Marine water quality monitoring data intelligent processing and service platform based on cloud-side cooperation

Abstract

The invention relates to the technical field of water quality online monitoring, in particular to an intelligent processing and service platform for ocean water quality monitoring data based on cloud-side collaboration. The method comprises the steps that an edge sensing preprocessing unit collects multi-point sensor data and uploads the data to a cloud after format standardization processing; the cloud multi-source fusion unit outputs a layered water quality spatial and temporal distribution result through a layered dynamic weighted interpolation model in combination with Bayesian optimization and cross validation; an edge end in the collaborative decision-making unit feeds back measured data to correct model parameters, and performs anomaly detection and early warning based on a One-Class SVM (Support Vector Machine) algorithm; and the intelligent information service unit is based on a cloud edge collaborative architecture, the cloud end provides global data query and visualization service, and the edge end pushes local abnormity early warning information. Through cloud-side cooperation, efficient cooperation of local rapid response and global accurate analysis is realized, and the accuracy and practicability of ocean water quality monitoring are improved.

Description

Technical Field

[0001] This invention relates to the field of online water quality monitoring technology, and more specifically, to a cloud-edge collaborative intelligent processing and service platform for marine water quality monitoring data. Background Technology

[0002] Online marine water quality monitoring technology is a technology used to collect key water quality parameters (such as water temperature, pH value, dissolved oxygen concentration, ammonia nitrogen concentration, salinity, etc.) in real time in the monitoring area, and to achieve marine water quality status assessment and related services through data analysis and processing. Its core objective is to accurately grasp the marine water quality status and provide data support for marine water quality management and control.

[0003] In existing technologies, it is difficult to effectively combine environmental factors such as ocean currents when fusing multi-source data, resulting in insufficient accuracy in predicting regional water quality parameters. Therefore, a cloud-edge collaborative intelligent processing and service platform for marine water quality monitoring data is needed. Summary of the Invention

[0004] The purpose of this invention is to provide a cloud-edge collaborative intelligent processing and service platform for marine water quality monitoring data, in order to solve the problem mentioned in the background art that in the existing technology, it is difficult to effectively combine environmental factors such as ocean currents when fusing multi-source data, resulting in insufficient accuracy in the prediction of regional water quality parameters.

[0005] To achieve the above objectives, the present invention aims to provide a cloud-edge collaborative intelligent processing and service platform for marine water quality monitoring data, comprising:

[0006] An edge sensing preprocessing unit is deployed at edge nodes to collect water quality data, location, monitoring radius, and ocean current data from multiple sensors. Through outlier detection, max-min normalization, and hierarchical processing, standardized hierarchical data is obtained and uploaded to the cloud.

[0007] The cloud-based multi-source fusion unit is used to receive standardized hierarchical data, calculate the predicted values ​​of regional water quality parameters through a hierarchical dynamic weighted interpolation model, obtain the optimal parameters of the hierarchical dynamic weighted interpolation model through Bayesian optimization, and output the hierarchical spatiotemporal distribution results of water quality through leave-one-out cross-validation.

[0008] A collaborative decision-making unit, wherein the edge terminal collects measured data and feeds it back to the cloud to correct the parameters of the hierarchical dynamic weighted interpolation model, and performs water quality anomaly detection and rapid early warning based on the One-Class SVM algorithm;

[0009] The intelligent information service unit is based on a cloud-edge collaborative architecture. The cloud provides global water quality data query and historical trend visualization services, while the edge pushes local anomaly warning information.

[0010] As a further improvement to this technical solution, the edge-aware preprocessing unit obtains standardized hierarchical data through outlier detection, max-min normalization, and hierarchical processing. The specific steps involved are as follows:

[0011] Acquire raw data from multiple sensors within the coverage area of ​​the edge node. The raw data includes the water quality parameter q of the i-th sensor at time t. i (t), location coordinates, monitoring radius, and ocean current data for the corresponding area;

[0012] Based on the 3σ principle, water quality parameter q i (t) Perform outlier detection and correct outliers using local interpolation;

[0013] The water quality parameters after outlier correction are normalized using the max-min method to obtain the standardized measurement value q of the i-th sensor at time t. i ″ (t);

[0014] Based on the typical stratification characteristics of the monitoring area, the sensors are stratified according to the depth in their location coordinates to obtain standardized stratified data. Each layer of data includes the standardized measurement values, location coordinates, monitoring radius, and ocean current data of the corresponding area of ​​the sensor within that layer.

[0015] As a further improvement to this technical solution, the specific steps involved in the cloud-based multi-source fusion unit calculating the predicted values ​​of regional water quality parameters through a hierarchical dynamic weighted interpolation model are as follows:

[0016] The hierarchical dynamic weighted interpolation model consists of a hierarchical weight calculation module and an interpolation calculation module. The hierarchical weight calculation module receives standardized hierarchical data uploaded by edge nodes, processes each layer of data independently, and determines the target location P within that layer.

[0017] For the target location P, filter all sensors within this layer that satisfy the monitoring radius constraint;

[0018] Based on the ocean current data of this layer, calculate the effective influence position of sensor i;

[0019] Calculate the distance d between the target position P and the effective position of sensor i. i And determine the directional consistency factor:

[0020] Define the parameter space Θ to be optimized, and combine it with the distance d. i And the orientation consistency factor, calculate the weight w of sensor i to target position P. i As the output of the hierarchical weight calculation module, the parameter space Θ includes the attenuation coefficient α and the Gaussian coefficient σ;

[0021] Based on weight w i The standardized measurement values ​​are weighted and summed using the interpolation calculation module to calculate the predicted water quality parameters at point P.

[0022] Repeat the above steps for all locations within the target area to obtain the regional water quality parameter distribution of that layer, and then integrate the results of each layer to form the complete regional water quality parameter distribution.

[0023] As a further improvement to this technical solution, the specific steps involved in obtaining the optimal parameters of the hierarchical dynamic weighted interpolation model through Bayesian optimization are as follows:

[0024] S1.1 Select the dataset of known real-area water quality distribution from the historical monitoring data of this layer as the training sample set S model , which contains N m One sample point;

[0025] S1.2 Based on the attenuation coefficient α and Gaussian coefficient σ in the parameter space Θ to be optimized, the root mean square error between the measured value and the true value of the hierarchical dynamic weighted interpolation model is used as the objective function f(α,σ).

[0026] S1.3. Bayesian optimization is used to optimize the parameter space to obtain the optimal parameters that minimize the objective function f(α,σ). The optimal parameters include the optimal decay coefficient α. * and the optimal Gaussian coefficient σ * .

[0027] As a further improvement to this technical solution, the specific steps involved in outputting the stratified spatiotemporal distribution results of water quality through leave-one-out cross-validation are as follows:

[0028] Hiding sensor i layer by layer, based on the data from the remaining sensors in that layer and the optimal parameter α. * σ * Predicted water quality parameter Q at sensor i location pred (i), and calculate the error e of sensor i. i ;

[0029] Based on the error of a single sensor, calculate the average error ē of all sensors. If the average error ē is not greater than the preset error threshold ε, the model parameters are deemed valid; otherwise, return to S1.3 to re-optimize the parameters.

[0030] The optimal parameters determined by the algorithm are substituted into the hierarchical dynamic weighted interpolation model to calculate the spatiotemporal distribution of water quality in the target area of ​​the layer, resulting in the layered spatiotemporal distribution Q(x,y,z,t), which is the predicted value of water quality parameters at any location (x,y,z) at any time t.

[0031] As a further improvement to this technical solution, the specific implementation steps for collecting measured data at the edge of the collaborative decision-making unit and feeding it back to the cloud to correct the parameters of the hierarchical dynamic weighted interpolation model are as follows:

[0032] S2.1 Select typical sampling points within the monitoring area, collect measured data and record environmental information, and after standardization processing at the edge, generate feedback data and upload it to the cloud.

[0033] S2.2 The cloud platform locates the level to which the sampling point belongs and calculates the predicted values ​​of water quality parameters based on the feedback data;

[0034] S2.3 Calculate the prediction error ΔQ of water quality parameters. If the prediction error ΔQ is greater than the dynamic error threshold ε′, the parameter correction mechanism is triggered.

[0035] S2.4 Based on the objective function f(α,σ), construct a modified objective function f′(α,σ) with measured constraints;

[0036] S2.5. Update the model parameters based on the Bayesian optimization framework to obtain the corrected parameters α′ and σ′;

[0037] S2.6 Verify the validity of the corrected parameters and update the hierarchical dynamic weighted interpolation model.

[0038] As a further improvement to this technical solution, the specific steps involved in water quality anomaly detection and rapid early warning based on the One-Class SVM algorithm are as follows:

[0039] S3.1 Collect water quality data under historical normal conditions at the edge and construct a training sample set;

[0040] S3.2. Construct a One-Class SVM model using radial basis kernel functions;

[0041] S3.3 Training a One-Class SVM model to determine model parameters;

[0042] S3.4 Detect anomalies in real-time monitoring data;

[0043] S3.5 Trigger an anomaly warning and upload the warning information.

[0044] As a further improvement to this technical solution, the specific steps for training the One-Class SVM model in S3.3 are as follows:

[0045] A one-class SVM model is constructed using the radial basis function (RBF kernel), and the kernel function is calculated.

[0046] The model parameters are determined by solving the optimization problem;

[0047] The original optimization problem is transformed into a dual problem, and the Lagrange function is constructed.

[0048] Solving the dual problem yields the Lagrange multipliers, i.e., the optimal multipliers.

[0049] Calculate the hyperplane parameter ρ;

[0050] Determine the final model parameters, including support vectors and optimal multipliers. And the hyperplane parameter ρ.

[0051] As a further improvement to this technical solution, the intelligent information service unit provides global water quality data query and historical trend visualization services in the cloud as follows:

[0052] A distributed data storage architecture is built in the cloud to classify and store the hierarchical spatiotemporal distribution results of water quality Q(x,y,z,t), the standardized hierarchical data uploaded by edge nodes, and the abnormal early warning records.

[0053] Deploy RESTful API interface services in the cloud to receive query requests from user terminals and parse request parameters;

[0054] The distributed query engine is invoked based on the request parameters. The target data is quickly located through a multi-dimensional index table, and the raw data is aggregated and calculated to generate a structured query result set.

[0055] The cloud-based visualization engine automatically selects the display method based on the type of query results.

[0056] The cloud performs permission verification on the visualization results and generates an encrypted visualization page or data interface response;

[0057] The encrypted visualization results are pushed to the user's terminal, supporting interactive operation, and user query logs are recorded for service optimization.

[0058] As a further improvement to this technical solution, the specific implementation method of pushing local anomaly warning information at the edge of the intelligent information service unit is as follows:

[0059] The edge device receives anomaly detection results and warning levels from edge nodes in real time based on the One-Class SVM algorithm;

[0060] The early warning information is encapsulated, including the location (x, y, z) of the anomaly, time t, the type of anomaly parameter and standardized measurement value, the degree of deviation from the normal boundary, and preliminary handling suggestions generated based on historical anomaly cases;

[0061] The target audience and method of notification should be determined according to the warning level;

[0062] Receive user confirmation feedback, record the processing status and synchronize it to the cloud for associated storage with the stratified spatiotemporal distribution results of water quality. If the push fails, start local caching. After the network is restored, push the warning information in order and mark it as delayed push.

[0063] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0064] 1. The intelligent processing and service platform for marine water quality monitoring data based on cloud-edge collaboration of the present invention adopts a cloud-edge collaborative architecture. The edge end performs standardized processing on the data and realizes rapid anomaly detection and early warning based on the One-Class SVM algorithm. The cloud end obtains accurate spatiotemporal distribution results of water quality through hierarchical dynamic weighted interpolation model combined with Bayesian optimization and cross-validation. The edge end feeds back measured data to correct model parameters, thereby improving the accuracy and reliability of marine water quality monitoring.

[0065] 2. In the intelligent processing and service platform for marine water quality monitoring data based on cloud-edge collaboration involved in this invention, the collaborative decision-making unit achieves rapid response to local anomalies through collaborative operation between the edge and the cloud, and ensures the accuracy of global data processing through model parameter correction. At the same time, the intelligent information service unit provides hierarchical information services, enhancing the efficiency and adaptability of the platform in practical applications. Attached Figure Description

[0066] Figure 1 This is an overall flowchart of the present invention;

[0067] The meanings of the various markings in the diagram are as follows:

[0068] 1. Edge perception preprocessing unit; 2. Cloud multi-source fusion unit; 3. Collaborative decision-making unit; 4. Intelligent information service unit. Detailed Implementation

[0069] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0070] Example

[0071] Please see Figure 1 As shown, a cloud-edge collaborative intelligent processing and service platform for marine water quality monitoring data is provided, including:

[0072] Edge sensing preprocessing unit 1 is deployed at the edge node. It collects water quality data, location, monitoring radius and ocean current data from multiple sensors. Through outlier detection, max-min normalization and hierarchical processing, it obtains standardized hierarchical data and uploads it to the cloud.

[0073] In this embodiment, the specific steps involved in obtaining standardized hierarchical data by the edge-aware preprocessing unit 1 through outlier detection, max-min normalization, and hierarchical processing are as follows:

[0074] Acquire raw data from multiple sensors within the coverage area of ​​the edge nodes;

[0075] This includes the water quality parameter measurement value q of the i-th sensor at time t. i (t) (such as pH value, dissolved oxygen concentration, etc.), location coordinates (x) i ,y i ,z i (x) i For longitude, y i For latitude, z i (depth), monitoring radius r i And the ocean current data (direction θ, velocity v) for the corresponding region, i = 1, 2, ..., N, where N is the total number of sensors;

[0076] Outlier detection is performed based on the 3σ principle, and outliers are corrected using local interpolation.

[0077] Calculate the mean μ and standard deviation σ of n consecutive measurements taken from the same sensor. err If a certain measured value q i (t) satisfies |q i (t)-μ|>3σ err If so, it is considered an outlier;

[0078] Outliers are corrected using local interpolation, with the following formula:

[0079]

[0080] Where, q i '(t) represents the corrected water quality parameter value at time t for a sensor with outliers; q j (t) represents the water quality parameter measurement (such as pH value, dissolved oxygen concentration, etc.) collected at time t by normal sensor j adjacent to the sensor with the outlier value; q k (t) represents the water quality parameter measurement value collected at time t by normal sensor k adjacent to the sensor with the outlier value; q l (t) represents the water quality parameter measurement value collected by normal sensor l adjacent to the sensor with the abnormal value at time t;

[0081] The water quality parameters after outlier correction were normalized using the max-min method to obtain the standardized measurement value q. i ″ (t);

[0082] By depth z i Stratification: Dividing the monitoring area into different levels based on typical stratification characteristics.

[0083] The shallow sea area is divided into two layers: 0-10m and 10-30m. The sensors are then positioned according to their depth z. i The data is categorized into corresponding levels to form standardized hierarchical data. Each layer of data includes standardized measurements, location coordinates, monitoring radius, and ocean current data for the corresponding region from the sensors within that layer.

[0084] The cloud-edge collaborative intelligent processing and service platform for marine water quality monitoring data also includes a cloud-based multi-source fusion unit 2. The cloud-based multi-source fusion unit 2 is deployed in the cloud, receives standardized hierarchical data, calculates the predicted values ​​of regional water quality parameters through a hierarchical dynamic weighted interpolation model, obtains the optimal parameters of the model through Bayesian optimization, and outputs hierarchical spatiotemporal distribution results of water quality through leave-one-out cross-validation.

[0085] In this embodiment, the specific steps involved in the cloud-based multi-source fusion unit 2 calculating the predicted values ​​of regional water quality parameters through a hierarchical dynamic weighted interpolation model are as follows:

[0086] The hierarchical dynamic weighted interpolation model consists of a hierarchical weight calculation module and an interpolation calculation module. The hierarchical weight calculation module receives standardized hierarchical data uploaded by edge nodes, processes each layer of data independently, and determines the target location P within that layer.

[0087] P(x,y,z,t)

[0088] Where x is longitude, y is latitude, z is depth, and t is the target time;

[0089] For the target location P, filter all sensors within this layer that satisfy the monitoring radius constraint;

[0090] Wherein, the corrected distance d between sensor i and P i ≤r i (r i (where w is the monitoring radius of sensor i), if the condition is not met, then the weight w of the sensor is... i =0;

[0091] Based on the ocean current data of this layer, calculate the effective influence position of sensor i;

[0092] The formula is:

[0093] x i ′=xi +v·Δt i ·sinθ

[0094] y i ′=y i +v·Δt i ·cosθ

[0095] Where, x i ′、y i ′ represents the corrected longitude and latitude; t i Let Δt be the sampling time of sensor i, t be the target time, and Δt be the sampling time of sensor i. i =tt i For time difference;

[0096] Calculate the distance d between the target position P and the effective position of sensor i. i And determine the directional consistency factor:

[0097] Distance formula:

[0098]

[0099] Where, d i The distance between the target position P and the effective position of sensor i;

[0100] Directional consistency factor:

[0101] f(θ,θ p )=1+cos(θ-θ p )

[0102] Where, f(θ,θ) p θ is the orientation consistency factor of the target position P. p The angle is the direction from the effective position of sensor i to P;

[0103] Define the parameter space to be optimized as θ, and combine it with the distance d. i And the orientation consistency factor, calculate the weight w of sensor i to target position P. i As the output of the hierarchical weight calculation module, the parameter space Θ includes the attenuation coefficient α and the Gaussian coefficient σ;

[0104]

[0105] Where α∈[0.5,5] is the attenuation coefficient, and σ∈[10,500] is the Gaussian coefficient;

[0106] Based on weight w i The standardized measurement values ​​are weighted and summed using the interpolation module to calculate the predicted water quality parameters at point P.

[0107]

[0108] Where Q(P) is the predicted value of the water quality parameter, q i ″ The water quality parameter measurement value of sensor i is normalized;

[0109] Repeat the above steps for all locations within the target area to obtain the regional water quality parameter distribution of that layer, and then integrate the results of each layer to form the complete regional water quality parameter distribution.

[0110] In this embodiment, the specific steps involved in obtaining the optimal parameters of the hierarchical dynamic weighted interpolation model through Bayesian optimization are as follows:

[0111] S1.1 Select the dataset of known real-area water quality distribution from the historical monitoring data of this layer as the training sample set S model , which contains N m One sample point;

[0112] The information of sample point s is its position coordinates (x, y, s). s ,y s ,z s ), time t s Actual water quality parameter value Q true (s) corresponds to sensor data, s = 1, 2, ..., N m N m This represents the number of sample points.

[0113] S1.2. Based on the attenuation coefficient α and Gaussian coefficient σ in the parameter space Θ to be optimized, the root mean square error between the measured value and the true value of the hierarchical dynamic weighted interpolation model is used as the objective function, i.e.:

[0114]

[0115] Where f(α,σ) is the objective function, representing the root mean square error between the model's predicted value and the true value under parameters α and σ (the smaller the value, the higher the model accuracy); the parameter space Θ is α∈[0.5,5], σ∈[10,500], α is the decay coefficient, and σ is the Gaussian coefficient; Q pred (s;α,σ) represents the predicted water quality parameters of sample point s obtained by a hierarchical dynamic weighted interpolation model based on parameters α and σ.

[0116] S1.3. Bayesian optimization is used to optimize the parameter space to obtain the optimal parameters that minimize the objective function f(α,σ). The optimal parameters include the optimal decay coefficient α. * and the optimal Gaussian coefficient σ * ;

[0117] Specifically, the steps for optimizing the parameter space using Bayesian optimization are as follows:

[0118] Initialize the Gaussian process model with n randomly selected initial parameters (α1, σ1), ..., (α...). n ,σ n ) and their corresponding objective function values ​​f(α1,σ1),…,f(α n ,σ n Using the training data, construct the posterior probability distribution p(f|α,σ,S) of the objective function. model );

[0119] Based on the posterior distribution, the expected improvement criterion is adopted:

[0120] EI(α,σ)=E[max(0,f best -f(α,σ))]

[0121] Where EI(α,σ) is the expected improvement value, representing the expected degree to which the objective function f is better than the current optimal value when parameters α and σ are used (the larger the value of EI(α,σ), the more likely this parameter combination is to improve the model accuracy); f best This represents the current optimal objective function value.

[0122] Select the next set of parameters to be evaluated, repeat the iteration until the maximum number of iterations is reached, and output the optimal parameter α. * σ * ;

[0123] In this embodiment, the specific steps involved in outputting the stratified spatiotemporal distribution results of water quality through leave-one-out cross-validation are as follows:

[0124] Hiding sensor i layer by layer, based on the data from the remaining sensors in that layer and the optimal parameter α. * σ * Predicted water quality parameter Q at sensor i location pred (i), and calculate the error e of sensor i. i :

[0125] Each time, a sensor i is hidden from the sensor set of that layer, using the data from the remaining sensors and the optimal parameter α. * σ * The water quality parameter Q at the location of the hidden sensor is predicted using a hierarchical dynamic weighted interpolation model. pred (i);

[0126] Calculate the predicted value and the sensor's true value q″ i error e i =|Q pred (i)-q″ i |;

[0127] Calculate the average error of all sensors based on the error of a single sensor.

[0128]

[0129] in, The average error of all sensors in this layer is used to evaluate the overall prediction accuracy of the hierarchical dynamic weighted interpolation model in this layer; e i The error of the i-th sensor (i = 1, 2, ..., N) sen N sen (This represents the total number of sensors in this layer);

[0130] If ē ≤ preset error threshold ε (set according to sensor accuracy, such as 0.1 times the maximum allowable error of the sensor), then the model parameters are deemed valid; otherwise, return to S1.3 to re-optimize the parameters.

[0131] The optimal parameter α * α * Substituting into the hierarchical dynamic weighted interpolation model, calculations are performed on all spatiotemporal points within the target area of ​​the layer to obtain the spatiotemporal distribution result of water quality Q(x,y,z,t), which is the predicted value of water quality parameters at any location (x,y,z) at any time t.

[0132] The cloud-edge collaborative intelligent processing and service platform for marine water quality monitoring data also includes a collaborative decision-making unit 3. In the collaborative decision-making unit 3, the edge end collects measured data and feeds it back to the cloud to correct the parameters of the hierarchical dynamic weighted interpolation model, and performs water quality anomaly detection and rapid early warning based on the One-Class SVM algorithm.

[0133] In this embodiment, the specific steps for collecting measured data at the edge and feeding it back to the cloud to correct the parameters of the hierarchical dynamic weighted interpolation model are as follows:

[0134] S2.1 Select typical sampling points within the monitoring area, collect measured data and record environmental information, and after standardization processing at the edge, generate feedback data and upload it to the cloud.

[0135] Select representative sampling points within the monitoring area (such as areas where ocean currents converge, sensor-sparse areas, or areas with variable water quality), and collect water quality parameter measurements Q at these points at time t. meas (x,y,z,t) (including pH value, dissolved oxygen concentration, etc.), where (x,y,z) are the latitude, longitude and depth of the sampling point;

[0136] Simultaneously record ocean current data at the sampling point at that moment: ocean current direction θ meas (Angle to true north, ranging from 0-360°) and ocean current velocity v meas (Unit: m / s);

[0137] The edge processing unit performs format standardization on the measured data to generate feedback data.

[0138] Data meas =[x,y,z,t,Q meas ,θ meas ,v meas ]

[0139] S2.2 The cloud platform locates the level to which the sampling point belongs and calculates the predicted values ​​of water quality parameters based on the feedback data;

[0140] Cloud receives data meas Then, the monitoring layer to which the sampling point belongs (such as the 0-10m layer, the 10-30m layer) is determined based on the depth z;

[0141] Invoke the current hierarchical dynamic weighted interpolation model, using the current optimal parameter α of this layer. * σ * Calculate the predicted water quality parameter Q at the sampling point at time t. pred (x,y,z,t;α * ,σ * );

[0142] S2.3 Calculate the prediction error of water quality parameters and determine whether parameter correction is triggered:

[0143] Calculate the absolute error between the measured and predicted values: ΔQ = |Q meas -Q pred |;

[0144] Set the dynamic error threshold ε′=1.2·ē, where ē is the average error obtained by cross-validation of this layer (the unit is consistent with the water quality parameter, such as mg / L);

[0145] If ΔQ>ε′, the current model parameters need to be corrected, triggering the parameter correction mechanism, i.e., executing S2.4-S2.6; otherwise, the measured data is stored as normal samples in the historical database, and the parameter correction mechanism is not triggered.

[0146] S2.4. Based on the objective function f(α,σ), construct a modified objective function f′(α,σ) with measured constraints:

[0147] Based on the original objective function f(α,σ), by adding the constraint term of measured data, the modified objective function is obtained:

[0148] f′(α,σ)=f(α,σ)+λ·(Q pred (x,y,z,t;α,σ)-Q meas ) 2

[0149] Where λ is the penalty coefficient (λ>0), and when the measured data is collected by a high-precision instrument, λ is taken as 5-10, and when the measurement is done manually, λ is taken as 1-3;

[0150] S2.5. Update model parameters based on the Bayesian optimization framework:

[0151] With the current optimal parameter (α) * ,σ * Using α as the initial point, iterative optimization is performed in the parameter space Θ(α∈[0.5,5],σ∈[10,500]).

[0152] A Gaussian process model is constructed, using historical sample errors and measured errors as training data, to fit the posterior distribution p(f′|α,σ,Data) of the objective function f′(α,σ). meas );

[0153] The expected improvement criterion is EI(α,σ)=E[max(0,f′). best -f′(α,σ))] selects the next set of evaluation parameters, where f′ best This represents the current optimal objective function value.

[0154] After 5-10 iterations, the corrected parameter α′=α is obtained. * +Δα and σ′=σ * +Δσ, where Δα and Δσ are parameter adjustment amounts;

[0155] S2.6 Verify the validity of the corrected parameters and update the hierarchical dynamic weighted interpolation model;

[0156] The predicted values ​​for the historical sample set S of this layer are recalculated using the corrected parameters (α′, σ′), resulting in a new average error.

[0157] like If the correction is deemed valid, (α′,σ′) is set as the new optimal parameter for this layer.

[0158] like The correction is deemed invalid, and the original parameter (α) is retained. * ,σ * Only the actual test data is added to the historical sample set S for subsequent optimization;

[0159] In this embodiment, the specific steps involved in water quality anomaly detection and rapid early warning based on the One-Class SVM algorithm are as follows:

[0160] S3.1. Collect historical water quality data under normal conditions at the edge and construct a training sample set:

[0161] Water quality data that were historically classified as normal within the coverage area of ​​the edge nodes were selected as training samples, and the training sample set is denoted as . Where N o The number of samples in the training sample set;

[0162] Each sample Let t be the time (t=1,2,…,N) o The multidimensional water quality parameter vector, i.e. D represents the number of water quality parameter types (such as pH value, dissolved oxygen concentration, etc.), q″ d (t) represents the standardized measurement of the d-th parameter at time t;

[0163] S3.2. Construct a One-Class SVM model using radial basis kernel functions;

[0164] The kernel function formula is:

[0165]

[0166] in, For the training sample set S ocsvm The two samples in the diagram correspond to normal water quality data at times k and l, respectively. γ>0 is the kernel function parameter (controlling the decay rate of similarity between samples); ‖*‖ is the Euclidean distance. Represents the squared Euclidean distance between the standardized water quality parameter vectors at two different times;

[0167] S3.3 Training a One-Class SVM model to determine model parameters;

[0168] Specifically, the model parameters are determined by solving the following optimization problem:

[0169]

[0170] Where w is the normal vector of the hyperplane. The mapping from samples to a high-dimensional feature space (implicitly implemented by a kernel function), where ρ is the hyperplane offset and ξ is the hyperplane offset. k is a slack variable (allowing a small number of normal samples to fall outside the boundary), and v∈(0,1] is a regularization parameter (controlling the upper bound of the proportion of abnormal samples);

[0171] The original optimization problem is transformed into a dual problem, and the Lagrange function is constructed.

[0172] formula:

[0173]

[0174] Where, α k ≥0,δ k≥0 represents a Lagrange multiplier, α k Corresponding constraints δ k Corresponding constraint ξ k ≥0; L is the Lagrangian function, which transforms a constrained optimization problem into an unconstrained problem, making it easier to solve;

[0175] After simplification using the KKT conditions, the dual problem becomes:

[0176]

[0177] Solving the duality problem yields the Lagrange multipliers:

[0178] The Sequence Minimum Optimization (SMO) algorithm is used to solve the dual problem above, iteratively selecting two multipliers α. k α l Optimize until the convergence condition is met (e.g., error less than 10). -3 );

[0179] Obtain the optimal multiplier in The samples are support vectors, corresponding to typical samples in ocean water quality data that play a key role in the normal boundary (such as extreme points of normal fluctuations).

[0180] Calculate the hyperplane parameter ρ:

[0181] Select a support vector (satisfy According to the KKT conditions, we can obtain...

[0182]

[0183] ρ represents the offset of the hyperplane in high-dimensional space, and its value is determined by the support vectors, reflecting the central position of normal ocean water quality data;

[0184] Determine the final model parameters:

[0185] After training, the model parameters are support vectors and optimal multipliers. And the hyperplane offset ρ, used for subsequent anomaly detection in real-time water quality data;

[0186] S3.3, Perform anomaly detection on real-time monitoring data:

[0187] Water quality parameters are acquired in real time at the edge and then standardized to form real-time samples.

[0188] Calculate the distance (decision function value) from the real-time sample to the hyperplane:

[0189]

[0190] in, The distance between the real-time sample and the normal boundary, if Then the sample is judged to be normal; if Then it is judged as abnormal;

[0191] S3.4 Trigger an anomaly warning and upload the warning information:

[0192] When an abnormal sample is detected for 3 consecutive moments (or 5 times cumulatively within 10 minutes), a rapid early warning mechanism is triggered.

[0193] The warning information includes the location (x, y, z) of the anomaly, the time t, the type and value of the anomaly parameter (e.g., "Dissolved oxygen concentration is abnormal, current value is 2.1 mg / L");

[0194] The edge device pushes early warning information to the local monitoring terminal in real time and uploads it to the cloud simultaneously for global decision-making reference.

[0195] The cloud-edge collaborative intelligent processing and service platform for marine water quality monitoring data also includes an intelligent information service unit 4. The intelligent information service unit 4 is based on a cloud-edge collaborative architecture. The cloud provides global water quality data query and historical trend visualization services, while the edge pushes local anomaly warning information.

[0196] In this embodiment, the cloud-based global water quality data query and historical trend visualization service is implemented as follows:

[0197] A distributed data storage architecture is built in the cloud. The hierarchical spatiotemporal distribution results of water quality Q(x,y,z,t), standardized hierarchical data uploaded by edge nodes, and abnormal early warning records are classified and stored according to the time dimension (year / month / day / hour), spatial dimension (edge ​​node monitoring area coding), and parameter dimension (5 core parameters such as water temperature and pH value). Multi-dimensional index tables (such as spatiotemporal index and parameter index) are established.

[0198] Deploy RESTful API interface services in the cloud to receive query requests from user terminals (such as web and mobile terminals), parse request parameters, including query time range, target monitoring area set, water quality parameters of interest, and data granularity (raw data / hourly average / daily average);

[0199] The distributed query engine is invoked based on the request parameters. The target data is quickly located through a multi-dimensional index table. The original data is aggregated and calculated (such as mean, maximum value, and frequency of exceeding the standard) to generate a structured query result set (including timestamp, region identifier, parameter value, and data credibility).

[0200] The cloud-based visualization engine automatically selects the display method based on the type of query results:

[0201] Spatial distribution display: For the same parameters in multiple regions, a marine heat map is used (using an electronic map of the sea area as the base map, and using color gradients to represent spatial differences in parameter values).

[0202] Multi-parameter correlation display: For multiple parameters in a single region, a radar chart is used (with parameter type as the axis and axis length representing the relative deviation of parameters).

[0203] Anomaly warning display: Linking anomaly warning records with corresponding regional water quality data, using a labeled chart (with the time axis as the horizontal axis, labeling the time of anomaly occurrence, type and handling status);

[0204] The cloud performs permission verification on the visualization results (assigning data access permissions based on user roles, such as administrators being able to view data across the entire domain, while ordinary users are only allowed access to data in public monitoring areas), and generates encrypted visualization pages or data interface responses.

[0205] The encrypted visualization results are pushed to the user terminal, supporting interactive operations (such as time range zooming, region filtering, parameter switching, and data export), while recording user query logs (including query content, access time, and terminal identifier) ​​for service optimization.

[0206] In this embodiment, the specific implementation method for pushing local anomaly warning information at the edge is as follows:

[0207] The edge terminal receives the anomaly judgment results and early warning levels (Level 1: water quality parameters are seriously exceeding the standard and require emergency treatment; Level 2: parameters deviate from the appropriate range and require attention and control; Level 3: abnormal parameter fluctuations and require continuous monitoring) from the edge nodes in real time.

[0208] The early warning information is packaged, including the time of the early warning, the latitude, longitude and depth of the monitoring area, the type of abnormal parameters and standardized measurement values, the degree of deviation from the normal boundary, the basis for judging the anomaly, and preliminary handling suggestions generated based on historical anomaly cases;

[0209] Determine the target audience and method of push notifications based on the warning level:

[0210] Level 1 warning: Pushed to regional administrators, technical maintenance personnel and emergency command center, using real-time push (interval ≤ 10s, until receipt confirmation);

[0211] Level 2 alert: Pushed to regional administrators and technical maintenance personnel, using timed push (once every 30 seconds, for a total of 3 times);

[0212] Level 3 alert: Only pushed to technical operations and maintenance personnel, using hourly batch push;

[0213] Receive user confirmation feedback, record the processing status (unprocessed / processing / resolved) and synchronize it to the cloud for associated storage with the layered spatiotemporal distribution results of water quality. If the push fails, start local caching. After the network is restored, push the warning information in order and mark it as delayed push.

[0214] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention.

Claims

1. A cloud-edge collaborative intelligent processing and service platform for marine water quality monitoring data, characterized in that, include: Edge sensing preprocessing unit (1), the edge sensing preprocessing unit (1) is deployed at the edge node, collects water quality data, location, monitoring radius and ocean current data from multiple sensors, and obtains standardized layered data and uploads it to the cloud through outlier detection, max-min normalization processing and layered processing; The cloud-based multi-source fusion unit (2) is used to receive standardized hierarchical data, calculate the predicted values ​​of regional water quality parameters through a hierarchical dynamic weighted interpolation model, obtain the optimal parameters of the hierarchical dynamic weighted interpolation model through Bayesian optimization, and output the hierarchical spatiotemporal distribution results of water quality through leave-one-out cross-validation. Collaborative decision-making unit (3), wherein the edge end collects measured data and feeds it back to the cloud to correct the parameters of the hierarchical dynamic weighted interpolation model, and performs water quality anomaly detection and rapid early warning based on the One-Class SVM algorithm; The intelligent information service unit (4) is based on a cloud-edge collaborative architecture. The cloud provides global water quality data query and historical trend visualization services, and the edge pushes local abnormal warning information.

2. The intelligent processing and service platform for marine water quality monitoring data based on cloud-edge collaboration as described in claim 1, characterized in that: The specific steps involved in the edge-aware preprocessing unit (1) to obtain standardized hierarchical data through outlier detection, max-min normalization, and hierarchical processing are as follows: Acquire raw data from multiple sensors within the coverage area of ​​the edge node. The raw data includes the water quality parameter q of the i-th sensor at time t. i (t), location coordinates, monitoring radius, and ocean current data for the corresponding area; Based on the 3σ principle, water quality parameter q i (t) Perform outlier detection and correct outliers using local interpolation; The water quality parameters after outlier correction are normalized using the max-min method to obtain the standardized measurement value q of the i-th sensor at time t. i "(t); Based on the typical stratification characteristics of the monitoring area, the sensors are stratified according to the depth in their location coordinates to obtain standardized stratified data. Each layer of data includes the standardized measurement values, location coordinates, monitoring radius, and ocean current data of the corresponding area of ​​the sensor within that layer.

3. The intelligent processing and service platform for marine water quality monitoring data based on cloud-edge collaboration as described in claim 2, characterized in that: The specific steps involved in the cloud-based multi-source fusion unit (2) calculating the predicted values ​​of regional water quality parameters through a hierarchical dynamic weighted interpolation model are as follows: The hierarchical dynamic weighted interpolation model consists of a hierarchical weight calculation module and an interpolation calculation module. The hierarchical weight calculation module receives standardized hierarchical data, processes each layer of data independently, and determines the target position P within that layer. For the target location P, filter all sensors within this layer that satisfy the monitoring radius constraint; Based on the ocean current data of this layer, calculate the effective influence position of sensor i; Calculate the distance d between the target position P and the effective position of sensor i. i And determine the direction consistency factor; Define the parameter space Θ to be optimized, and combine it with the distance d. i And the orientation consistency factor, calculate the weight w of sensor i to target position P. i As the output of the hierarchical weight calculation module, the parameter space Θ includes the attenuation coefficient α and the Gaussian coefficient σ; Based on weight w i The standardized measurement values ​​are weighted and summed using the interpolation calculation module to calculate the predicted water quality parameters at point P. Repeat the above steps for all locations within the target area to obtain the regional water quality parameter distribution of that layer, and then integrate the results of each layer to form the complete regional water quality parameter distribution.

4. The intelligent processing and service platform for marine water quality monitoring data based on cloud-edge collaboration as described in claim 1, characterized in that: The specific steps involved in obtaining the optimal parameters of the hierarchical dynamic weighted interpolation model through Bayesian optimization are as follows: S1.1 Select the dataset of known real-area water quality distribution from the historical monitoring data of this layer as the training sample set S model , which contains N m One sample point; S1.2 Based on the attenuation coefficient α and Gaussian coefficient σ in the parameter space Θ to be optimized, the root mean square error between the measured value and the true value of the hierarchical dynamic weighted interpolation model is used as the objective function f(α,σ). S1.

3. Bayesian optimization is used to optimize the parameter space to obtain the optimal parameters that minimize the objective function f(α,σ). The optimal parameters include the optimal decay coefficient α. * and the optimal Gaussian coefficient σ * .

5. The intelligent processing and service platform for marine water quality monitoring data based on cloud-edge collaboration as described in claim 4, characterized in that: The specific steps involved in outputting the stratified spatiotemporal distribution results of water quality through leave-one-out cross-validation are as follows: By hierarchically concealing sensor i in layers, and using the data from the remaining sensors in that layer and the optimal parameters, the predicted water quality parameter Q at the location of sensor i is calculated. pred (i), and calculate the error e of sensor i. i ; Based on the error of a single sensor, calculate the average error ē of all sensors. If the average error ē is not greater than the preset error threshold ε, the model parameters are deemed valid; otherwise, return to S1.3 to re-optimize the parameters. The optimal parameters determined by the algorithm are substituted into the hierarchical dynamic weighted interpolation model to calculate the spatiotemporal distribution of water quality in the target area of ​​the layer, resulting in the layered spatiotemporal distribution Q(x,y,z,t), which is the predicted value of water quality parameters at any location (x,y,z) at any time t.

6. The intelligent processing and service platform for marine water quality monitoring data based on cloud-edge collaboration as described in claim 5, characterized in that: The specific implementation steps of the collaborative decision-making unit (3) in which the edge end collects measured data and feeds it back to the cloud to correct the parameters of the hierarchical dynamic weighted interpolation model are as follows: S2.1 Select typical sampling points within the monitoring area, collect measured data and record environmental information, and after standardization processing at the edge, generate feedback data and upload it to the cloud. S2.2 The cloud platform locates the level to which the sampling point belongs and calculates the predicted values ​​of water quality parameters based on the feedback data; S2.3 Calculate the prediction error ΔQ of water quality parameters. If the prediction error ΔQ is greater than the dynamic error threshold ε... ′ If so, the parameter correction mechanism will be triggered; S2.

4. Based on the objective function f(α,σ), construct a modified objective function f with measured constraints. ′ (α,σ); S2.

5. Update the model parameters based on the Bayesian optimization framework to obtain the corrected parameters α′ and σ′; S2.6 Verify the validity of the corrected parameters and update the hierarchical dynamic weighted interpolation model.

7. The intelligent processing and service platform for marine water quality monitoring data based on cloud-edge collaboration as described in claim 6, characterized in that: The specific steps involved in water quality anomaly detection and rapid early warning based on the One-Class SVM algorithm are as follows: S3.1 Collect water quality data under historical normal conditions at the edge and construct a training sample set; S3.

2. Construct a One-Class SVM model using radial basis kernel functions; S3.3 Training a One-Class SVM model to determine model parameters; S3.4 Detect anomalies in real-time monitoring data; S3.5 Trigger an anomaly warning and upload the warning information.

8. The intelligent processing and service platform for marine water quality monitoring data based on cloud-edge collaboration as described in claim 7, characterized in that: In S3.3, the specific steps for training the One-Class SVM model are as follows: The model parameters are determined by solving the optimization problem; The original optimization problem is transformed into a dual problem, and the Lagrange function is constructed. Solving the dual problem yields the Lagrange multipliers, i.e., the optimal multipliers. Calculate the hyperplane parameter ρ; Determine the final model parameters, including support vectors and optimal multipliers. And the hyperplane parameter ρ.

9. The intelligent processing and service platform for marine water quality monitoring data based on cloud-edge collaboration as described in claim 8, characterized in that: The intelligent information service unit (4) provides global water quality data query and historical trend visualization services in the cloud. The specific implementation method is as follows: A distributed data storage architecture is built in the cloud to classify and store the hierarchical spatiotemporal distribution results of water quality Q(x,y,z,t), the standardized hierarchical data uploaded by edge nodes, and the abnormal early warning records. Deploy RESTful API interface services in the cloud to receive query requests from user terminals and parse request parameters; The distributed query engine is invoked based on the request parameters. The target data is quickly located through a multi-dimensional index table, and the raw data is aggregated and calculated to generate a structured query result set. The cloud-based visualization engine automatically selects the display method based on the type of query results. The cloud performs permission verification on the visualization results and generates an encrypted visualization page or data interface response; The encrypted visualization results are pushed to the user's terminal, supporting interactive operation, and user query logs are recorded for service optimization.

10. The intelligent processing and service platform for marine water quality monitoring data based on cloud-edge collaboration as described in claim 9, characterized in that: The specific implementation method of the intelligent information service unit (4) pushing local anomaly warning information at the edge is as follows: The edge device receives anomaly detection results and warning levels from edge nodes in real time based on the One-Class SVM algorithm; The early warning information is encapsulated, including the location (x, y, z) of the anomaly, time t, the type of anomaly parameter and standardized measurement value, the degree of deviation from the normal boundary, and preliminary handling suggestions generated based on historical anomaly cases; The target audience and method of notification should be determined according to the warning level; Receive user confirmation feedback, record the processing status and synchronize it to the cloud for associated storage with the stratified spatiotemporal distribution results of water quality. If the push fails, start local caching. After the network is restored, push the warning information in order and mark it as delayed push.

Images