC s (t k );

[0195] If the measured concentration exceeds the set abnormal threshold C th , generate a new abnormal event report and return to step S6 for subsequent processing; otherwise, continue to S6.3;

[0196] S6.3, the data fusion module converts the pollution diffusion trajectory concentration prediction sequence C generated by S2 pred (t k ) and the supplementary concentration data C collected in step S6.2 sup (t k) are merged into fusion data pairs {t k ,C pred (t k ),C sup (t k )}, calculate the fusion error:

[0197]

[0198] When there exists any moment that satisfies E sup (t k )>0.3, a fusion quality warning will be triggered and transmitted to the fusion analysis module of S7, otherwise the supplementary data will be added to the training samples for subsequent recognition process updates;

[0199] S7, fuses and analyzes the pollution diffusion trajectory with the supplementary collected water quality data, and updates the input data structure of the artificial intelligence recognition process based on the fusion analysis, so that the recognition process can adapt to the current changes in water quality and diffusion environment.

[0200] The pollution diffusion trend simulation model is a diffusion prediction model constructed based on two-dimensional hydrodynamic equations. The simulation input parameters include the location of the pollution source, the water velocity field, and the boundary conditions of the water body. The diffusion prediction results are fused with the sensor sampling data using a weighted fusion mechanism, and the weights are dynamically adjusted based on the distance to the pollution source, data credibility, and sampling timeliness. The image acquisition equipment in S4 includes a fixed shore high-definition camera and a low-altitude flying drone system equipped with a visual positioning module.

[0201] Multi-point sensors are used to collect water temperature T, pH, conductivity σ, turbidity NTU, and dissolved oxygen concentration DO in real time to form a multi-dimensional data vector, providing high-density and high-timeliness basic data support for subsequent analysis. Compared with traditional single-point sampling, it can cover a wide range of water areas, capture the dynamic change trends of water quality parameters, and avoid misjudgment due to local data loss.

[0202] Based on flow velocity, water depth, and boundary morphology, the instantaneous flow field equation is constructed to quantify the spatial distribution characteristics of the water flow vector field. The hydrodynamic model depicts the physical environment in which pollutants propagate, providing dynamic boundary conditions for pollution diffusion simulations and addressing the inability of traditional static models to adapt to changes in water flow.

[0203] By introducing a normalized distance function and a local hydrodynamic intensity index, a multi-parameter joint judgment function is established to comprehensively assess the physical plausibility of suspected pollution points. This effectively distinguishes true pollution events from sensor noise interference, avoids the limitations of a single threshold judgment criterion, and improves the robustness of anomaly detection.

[0204] Based on the one-dimensional steady-state diffusion equation, the concentration distribution array is generated by combining flow velocity and attenuation coefficient to quantify the propagation of pollutants along the water flow direction. Through discretized node calculations, the complex diffusion process is transformed into a parseable numerical model, providing high-precision spatiotemporal trajectory prediction data for AI recognition.

[0205] Through concentration threshold screening and diffusion range index calculation, the boundaries of the pollution impact area are dynamically delineated. The abstract diffusion model is converted into a visual spatial coverage, providing an intuitive decision-making basis for emergency response and significantly improving the spatial perception of pollution incidents.

[0206] Normalized concentration values ​​and error indicators are concatenated into feature vectors, and a multilayer perceptron model is used to achieve probabilistic classification of pollution events. By integrating simulation data with real-time observations, the AI ​​model's ability to generalize complex pollution patterns is enhanced, reducing the risk of misjudgment due to data conflicts.

[0207] Dynamically adjust flight altitude and imaging parameters based on the ground sampling distance formula to ensure image resolution meets pollution feature identification requirements. Adaptive device configuration solves image blur or incomplete coverage issues caused by traditional fixed-viewing angle acquisition, improving image data usability.

[0208] Grayscale normalization and gamma correction algorithms are used to optimize image contrast and eliminate the interference of uneven illumination on feature extraction. Mathematical transformations are used to enhance the visual saliency of contaminated areas, providing high signal-to-noise ratio image input for subsequent analysis and reducing false detection rates in complex environments.

[0209] A binary mask based on hue and saturation is constructed in HSV space, and the objective quantification of pollution characteristics is achieved by calculating the proportion of polluted pixels. This converts subjective visual judgments into reproducible mathematical indicators, avoiding subjective bias in human interpretation and improving the standardization of pollution identification.

[0210] Generate state variables based on the results of continuous time-series image analysis and construct anomaly record vectors containing location and timestamps. This enables automated tagging and tracing of sensor faults, providing precise location information for equipment maintenance and reducing manual inspection costs.

[0211] By using neighborhood radius retrieval and distance-priority strategies to call backup devices, we ensure continuous monitoring of abnormal areas. Leveraging the spatial coverage advantages of redundant sensor networks, we address data interruptions caused by single-point failures and ensure the continuous operational reliability of the monitoring system.

[0212] By dynamically evaluating the consistency between the diffusion model and measured data through fusion error calculation, adaptive adjustment of model parameters is triggered. This achieves closed-loop optimization of the AI ​​recognition process, enabling the system to continuously adapt to changes in the water environment and improving the stability and prediction accuracy of long-term monitoring.

[0213] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent monitoring method for drinking water sources based on artificial intelligence, characterized in that: include: S1, sensors deployed in drinking water source areas collect water quality parameters and hydrological and hydraulic information in real time; S2, based on the collected hydrological and hydraulic information, establishes a pollution diffusion simulation model. When the sensor detects abnormal water quality parameters at any monitoring point, the pollution diffusion simulation is carried out with the current monitoring point as the initial condition to predict the future diffusion trajectory of the pollution and the affected area; S3, the generated pollution diffusion trajectory and impact area are input into the artificial intelligence identification process as time series data, and the artificial intelligence identification process is integrated with the collected water quality parameters to generate pollution identification results; S4: Once the AI ​​recognition process identifies a potential pollution incident, shore-based fixed cameras or drones are used to capture images of the affected area. S5, if no contamination features are found in the image acquisition, the sensor abnormality marking program will be triggered to record the status information of the relevant sensors for subsequent re-inspection; S6, for sensor locations marked as abnormal, dispatching backup sensors in the adjacent sensor network to perform coverage supplementary monitoring and collect supplementary water quality data; S7, fuses and analyzes the pollution diffusion trajectory with the supplementary collected water quality data, and updates the input data structure of the artificial intelligence recognition process based on the fusion analysis, so that the recognition process can adapt to the current changes in water quality and diffusion environment.

2. The method for intelligent monitoring of drinking water sources based on artificial intelligence according to claim 1, characterized in that: In S1, collecting water quality parameters and hydrological and hydraulic information of drinking water sources further includes: S1.1, using multiple water quality sensors deployed in the drinking water source area to collect basic water quality parameters in real time, including: water temperature T, pH, conductivity σ, turbidity NTU, and dissolved oxygen concentration DO; The collected data is recorded as vector Q i =[T i ,pH i ,σ i ,NTU i ,DO i ]; Set the initial threshold vector Q ref =[T0,pH0,σ0,NTU0,DO0] and the allowable fluctuation coefficient δ = [δ T ,δ pH ,δ σ ,δ NTU ,δ DO ], conduct preliminary screening and judgment: Young Being: |Q ij -Q ref,j |>δ j Q ref,j , Among them, Q ij is the jth water quality parameter value at the i-th monitoring point, δ j is the allowable fluctuation coefficient of the jth water quality parameter, Q ref,j is the reference threshold value of j water quality parameter; Mark the data as initially suspected abnormal data for further analysis by the subsequent pollution judgment module; S1.

2. Collect hydrological and hydraulic information through flowmeters and water level gauges installed at river sections and nodes. The information includes: Velocity vector v = [u(X, Y, t), v(X, Y, t)], where u and v are the instantaneous water velocity in the X and Y directions, and (X, Y, t) are the two-dimensional spatial coordinates and time variables; Water depth h(X,Y,t), which represents the height of the water column at each coordinate point; Boundary morphology Γ(X,Y), is the distribution of the water area boundary curve on the two-dimensional plane (X,Y); Inflow and outflow flow Q in With Q out ; Based on the above information, the instantaneous flow field equation is constructed: Among them, V(X,Y,t) is the divergence of the water flow vector field, is a two-dimensional gradient operator, u(X,Y,t) is the velocity component of the water flow in the X direction, and v(X,Y,t) is the velocity component of the water flow in the Y direction; S1.3, the water quality parameter Q obtained in S1.1 i The local water flow vector V calculated in S1.2 i Conduct joint judgment to determine whether there is real driving force behind the suspected pollution; Define the following joint decision function: P i =α·Norm(Q i -Q ref )+β·||V i ||, Among them, i is the joint anomaly determination function of monitoring point i, Norm(·) is the normalized distance function, α and β are the empirical weight coefficients, ||V i || is the local hydrodynamic strength; When i >Θ, where Θ is the empirically set joint anomaly threshold, and the monitoring point i is marked as a pollution candidate point.

3. The method for intelligent monitoring of drinking water sources based on artificial intelligence according to claim 1, characterized in that: In said S2, establishing the pollution diffusion simulation model further comprises: S2.1, according to the pollution candidate point i marked in S1.3, extract the spatial coordinates (X, Y) and time t, and define the initial pollution concentration C0 = Q ij , where Q ij is the j-th water quality parameter value at the i-th monitoring point; Set the simulation starting point to (X i ,Y i ), the starting time is t i , providing clear initial conditions for subsequent concentration diffusion calculations; S2.2, according to S2.1 (X i ,Y i ,t i , C0), combined with the flow velocity u and flow rate Q of S1.2 in , background concentration C p and attenuation coefficient K, a one-dimensional steady-state diffusion model is established: Among them, C(x) is the pollution concentration at x meters along the flow direction from the initial point, u is the flow velocity, K is the comprehensive attenuation coefficient of the pollutant, C p is the background concentration; According to the length of the river section L, N = L / Δx groups of discrete nodes are divided according to the step size Δx, and C(x k ) Generate concentration array Provide data basis for subsequent trajectory input; S2.3, using the concentration array generated in S2.2, set the concentration threshold C th , when node k has C(x k )>C th When the corresponding position x k and upstream and downstream adjacent nodes are included in the impact area; By judging max k C(x k ) and C th Ratio, generating the diffusion range index R: R={x|C(x)>C th }, and output the final pollution trajectory {(x k ,C(x k )),R}, providing well-defined diffusion path data for S3 input.

4. The method for intelligent monitoring of drinking water sources based on artificial intelligence according to claim 1, characterized in that: Said S3 further comprises: S3.1, the predicted value of pollution concentration C pred (t i ) to perform normalization processing and calculate the normalized concentration value C nrm (t i ), execute as follows: Among them, C pred (t i ) is the time t i The predicted concentration, C min is the historical minimum predicted concentration, C max is the maximum predicted concentration in history, t i is the prediction time series; If any C nrm >0.8, marked as high risk; S3.2, combined with real-time observation of concentration C obs (t i ), calculate the relative error E at each moment i And judge the consistency, and calculate according to the following formula: Among them, C obs (t i ) is the time t i Observed concentration of If there is E i >0.5, marking the current sequence as inconsistent in prediction; If for all i, E i <0.2, marked as reliable prediction; S3.3, normalized concentration value vector C in S3.1 nrm =[C nrm (t1),…,C nrm (t n )], and the error index vector E=[E1,…,E n ] is concatenated into a feature vector x = [C nrm ,E], input the multi-layer perceptron model to perform classification, and the internal operation of the model is as follows: z=Wx+b, Among them, W is the weight matrix, b is the bias vector, z is the unnormalized output vector, y j is the probability output of the jth class; When the pollution category probability y pollution When it is greater than 0.7, it is determined to be a pollution event and an early warning is output.

5. The method for intelligent monitoring of drinking water sources based on artificial intelligence according to claim 1, characterized in that: In S4, when the artificial intelligence recognition process determines that a potential pollution event has occurred, calling a shore-based fixed camera device or a drone to collect images of the involved area further includes: S4.1, the device parameter setting module calculates the ground sampling distance GSD according to the following formula and generates raw image data: Where h is the flight altitude, p is the camera sensor pixel size, and f is the lens focal length; When the calculated GSD is greater than the preset maximum allowable value GSD max , adjust the flight altitude to: To meet the resolution requirements and complete the original image acquisition; S4.2, the image preprocessing module performs grayscale normalization on the original image data generated in S4.1 according to the following formula: Among them, I(x,y) is the pixel gray value, I min with I max is the minimum and maximum grayscale value of the image; When the image grayscale standard deviation σ is lower than the threshold σ th When , calculate the Gamma correction index: Where μ is the mean grayscale value of the image, and applies: I′(x,y)=255×[I n (x,y)] γ , generate pre-processed images for subsequent analysis; S4.3, the feature extraction module converts the pre-processed image generated in S4.2 into HSV space and generates a binary mask according to the following formula: Among them, H(x,y) and S(x,y) are the hue and saturation of the pixel, H th With S th is the judgment threshold; Calculate the proportion of polluted pixels: Among them, N total is the total number of pixels in the image; When M is greater than the preset ratio threshold M th When , it is determined that there are pollution features in the area and the recognition result is output.

6. The method for intelligent monitoring of drinking water sources based on artificial intelligence according to claim 1, characterized in that: Said S5 further comprises: S5.1, based on the proportion of contaminated pixels calculated in step S4.3: When the ratio values ​​at consecutive sampling moments all satisfy M<M th When , it is determined that no pollution features are found in the current image acquisition and the abnormal marking process is entered; S5.2, the marking process generates a state variable according to the determination result: Among them, the flag value 1 indicates that the sensor data is abnormal, and the recording module is called later; S5.3, the recording module records the state variable S generated in step S5.2 status The sensor identifier ID s 、Position coordinates (x s ,y s ) and timestamp t mark Assemble into an abnormal record vector as follows: V rec =[ID s ,x s ,y s ,t mark ,S status ], written into the log storage for subsequent review and scheduling.

7. The method for intelligent monitoring of drinking water sources based on artificial intelligence according to claim 1, characterized in that: In S6, for the sensor location marked as abnormal, dispatching spare sensors in the adjacent sensor network to perform coverage supplementary monitoring and collect supplementary water quality data, further includes: S6.1, the scheduling module calculates the position P of each abnormal main sensor s =(x s ,y s ) in the neighborhood radius D max Retrieve the list of backup sensors and calculate the distance between each backup sensor j and the abnormal location: Among them, (x j ,y j ) is the position coordinate of the backup sensor j; When there is d j ≤D max When , the spare sensor is marked as a candidate. If the number of candidates is greater than zero, the sensor with the smallest distance is selected for call monitoring; S6.2, the supplementary monitoring module starts the sampling process for the selected backup sensor and calculates the supplementary collection concentration C sup (t k ), the sampling time is t k ; At the same time, record the sensor temperature T at that moment s (t k ) and electrical conductivity EC s (t k ); If the measured concentration exceeds the set abnormal threshold C th , generate a new abnormal event report and return to step S6 for subsequent processing; otherwise, continue to S6.3; S6.3, the data fusion module converts the pollution diffusion trajectory concentration prediction sequence C generated by S2 pred (t k ) and the supplementary concentration data C collected in step S6.2 sup (t k ) are merged into fusion data pairs {t k ,C pred (t k ),C sup (t k )}, calculate the fusion error: When there exists any moment that satisfies E sup (t k )>0.3, a fusion quality warning will be triggered and transmitted to the fusion analysis module of S7, otherwise the supplementary data will be added to the training samples for subsequent recognition process updates.

8. The method for intelligent monitoring of drinking water sources based on artificial intelligence according to claim 1, characterized in that: The pollution diffusion trend simulation model is a diffusion prediction model constructed based on a two-dimensional hydrodynamic equation, wherein the simulation input parameters include the location of the pollution source, the water flow velocity field and the water body boundary conditions.

9. The method for intelligent monitoring of drinking water sources based on artificial intelligence according to claim 1, characterized in that: The diffusion prediction results are fused with the sensor sampling data using a weighted fusion mechanism, and the weights are dynamically adjusted based on the distance to the pollution source, the data credibility, and the sampling timeliness.

10. The method for intelligent monitoring of drinking water sources based on artificial intelligence according to claim 1, characterized in that: The image acquisition equipment in the S4 includes a fixed shore high-definition camera and a low-altitude flying drone system equipped with a visual positioning module.

Images