Multi-element synchronous rapid prediction method and device for water environment pollution and storage medium

Abstract

The invention discloses a multi-element synchronous rapid prediction method and device for water environment pollution and a storage medium, and the method comprises the steps: selecting i water environment parameters # imgabs0 # which need to be predicted, and taking j non-linearly related hydrological water quality and meteorological parameters xj as input layer neurons of a neural network model; the value of the current data hydrology, water quality and meteorological parameter xj corresponds to the value of the water environment element # imgabs1 # corresponding to the next moment, and a training set and a test set are constructed; inputting the training set and the test set into a neural network model to obtain network parameters, and training to obtain a water environment multi-element synchronous prediction model; and taking the hydrology and water quality and meteorological parameters at the current moment as input of the water environment multi-element synchronous prediction model, and outputting predicted values of various water environment elements at the next moment by the prediction model. According to the method, the water quality multi-parameter prediction process is simplified, and the synchronism and timeliness of multi-parameter prediction are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of multi-factor prediction of water environment, and particularly to a method, device and storage medium for rapid synchronous prediction of multi-factors of water environment based on deep learning. Background Art

[0002] The prediction of seawater quality parameters is a key task in marine environmental monitoring and climate research. Accurately predicting multiple water quality indicators is of great significance for marine ecological protection, resource development, and disaster warning. Traditional prediction methods rely on physical and statistical models, which are difficult to effectively capture the complex non-linear relationships and spatio-temporal dynamic changes in seawater quality, resulting in great prediction difficulty, low accuracy, few predictable quantities, and disadvantages such as poor timeliness and untimely prediction. They cannot provide timely and accurate warning indicators for business departments, restricting their application in practice.

[0003] Deep learning is a robust machine algorithm developed on the basis of statistical machine learning and artificial neural networks, combined with big data and high computing power. This algorithm performs pattern recognition and prediction through multi-layer neural networks, has the ability to automatically extract features and establish complex non-linear relationships, and shows significant advantages in non-linear prediction. Through learning a large amount of historical data, deep learning can achieve high-precision prediction. However, the current seawater quality prediction methods based on deep learning are mainly limited to single-parameter prediction and cannot meet the needs of multi-factor synchronous prediction.

[0004] Patent document CN117892767A proposes a method for predicting seawater temperature based on deep learning. This method constructs a neural network model for temperature prediction by selecting environmental parameters related to seawater surface temperature. Although this method effectively improves the temperature prediction accuracy, it can only predict a single seawater surface temperature and does not consider the comprehensive synchronous prediction of multiple water quality parameters. With the continuous improvement of the timeliness and synchrony of the comprehensive warning of water quality elements, the above single-factor prediction method cannot well meet the application requirements. Summary of the Invention

[0005] The purpose of the present invention is to overcome the deficiencies of the above-mentioned prior art, and provide a method, device and storage medium for rapid synchronous prediction of multi-factors of water environment based on deep learning, so as to simplify the prediction process of multiple water quality parameters and improve the synchrony and timeliness of multi-parameter prediction.

[0006] To achieve the above purpose, the technical solution of the present invention is: In the first aspect, the present invention provides a method for synchronous rapid prediction of multi-factors of water environment pollution, including: Select the i water environment parameters affecting the required prediction , and jNon-linearly correlated hydrological, water quality and meteorological parameters x j As the input layer neurons of the neural network model; i ≥2, j ≥2; Correspond the current hydrological, water quality and meteorological parameters of the data x j with the corresponding water environment elements at the next moment to construct a training set and a test set; Input the training set and the test set into the neural network model to obtain network parameters, so as to train a synchronous prediction model for multiple water environment elements; Use the hydrological, water quality and meteorological parameters at the current moment as the input of the synchronous prediction model for multiple water environment elements; the synchronous prediction model for multiple water environment elements outputs the predicted values of multiple water environment elements at the next moment.

[0007] Optionally, the neural network model is a multi-layer neural network model, which is composed of an input layer, a hidden layer, a feedback control layer, an error input layer and an output layer in sequence; The input layer is used to receive the externally input hydrological, water quality and meteorological parameters; The hidden layer consists of a first fully connected layer and a second fully connected layer; the first fully connected layer and the second fully connected layer are responsible for feature extraction and transformation of the input data, and use an activation function to increase the non-linear representation ability of the network; The feedback control layer improves the prediction accuracy of the neural network model through a dynamic error feedback mechanism.

[0008] Optionally, the first fully connected layer is used to extract key features from the input hydrological, water quality and meteorological parameters x j and perform preliminary information fusion using Equation (1):

[0009] where, is the output predicted value of the first fully connected layer, is the weight matrix of the first fully connected layer, is the bias term of the first fully connected layer; is the activation function.

[0010] Optionally, the second fully connected layer further learns the feature relationship based on the output predicted value of the first fully connected layer:

[0011] where, is the output predicted value of the second fully connected layer, is the weight matrix of the second fully connected layer, is the bias term of the second fully connected layer.

[0012] Optionally, the feedback control layer adjusts the predicted value of the network through an additional error input layer; the additional error input layer is the output predicted value of the second fully connected layer .

[0013] Optionally, the feedback control layer adjusts the predicted value of the network through an additional error input layer, including: Error input layer calculation step, calculating the true value and the predicted value The error between:

[0014] Feedback control layer calculation step, the error is fed as an additional input into a separate feedback control layer:

[0015] where, is the output of the feedback control layer, is the weight matrix of the feedback layer, is the bias term of the feedback layer, activation function; Feedback control adjustment step, adding the output of the feedback layer to the output predicted value of the second fully connected layer:

[0016] is the output of the final hidden layer; Final calculation prediction step, the final output layer is based on the adjusted hidden state:

[0017] is the weight of the output layer, is the bias term of the output layer, is the final predicted result obtained.

[0018] Optionally, the prediction performance of the multi-factor synchronous prediction model for the water environment and the stopping condition of model training are judged by using the goodness-of-fit index R 2 and the mean relative error MRE to make a judgment.

[0019] Optionally, the prediction performance of the multi-element synchronous prediction model for the water environment and the stopping condition of model training are judged by using a goodness-of-fit index R 2 and the mean relative error MRE as follows: By using the goodness-of-fit index R 2 and MRE compare the prediction results output by the neural network model with the measured data in the test set. When and the neural network model meets the prediction accuracy, stop training. If not, adjust the bias of the neural network model using the backpropagation algorithm and continue to loop until the neural network model meets the prediction accuracy; M is the goodness-of-fit index threshold, and N is the mean relative error threshold.

[0020] In a second aspect, the present invention provides an electronic device, including a processor, a communication interface, a memory, and a communication bus. Among them, the processor, the communication interface, and the memory complete communication with each other through the communication bus; The memory is used to store a computer program; The processor, when executing the program stored on the memory, implements the steps of the multi-element synchronous rapid prediction method for water environment pollution as described in any one of the above.

[0021] In a third aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the steps of the multi-element synchronous rapid prediction method for water environment pollution as described in any one of the above.

[0022] Compared with the prior art, the beneficial effects of the present invention are as follows: The multi-element rapid synchronous prediction method for water environment based on deep learning provided by the embodiments of the present application uses multiple non-linearly related hydrological, water quality, and meteorological parameters as input parameters, and simultaneously rapidly predicts multiple water quality parameters. It has the advantages of simple and rapid implementation of multi-parameter synchronous prediction output based on multiple identical input parameters and using the same model, which not only simplifies the process of multi-parameter water quality prediction, but also improves the synchronization and timeliness of multi-parameter prediction, and provides decision-making and management for fields such as marine ecological protection and disaster prediction and early warning. Description of the Drawings

[0023] Figure 1 is a flowchart of the multi-element rapid synchronous prediction method for water environment based on deep learning provided by the embodiments of the present application; Figure 2 is a schematic diagram of the neural network structure based on deep learning; Figure 3 is a fitting diagram of synchronous prediction of three water quality parameters for a certain island based on deep learning; Figure 4 It is a long - term sequence observation graph for the synchronous prediction of three water quality parameters on a certain island based on deep learning; Figure 5 It is a schematic diagram of the composition of the electronic device provided by the embodiment of the present application. Detailed implementation manners

[0024] Embodiment: The technical solutions of the present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0025] Refer to Figure 1 As shown, the multi - element rapid synchronous prediction method for water environment based on deep learning provided by this embodiment mainly includes the following steps: 110. Select i water environment parameters that affect the required prediction, and j non - linearly related hydrological, water quality and meteorological parameters x j as the input - layer neurons of the neural network model; i ≥2, j ≥2; 120. Correlate the values of the current - data hydrological, water quality and meteorological parameters x j with the values of the water environment elements corresponding to the next moment to construct a training set and a test set; 130. Input the training set and the test set into the neural network model to obtain network parameters, so as to train a multi - element synchronous prediction model for the water environment; 140. Use the hydrological, water quality and meteorological parameters at the current moment as the input of the multi - element synchronous prediction model for the water environment, and the multi - element synchronous prediction model for the water environment outputs the predicted values of multiple water environment elements at the next moment.

[0026] Thus, this method uses multiple non - linearly related hydrological, water quality and meteorological parameters as input parameters, and simultaneously and rapidly predicts multiple water quality parameters. It has the advantages of being based on multiple identical input parameters, using the same model, and simply and rapidly realizing the synchronous prediction output of multiple parameters. It not only simplifies the process of multi - parameter water quality prediction, but also improves the synchronism and timeliness of multi - parameter prediction, providing decision - making and management for fields such as marine ecological protection and disaster prediction and early warning.

[0027] In a specific embodiment, as Figure 2As shown, the neural network model is a multi-layer neural network model, which is successively composed of an input layer, a hidden layer, a feedback control layer, an error input layer, and an output layer. Among them, the input layer is used to receive external input hydrological and water quality and meteorological parameters, such as environmental variables such as flow velocity, flow direction, pH, and chlorophyll. The hidden layer is the core of the multi-layer neural network model and is composed of a first fully connected layer and a second fully connected layer. The first fully connected layer and the second fully connected layer are responsible for feature extraction and transformation of the input data, and use activation functions (such as ReLU, Sigmoid) to increase the non-linear representation ability of the network, extract high-dimensional features for subsequent prediction; the feedback control layer mainly improves the prediction accuracy of the model through the dynamic error feedback mechanism of the error input layer, reduces the cumulative error, and makes the model more adaptable and robust in time series prediction tasks. In this way, by adopting a multi-layer neural network model and combining a dynamic error feedback control mechanism, the complex non-linear relationship between multiple key water quality parameters (such as dissolved oxygen, nutrients, oxygen demand, etc.) can be effectively learned. Different from traditional regression models, this neural network model extracts features through two fully connected networks and adjusts the prediction error through the feedback control layer, making it more stable in long-term prediction.

[0028] In a specific embodiment, the first fully connected layer mainly extracts key features from the input hydrological and water quality and meteorological parameters x j (such as hydrometeorological parameters such as temperature, humidity, and wind speed), and performs preliminary information fusion using Equation (1).

[0029]

[0030] Among them, is the output prediction value of the first fully connected layer, is the weight matrix of the first fully connected layer, x j is the input feature of the first fully connected layer, is the bias term of the first fully connected layer, is the activation function. The activation function is used to increase the non-linear ability of the network, so that the model can fit more complex patterns.

[0031] The second fully connected layer is based on the output prediction value of the first fully connected layer, and further learns more complex feature relationships to improve the expression ability of the model.

[0032]

[0033] Among them, is the output prediction value of the second fully connected layer, which is used as the input of the initial prediction value in the additional error input layer in the following feedback control layer; is the weight matrix of the second fully connected layer, is the bias term of the second fully connected layer.

[0034] In a specific embodiment, the feedback control layer adjusts the predicted value of the network through an additional error input layer to adapt to changing input patterns. The calculation process is as follows: Error input layer calculation steps, calculate the true value and the predicted value The error between them:

[0035] Feedback control layer calculation steps, the error is fed as an additional input into a separate feedback control layer:

[0036] where, is the output of the feedback control layer, is the weight matrix of the feedback layer, is the bias term of the feedback layer, is the activation function; the feedback signal is transformed through a neural network layer to extract useful information from the error; Feedback control adjustment steps, add the output of the feedback layer to the output predicted value of the second fully connected layer:

[0037] is the output of the final hidden layer; Final calculation prediction steps, the final output layer is based on the adjusted hidden state:

[0038] is the weight of the output layer, is the bias term of the output layer, is the final predicted result obtained.

[0039] In this way, through the above steps of operation, the error accumulation can be effectively reduced, and the time series prediction ability and generalization can be enhanced. In time series prediction tasks, error accumulation is a common problem, which will lead to a decrease in accuracy during long-term prediction. This neural network model adopts an error adaptive mechanism, by calculating the prediction error during the training process and feeding it back as a new input, enabling the model to dynamically adjust the weights and reduce the propagation effect of errors.

[0040] In a specific embodiment, the prediction performance of the multi-factor synchronous prediction model for water environment and the stopping condition of model training are judged by using the goodness-of-fit index R 2 and the mean relative error: The goodness-of-fit index R 2 The calculation formula is:

[0041] is the actual observed value, is the predicted value, is the average of the true values. The mean relative error MRE The calculation formula is:

[0042] is the actual observed value, is the predicted value, and m is the number of test samples.

[0043] By using the goodness-of-fit index R 2 and MRE, the prediction results output by the neural network model are compared with the measured data in the test set. When and the model meets the prediction accuracy and stops training. At this time, the parameters of the model are the optimal parameters of the neural network model structure. If not satisfied, the bias of the model is adjusted using the backpropagation algorithm and the loop continues. In this way, by setting an early stopping mechanism (R²>0.85 and MRE<30%), overfitting is prevented and the stability and reliability of the prediction are improved.

[0044] The following further illustrates the multi-factor rapid synchronous prediction method for water environment based on deep learning provided in this embodiment in combination with an application scenario example: Step 1: Select the environmental parameters that affect the water environment factors to be predicted on a certain island. And use the characteristic parameters as the input layer neurons of the multi-factor synchronous prediction neural network model for water environment. The relevant parameters used in the present invention include temperature, salinity, chlorophyll, pH, flow velocity, flow direction, turbidity, wind speed, and wind direction.

[0045] Step 2: Correlate the characteristic parameter values in the historical dataset with the values of DO (dissolved oxygen), BOD (biochemical oxygen demand), and COD (chemical oxygen demand) corresponding to the next moment to construct a training set and a test set.

[0046] Step 3: Input the training set and the test set into the neural network model to obtain network parameters. According to the trained model, input the characteristic parameters at the current moment, and the predicted values of the three water environment factors of DO, BOD, and COD at the next moment can be obtained.

[0047] The model parameters of the deep learning network used are shown in Table 1: Table 1. Model parameters of the deep neural network

[0048] Figure 3 It is a fitting graph of three water quality parameters, DO, BOD, and COD, for a certain island. By observing Figure 3 , it is found that the model has a high fitting degree for the three water quality parameters, and the R2 can reach above 0.9 for all, and the MRE is less than 30%.

[0049] Figure 4 It is a comparison graph of the long-term sequence observation values and predicted values of three water quality parameters, DO, BOD, and COD, for a certain island. By observing Figure 4 , it is found that the model has a high accuracy for synchronously predicting multiple water quality parameters. It can be applied to the synchronous prediction of multiple elements in the water environment for a long time series.

[0050] In summary, compared with the prior art, the method for fast synchronous prediction of multiple water environment elements based on deep learning provided in this embodiment has the following technical advantages: (1). The combination of deep learning and feedback control improves the water quality prediction accuracy The present invention adopts a multi-layer neural network (MLP) and combines a dynamic error feedback control mechanism, which can effectively learn the complex non-linear relationships among multiple key water quality parameters (such as: dissolved oxygen, nutrients, oxygen demand, etc.). Different from the traditional regression model, this model extracts features through two fully connected layers and adjusts the prediction error through the feedback control layer to make it more stable in long-term prediction.

[0051] (2). Reducing error accumulation and enhancing the time series prediction ability The present invention can effectively reduce error accumulation and enhance the time series prediction ability and generalization. In time series prediction tasks, error accumulation is a common problem, which will lead to a decrease in accuracy during long-term prediction. This model adopts an error adaptive mechanism, calculates the prediction error during the training process and feeds it back as a new input, so that the model can dynamically adjust the weights and reduce the propagation effect of errors. At the same time, an early stopping mechanism (R²>0.85 and MRE<30%) is set to prevent overfitting and improve the stability and reliability of the prediction.

[0052] (3). Efficient multi-objective prediction, suitable for water quality monitoring and real-time analysis The present invention supports predicting multiple water quality indicators simultaneously using the same input feature parameters, eliminating the need to separately establish multiple models, thereby improving the computational efficiency and data utilization rate. It is particularly suitable for predicting the changes of multiple water quality parameters using low-cost hydrometeorological parameters and providing comprehensive early warnings in a timely manner. The prediction results of the model can be evaluated through visual analysis (scatter plots, time series curves). Finally, the model can export the prediction results as CSV files for easy storage and analysis, and is applicable to application scenarios such as water environment management, pollution monitoring, and ecological assessment.

[0053] As Figure 5 shown, an embodiment of the present application further provides an electronic device, including a processor 111, a communication interface 112, a memory 113, and a communication bus 114. Among them, the processor 111, the communication interface 112, and the memory 113 communicate with each other through the communication bus 114; the memory 113 is used to store computer programs; when the processor 111 executes the programs stored on the memory 113, it implements the steps of the method for rapid synchronous prediction of multiple water environment elements based on deep learning provided by any one of the foregoing method embodiments.

[0054] An embodiment of the present application further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the steps of the method for rapid synchronous prediction of multiple water environment elements based on deep learning provided by any one of the foregoing method embodiments.

[0055] The above embodiments are only used to illustrate the technical concept and features of the present invention, and the purpose is to enable those of ordinary skill in the art to understand the content of the present invention and implement it accordingly. However, the protection scope of the present invention cannot be limited thereby. Any equivalent changes or modifications made according to the essence of the content of the present invention should be covered within the protection scope of the present invention.

Claims

1. A method for rapid simultaneous prediction of multiple factors of water environment pollution, characterized in that: include: Select the factors that affect the forecast you want. i Water environment parameters , and j Nonlinearly related hydrological, water quality and meteorological parameters x j As the input layer neurons of the neural network model; i ≥2, j ≥2; Current data on hydrological, water quality and meteorological parameters x j The value of and the water environment element corresponding to the next moment The values ​​of are matched to construct the training set and the test set; The training set and the test set are input into the neural network model to obtain the network parameters, so as to train and obtain the water environment multi-factor synchronous prediction model; The hydrological, water quality and meteorological parameters at the current moment are used as inputs of the water environment multi-factor synchronous prediction model; the water environment multi-factor synchronous prediction model outputs the predicted values ​​of multiple water environment elements at the next moment.

2. The method for rapid simultaneous prediction of multiple factors of water environment pollution according to claim 1, characterized in that: The neural network model is a multi-layer neural network model, which is composed of an input layer, a hidden layer, a feedback control layer, an error input layer and an output layer in sequence; The input layer is used to receive external input hydrological, water quality and meteorological parameters; The hidden layer is composed of a first fully connected layer and a second fully connected layer; the first fully connected layer and the second fully connected layer are responsible for extracting and transforming features of input data, and using activation functions to increase the nonlinear representation capability of the network; The feedback control layer improves the prediction accuracy of the neural network model through a dynamic error feedback mechanism.

3. The method for rapid simultaneous prediction of multiple factors of water environment pollution according to claim 2, characterized in that: The first fully connected layer is used to obtain the input hydrological, water quality and meteorological parameters x j Extract key features and use formula (1) to perform preliminary information fusion: ； in, is the output prediction value of the first fully connected layer, is the weight matrix of the first fully connected layer, is the bias term of the first fully connected layer; is the activation function.

4. The method for rapid simultaneous prediction of multiple factors of water environment pollution according to claim 3, characterized in that: The output prediction value of the second fully connected layer at the first fully connected layer Based on this, we can further learn the feature relationship: ； in, is the output prediction value of the second fully connected layer, is the weight matrix of the second fully connected layer, is the bias term of the second fully connected layer.

5. The method for rapid simultaneous prediction of multiple factors of water environment pollution according to claim 4, characterized in that: The feedback control layer adjusts the prediction value of the network through an additional error input layer; the additional error input layer is the output prediction value of the second fully connected layer .

6. The method for rapid simultaneous prediction of multiple factors of water environment pollution according to claim 5, characterized in that: The feedback control layer adjusts the network's predictions through an additional error input layer including: Error input layer calculation steps to calculate the true value and predicted values The error between: ； Feedback control layer calculation steps, error As an additional input is fed into a separate feedback control layer: ； in, is the output of the feedback control layer, is the weight matrix of the feedback layer, is the bias term of the feedback layer, Activation function; The feedback control adjustment step is to adjust the output of the feedback layer The output prediction value of the second fully connected layer To sum: ； is the output of the final hidden layer; Finally, the prediction step is calculated, and the final output layer is based on the adjusted hidden state: ； is the weight of the output layer, is the bias term of the output layer, The final prediction result.

7. The method for rapid simultaneous prediction of multiple factors of water environment pollution according to claim 1, characterized in that: The prediction performance of the water environment multi-factor synchronous prediction model and the stopping condition of model training adopt the fitting index R 2 and the mean relative error MRE To make a judgment.

8. The method for rapid simultaneous prediction of multiple factors of water environment pollution according to claim 7, characterized in that: The prediction performance of the water environment multi-factor synchronous prediction model and the stopping condition of model training adopt the fitting index R 2 and the mean relative error MRE To make judgments, including: Through the goodness of fit index R 2 and MRE The prediction results output by the neural network model are compared with the measured data in the test set. and When the neural network model meets the prediction accuracy, the training is stopped. If it does not meet the accuracy, the back propagation algorithm is used to adjust the deviation of the neural network model, and the cycle continues until the neural network model meets the prediction accuracy. M is the fitting index threshold, and N is the average relative error threshold.

9. An electronic device, characterized in that: It includes a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus; Memory, used to store computer programs; The processor is used to implement the steps of the method for synchronously and rapidly predicting multiple factors of water environment pollution as described in any one of claims 1 to 8 when executing the program stored in the memory.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for rapid simultaneous prediction of multiple factors of water environment pollution according to any one of claims 1 to 8 are implemented.

Images