A method for constructing groundwater level monitoring and early warning model based on big data

Abstract

The present invention relates to the field of data processing technology, and in particular to a method for constructing a groundwater level monitoring and early warning model based on big data. The method for constructing a groundwater level monitoring and early warning model based on big data provided by the present invention uses a first training sample set to train a generative adversarial network including a Lagrangian unit and a time series model to expand the first training sample set, thereby improving the extensiveness of the model, the quality and authenticity of the sample; and based on the neural network trained by the bioelectric current dynamic balance method, features are extracted from the training samples in the target training sample set, thereby improving the efficiency and accuracy of feature extraction; finally, a classifier is trained to improve the adaptability and robustness of the model, thereby improving the accuracy and efficiency of groundwater level monitoring and early warning.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a method for constructing a groundwater level monitoring and early warning model based on big data. Background Art

[0002] With the impact of global climate change and human activities, the management and protection of groundwater resources have become increasingly important. Changes in groundwater levels are directly related to the sustainable use of water resources, agricultural irrigation, urban water supply security, and the balance of the ecological environment. Especially in areas where extreme climatic conditions such as droughts and floods occur frequently, an effective groundwater level monitoring and early warning system is essential to prevent natural disasters, reduce losses, and protect people's lives.

[0003] However, traditional groundwater level monitoring methods often rely on scattered monitoring sites or regular manual measurements, which are not only time-consuming and labor-intensive, but also difficult to achieve real-time monitoring and large-scale data collection. In addition, due to the complexity and dynamics of groundwater systems, traditional methods face many challenges in data processing and analysis, and prediction model construction, such as data discontinuity, monitoring point limitations, and insufficient model accuracy and generalization capabilities. Summary of the invention

[0004] In view of this, the object of the present invention is to provide a groundwater level monitoring and early warning method based on big data.

[0005] In a first aspect, an embodiment of the present invention provides a method for constructing a groundwater level monitoring and early warning model based on big data, the method is as follows:

[0006] Obtaining an original training sample set, and labeling the original training samples in the original training sample set to obtain a first training sample set; wherein the original training samples are obtained by collecting data through ground monitoring stations, groundwater level sensor networks, and auxiliary remote sensing equipment; the original training samples include multiple parameters, and the parameters include at least groundwater level depth, groundwater level change rate, ground temperature, ground humidity, rainfall, and groundwater temperature;

[0007] Using the first training sample set to train a generative adversarial network including a Lagrangian unit and a time series model to expand the first training sample set to obtain a target training sample set;

[0008] A pre-built neural network is used to extract features from the training samples in the target training sample set to obtain a target feature set; the neural network is trained based on the bioelectrical dynamic balance method;

[0009] The target feature set is used to train the classifier, and the groundwater level monitoring and early warning results are output to obtain a groundwater level monitoring and early warning model; wherein the groundwater level monitoring and early warning model is used to monitor and warn the groundwater level.

[0010] Combined with the first aspect, the generative adversarial network includes a generator network, a discriminator network and a time series model; the generator network is integrated with a Lagrangian unit;

[0011] The step of applying the first training sample set to train a generative adversarial network including a Lagrangian unit and a time series model to expand the first training sample set to obtain a target training sample set includes:

[0012] Using the first training sample set to train the generator network, the time sequence model and the discriminator network in sequence to generate new training samples;

[0013] The Lagrangian unit is used to optimize the newly added training samples, and the quality of the optimized newly added training samples is evaluated;

[0014] Determine the newly added training samples whose quality is greater than a preset quality assessment threshold as target newly added training samples;

[0015] The newly added training samples are added to the first training sample set to obtain an expanded target training sample set.

[0016] In combination with the first aspect, the method for constructing a neural network includes:

[0017] Initialize the network parameters, weights and biases of the initial neural network;

[0018] Assigning the network parameters of the initial neural network to initial potential values;

[0019] Inputting the initial potential value into the initial neural network to obtain a first output result;

[0020] The initial neural network is optimized according to the first output result until a neural network that meets preset conditions is obtained.

[0021] In combination with the first aspect, the step of optimizing the initial neural network according to the first output result until a neural network that meets preset conditions is obtained includes:

[0022] Calculate the loss function of the neural network according to the first output result;

[0023] According to the loss function, the potential adjustment amount is calculated;

[0024] According to the potential adjustment amount and the preset adaptive factor, the network parameters, weights and biases of the neural network are updated until a neural network that meets the preset conditions is obtained.

[0025] In combination with the first aspect, the classifier includes a multi-layer neural network, each layer of the neural network includes a plurality of neurons, and the plurality of neurons are cross-connected;

[0026] The steps of using the target feature set to train the classifier and obtain the groundwater level monitoring and early warning model include:

[0027] The target feature set is input into the classifier, and the parameters of the classifier are optimized by the proximal gradient descent method set simulated annealing ensemble method;

[0028] The multi-layer neural network is trained in sequence based on the target feature set, and the groundwater level monitoring and early warning results are output to obtain the groundwater level monitoring and early warning model.

[0029] In combination with the first aspect, the step of using the target feature set to train the classifier includes:

[0030] Initialize the classifier’s network parameters, weights, and biases;

[0031] Input the target feature set into the classifier and output a second output result;

[0032] Calculate a second loss function of the classifier according to the second output result;

[0033] The network parameters, weights, biases and current temperature parameters of the classifier are adaptively updated according to the second loss function, the proximal term and the current temperature parameter until a classifier that meets the preset conditions is obtained.

[0034] In a second aspect, the present application provides a groundwater level monitoring and early warning method based on big data, the method comprising:

[0035] Obtain actual sample sets collected by monitoring facilities;

[0036] Preprocess the actual sample set to obtain the target sample set;

[0037] Input the target sample set into the pre-trained groundwater level monitoring and early warning model, and output the groundwater level monitoring and early warning results;

[0038] Among them, the pre-trained groundwater level monitoring and early warning model is constructed based on the above method.

[0039] In a third aspect, the present application provides a device for constructing a groundwater level monitoring and early warning model based on big data, the device comprising:

[0040] The acquisition module is used to acquire an original training sample set and annotate the original training samples in the original training sample set to obtain a first training sample set; wherein the original training samples are obtained by collecting data through ground monitoring stations, groundwater level sensor networks, and auxiliary remote sensing equipment; the original training samples include multiple parameters, and the parameters include at least groundwater level depth, groundwater level change rate, ground temperature, ground humidity, rainfall, and groundwater temperature;

[0041] An expansion module, used for applying the first training sample set to train a generative adversarial network including a Lagrangian unit and a time series model, so as to expand the first training sample set and obtain a target training sample set;

[0042] A feature extraction module is used to extract features from training samples in a target training sample set using a pre-built neural network to obtain a target feature set; the neural network is trained based on a bioelectrical dynamic balance method;

[0043] The prediction module is used to train the classifier using the target feature set, output the groundwater level monitoring and early warning results, and obtain the groundwater level monitoring and early warning model; wherein the groundwater level monitoring and early warning model is used to monitor and warn the groundwater level.

[0044] In a fourth aspect, the present application provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method as described above.

[0045] In a fifth aspect, the present application provides a storage medium, in which computer program instructions are stored. When the computer program instructions are read and executed by a processor, the above method is executed.

[0046] The embodiments of the present invention bring the following beneficial effects: the present invention provides a method for constructing a groundwater level monitoring and early warning model based on big data, trains a generative adversarial network including a Lagrangian unit and a time series model with a first training sample set to expand the first training sample set, thereby improving the extensiveness of the model, the quality and authenticity of the samples; and extracts features of the training samples in the target training sample set based on a neural network trained by a bioelectric current dynamic balance method, thereby improving the efficiency and accuracy of feature extraction; finally, trains a classifier to improve the adaptability and robustness of the model, thereby improving the accuracy and efficiency of groundwater level monitoring and early warning.

[0047] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description, claims and drawings.

[0048] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without paying any creative work.

[0050] Figure 1 A schematic flow chart of a method for constructing a groundwater level monitoring and early warning model based on big data provided by an embodiment of the present invention;

[0051] Figure 2 A flow chart of a groundwater level monitoring and early warning method based on big data provided by an embodiment of the present invention;

[0052] Figure 3 A schematic diagram of the structure of a device for building a groundwater level monitoring and early warning model based on big data provided by an embodiment of the present invention;

[0053] Figure 4 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0055] In order to facilitate the understanding of this embodiment, the application scenarios and design concepts of the embodiments of the present application are briefly introduced below.

[0056] Groundwater level monitoring and early warning methods are widely used in the geological field. The identification models in the methods provided by the existing technologies have low recognition accuracy due to the limitations of data expansion methods and the limited generalization ability of model optimization and parameter adjustment methods.

[0057] Based on this, an embodiment of the present application provides a method for constructing a groundwater level monitoring and early warning model based on big data to improve the accuracy of groundwater level monitoring and early warning, thereby improving prediction efficiency.

[0058] Example 1

[0059] The present application provides a method for constructing a groundwater level monitoring and early warning model based on big data, combining Figure 1 As shown, the method includes:

[0060] S110, obtaining an original training sample set, and annotating the original training samples in the original training sample set to obtain a first training sample set; wherein the original training samples are obtained by collecting data through ground monitoring stations, groundwater level sensor networks, and auxiliary remote sensing equipment; the original training samples include multiple parameters, and the parameters include at least groundwater level depth, groundwater level change rate, ground temperature, ground humidity, rainfall, and groundwater temperature.

[0061] S120, using the first training sample set to train a generative adversarial network including a Lagrangian unit and a time series model to expand the first training sample set to obtain a target training sample set.

[0062] S130, applying a pre-built neural network to extract features from training samples in a target training sample set to obtain a target feature set; the neural network is trained based on a bioelectrical dynamic balance method.

[0063] S140, using the target feature set to train the classifier, outputting the groundwater level monitoring and early warning results, and obtaining a groundwater level monitoring and early warning model; wherein the groundwater level monitoring and early warning model is used to monitor and warn the groundwater level.

[0064] The method for constructing a groundwater level monitoring and early warning model provided in the present application labels each original training sample to obtain a first training sample set, trains a generative adversarial network including a Lagrangian unit and a time series model to generate new training samples that combine time series factors and have high quality and authenticity to expand the first training sample set to obtain a target training sample set, thereby increasing the sample size of the training model to improve the generalization ability and accuracy of the model, and finally trains a neural network and classifier with high accuracy to obtain a groundwater level monitoring and early warning model that can accurately predict groundwater level demand.

[0065] The original training sample set in step S110 is obtained by collecting data from ground monitoring stations, groundwater level sensor networks, and auxiliary remote sensing equipment. The sensor system includes but is not limited to ground monitoring stations and groundwater level sensor networks. The corresponding sensors are configured according to the parameters to be monitored and the application scenarios. Each original training sample includes multiple parameters. For example, sample A includes:

[0066] a x : depth of groundwater level;

[0067] a y : Groundwater level change rate;

[0068] a z : ground temperature;

[0069] a α : ground humidity;

[0070] a β : rainfall;

[0071] a γ : soil moisture content;

[0072] a δ : Geographic location information (latitude and longitude);

[0073] a ∈ : Collection time;

[0074] a b : Soil electrical conductivity;

[0075] a η : Groundwater temperature.

[0076] After the original training sample is obtained, the original training sample is labeled, that is, a label is added to the original training sample to obtain a first training sample.

[0077] In this embodiment, the label categories include: Category 1: normal water level; Category 2: low / high water level.

[0078] Combining the above examples, after labeling the original training samples, we can obtain: the first training sample A' is:

[0079] a x =5m, a y =0.01m / s, a z =25℃、a α =60%, a β =5mm, a γ =20%, a δ =(30.123,120.123),a ∈ =20230205,a b =0.8, a η =22°C, and the first training sample A' belongs to the category: Category 2 - low / high water level.

[0080] It is understandable that in actual application, the parameters of the sample can be increased or decreased according to the application environment and scenario. Usually, the number of sample parameters during use is greater than 10.

[0081] It is worth noting that before labeling the original training samples, it is usually necessary to preprocess the original training samples to standardize each original training sample to ensure that the input original training samples are on the same scale. The standardization preprocessing process is:

[0082]

[0083] Among them, bx′ i is the original training sample after standardization, bx i is the original training sample, μ bx and σ bx are the mean and standard deviation of the original training samples respectively.

[0084] In combination with the first aspect, the generative adversarial network includes a generator network and a discriminator network; the generator network integrates a Lagrangian unit and a time series model. Step S120 uses the first training sample set to train the generative adversarial network including the Lagrangian unit and the time series model to expand the first training sample set to obtain a target training sample set, including:

[0085] S121, using the first training sample set to train the generator network and the discriminator network in sequence to generate new training samples.

[0086] In this embodiment, the generative adversarial network includes a generator network, a discriminator network and a time series model. The generative adversarial network that integrates stream of consciousness simulation is used to expand the samples, and can generate new water level monitoring and early warning scenarios by simulating the human stream of consciousness.

[0087] Specific:

[0088] First, define G as the generator network and D as the discriminator network;

[0089] Secondly, a generator network is constructed to simulate the human consciousness stream, which generates new additional training samples by receiving random noise and the existing first training samples. Specifically: Let cz i is the input random noise vector, cx is the first training sample. The goal of the generator network G is to learn a mapping: G(cz i ,cx)→cg i ;

[0090] Among them, cg i is the newly generated training sample.

[0091] The objective function of the generator network G can be expressed as:

[0092]

[0093] Among them, D is the discriminator network, and V(D,G) represents the value function between the generator network G and the discriminator network D.

[0094] Furthermore, for min G max D V(D, G) The calculation of the term can be expressed as:

[0095]

[0096] Where E represents the expected sign, cx~p data (cx) indicates that the first training sample follows its own specific distribution, D() is the discriminator symbol, TSM() is the function of the time series model, H is the historical sample, cg′ i is the newly added training sample generated after considering the influence of time series; t refers to time point t, that is, the tth iteration; ct i is the output of the time series model, N is the number of the first training samples, cx i is the i-th first training sample; is the time series influence term; ||||2 is the L2 norm symbol; λ is the regularization parameter. In this embodiment, λ=0.1.

[0097] It is worth mentioning that the time series model is integrated in the generator network G. During the training process of the generator network G, the time series analysis of historical samples and the currently generated new training samples is integrated by introducing time series influence terms to achieve fine control over the dynamic evolution process of the generated samples.

[0098] Specifically, a time series model TSM is introduced to analyze the historical sample H and the current generated sample cg i Time correlation. Let t be the current time point, the output ct of TSM i represents the influencing factor at time point t. Then the calculation formula of TSM can be expressed as:

[0099] ct i =TSM(H,cg i , t) = α t ×H(t)+β t ×cg i ;

[0100] Among them, α t and β t is a time-dependent weight coefficient used to balance the impact of historical samples and current samples; H(t) is the historical sample at the tth iteration, cg i To add new training samples.

[0101] Integrate ct in the generator network Gi , adjust the generation process to reflect the influence of time series, which can be expressed as:

[0102] G(cz i ,cx,ct i )→cg′ i ;

[0103] Among them, cz i is the input random noise vector, cx is the first training sample, ct i is the output of the time series model, cg′ i It is a new training sample generated after considering the influence of time series.

[0104] Next, a discriminator network D is constructed to distinguish the generated new training samples from the first training samples, so as to train the generator network G to better simulate the real scene. Specifically, the goal of the discriminator network D is to distinguish the new training samples cg i And the first training sample cx. Its loss function is:

[0105]

[0106] Among them, L D is the loss function of the discriminator network D, cg i ~p g (cg i ) indicates that the newly added training samples obey their own specific distribution.

[0107] Furthermore, for L D In The calculation of the term can be expressed as:

[0108]

[0109] Where M is the number of generated training samples, cz j is the jth random noise vector, D(G(cz j , cx)) is the output of the discriminator when the generator inputs the jth random noise vector, and λ is the regularization parameter.

[0110] S122, optimizing the newly added training samples by using Lagrangian units, and performing quality assessment on the optimized newly added training samples.

[0111] In this embodiment, the Lagrangian unit is integrated into the generator network G, and the generation process is guided by physical principles to enhance the physical accuracy and diversity of the generated new training samples. At the same time, the Lagrangian equation is used to model and optimize the dynamic changes of the new training samples to ensure that the new training samples meet the constraints of physics and the real world.

[0112] Specifically, the generated new training sample cg i As the state of the dynamic system, define the Lagrangian function To describe its dynamic behavior, is the time derivative of the state. Then the Lagrange equation is:

[0113]

[0114] In this embodiment, the Lagrangian function Ls is set to Ls=T(cg i )-V(cg i ), where T is the kinetic energy of the dynamic system and V is the potential energy.

[0115] Among them, the kinetic energy function T(cg i ) can be calculated as:

[0116]

[0117] Herein, m represents a virtual quality parameter, and its value is set according to the actual application scenario. In this embodiment, m=1.

[0118] Furthermore, the potential energy function V(cg i ) can be calculated as:

[0119]

[0120] Among them, k is a parameter, cg target It is the state of the target generating new training samples. In this embodiment, k=0.1.

[0121] During stochastic gradient optimization, for each newly generated training sample cg i , the optimization method using stochastic gradient descent can be expressed as:

[0122]

[0123] in, is the newly added training sample at the t+1th iteration, is the newly added training sample at the tth iteration, is the gradient of the loss function for the newly added training sample, For cg i The first-order derivative of , η is the learning rate, in this embodiment, η = 0.01.

[0124] Furthermore, the gradient The calculation can be expressed as:

[0125]

[0126] In the specific calculation, considering Ls = TV, it can be further expanded as follows:

[0127]

[0128] in, It's cg i The acceleration can be obtained by i The time second derivative of is calculated.

[0129] Furthermore, the above steps are iterated until the generator network G can generate high-quality and diverse new training samples. Specifically, for the joint training of the generator network G and the discriminator network D, the following steps are performed alternately:

[0130] Fix G, update D:

[0131] Fix D, update G:

[0132] Among them, D (t) is the discriminator function of the tth iteration, G (t) is the generator function of the tth iteration, D (t+1) is the discriminator function of the t+1th iteration, G (t+1) is the generator function of the t+1th iteration, α nc and β nc are the learning rate of the discriminator and the learning rate of the generator, respectively.

[0133] Preferably, α nc =0.0002; β nc =0.0001.

[0134] S123: Determine the newly added training samples whose quality is greater than a preset quality evaluation threshold as target newly added training samples.

[0135] Finally, the quality of the generated new training data is evaluated to obtain high-quality target new training samples. Specifically:

[0136] Let CQ(cg i ) is a quality evaluation function used to evaluate the generated new training samples cg i The average quality of all newly generated training samples is calculated. The calculation method can be expressed as:

[0137]

[0138] Among them, CQ is the quality score of the newly added training samples, is the average quality of the newly added training samples.

[0139] Afterwards, the newly added training samples whose quality scores are greater than the preset quality assessment threshold are added to obtain a plurality of high-quality target newly added training samples that can better simulate the real data.

[0140] S124, adding the newly added training samples to the first training sample set to obtain an expanded target training sample set.

[0141] In this embodiment, a generative adversarial network including a Lagrangian unit and a time series model is trained to expand the first training sample set to obtain a target training sample set with a larger number of training samples, so as to improve the data generalization ability of the training samples, thereby improving the adaptability and generalization ability of the groundwater level monitoring and early warning model to unseen data, and further improving the prediction accuracy of the groundwater level monitoring and early warning model.

[0142] Afterwards, the target training sample set is used to train the neural network trained based on the biocurrent dynamic balance method to extract features from the training samples in the target training sample set. Since the neural network is trained based on the biocurrent dynamic balance method by dynamically adjusting the parameters of the neural network, it has better network adaptability and feature extraction efficiency compared to the neural network optimized by the traditional optimization method in the prior art, so that the training samples can be better extracted, thereby improving the prediction accuracy of the groundwater level monitoring and early warning model. In this embodiment, a biocurrent optimization algorithm is used to simulate the process of biocurrent dynamic balance to dynamically adjust the weights and biases of the neural network. At the same time, the biocurrent optimization algorithm combines the feedback of the neural network performance to dynamically adjust the parameter update strategy, similar to how biological cells adjust their potential according to the external environment. In addition, the biocurrent optimization algorithm maintains the diversity of parameter updates to avoid falling into local optimal solutions, similar to how organisms maintain potential diversity in different environments.

[0143] In combination with the first aspect, the method for constructing the neural network includes:

[0144] S210, initializing network parameters, weights and biases of the initial neural network.

[0145] Specifically, the parameters of the neural network θ = {w, b} are initialized using a standard normal distribution and can be expressed as:

[0146] w~N(0,1), b~N(0,1);

[0147] Among them, ~ represents the symbol that obeys a certain distribution, and N(0,1) represents the normal distribution in the interval (0,1).

[0148] S220, assigning network parameters of the initial neural network to initial potential values.

[0149] Each parameter is assigned an initial potential value to simulate the potential difference between the inside and outside of the cell. Specifically, each parameter is assigned an initial potential value E, the initial potential E i The assignment formula is:

[0150] E i =V0·tanh(w i +b i );

[0151] Among them, E i is the initial potential, V0 is the initial reference value of the potential, and the tanh function simulates the nonlinear response characteristics of the biological cell potential to ensure that the potential value changes within a certain range; w i is the weight of the i-th parameter, b i is the bias of the i-th parameter.

[0152] In this embodiment, the normal cell membrane potential is assumed, V0 = 70 mV.

[0153] S230, inputting the initial potential value into the initial neural network to obtain a first output result.

[0154] Use the current parameters to perform a forward propagation and calculate the first output result. Specifically, the forward propagation of the initial neural network can be expressed as:

[0155] a=Sig(w×x+b);

[0156] Among them, Sig() is the Sigmoid activation function, x is the input training sample, w is the weight parameter, and b is the bias parameter.

[0157] S240, optimizing the initial neural network according to the first output result until a neural network that meets preset conditions is obtained, specifically including:

[0158] S241, calculating the loss function of the neural network according to the first output result.

[0159] The loss function is calculated based on the first output result of the initial neural network and the actual value. Specifically, the loss function L uses the mean square error, which can be expressed as:

[0160]

[0161] Among them, y i and are the true value and the predicted value respectively, and R(E) is the regularization term.

[0162] S242, calculating the potential adjustment amount according to the loss function.

[0163] Specifically, the influence of environmental changes on cell points is simulated according to the loss function, so as to calculate the point adjustment amount. Specifically, the potential adjustment amount ΔE reflects the change of the loss function and can be expressed as:

[0164]

[0165] in, is the gradient of the loss function L, and α is the learning rate, which can be dynamically adjusted according to the training speed and the performance of the neural network. By dynamically adjusting the learning rate, different learning measurements can be taken at different training stages to achieve rapid convergence in the early stage and detailed optimization in the later stage, so as to improve the overall training effect of the neural network.

[0166] Specifically, define a learning rate adjustment factor λ lr , according to the change rate of the loss function and the performance index P, it can be expressed as:

[0167]

[0168] Among them, σ lr is the sensitivity coefficient of the adjustment factor, P lr is the set performance threshold.

[0169] Furthermore, according to the learning rate adjustment factor λ lr The way to adjust the learning rate α can be expressed as:

[0170] α new =α base ×λ lr ;

[0171] Among them, α base is the base learning rate, α new is the adjusted learning rate.

[0172] S243, updating the network parameters, weights and biases of the neural network according to the potential adjustment amount and the preset adaptive factor, until a neural network that meets the preset conditions is obtained.

[0173] According to the environmental feedback, the potential value of each parameter is adjusted. Specifically, the method of updating the potential value according to ΔE can be expressed as:

[0174]

[0175] in, is the electrical position before adjustment, is the adjusted electrical position.

[0176] According to the adjusted potential value, the network parameters are updated. Specifically, the parameter update adopts the potential influence mechanism, which can be expressed as:

[0177]

[0178]

[0179] Among them, f() is the potential influence function, γ is the parameter adjustment coefficient, λ s is the adaptive factor; R(E) is the regularization term, η is the coefficient that controls the influence of regularization, is the updated weight parameter, is the weight parameter before updating, To find the sign of the partial derivative, b i is the i-th bias parameter.

[0180] Furthermore, the regularization term R(E) is based on the potential value of the parameter and aims to control the complexity of the weight. Through this regularization, the model will take into account the balance of potential while optimizing the parameters, thereby preventing overfitting and improving the generalization ability of the model. Specifically, the calculation method of the regularization term R(E) can be expressed as:

[0181]

[0182] Among them, ρ is the regularization coefficient, which controls the strength of the regularization term; E i is the initial potential.

[0183] Adaptive factor λ s The purpose is to automatically adjust the sensitivity of the potential regulation according to the training process of the neural network to improve the efficiency and adaptability of the algorithm, make the parameter update more adaptive, and be able to quickly find the optimization direction in the early stage of training, and make detailed adjustments in the later stage of training to prevent over-regulation. s The change in loss function depending on several consecutive iterations can be expressed as:

[0184] λ s =exp(-β d ΔL avg );

[0185] Where, ΔL avg is the average value of the difference between the loss functions in recent iterations, β d is the attenuation coefficient.

[0186] The potential influence function f() is set as a nonlinear function, which can be expressed as: f(xs) = log(1+e xs );

[0187] Among them, xs is the independent variable of the potential influence function f().

[0188] Use the updated parameters for forward propagation and recalculate the loss function of the neural network to evaluate the network performance, making dynamic adjustments when the preset conditions cannot be met.

[0189] After that, random factors are introduced to ensure that the parameter update will not fall into the local optimal measurement. Specifically, in order to prevent the parameters from converging to the local optimum, a random perturbation δ is introduced, and the calculation method of the parameters can be expressed as:

[0190]

[0191]

[0192] Furthermore, the calculation method of the random perturbation δ can be expressed as:

[0193] δ=ξ×tandn();

[0194] in, To introduce random factors, the updated weight parameters are: To introduce random factors into the updated bias parameters, ξ is the perturbation intensity coefficient, and randn() generates random numbers from a standard normal distribution. This random perturbation increases the algorithm's exploration ability and avoids the parameter update process being too single and deterministic.

[0195] By repeating the above steps S230-S243 until a neural network that meets the preset conditions is obtained, as an implementable method, the preset condition is to reach a preset maximum number of iterations. As another implementable method, the loss function of the neural model is less than a preset threshold. It can be selected according to actual use requirements and is not limited here.

[0196] After expanding and extracting features from the first training sample set, the target training sample set after feature extraction is obtained. The target feature set is input into the classifier for classification. The classifier is a multi-level neural network structure, each level contains multiple high-order neurons, which can handle more complex nonlinear relationships. At the same time, cross-layer connections are introduced in the classifier to allow neurons in different layers to directly influence each other, so as to enhance the network's expression ability.

[0197] In addition, the classifier provided in this application adopts an improved proximal gradient descent algorithm to optimize the weights of the neural network, and introduces a proximal term algorithm to more effectively process large-scale and coefficient data sets, which is beneficial to improving training efficiency and stability.

[0198] Furthermore, a simulated annealing strategy is integrated to avoid falling into the local minimum during the optimization process. The simulated annealing strategy helps the network explore a wider parameter space by gradually reducing the temperature parameter during the training process, and encourages the neural network to explore new solutions by temporarily increasing the temperature during the training phase, thereby improving the generalization ability of the classifier.

[0199] S130 uses the target feature set to train the classifier to obtain a groundwater level monitoring and early warning model, including:

[0200] S131, inputting the target feature set into the classifier, and optimizing the parameters of the classifier by using the simulated annealing ensemble method of the proximal gradient descent method set;

[0201] S132, training the multi-layer neural network in sequence based on the target feature set, outputting the groundwater level monitoring and early warning results, and obtaining the groundwater level monitoring and early warning model.

[0202] The training process of the classifier is as follows:

[0203] S310, initializing the network parameters, weights and biases of the classifier.

[0204] Specifically, the improved high-order neural network architecture used in the present invention consists of L layers, each layer containing n l neurons, where l = 1, 2, ..., L; where L is the number of layers in the neural network; n l is the number of neurons in the lth layer.

[0205] In this embodiment, the neuron output e in the lth layer lx It can be expressed as:

[0206]

[0207] Among them, Sig() is the Sigmoid activation function, w lj is the weight, b l is the bias term, e (l-1)j is the output of layer l-1.

[0208] S320, input the target feature set into the classifier, and output a second output result.

[0209] The output of the neuron can be further refined after square calculation as follows:

[0210]

[0211] The squares of weights and bias terms are added to the calculation to emphasize the importance of higher-order relationships.

[0212] The improved high-order neural network contains cross-layer connections. For cross-layer connections, a cross-connection matrix C is defined lk , where l and k represent different layers. The output of the cross-layer connection can be expressed as:

[0213]

[0214] Among them, C lkis the jth cross-connection matrix, e kj is the output of the kth layer, n k is the number of neurons in the kth layer, is the output of the cross-layer connection.

[0215] For the output of the cross-layer connection After square calculation, it can be specifically expressed as:

[0216]

[0217] In this way, the expressiveness of the model is enhanced by the square term.

[0218] S330, calculating a second loss function of the classifier according to the second output result.

[0219] S340, adaptively updating the network parameters, weights, biases and current temperature parameters of the classifier according to the second loss function, the proximal term and the current temperature parameter until a classifier that meets the preset conditions is obtained.

[0220] Furthermore, during the training of the classifier, the proximal gradient descent method is used to optimize the parameters, that is, the proximal term is introduced. Specifically, f(e) is the second loss function, θ is the parameter of the classifier, and the proximal term p(θ) is used for regularization. The parameter update formula is:

[0221]

[0222] Among them, η re is the learning rate, in this embodiment η re =0.01;θ old is the classifier parameter before updating, θ new is the updated classifier parameter; ||||2 is the L2 norm symbol.

[0223] Furthermore, the proximal term p(θ) is an L1 regularization term used to promote sparsity and can be expressed as:

[0224] p(θ)=λ e ×||θ||1;

[0225] Among them, λ e is the regularization parameter, which controls the strength of the regularization term; ||||1 is the L1 norm.

[0226] At the same time, the simulated annealing strategy is integrated in the classifier. According to the simulated annealing process, the temperature parameter T is defined to control the search range. The parameter update formula of the classifier takes into account the influence of temperature and can be expressed as:

[0227]

[0228] Among them, ∈ is random noise, which is used to simulate the annealing process; T is the temperature parameter; θ old is the classifier parameter before updating, θ new is the updated classifier parameter, The gradient before and after the second loss function update.

[0229] Furthermore, the temperature parameter T is dynamically adjusted according to the training rounds, which can be expressed as:

[0230] T(t) = T0 × exp(-κ × t);

[0231] Where T0 is the initial temperature, T(t) is the temperature of the tth iteration, κ is the decay rate, and t is the current training round.

[0232] In the adaptive learning rate adjustment, the learning rate is adjusted dynamically. In this embodiment, let α(t) be the learning rate at time t, and the learning rate adjustment strategy can be expressed as:

[0233]

[0234] Wherein, α0 is the initial learning rate. In this embodiment, α0=0.1; Is the adjustment coefficient, which affects the speed at which the learning rate decreases.

[0235] In another embodiment, the learning rate adjustment adds a feedback mechanism based on the training error, which can be expressed as:

[0236]

[0237] Where Error(t) is the training error at time t, γ r is a parameter used to adjust the effect of error.

[0238] In this embodiment, the first training sample set is obtained by labeling the original training samples in the original training sample set, and new training samples are generated based on the training of the generative adversarial network to expand the first training samples, thereby improving the extensiveness of the data from the dimension of the number of training samples and thus improving the generalization ability of the model; then, the pre-trained neural network is applied to extract features of the expanded target training sample set, so as to dynamically adjust the parameters of the neural network through simulation points, thereby improving the generalization ability and prediction accuracy from the feature dimension; and then, classification is performed through the pre-trained classifier, and the overall generalization ability and prediction accuracy of the groundwater level monitoring and early warning model are further improved through cross-connections between the neural networks at each layer, simulated annealing strategies, and the introduction of proximal terms, so as to obtain accurate prediction results.

[0239] Among them, the monitoring and early warning results can be whether the groundwater level is normal or a specific early warning level. The groundwater level monitoring and early warning model can be trained according to the use requirements to obtain the target prediction results. For example, when the prediction result is an early warning level, the early warning level in groundwater level monitoring includes but is not limited to normal, attention, warning, and severe levels. At this time, the prediction result is a specific early warning level. This classification can help relevant departments and individuals take corresponding measures, such as adjusting water resource utilization strategies, implementing water volume regulation, or taking emergency replenishment measures to prevent potential water resource problems.

[0240] In the second aspect, the present application provides a groundwater level monitoring and early warning method based on big data, combined with Figure 2 As shown, the method includes:

[0241] S410, obtaining an actual sample set collected by the monitoring facility.

[0242] S420, preprocessing the actual sample set to obtain a target sample set.

[0243] S430, inputting the target sample set into the pre-trained groundwater level monitoring and early warning model, and outputting the groundwater level monitoring and early warning results.

[0244] Among them, the pre-trained groundwater level monitoring and early warning model is constructed based on the above method.

[0245] The collected and pre-processed target sample sets are input into the above-mentioned groundwater level monitoring and early warning model with good generalization ability and accurate prediction to obtain the predicted water level results, so that geological staff can judge whether emergency recharge measures are needed.

[0246] It can be understood that by connecting the groundwater level monitoring and early warning model with the natural resources monitoring system to assist in carrying out water level early warning tasks, application scenarios include but are not limited to: water level monitoring, water level early warning, water level prediction or emergency alarm.

[0247] The monitoring facilities include ground monitoring stations and groundwater level sensor networks, and the corresponding sensors are configured according to the parameters and application scenarios to be monitored. For different monitoring needs, various types of sensors can be deployed, such as groundwater level sensors, soil moisture sensors, thermometers and hygrometers, rain gauges, and other environmental monitoring equipment. These sensors work together to provide comprehensive groundwater and environmental parameter data to support the accurate operation of the groundwater level monitoring and early warning system.

[0248] In the third aspect, the present application provides a device for constructing a groundwater level monitoring and early warning model based on big data, combined with Figure 3As shown, the device includes: an acquisition module 10, an expansion module 20, a feature extraction module 30 and a prediction module 40.

[0249] The acquisition module 10 is used to acquire an original training sample set and annotate the original training samples in the original training sample set to obtain a first training sample set; wherein the original training samples are obtained by collecting data through ground monitoring stations, groundwater level sensor networks, and auxiliary remote sensing equipment; the original training samples include multiple parameters, and the parameters include at least groundwater level depth, groundwater level change rate, ground temperature, ground humidity, rainfall and groundwater temperature.

[0250] The expansion module 20 is used to apply the first training sample set to train the generative adversarial network including the Lagrangian unit and the time series model, so as to expand the first training sample set and obtain the target training sample set.

[0251] The feature extraction module 30 is used to apply a pre-built neural network to extract features from the training samples in the current training sample set and the target training sample set to obtain a target feature set; the neural network is trained based on the bioelectrical dynamic balance method.

[0252] The prediction module 40 is used to train the classifier using the target feature set, output the groundwater level monitoring and prediction results, and obtain the groundwater level monitoring and early warning model; wherein the groundwater level monitoring and early warning model is used to monitor and warn the groundwater level.

[0253] In a fourth aspect, the present application provides an electronic device, in combination with Figure 4 As shown, the electronic device includes a memory 44 and a processor 41, the memory 44 stores a computer program, and the processor 41 executes the computer program to implement the following Figure 1 , Figure 2 Any method. Figure 4 In the illustrated embodiment, the electronic device further includes a bus 42 and a communication interface 43 , wherein the processor 41 , the communication interface 43 and the memory 44 are connected via the bus 42 .

[0254] Among them, the memory 44 may include a high-speed random access memory (RAM), and may also include a non-volatile memory (non-volatile memory), such as at least one disk storage. The communication connection between the system network element and at least one other network element is realized through at least one communication interface 43 (which may be wired or wireless), and the Internet, wide area network, local area network, metropolitan area network, etc. can be used. The bus 42 can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc., and can also be an AMBA (Advanced Microcontroller Bus Architecture, on-chip bus standard) bus, wherein AMBA defines three types of buses, including APB (Advanced Peripheral Bus) bus, AHB (Advanced High-performance Bus) bus and AXI (Advanced eXtensible Interface) bus. The bus 42 can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 4 Only one bidirectional arrow is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0255] The processor 41 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the hardware integrated logic circuit or software instructions in the processor 41. The above processor 41 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor. The steps of the method disclosed in the embodiment of the present application can be directly embodied as a hardware decoding processor to execute, or the hardware and software modules in the decoding processor can be executed. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor 41 reads the information in the memory and completes the above-mentioned method in combination with its hardware.

[0256] In a fifth aspect, this embodiment further provides a storage medium storing a computer program, which is executed by a processor according to the above method.

[0257] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the system and device described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0258] In addition, in the description of the embodiments of the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, and it can be the internal communication of two components. For those skilled in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0259] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program codes.

[0260] In the description of the present invention, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc., indicating the orientation or positional relationship, are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", and "third" are used for descriptive purposes only, and cannot be understood as indicating or implying relative importance.

[0261] Finally, it should be noted that the above embodiments are only specific implementations of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The protection scope of the present invention is not limited thereto. Although the present invention is described in detail with reference to the above embodiments, those skilled in the art should understand that any person skilled in the art can still modify the technical solutions recorded in the above embodiments within the technical scope disclosed by the present invention, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A method for constructing a groundwater level monitoring and early warning model based on big data, characterized in that: The method comprises: Acquire an original training sample set, and annotate the original training samples in the original training sample set to obtain a first training sample set; wherein the original training samples are obtained by collecting data through ground monitoring stations, a groundwater level sensor network, and auxiliary remote sensing equipment; the original training samples include multiple parameters, and the parameters include at least groundwater level depth, groundwater level change rate, ground temperature, ground humidity, rainfall, and groundwater temperature; Applying the first training sample set to train a generative adversarial network including a Lagrangian unit and a time series model to expand the first training sample set to obtain a target training sample set; Using a pre-built neural network to extract features from the training samples in the target training sample set to obtain a target feature set; the neural network is trained based on a bioelectrical dynamic balance method; The target feature set is used to train the classifier, output the groundwater level monitoring and early warning results, and obtain the groundwater level monitoring and early warning model; wherein the groundwater level monitoring and early warning model is used to monitor and warn the groundwater level; The process of training the neural network based on the bioelectrical dynamic balance method is as follows: Initialize the network parameters, weights and biases of the initial neural network; For each network parameter of the initial neural network, the network parameter is assigned an initial potential value; wherein the assignment formula of the initial potential value is: ; is the initial potential value, It is the initial reference value of the potential. The tanh function is used to simulate the nonlinear response characteristics of the biological cell potential to ensure that the potential value changes within a certain range. is the weight of the i-th parameter, is the bias of the i-th parameter; The initial potential value is input into the initial neural network to perform forward propagation to obtain a first output result; wherein the forward propagation is expressed as: ; is the Sigmoid activation function, x is the input training sample, w is the weight parameter, and b is the bias parameter; According to the first output result, the loss function and potential adjustment amount of the neural network are calculated; wherein the potential adjustment amount The change of the loss function is expressed as: ;in, is the loss function The gradient of is the learning rate; According to the potential adjustment amount and the preset adaptive factor, the network parameters, weights and biases of the neural network are updated until a neural network that meets the preset conditions is obtained; wherein updating the network parameters according to the potential adjustment amount is expressed as: ； ； in, is the potential influence function, is the parameter adjustment coefficient, is the adaptive factor; R(E) is the regularization term, is the coefficient that controls the effect of regularization, is the updated weight parameter, is the weight parameter before updating, To find the sign of the partial derivative, is the i-th bias parameter; is the electrical position before adjustment, is the potential value after adjustment; Wherein, the generative adversarial network includes a generator network, a discriminator network and the time series model; the generator network is integrated with the Lagrangian unit; The step of applying the first training sample set to train a generative adversarial network including a Lagrangian unit and a time series model to expand the first training sample set to obtain a target training sample set includes: The generator network, the time series model and the discriminator network are trained in sequence using the first training sample set to generate new training samples; wherein the calculation formula of the time series model is: ;in, is the output of the time series model, and is a time-dependent weight coefficient used to balance the impact of historical samples and current samples; is the historical sample at the tth iteration, To add new training samples; Integrating the output of the time series model in the generator network and adjusting the generation process to reflect the influence of the time series is expressed as: ;in, is the input random noise vector, is the first training sample, is the output of the time series model, It is the newly added training samples generated after considering the influence of time series; The newly added training samples are optimized by using the Lagrangian unit, and the quality of the newly added training samples after optimization is evaluated; wherein the newly added training samples generated are As the state of the dynamic system, the Lagrangian function Set to ,in is the kinetic energy of the dynamic system, is the potential energy; where the kinetic energy function The calculation method is expressed as: ;in, represents the virtual quality parameter, is the time derivative of the state; Potential Energy Function The calculation method is expressed as: ;in, =0.1, It is the state of the target generating new training samples; Determine the newly added training samples that are greater than a preset quality assessment threshold as target newly added training samples; The newly added training samples are added to the first training sample set to obtain an expanded target training sample set.

2. The method according to claim 1, characterized in that The classifier includes a multi-layer neural network, each layer of the neural network includes a plurality of neurons, and the plurality of neurons are cross-connected; The step of using the target feature set to train the classifier to obtain a groundwater level monitoring and early warning model includes: Inputting the target feature set into the classifier, and optimizing the parameters of the classifier by using a proximal gradient descent method set simulated annealing ensemble method; The multi-layer neural network is trained in sequence based on the target feature set, and the groundwater level monitoring and early warning results are output to obtain the groundwater level monitoring and early warning model.

3. The method according to claim 2, characterized in that The step of applying the target feature set to train the classifier comprises: Initializing network parameters, weights and biases of the classifier; Inputting the target feature set into the classifier, and outputting a second output result; Calculate a second loss function of the classifier according to the second output result; The network parameters, the weights, the biases and the current temperature parameters of the classifier are adaptively updated according to the second loss function, the proximal term and the current temperature parameters until a classifier that meets preset conditions is obtained.

4. A method for groundwater level monitoring and early warning based on big data, characterized in that: The method comprises: Obtain actual sample sets collected by monitoring facilities; Preprocessing the actual sample set to obtain a target sample set; Inputting the target sample set into the pre-trained groundwater level monitoring and early warning model, and outputting groundwater level monitoring and early warning results; Wherein, the pre-trained groundwater level monitoring and early warning model is constructed based on the method described in any one of claims 1-3.

5. A device for constructing a groundwater level monitoring and early warning model based on big data, characterized in that: The device is used to perform the method according to any one of claims 1 to 3; the device comprises: An acquisition module is used to acquire an original training sample set and annotate the original training samples in the original training sample set to obtain a first training sample set; wherein the original training samples are obtained by collecting data through ground monitoring stations, groundwater level sensor networks, and auxiliary remote sensing equipment; the original training samples include multiple parameters, and the parameters include at least groundwater level depth, groundwater level change rate, ground temperature, ground humidity, rainfall, and groundwater temperature; An expansion module, used for applying the first training sample set to train a generative adversarial network including a Lagrangian unit and a time series model, so as to expand the first training sample set and obtain a target training sample set; A feature extraction module, used to apply a pre-built neural network to extract features from the training samples in the target training sample set to obtain a target feature set; the neural network is trained based on a bioelectrical dynamic balance method; The prediction module is used to train the classifier using the target feature set, output the groundwater level monitoring and early warning results, and obtain the groundwater level monitoring and early warning model; wherein the groundwater level monitoring and early warning model is used to monitor and warn the groundwater level.

6. An electronic device, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method according to any one of claims 1 to 3.

7. A storage medium, characterized in that: The storage medium stores computer program instructions, and when the computer program instructions are read and executed by a processor, the method according to any one of claims 1 to 3 is executed.

Images