Environmental protection water-stop curtain leakage risk early warning method and system based on deep learning

Abstract

The present invention discloses an environmental protection water-stop curtain leakage risk warning method and system based on deep learning, including: S1: collecting soil parameters, leakage, groundwater level, soft soil foundation deformation data, and preprocessing; S2: splicing and feature extraction of the preprocessed soil parameter data to obtain soil parameter features; S3: splicing and feature extraction of the preprocessed leakage data in chronological order to obtain leakage time series features; S4: constructing a joint time series sequence based on the preprocessed groundwater level and soft soil foundation deformation data, and extracting the water level deformation joint time series features in combination with the cross-correlation function; S5: performing feature fusion and dimensionality reduction processing on the soil parameter features, leakage time series features and water level deformation joint time series features, obtaining warning features and issuing warnings. The present invention can solve the problem that the warning of soft soil foundation water-stop curtain leakage risk does not fully consider the soft soil foundation characteristics and the insufficient warning accuracy caused by various indicators.

Description

Technical Field

[0001] The present invention relates to the field of water-stop curtains, and in particular to a deep learning-based early warning method and system for leakage risk of environmentally friendly water-stop curtains. Background Art

[0002] As an important anti-seepage structure in foundation pit engineering, the anti-seepage performance of the water-stop curtain is directly related to the stability of the foundation pit and the safety of the surrounding environment. The early warning of the leakage risk of the water-stop curtain refers to the early warning and assessment of the leakage risk that may occur during the construction and operation of the water-stop curtain, so as to take timely measures to prevent and control it. The traditional method of early warning of the leakage risk of the water-stop curtain mainly relies on leakage monitoring and groundwater level monitoring. The leakage monitoring monitors the leakage of the water-stop curtain by installing flow meters and other equipment, and issues an early warning when the leakage exceeds a certain threshold; the groundwater level monitoring monitors the changes in the groundwater level by setting up groundwater level observation wells to indirectly evaluate the anti-seepage performance of the water-stop curtain.

[0003] However, for soft soil foundations, traditional methods often only focus on single indicators such as leakage or groundwater level, but fail to comprehensively consider the impact of multiple factors such as soft soil foundation deformation and groundwater seepage field changes on the anti-seepage performance of the water-stop curtain; and the soft soil foundation has the characteristics of low strength and high compressibility, which puts higher requirements on the anti-seepage performance of the water-stop curtain. Therefore, for the early warning of water-stop curtain leakage risk under soft soil foundation conditions, it is necessary to develop a new method and technical means that comprehensively considers multiple factors and fully considers the characteristics of the soft soil foundation. Summary of the invention

[0004] In view of this, the present invention proposes an environmentally friendly water-stop curtain leakage risk warning method and system based on deep learning, aiming to solve the problem that the soft soil foundation water-stop curtain leakage risk warning does not fully consider the soft soil foundation characteristics and the insufficient warning accuracy caused by various indicators.

[0005] The environmental protection water-stop curtain leakage risk early warning method based on deep learning includes:

[0006] S1: Collect soil parameters, seepage, groundwater level, and soft soil foundation deformation data, and pre-process them to obtain pre-processed soil parameters, seepage, groundwater level, and soft soil foundation deformation data; specifically include:

[0007] S11: Obtain soil parameter data, including soil water content, porosity, saturation, liquid limit and plastic limit index; obtain and record the hourly water leakage of the water-stop curtain through the water-stop curtain flow meter to obtain leakage data; record the hourly groundwater level depth through the water level meter to obtain groundwater level data; obtain the hourly vertical settlement and horizontal displacement of the soil through the settlement observation pile and the inclinometer installed at the bottom of the soft soil foundation pit, and obtain the soft soil foundation deformation data;

[0008] S12: For the soil parameters, seepage, groundwater level, and soft soil foundation deformation data, the 3σ principle is used to remove outliers, and then the maximum and minimum value method is used for normalization to obtain the pre-processed soil parameters, seepage, groundwater level, and soft soil foundation deformation data;

[0009] S2: splice the preprocessed soil parameter data to obtain soil parameter data vectors; then construct a soil parameter feature extraction network, combine the attention mechanism to calculate the layer weights of each hidden layer, extract features from the soil parameter data vectors, and obtain soil parameter features; specifically include:

[0010] S21: splice the preprocessed soil parameter data to obtain the soil parameter data vector, which is calculated as follows:

[0011] ;

[0012] in, is the soil parameter data vector, For data splicing operation, They are the water content, void ratio, saturation, liquid limit and plastic limit index of the soil in the pre-processed soil parameter data;

[0013] S3: splicing the preprocessed leakage data in chronological order to obtain a leakage time series; extracting leakage stable state characteristics and leakage change trend characteristics according to the leakage time series, and then constructing leakage time series characteristics;

[0014] S4: Based on the pre-processed groundwater level and soft soil foundation deformation data, a joint time series is constructed, and the water level deformation delay is calculated using the cross-correlation function; then, based on the joint time series and the water level deformation delay, the joint time series characteristics of the water level deformation are extracted; the specific steps include:

[0015] S41: Based on the pre-processed groundwater level and soft soil foundation deformation data, a joint time series is constructed, and the water level deformation time delay of the pre-processed groundwater level data and the pre-processed soft soil foundation deformation data is calculated using a cross-correlation function. The calculation method is:

[0016] ;

[0017] in, is a joint time series, is the preprocessed groundwater level data, is the preprocessed soft soil foundation deformation data, is the water level deformation time delay, is the cross-correlation function operation;

[0018] S42: Extract the joint time series characteristics of water level deformation according to the joint time series sequence and the water level deformation time delay. The calculation method is:

[0019] ;

[0020] in, It is the joint time series characteristics of water level deformation;

[0021] S5: The soil parameter characteristics, leakage time series characteristics and water level deformation joint time series characteristics are subjected to feature fusion and dimensionality reduction processing to obtain warning features; the warning features are then used to carry out environmental protection water-stop curtain leakage warning. Traditional water level and deformation data analysis methods often process the two separately, ignoring the interaction and time delay relationship between them. This invention combines GRU and MLP, proposes to conduct joint time series modeling of groundwater level and soft soil foundation deformation data, and uses the cross-correlation function (CCF) to calculate the water level deformation delay, capture the time delay relationship between groundwater level changes and soft soil foundation deformation, and provide time dimension information for subsequent feature extraction.

[0022] Furthermore, the S2 step further includes:

[0023] S22: Construct a soil parameter feature extraction network, extract features from the soil parameter data vector, and obtain soil parameter features. The calculation method is:

[0024] ;

[0025] in, is the preliminary soil parameter characteristics of the i-th hidden layer, i is the hidden layer index, is the hyperbolic tangent function, is the weight matrix of the i-th hidden layer, is the preliminary soil parameter characteristics of the i-1th hidden layer, is the bias vector of the i-th hidden layer; is the layer weight of the i-th hidden layer, is the attention mechanism operation, are the preliminary soil parameter characteristics of the first, second, ..., nth hidden layers, respectively, n is the total number of hidden layers, is the soil parameter characteristic, is the soil parameter weight matrix, is the soil parameter bias vector;

[0026] For the formula , when i=1, ; is the preliminary soil parameter characteristics of the first hidden layer, is the weight matrix of the first hidden layer, is the bias vector of the first hidden layer.

[0027] The soil parameter feature extraction network proposed in the present invention performs deep feature extraction on soil parameter data vectors through multiple hidden layers, and then calculates the layer weight of each hidden layer output by combining the attention mechanism, and dynamically adjusts the contribution of each layer feature in the final soil parameter feature, thereby highlighting the feature information that is more important for leakage risk prediction, making the extracted soil parameter features more comprehensive and accurate.

[0028] Furthermore, the S3 step includes:

[0029] The preprocessed leakage data are spliced ​​in chronological order to obtain the leakage time series; the mean and standard deviation of the leakage time series are calculated, and the steady-state characteristics of the leakage are extracted; the slope and duration of the leakage change trend are calculated, and the leakage change trend characteristics are extracted; the leakage time series characteristics are constructed based on the steady-state characteristics of the leakage and the leakage change trend characteristics.

[0030] Furthermore, the S3 step also includes:

[0031] S31: splicing the pre-processed leakage data in chronological order to obtain a leakage time series, which is calculated as follows:

[0032] ;

[0033] in, is the leakage time series, are the pre-processed leakage data at the 1st moment, the 2nd moment, ..., and the Tth moment, is the total number of moments;

[0034] S32: Calculate the mean and standard deviation of the leakage time series and extract the steady-state characteristics of the leakage. The calculation method is:

[0035] ;

[0036] in, is the mean of the leakage time series, is the preprocessed leakage data at the tth moment, t is the time index, is the standard deviation of the leakage time series, is the steady-state characteristic of leakage at time t, For long short-term memory network operations;

[0037] S33: Calculate the slope and duration of the leakage change trend, and extract the leakage change trend characteristics. The calculation method is:

[0038] ;

[0039] in, is the slope of the leakage change trend, is the duration of the change, is an indicator function. When the conditions in the brackets are met, ,otherwise ; is the adaptive threshold value of the changing trend, For multi-layer perceptron operation, is the leakage change trend characteristic at the tth moment, It is a gated recurrent unit operation;

[0040] S34: According to the stable state characteristics of leakage and the trend characteristics of leakage, the time series characteristics of leakage are constructed. The calculation method is:

[0041] ;

[0042] in, is the leakage time series characteristic, This is a vector concatenation operation.

[0043] Most traditional leakage feature extraction methods only focus on the absolute value of the leakage or simple statistics, which makes it difficult to fully reflect the stable state and changing trend of the leakage. The present invention calculates the mean and standard deviation of the leakage time series, and uses a long short-term memory network (LSTM) to process the standardized leakage data, thereby realizing the standardized processing of the leakage data, eliminating the influence of the data magnitude and fluctuation range on feature extraction, and innovatively extracting the stable state characteristics of the leakage. At the same time, the present invention introduces an indicator function and an adaptive threshold, calculates the slope of the leakage trend, and combines the change trend threshold and the change duration adaptively determined by a multi-layer perceptron (MLP), and uses a gated recurrent unit (GRU) to extract the change trend characteristics of the leakage, captures the dynamic change process of the leakage, and provides strong support for predicting the development trend of leakage risks, and lays a solid foundation for the subsequent feature fusion and dimensionality reduction processing and the construction of an early warning model.

[0044] Furthermore, the step S5 further includes:

[0045] S51: The soil parameter characteristics, leakage time series characteristics and water level deformation joint time series characteristics are fused to obtain fusion characteristics. The calculation method is:

[0046] ;

[0047] in, For fusion features;

[0048] S52: Perform dimension reduction processing on the fused features to obtain warning features. The calculation method is:

[0049] ;

[0050] in, As a warning feature, is the LeakyReLU function, is the dimension reduction weight matrix, is the dimension reduction bias vector;

[0051] S53: Use early warning features to warn of water leakage in water-stop curtains. The calculation method is:

[0052] ;

[0053] in, The probability of water leakage from the water-stopping curtain; is the sigmoid function, when When the water leakage warning is issued at the water stop curtain flow meter.

[0054] The present invention fuses soil parameter characteristics, leakage time series characteristics and water level deformation joint time series characteristics through a multi-layer perceptron (MLP). At the same time, the present invention uses MLP with a LeakyReLU activation function to perform dimensionality reduction processing on the fused features. The LeakyReLU function is used as an activation function, which can give a certain gradient to the negative value while retaining the positive value of the feature, avoiding the "dead zone" problem of the ReLU function and improving the generalization ability of the model.

[0055] The present invention also discloses an environmentally friendly water-stopping curtain leakage risk early warning system based on deep learning, including:

[0056] Data acquisition and preprocessing module: collect soil parameters, seepage, groundwater level, and soft soil foundation deformation data, and preprocess them to obtain preprocessed soil parameters, seepage, groundwater level, and soft soil foundation deformation data;

[0057] Soil parameter feature extraction module: splice the preprocessed soil parameter data to obtain soil parameter data vector; then construct a soil parameter feature extraction network, extract features from the soil parameter data vector, and obtain soil parameter features;

[0058] Leakage time series feature extraction module: splice the preprocessed leakage data in chronological order to obtain a leakage time series; extract the leakage steady state features and leakage change trend features according to the leakage time series, and then construct the leakage time series features;

[0059] Water level deformation joint time series feature extraction module: Based on the pre-processed groundwater level and soft soil foundation deformation data, a joint time series sequence is constructed, and the water level deformation delay is calculated using the cross-correlation function; then, based on the joint time series sequence and the water level deformation delay, the water level deformation joint time series feature is extracted;

[0060] Feature fusion and early warning module: The soil parameter characteristics, leakage time series characteristics and water level deformation joint time series characteristics are fused and reduced in dimension to obtain early warning characteristics; the early warning characteristics are then used to issue an early warning for water leakage in the environmental protection water-stop curtain.

[0061] Compared with the prior art, the present invention has the following beneficial effects:

[0062] (1) The present invention constructs a soil parameter feature extraction network, and by incorporating key parameters such as the moisture content and porosity of soft soil foundation soil, deeply explores the potential relationship between these parameters and leakage risk. At the same time, the deep learning model is combined to extract the features of leakage, soft soil foundation deformation and groundwater level, which effectively improves the sensitivity of the early warning model to the characteristics of soft soil foundation and the early warning accuracy.

[0063] (2) In terms of soil parameter feature extraction, the present invention innovatively introduces an attention mechanism to calculate the layer weights of the multi-layer perceptron, dynamically adjusts the contribution of each hidden layer feature, highlights the features that are more important for leakage risk prediction, effectively suppresses the interference of redundant information, and improves the accuracy and efficiency of feature extraction.

[0064] (3) The present invention combines the mean and standard deviation to calculate the steady-state characteristics of the leakage volume, and uses the slope and duration of the change to characterize the changing trend characteristics of the leakage volume. It comprehensively considers the statistical characteristics and dynamic changes of the leakage volume time series, and provides strong support for accurately capturing abnormal fluctuations in the leakage volume.

[0065] (4) The present invention conducts joint time series modeling of groundwater level and soft soil foundation deformation data, and uses cross-correlation function to calculate water level deformation delay, capturing the time delay relationship between groundwater level change and soft soil foundation deformation, providing time dimension information for the extraction of joint time series characteristics of water level deformation, and enhancing the robustness and reliability of early warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 A schematic diagram of the process of the environmental protection water-stop curtain leakage risk early warning method based on deep learning provided by the present invention;

[0067] Figure 2 A schematic diagram of the algorithm flow for extracting soil parameter characteristics provided by the present invention;

[0068] Figure 3 A schematic diagram of the algorithm flow for extracting the time series characteristics of leakage provided by the present invention. DETAILED DESCRIPTION

[0069] The present invention is further described below in conjunction with the accompanying drawings, but the present invention is not limited in any way. Any changes or substitutions made based on the teachings of the present invention belong to the protection scope of the present invention.

[0070] Example 1: Environmental protection water-stop curtain leakage risk early warning method based on deep learning, such as Figure 1 As shown, the following steps are included:

[0071] S1: collect soil parameters, seepage, groundwater level, and soft soil foundation deformation data, and preprocess them to obtain preprocessed soil parameters, seepage, groundwater level, and soft soil foundation deformation data;

[0072] S11: Obtain soil parameter data, including soil water content, porosity, saturation, liquid limit and plastic limit index; obtain and record the hourly water leakage of the water-stop curtain through the water-stop curtain flow meter to obtain leakage data; record the hourly groundwater level depth through the water level meter to obtain groundwater level data; obtain the hourly vertical settlement and horizontal displacement of the soil through the settlement observation pile and the inclinometer installed at the bottom of the soft soil foundation pit, and obtain the soft soil foundation deformation data;

[0073] S12: For the soil parameters, seepage, groundwater level, and soft soil foundation deformation data, the 3σ principle is used to remove outliers, and then the maximum and minimum method is used for normalization to obtain the pre-processed soil parameters, seepage, groundwater level, and soft soil foundation deformation data.

[0074] S2: splicing the preprocessed soil parameter data to obtain soil parameter data vectors; then constructing a soil parameter feature extraction network to extract features from the soil parameter data vectors to obtain soil parameter features; Figure 2 As shown, the following steps are included:

[0075] S21: splice the preprocessed soil parameter data to obtain the soil parameter data vector, which is calculated as follows:

[0076] ;

[0077] in, is the soil parameter data vector, For data splicing operation, They are the water content, void ratio, saturation, liquid limit and plastic limit index of the soil in the pre-processed soil parameter data;

[0078] S22: Construct a soil parameter feature extraction network, extract features from the soil parameter data vector, and obtain soil parameter features. The calculation method is:

[0079] ;

[0080] in, is the preliminary soil parameter characteristics of the i-th hidden layer, i is the hidden layer index, is the hyperbolic tangent function, is the weight matrix of the i-th hidden layer, is the preliminary soil parameter characteristics of the i-1th hidden layer, is the bias vector of the i-th hidden layer; is the layer weight of the i-th hidden layer, is the attention mechanism operation, are the preliminary soil parameter characteristics of the first, second, ..., nth hidden layers, respectively, n is the total number of hidden layers, is the soil parameter characteristic, is the soil parameter weight matrix, is the soil parameter bias vector;

[0081] In particular, for the formula , when i=1, ; is the preliminary soil parameter characteristics of the first hidden layer, is the weight matrix of the first hidden layer, is the bias vector of the first hidden layer.

[0082] The soil parameter feature extraction network proposed in the present invention performs deep feature extraction on soil parameter data vectors through multiple hidden layers, and then calculates the layer weight of each hidden layer output by combining the attention mechanism, and dynamically adjusts the contribution of each layer feature in the final soil parameter feature, thereby highlighting the feature information that is more important for leakage risk prediction, making the extracted soil parameter features more comprehensive and accurate.

[0083] S3: splicing the preprocessed leakage data in chronological order to obtain a leakage time series; extracting leakage stable state characteristics and leakage change trend characteristics according to the leakage time series, and then constructing leakage time series characteristics;

[0084] In particular, when the requirements for early warning of water leakage of the water-stop curtain on soft soil foundation are relatively strict, the following steps can be adopted before step S31 to enhance the accuracy of early warning:

[0085] The pre-processed leakage data is amplified by using a nonlinear variation function to obtain the amplified leakage data. The calculation method is:

[0086] ;

[0087] in, is the amplified leakage data, is the symbolic function, To take the absolute value, It is an adjustment parameter used to control the degree of nonlinear transformation. When γ>1, the transformation function is a convex function, which amplifies small data. When γ<1, the transformation function is a concave function, which reduces large data. When γ=1, the transformation function degenerates into a linear function, that is, no amplification is performed.

[0088] Get the amplified leakage data Afterwards, use replace , perform operations from step S31 to step S34; if Figure 3 As shown, the following steps are included:

[0089] S31: splicing the pre-processed leakage data in chronological order to obtain a leakage time series, which is calculated as follows:

[0090] ;

[0091] in, is the leakage time series, are the pre-processed leakage data at the 1st moment, the 2nd moment, ..., and the Tth moment, is the total number of moments;

[0092] S32: Calculate the mean and standard deviation of the leakage time series and extract the steady-state characteristics of the leakage. The calculation method is:

[0093] ;

[0094] in, is the mean of the leakage time series, is the preprocessed leakage data at the tth moment, t is the time index, is the standard deviation of the leakage time series, is the steady-state characteristic of leakage at time t, For long short-term memory network operations;

[0095] S33: Calculate the slope and duration of the leakage change trend, and extract the leakage change trend characteristics. The calculation method is:

[0096] ;

[0097] in, is the slope of the leakage change trend, is the duration of the change, is an indicator function. When the conditions in the brackets are met, ,otherwise ; is the adaptive threshold value of the changing trend, For multi-layer perceptron operation, is the leakage change trend characteristic at the tth moment, It is a gated recurrent unit operation;

[0098] S34: According to the stable state characteristics of leakage and the trend characteristics of leakage, the time series characteristics of leakage are constructed. The calculation method is:

[0099] ;

[0100] in, is the leakage time series characteristic, This is a vector concatenation operation.

[0101] Most traditional leakage feature extraction methods only focus on the absolute value of the leakage or simple statistics, which makes it difficult to fully reflect the stable state and changing trend of the leakage. The present invention calculates the mean and standard deviation of the leakage time series, and uses a long short-term memory network (LSTM) to process the standardized leakage data, thereby realizing the standardized processing of the leakage data, eliminating the influence of the data magnitude and fluctuation range on feature extraction, and innovatively extracting the stable state characteristics of the leakage. At the same time, the present invention introduces an indicator function and an adaptive threshold, calculates the slope of the leakage trend, and combines the change trend threshold and the change duration adaptively determined by a multi-layer perceptron (MLP), and uses a gated recurrent unit (GRU) to extract the change trend characteristics of the leakage, captures the dynamic change process of the leakage, and provides strong support for predicting the development trend of leakage risks, and lays a solid foundation for the subsequent feature fusion and dimensionality reduction processing and the construction of an early warning model.

[0102] For example, suppose the leakage data of a project is collected within one month, and the preprocessed leakage data is obtained after preprocessing. , after calculating the mean , standard deviation , adaptive threshold value of changing trend , slope , the duration of the change ;but:

[0103]

[0104] S4: Based on the pre-processed groundwater level and soft soil foundation deformation data, a joint time series is constructed, and the water level deformation time delay is calculated using the cross-correlation function; then, based on the joint time series and the water level deformation time delay, the joint time series characteristics of the water level deformation are extracted;

[0105] S41: Based on the pre-processed groundwater level and soft soil foundation deformation data, a joint time series is constructed, and the water level deformation time delay of the pre-processed groundwater level data and the pre-processed soft soil foundation deformation data is calculated using a cross-correlation function. The calculation method is:

[0106] ;

[0107] in, is a joint time series, is the preprocessed groundwater level data, is the preprocessed soft soil foundation deformation data, is the water level deformation time delay, is the cross-correlation function operation;

[0108] S42: Extract the joint time series characteristics of water level deformation according to the joint time series sequence and the water level deformation time delay. The calculation method is:

[0109] ;

[0110] in, It is the joint time series characteristics of water level deformation.

[0111] Traditional water level and deformation data analysis methods often treat the two separately, ignoring the interaction and time delay relationship between them. This paper combines GRU and MLP, proposes to conduct joint time series modeling of groundwater level and soft soil foundation deformation data, and uses the cross-correlation function (CCF) to calculate the water level deformation delay, capturing the time delay relationship between groundwater level changes and soft soil foundation deformation, and providing time dimension information for subsequent feature extraction.

[0112] S5: The soil parameter characteristics, leakage time series characteristics and water level deformation joint time series characteristics are fused and reduced in dimension to obtain early warning characteristics. The early warning characteristics are then used to issue early warning for water leakage in the water-stop curtain.

[0113] S51: The soil parameter characteristics, leakage time series characteristics and water level deformation joint time series characteristics are fused to obtain fusion characteristics. The calculation method is:

[0114] ;

[0115] in, For fusion features;

[0116] S52: Perform dimension reduction processing on the fused features to obtain warning features. The calculation method is:

[0117] ;

[0118] Among them, is the early warning feature, is the LeakyReLU function, is the dimensionality reduction weight matrix, is the dimensionality reduction bias vector;

[0119] S53: Use the early warning feature to carry out the leakage warning of the cut-off wall, and the calculation method is:

[0120] ;

[0121] Among them, is the probability of leakage of the cut-off wall; is the sigmoid function. When occurs, a leakage warning at the cut-off wall flowmeter is issued.

[0122] In the present invention, a multi-layer perceptron (MLP) is used to fuse the soil parameter characteristics, the leakage volume time series characteristics and the water level deformation joint time series characteristics. At the same time, the present invention uses an MLP with a LeakyReLU activation function to perform dimensionality reduction processing on the fused features; the LeakyReLU function, as an activation function, can give a certain gradient to negative values while retaining the positive values of the features, avoiding the "dead zone" problem of the ReLU function and improving the generalization ability of the model.

[0123] Furthermore, different early warning intensity classification criteria can also be set according to the value of the early warning probability p. For example:

[0124] When 0 ≤ p ≤ 0.3, it is determined as "no warning", indicating that the current cut-off wall state is good and there is no leakage risk;

[0125] When 0.3 < p ≤ 0.5, it is determined as "low early warning level", indicating that there are slight leakage signs and monitoring needs to be strengthened;

[0126] When 0.5 < p ≤ 0.7, it is determined as "medium early warning level", indicating that the leakage situation is relatively obvious and preliminary anti-seepage measures need to be taken;

[0127] When p > 0.7, it is determined as "high early warning level", indicating that the leakage is serious and urgent anti-seepage measures need to be taken immediately.

[0128] For example, assume that in a certain monitoring, the obtained early warning feature After MLP and sigmoid calculations, the water leakage probability p=0.6 is obtained. According to the classification standards, it is judged as the "medium warning level" at this time. Combined with the actual situation, it may be due to the increase in rainfall in recent times, which has caused the groundwater level to rise and the soft soil foundation to deform slightly, but the leakage has not yet reached a serious level. Therefore, the monitoring frequency should be strengthened and necessary anti-seepage materials and equipment should be prepared to deal with possible leakage.

[0129] Embodiment 2: The present invention also discloses an environmentally friendly water-stopping curtain leakage risk early warning system based on deep learning, comprising:

[0130] Data acquisition and preprocessing module: collect soil parameters, seepage, groundwater level, and soft soil foundation deformation data, and preprocess them to obtain preprocessed soil parameters, seepage, groundwater level, and soft soil foundation deformation data;

[0131] Soil parameter feature extraction module: splice the preprocessed soil parameter data to obtain soil parameter data vector; then construct a soil parameter feature extraction network, extract features from the soil parameter data vector, and obtain soil parameter features;

[0132] Leakage time series feature extraction module: splice the preprocessed leakage data in chronological order to obtain a leakage time series; extract the leakage steady state features and leakage change trend features according to the leakage time series, and then construct the leakage time series features;

[0133] Water level deformation joint time series feature extraction module: Based on the pre-processed groundwater level and soft soil foundation deformation data, a joint time series sequence is constructed, and the water level deformation delay is calculated using the cross-correlation function; then, based on the joint time series sequence and the water level deformation delay, the water level deformation joint time series feature is extracted;

[0134] Feature fusion and early warning module: The soil parameter characteristics, leakage time series characteristics and water level deformation joint time series characteristics are fused and reduced in dimension to obtain early warning characteristics; the early warning characteristics are then used to issue an early warning for water leakage in the environmental protection water-stop curtain.

[0135] It should be noted that the serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages and disadvantages of the embodiments. And the terms "including", "comprising" or any other variants thereof in this article are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "including a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0136] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes a number of instructions for a terminal device (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in multiple embodiments of the present invention.

[0137] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. The environmental protection water-stop curtain leakage risk early warning method based on deep learning is characterized by: The following steps are involved: S1: Collect soil parameters, seepage, groundwater level, and soft soil foundation deformation data, and pre-process them to obtain pre-processed soil parameters, seepage, groundwater level, and soft soil foundation deformation data; specifically include: S11: Obtain soil parameter data, including soil water content, porosity, saturation, liquid limit and plastic limit index; obtain and record the hourly water leakage of the water-stop curtain through the water-stop curtain flow meter to obtain leakage data; record the hourly groundwater level depth through the water level meter to obtain groundwater level data; obtain the hourly vertical settlement and horizontal displacement of the soil through the settlement observation pile and the inclinometer installed at the bottom of the soft soil foundation pit, and obtain the soft soil foundation deformation data; S12: For the soil parameters, seepage, groundwater level, and soft soil foundation deformation data, the 3σ principle is used to remove outliers, and then the maximum and minimum value method is used for normalization to obtain the pre-processed soil parameters, seepage, groundwater level, and soft soil foundation deformation data; S2: splice the preprocessed soil parameter data to obtain soil parameter data vectors; then construct a soil parameter feature extraction network, combine the attention mechanism to calculate the layer weights of each hidden layer, extract features from the soil parameter data vectors, and obtain soil parameter features; specifically include: S21: splice the preprocessed soil parameter data to obtain the soil parameter data vector, which is calculated as follows: ； in, is the soil parameter data vector, For data splicing operation, They are the water content, void ratio, saturation, liquid limit and plastic limit index of the soil in the pre-processed soil parameter data; S3: splicing the preprocessed leakage data in chronological order to obtain a leakage time series; extracting leakage stable state characteristics and leakage change trend characteristics according to the leakage time series, and then constructing leakage time series characteristics; S4: Based on the pre-processed groundwater level and soft soil foundation deformation data, a joint time series is constructed, and the water level deformation delay is calculated using the cross-correlation function; then, based on the joint time series and the water level deformation delay, the joint time series characteristics of the water level deformation are extracted; the specific steps include: S41: Based on the pre-processed groundwater level and soft soil foundation deformation data, a joint time series is constructed, and the water level deformation time delay of the pre-processed groundwater level data and the pre-processed soft soil foundation deformation data is calculated using a cross-correlation function. The calculation method is: in, is a joint time series, It is a vector concatenation operation; is the preprocessed groundwater level data, is the preprocessed soft soil foundation deformation data, is the water level deformation time delay, is the cross-correlation function operation; S42: Extract the joint time series characteristics of water level deformation according to the joint time series sequence and the water level deformation time delay. The calculation method is: ； in, It is the joint time series characteristics of water level deformation; Operate for multi-layer perceptron; It is a gated recurrent unit operation; S5: Perform feature fusion and dimensionality reduction processing on soil parameter characteristics, leakage time series characteristics and water level deformation joint time series characteristics to obtain early warning features; then use the early warning features to issue an early warning for water leakage of the environmental protection water-stop curtain.

2. The environmental protection water-stop curtain leakage risk early warning method based on deep learning according to claim 1 is characterized in that: The S2 step includes: S22: Construct a soil parameter feature extraction network, extract features from the soil parameter data vector, and obtain soil parameter features. The calculation method is: in, is the preliminary soil parameter characteristics of the i-th hidden layer, i is the hidden layer index, is the hyperbolic tangent function, is the weight matrix of the i-th hidden layer, is the preliminary soil parameter characteristics of the i-1th hidden layer, is the bias vector of the i-th hidden layer; is the layer weight of the i-th hidden layer, is the attention mechanism operation, are the preliminary soil parameter characteristics of the first, second, ..., and nth hidden layers, respectively, where n is the total number of hidden layers. is the soil parameter characteristic, is the soil parameter weight matrix, is the soil parameter bias vector; For the formula , when i=1, ; is the preliminary soil parameter characteristics of the first hidden layer, is the weight matrix of the first hidden layer, is the bias vector of the first hidden layer.

3. The environmental protection water-stop curtain leakage risk early warning method based on deep learning according to claim 2 is characterized in that: The S3 step includes: The preprocessed leakage data are spliced ​​in chronological order to obtain the leakage time series; the mean and standard deviation of the leakage time series are calculated, and the steady-state characteristics of the leakage are extracted; the slope and duration of the leakage change trend are calculated, and the leakage change trend characteristics are extracted; the leakage time series characteristics are constructed based on the steady-state characteristics of the leakage and the leakage change trend characteristics.

4. The environmental protection water-stop curtain leakage risk early warning method based on deep learning according to claim 3 is characterized in that: The S3 step includes: S31: splicing the pre-processed leakage data in chronological order to obtain a leakage time series, which is calculated as follows: ； in, is the leakage time series, are the pre-processed leakage data at the 1st moment, the 2nd moment, ..., and the Tth moment, is the total number of moments; S32: Calculate the mean and standard deviation of the leakage time series and extract the steady-state characteristics of the leakage. The calculation method is: in, is the mean of the leakage time series, is the preprocessed leakage data at the tth moment, t is the time index, is the standard deviation of the leakage time series, is the steady-state characteristic of leakage at time t, For long short-term memory network operations; S33: Calculate the slope and duration of the leakage change trend, and extract the leakage change trend characteristics. The calculation method is: in, is the slope of the leakage change trend, is the duration of the change, is an indicator function. When the conditions in the brackets are met, ,otherwise ; is the adaptive threshold value of the changing trend, For multi-layer perceptron operation, is the leakage change trend characteristic at the tth moment, It is a gated recurrent unit operation; S34: According to the stable state characteristics of leakage and the trend characteristics of leakage, the time series characteristics of leakage are constructed. The calculation method is: ； in, is the leakage time series characteristic, This is a vector concatenation operation.

5. The environmental protection water-stop curtain leakage risk early warning method based on deep learning according to claim 4 is characterized in that: The S5 step includes: S51: The soil parameter characteristics, leakage time series characteristics and water level deformation joint time series characteristics are fused to obtain fusion characteristics. The calculation method is: ； in, For fusion features; S52: Perform dimension reduction processing on the fused features to obtain warning features. The calculation method is: ； in, As a warning feature, is the LeakyReLU function, is the dimension reduction weight matrix, is the dimension reduction bias vector; S53: Use early warning features to warn of water leakage in environmentally friendly water-stop curtains. The calculation method is: ； in, To reduce the probability of water leakage of the environmental protection water-stop curtain, is the sigmoid function; when When the water leakage warning is issued at the water stop curtain flow meter.

6. Environmentally friendly water-stop curtain leakage risk early warning system based on deep learning, including: Data acquisition and preprocessing module: collect soil parameters, seepage, groundwater level, and soft soil foundation deformation data, and preprocess them to obtain preprocessed soil parameters, seepage, groundwater level, and soft soil foundation deformation data; Soil parameter feature extraction module: splice the preprocessed soil parameter data to obtain soil parameter data vector; then construct a soil parameter feature extraction network, extract features from the soil parameter data vector, and obtain soil parameter features; Leakage time series feature extraction module: splice the preprocessed leakage data in chronological order to obtain a leakage time series; extract the leakage steady state features and leakage change trend features according to the leakage time series, and then construct the leakage time series features; Water level deformation joint time series feature extraction module: Based on the pre-processed groundwater level and soft soil foundation deformation data, a joint time series sequence is constructed, and the water level deformation delay is calculated using the cross-correlation function; then, based on the joint time series sequence and the water level deformation delay, the water level deformation joint time series feature is extracted; Feature fusion and early warning module: perform feature fusion and dimensionality reduction processing on soil parameter features, leakage time series features and water level deformation joint time series features to obtain early warning features; Then use the early warning feature to warn of water leakage of the environmental protection water-stop curtain; To realize the environmental protection water-stop curtain leakage risk warning method based on deep learning as described in any one of claims 1-5.

Images