Lake water regimen monitoring method and system based on deep learning

Abstract

The invention is suitable for the technical field of water regimen monitoring, and provides a lake water regimen monitoring method and system based on deep learning, and the method comprises the following steps: classifying historical water regimen data based on water regimen abnormality types; determining a strong correlation water regimen parameter corresponding to each water regimen abnormity type, and determining a historical change curve graph; collecting real-time water regimen data, and predicting the real-time water regimen data based on deep learning and climate conditions to obtain a predicted change curve graph; splitting the prediction change curve graph based on the strong correlation water regimen parameters corresponding to each water regimen anomaly type to obtain a plurality of prediction change curve sub-graphs; and performing similarity calculation on the predicted change curve sub-graph and the corresponding historical change curve graph, and determining a potential anomaly type according to a similarity value. According to the method, calculation is carried out only through the change curve of the strongly correlated water regimen parameters, interference of other parameter curves is eliminated, interaction of multiple strongly correlated water regimen parameters is considered, and the water regimen anomaly type can be accurately identified.

Description

Technical Field

[0001] The present invention relates to the technical field of water regime monitoring, and in particular to a lake water regime monitoring method and system based on deep learning. Background Art

[0002] With global climate change and accelerating urbanization, lakes, as vital water resources and ecosystems, are experiencing increasingly complex and far-reaching changes in their water regimes. Lake water regimes not only impact the secure supply of water resources but also directly influence the surrounding ecological environment, agricultural irrigation, industrial production, and residents' lives. Traditional lake water regime monitoring relies primarily on manual inspections and periodic sampling and analysis. This approach is not only time-consuming and labor-intensive, but also struggles to achieve real-time, comprehensive monitoring of water regimes. Furthermore, traditional methods rely solely on the numerical value of a single water parameter to determine if an anomaly has occurred. However, anomalies in water regimes are often caused by multiple parameters, such as eutrophication and aquatic ecological abnormalities. Traditional methods struggle to accurately distinguish between different types of anomalies, resulting in inaccurate early warnings and the inability to implement effective response measures. Furthermore, these methods fail to fully utilize useful information from historical data and are insufficiently adaptable to complex water regime changes. Therefore, there is a need for a deep learning-based lake water regime monitoring method and system to address these issues. Summary of the Invention

[0003] In view of the shortcomings of the existing technology, the purpose of the present invention is to provide a lake water condition monitoring method and system based on deep learning to solve the problems existing in the above-mentioned background technology.

[0004] The present invention is implemented as follows: a lake water regime monitoring method based on deep learning, the method comprising the following steps: Classify historical water regime data based on water regime anomaly types. Each historical water regime data contains time series data of various water regime parameters. Analyze the historical water regime data of each category, determine the strongly correlated water regime parameters corresponding to each water regime anomaly type, and determine a historical change curve graph, which includes the change curves of all strongly correlated water regime parameters; Collect real-time water condition data, predict it based on deep learning and climate conditions, and obtain a predicted change curve chart, which contains the change curves of all water condition parameters; The predicted change curve is split based on the strongly correlated water regime parameters corresponding to each water regime anomaly type to obtain several predicted change curve sub-graphs; Calculate the similarity between the predicted change curve subgraph and the corresponding historical change curve graph, and determine the potential anomaly type based on the similarity value; Generate early warning information and monitoring plans based on potential abnormality types.

[0005] As a further solution of the present invention, the step of determining the strongly correlated water regime parameters corresponding to each type of water regime anomaly specifically includes: All historical water regime data in the same category are extracted and grouped according to water regime parameters, so that each group of data is time series data of the same water regime parameter; Based on the time series data, a curve graph of the corresponding water condition parameters is drawn, and the similarity between any two curve graphs in each group is calculated in turn to obtain the mean similarity value; The water regime parameter is determined to be a strongly correlated water regime parameter corresponding to the water regime anomaly type based on its similarity mean.

[0006] As a further solution of the present invention, the step of predicting real-time water condition data based on deep learning and climate conditions specifically includes: Define the input window. Each sample is all the characteristic data of a certain period in the past. The characteristic data includes water parameter values, climate data, and statistical characteristics. Define the prediction window, and the label corresponding to each sample is the water parameter status after a certain period of time in the past N hours; All samples are divided into training set, validation set and test set in chronological order; Input the training set data into the LSTM model, minimize the loss function through the backpropagation algorithm and optimizer, and iteratively update the model parameters; Perform early stopping on the validation set to monitor the model's performance on the validation set. When the validation set loss stops decreasing, stop training and restore the model parameters of the group with the best performance on the validation set. The final performance of the LSTM model is evaluated based on the test set, and the evaluation indicators include root mean square error, mean absolute error, and determination coefficient.

[0007] As a further solution of the present invention, the step of calculating the similarity between the predicted change curve subgraph and the corresponding historical change curve graph specifically includes: Determine the curve color according to the water regime parameter type, and color all the change curves in the predicted change curve subgraph and the historical change curve graph; Perform DTW distance calculation on the predicted change curve subgraph and the historical change curve graph, and normalize the distance value calculated by DTW to obtain the similarity value; All similarity values ​​obtained through the predicted change curve subgraph are summarized, and the largest one is taken as the final similarity value.

[0008] As a further solution of the present invention, the step of determining the potential anomaly type based on the similarity value specifically includes: Determine whether the similarity value of each predicted change curve subgraph is greater than the similarity threshold. If it is greater than, the corresponding water abnormality type is determined as a potential abnormality type; The potential anomaly types determined by a plurality of the predicted change curve subgraphs are aggregated to obtain an anomaly type set.

[0009] As a further solution of the present invention, the step of generating early warning information and monitoring solutions according to potential abnormality types specifically includes: Determine the anomaly level based on the similarity value, and generate water situation warning information based on the potential anomaly type and anomaly level; Determine the monitoring frequency, monitoring range, and newly added monitoring points for the corresponding strongly correlated water parameters based on the anomaly level; Determine the water situation treatment plan based on the potential abnormality type and abnormality level.

[0010] Another object of the present invention is to provide a lake water regime monitoring system based on deep learning, the system comprising: The historical water regime classification module is used to classify historical water regime data based on the type of water regime anomaly. Each historical water regime data contains time series data of various water regime parameters. The strongly correlated parameter module is used to analyze the historical water regime data of each category, determine the strongly correlated water regime parameters corresponding to each water regime anomaly type, and determine the historical change curve graph, which contains the change curves of all strongly correlated water regime parameters; The real-time water regime prediction module is used to collect real-time water regime data, predict the real-time water regime data based on deep learning and climate conditions, and obtain a predicted change curve chart, which contains the change curves of all water regime parameters; The prediction graph splitting module is used to split the prediction change curve graph based on the strongly correlated water regime parameters corresponding to each water regime anomaly type to obtain several prediction change curve subgraphs; The potential anomaly type module is used to calculate the similarity between the predicted change curve subgraph and the corresponding historical change curve graph, and determine the potential anomaly type based on the similarity value; The abnormal warning monitoring module is used to generate early warning information and monitoring plans based on potential abnormality types.

[0011] As a further solution of the present invention, the strongly correlated parameter module includes: The water regime parameter grouping unit is used to extract and group all historical water regime data of the same category according to the water regime parameters, so that each group of data is time series data of the same water regime parameter; A similarity mean determination unit is used to draw a curve graph corresponding to the water condition parameter based on the time series data, and sequentially calculate the similarity between any two curve graphs in each group to obtain a similarity mean; The water regime parameter determination unit is used to determine whether the water regime parameter is a strongly correlated water regime parameter corresponding to the water regime anomaly type based on the mean similarity value of the water regime parameter.

[0012] As a further solution of the present invention, the real-time water regime prediction module includes: The input window definition unit is used to define the input window. Each sample is all the characteristic data of a certain period of time in the past. The characteristic data includes water parameter values, climate data and statistical characteristics. The prediction window definition unit is used to define the prediction window. The label corresponding to each sample is the water condition parameter after a certain period of time in the past N hours. The collection classification unit is used to divide all samples into training set, validation set and test set in chronological order; The model learning unit is used to input the training set data into the LSTM model, minimize the loss function through the backpropagation algorithm and optimizer, and iteratively update the model parameters; The model optimization unit is used to perform early stopping on the validation set, monitor the performance of the model on the validation set, stop training when the validation set loss no longer decreases, and restore the model parameters of the group with the best performance on the validation set; The model evaluation unit is used to evaluate the final performance of the LSTM model based on the test set. The evaluation indicators include root mean square error, mean absolute error, and determination coefficient.

[0013] As a further solution of the present invention, the potential exception type module includes: The curve color adding unit is used to determine the curve color according to the water condition parameter type and color all the change curves in the predicted change curve sub-graph and the historical change curve graph; A similarity value calculation unit is used to perform DTW distance calculation on the predicted change curve subgraph and the historical change curve graph, and normalize the distance value calculated by DTW to obtain a similarity value; The similarity value determination unit is used to summarize all similarity values ​​obtained through the predicted change curve subgraph and take the largest one as the final similarity value.

[0014] Compared with the prior art, the present invention has the following beneficial effects: By analyzing historical water regime data, the strongly correlated water regime parameters corresponding to each type of water regime anomaly are identified. The predicted change curve is then split based on these strongly correlated water regime parameters, resulting in several subgraphs. Similarity is then calculated between these subgraphs and the corresponding historical change curves, and the potential anomaly type is determined based on these similarities. This approach uses only the curves of the strongly correlated water regime parameters for calculations, eliminating interference from other parameter curves. Furthermore, since the graph contains multiple curves, a single calculation allows for comparison of these multiple curves. This approach comprehensively considers the interactions and influences of multiple strongly correlated water regime parameters, enabling rapid and accurate identification of water regime anomaly types. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a flow chart of a lake water regime monitoring method based on deep learning.

[0016] Figure 2 Flowchart for determining strongly correlated water regime parameters in a deep learning-based lake water regime monitoring method.

[0017] Figure 3 This is a flowchart for predicting real-time water regime data in a deep learning-based lake water regime monitoring method.

[0018] Figure 4 This is a flowchart for similarity calculation in a deep learning-based lake water regime monitoring method.

[0019] Figure 5 Flowchart for identifying potential anomaly types in a deep learning-based lake water regime monitoring method.

[0020] Figure 6 Flowchart for generating early warning information and monitoring plans in a deep learning-based lake water regime monitoring method.

[0021] Figure 7 This is a structural diagram of a lake water condition monitoring system based on deep learning. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical solutions and advantages of the present invention clearer, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0023] The specific implementation of the present invention is described in detail below with reference to specific embodiments.

[0024] like Figure 1 As shown, an embodiment of the present invention provides a lake water condition monitoring method based on deep learning, the method comprising the following steps: S100, classifying historical water regime data based on water regime anomaly types, where each historical water regime data includes time series data of various water regime parameters; S200, analyzing historical water regime data of each category, determining strongly correlated water regime parameters corresponding to each water regime anomaly type, and determining a historical change curve graph, the historical change curve graph including change curves of all strongly correlated water regime parameters; S300: Collect real-time water regime data, predict the real-time water regime data based on deep learning and climate conditions, and obtain a predicted change curve graph, which includes change curves of all water regime parameters; S400, splitting the predicted change curve graph based on the strongly correlated water regime parameters corresponding to each water regime anomaly type to obtain a plurality of predicted change curve subgraphs; S500, calculating the similarity between the predicted change curve subgraph and the corresponding historical change curve graph, and determining the potential anomaly type based on the similarity value; S600: Generate early warning information and monitoring plans based on potential abnormality types.

[0025] In an embodiment of the present invention, historical water regime data is first retrieved. Each historical water regime data set contains time-series data for various water regime parameters, including pH, dissolved oxygen, transparency, water temperature, water level, flow rate, and so on. Each historical water regime data set is also labeled with a water regime anomaly type, including eutrophication, water ecological anomalies, sediment content anomalies, and water level anomalies. The historical water regime data is then categorized according to the anomaly type, with each category corresponding to the same anomaly type. The historical water regime data for each category is then analyzed sequentially to determine the strongly correlated water regime parameters corresponding to each anomaly type. Strongly correlated water regime parameters are parameters that clearly reflect or cause that anomaly type. A historical change curve is then plotted based on the determined strongly correlated water regime parameters. A single historical change curve reflects the change curves for all strongly correlated water regime parameters. Furthermore, real-time water regime data and climate conditions are collected, and deep learning methods are used to predict the real-time water regime data, resulting in a predicted change curve. A single predicted change curve contains the change curves for all water regime parameters. The predicted change curve graph is then split based on the strongly correlated water parameters corresponding to each type of water anomaly, resulting in several predicted change curve subgraphs. For example, if there are 10 types of water anomaly, 10 predicted change curve subgraphs will be generated, each containing the corresponding change curves for all strongly correlated water parameters. Next, a similarity calculation is performed between the predicted change curve subgraph and the corresponding historical change curve graph, and the potential anomaly type is determined based on the similarity value. In this way, only the change curves of the strongly correlated water parameters are used for calculation, eliminating interference from other parameter curves. Furthermore, the graph contains multiple curves, and a one-time similarity calculation is performed, comprehensively considering the interactions and influences of multiple strongly correlated water parameters, enabling the rapid and accurate identification of water anomaly types. Finally, early warning information and monitoring plans are generated based on the potential anomaly type, providing timely warnings and improving monitoring efficiency.

[0026] like Figure 2 As shown, as a preferred embodiment of the present invention, the step of determining the strongly correlated water condition parameters corresponding to each water condition anomaly type specifically includes: S201, extracting and grouping all historical water regime data of the same category according to water regime parameters, so that each group of data is time series data of the same water regime parameter; S202, plotting a curve graph corresponding to the water condition parameter based on the time series data, and sequentially calculating the similarity between any two curve graphs in each group to obtain a mean similarity value; S203: Determine, based on the mean value of the similarity of the water regime parameter, whether the water regime parameter is a strongly correlated water regime parameter corresponding to the water regime anomaly type.

[0027] In an embodiment of the present invention, in order to eliminate the interference of other parameters and automatically extract strongly correlated water condition parameters, all historical water condition data in the same category will be extracted and grouped according to the water condition parameters, so that each group of data is time series data of the same water condition parameter. For example, if there are 20 types of water condition parameters, there will be 20 groups of data. Then, a curve graph corresponding to the water condition parameter is drawn for each group of data. For example, if there are 50 historical water condition data in this category, there will be 50 curve graphs for each group of data. Then, the 50 curve graphs will be paired in pairs, and the similarity between any two curve graphs in each group will be calculated. Many similarity results will be obtained, and the mean similarity will be calculated. Then, it is determined whether the similarity mean is greater than the set value. If it is greater than, it means that the change trends of the 50 curves of the water condition parameter are similar, and the water condition parameter is a strongly correlated water condition parameter corresponding to the water condition anomaly type; if it is less than, it is not a strongly correlated water condition parameter.

[0028] like Figure 3 As shown, as a preferred embodiment of the present invention, the step of predicting real-time water condition data based on deep learning and climate conditions specifically includes: S301, defining an input window, where each sample is all feature data from a certain period of time in the past, including water parameter values, climate data, and statistical features; S302, defining a prediction window, where the label corresponding to each sample is the water condition parameter status of a certain period of time in the past delayed by N hours; S303, dividing all samples into training set, validation set and test set in chronological order; S304, input the training set data into the LSTM model, minimize the loss function through the back propagation algorithm and optimizer, and iteratively update the model parameters; S305, performing early stopping on the validation set to monitor the performance of the model on the validation set. When the validation set loss stops decreasing, stop training and restore the model parameters of the group with the best performance on the validation set. S306, evaluate the final performance of the LSTM model based on the test set, and the evaluation indicators include root mean square error, mean absolute error, and determination coefficient.

[0029] In an embodiment of the present invention, when performing predictions based on deep learning, historical water regime data and corresponding climate data must first be retrieved and preprocessed. This preprocessing includes handling missing values, outliers, and data standardization. The data is then converted into the "sample-label" pairs required by the deep learning model. An input window is defined, where each sample represents all feature data from a certain period in the past. Feature data includes water regime parameter values, climate data, and statistical features, such as mean and variance. A prediction window is also defined, where the label corresponding to each sample represents the water regime parameter N hours later from a certain period in the past, where N is a fixed value, such as 6 hours. The data is then divided into a training set (e.g., the earliest 70% of data), a validation set (the middle 15% of data), and a test set (the latest 15% of data) in chronological order. The chronological order must not be disrupted to avoid data confusion (using future information to predict the past). The model architecture is then selected. Since the data is time series data, an LSTM (Long Short-Term Memory) network can effectively capture temporal dependencies. Mean squared error or mean absolute error is used as the loss function to minimize the difference between the predicted and true values. The training set data is then fed into the LSTM model. Backpropagation and an optimizer are used to minimize the loss function, iteratively updating the model parameters. Early stopping is performed on the validation set to monitor the model's performance. When the validation set loss stops decreasing or even begins to increase (indicating overfitting), training is stopped and the model parameters that performed best on the validation set are restored. Finally, the final performance of the LSTM model is evaluated on the test set using a variety of comprehensive metrics.

[0030] like Figure 4 As shown in FIG. 1 , as a preferred embodiment of the present invention, the step of calculating the similarity between the predicted change curve subgraph and the corresponding historical change curve graph specifically includes: S501, determining the curve color according to the water regime parameter type, and coloring all the change curves in the predicted change curve subgraph and the historical change curve graph; S502, performing DTW distance calculation on the predicted change curve subgraph and the historical change curve graph, and normalizing the distance values ​​calculated by DTW to obtain similarity values; S503: Summarize all similarity values ​​obtained through the predicted change curve subgraph, and take the largest one as the final similarity value.

[0031] In an embodiment of the present invention, in order to perform similarity matching more accurately, a curve color is defined for each water condition parameter type, and then all change curves in the predicted change curve subgraph and the historical change curve graph are colored. The predicted change curve subgraph and the historical change curve graph are then subjected to DTW distance calculation, and the distance value calculated by DTW is normalized to obtain a similarity value, for example: similarity = 1 / (1+DTW distance). The DTW algorithm uses dynamic programming to find the optimal nonlinear alignment path between the two sequences, automatically compensating for offsets or local deformations on the time axis (such as the time lag of abnormal peaks), significantly improving the accuracy of similarity calculation and the robustness of anomaly detection. Finally, all similarity values ​​obtained through the predicted change curve subgraph are summarized, and the largest one is taken as the final similarity value for subsequent calculations.

[0032] like Figure 5 As shown, as a preferred embodiment of the present invention, the step of determining the potential abnormality type based on the similarity value specifically includes: S504, determining whether the similarity value of each predicted change curve subgraph is greater than a similarity threshold; if so, the corresponding water regime anomaly type is determined as a potential anomaly type; S505 , summarizing potential abnormality types determined by a plurality of the predicted change curve subgraphs to obtain an abnormality type set.

[0033] In this embodiment of the present invention, after the final similarity value of each predicted change curve subgraph is determined, a determination is made as to whether the similarity value exceeds a similarity threshold. This threshold is a pre-set value. When the similarity threshold is exceeded, a potential anomaly type is determined. The potential anomaly types determined for the ten predicted change curve subgraphs are then summarized. If more than one potential anomaly type is determined, a composite anomaly is present.

[0034] like Figure 6 As shown, as a preferred embodiment of the present invention, the step of generating early warning information and monitoring solutions according to potential abnormality types specifically includes: S601, determining the abnormality level based on the similarity value, and generating water situation warning information based on the potential abnormality type and abnormality level; S602, determining the monitoring frequency, monitoring range, and newly added monitoring points of the corresponding strongly correlated water regime parameters according to the anomaly level; S603: Determine a water situation treatment plan based on the potential abnormality type and abnormality level.

[0035] In an embodiment of the present invention, an abnormality level table is established, and each level in the abnormality level table corresponds to a similar value range. In this way, the abnormality level can be directly determined based on the similarity value, and then water situation warning information can be generated. In addition, the monitoring frequency, monitoring range and newly added monitoring points of the corresponding strongly correlated water situation parameters will be determined according to the abnormality level. The higher the level, the higher the risk, the higher the monitoring frequency, the larger the monitoring range, and the density of the monitoring points. Finally, the water situation treatment plan will be determined based on the potential abnormality type and abnormality level. Multiple sets of treatment plans are formulated in advance for each abnormality type, and each set of treatment plans is suitable for different levels.

[0036] like Figure 7 As shown, an embodiment of the present invention further provides a lake water condition monitoring system based on deep learning, the system comprising: The historical water regime classification module 100 is used to classify historical water regime data based on the type of water regime anomaly, where each historical water regime data contains time series data of various water regime parameters; A strongly correlated parameter module 200 is used to analyze the historical water regime data of each category, determine the strongly correlated water regime parameters corresponding to each water regime anomaly type, and determine a historical change curve graph, which includes change curves of all strongly correlated water regime parameters; The real-time water regime prediction module 300 is used to collect real-time water regime data, predict the real-time water regime data based on deep learning and climate conditions, and obtain a prediction change curve graph, which contains the change curves of all water regime parameters; The prediction graph splitting module 400 is used to split the prediction change curve graph based on the strongly correlated water regime parameters corresponding to each water regime anomaly type to obtain a plurality of prediction change curve subgraphs; The potential anomaly type module 500 is used to calculate the similarity between the predicted change curve subgraph and the corresponding historical change curve graph, and determine the potential anomaly type based on the similarity value; The abnormality warning monitoring module 600 is used to generate early warning information and monitoring solutions according to potential abnormality types.

[0037] In the embodiment of the present invention, the strongly correlated parameter module 200 includes: The water regime parameter grouping unit is used to extract and group all historical water regime data of the same category according to the water regime parameters, so that each group of data is time series data of the same water regime parameter; A similarity mean determination unit is used to draw a curve graph corresponding to the water condition parameter based on the time series data, and sequentially calculate the similarity between any two curve graphs in each group to obtain a similarity mean; The water regime parameter determination unit is used to determine whether the water regime parameter is a strongly correlated water regime parameter corresponding to the water regime anomaly type based on the mean similarity value of the water regime parameter.

[0038] In an embodiment of the present invention, the real-time water regime prediction module 300 includes: The input window definition unit is used to define the input window. Each sample is all the characteristic data of a certain period of time in the past. The characteristic data includes water parameter values, climate data and statistical characteristics. The prediction window definition unit is used to define the prediction window. The label corresponding to each sample is the water condition parameter after a certain period of time in the past N hours. The collection classification unit is used to divide all samples into training set, validation set and test set in chronological order; The model learning unit is used to input the training set data into the LSTM model, minimize the loss function through the backpropagation algorithm and optimizer, and iteratively update the model parameters; The model optimization unit is used to perform early stopping on the validation set, monitor the performance of the model on the validation set, stop training when the validation set loss no longer decreases, and restore the model parameters of the group with the best performance on the validation set; The model evaluation unit is used to evaluate the final performance of the LSTM model based on the test set. The evaluation indicators include root mean square error, mean absolute error, and determination coefficient.

[0039] In the embodiment of the present invention, the potential abnormality type module 500 includes: The curve color adding unit is used to determine the curve color according to the water condition parameter type and color all the change curves in the predicted change curve sub-graph and the historical change curve graph; A similarity value calculation unit is used to perform DTW distance calculation on the predicted change curve subgraph and the historical change curve graph, and normalize the distance value calculated by DTW to obtain a similarity value; The similarity value determination unit is used to summarize all similarity values ​​obtained through the predicted change curve subgraph and take the largest one as the final similarity value.

[0040] The above is only a detailed description of the preferred embodiments of the present invention, which is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

[0041] It should be understood that, although the various steps in the flow chart of each embodiment of the present invention are shown in sequence according to the indication of the arrows, these steps are not necessarily performed in sequence according to the order indicated by the arrows. Unless otherwise specified herein, the execution of these steps is not strictly limited in order, and these steps can be performed in other orders. Moreover, at least a portion of the steps in each embodiment may include a plurality of sub-steps or a plurality of stages, and these sub-steps or stages are not necessarily performed at the same time, but can be performed at different times, and the execution order of these sub-steps or stages is not necessarily performed in sequence, but can be performed in turn or alternately with at least a portion of other steps or sub-steps or stages of other steps.

[0042] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing the relevant hardware through a computer program. The program can be stored in a non-volatile computer-readable storage medium. When executed, the program can include the processes of the above-described method embodiments. Any reference to memory, storage, database, or other media used in the various embodiments provided herein may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct RAMbus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM).

[0043] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the disclosure in the specification and examples. This application is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the art not disclosed herein. The description and examples are to be considered merely as exemplary, and the true scope and spirit of the present disclosure are indicated by the claims.

Claims

1. A lake water regime monitoring method based on deep learning, characterized in that: The method comprises the following steps: Classify historical water regime data based on water regime anomaly types. Each historical water regime data contains time series data of various water regime parameters. Analyze the historical water regime data of each category, determine the strongly correlated water regime parameters corresponding to each water regime anomaly type, and determine a historical change curve chart, which only contains the change curves of all strongly correlated water regime parameters; Collect real-time water condition data, predict it based on deep learning and climate conditions, and obtain a predicted change curve chart, which contains the change curves of all water condition parameters; The predicted change curve is split based on the strongly correlated water regime parameters corresponding to each water regime anomaly type to obtain several predicted change curve sub-graphs; Calculate the similarity between the predicted change curve subgraph and the corresponding historical change curve graph, and determine the potential anomaly type based on the similarity value; Generate early warning information and monitoring plans based on potential abnormality types.

2. The lake water regime monitoring method based on deep learning according to claim 1, characterized in that: The step of determining the strongly correlated water regime parameters corresponding to each water regime anomaly type specifically includes: All historical water regime data in the same category are extracted and grouped according to water regime parameters, so that each group of data is time series data of the same water regime parameter; Based on the time series data, a curve graph of the corresponding water condition parameters is drawn, and the similarity between any two curve graphs in each group is calculated in turn to obtain the mean similarity value; The water regime parameter is determined to be a strongly correlated water regime parameter corresponding to the water regime anomaly type based on its similarity mean.

3. The lake water regime monitoring method based on deep learning according to claim 1, characterized in that: The step of predicting real-time water condition data based on deep learning and climate conditions specifically includes: Define the input window. Each sample is all the characteristic data of a certain period in the past. The characteristic data includes water parameter values, climate data, and statistical characteristics. Define the prediction window, and the label corresponding to each sample is the water parameter status after a certain period of time in the past N hours; All samples are divided into training set, validation set and test set in chronological order; Input the training set data into the LSTM model, minimize the loss function through the backpropagation algorithm and optimizer, and iteratively update the model parameters; Perform early stopping on the validation set to monitor the model's performance on the validation set. When the validation set loss stops decreasing, stop training and restore the model parameters of the group with the best performance on the validation set. The final performance of the LSTM model is evaluated based on the test set, and the evaluation indicators include root mean square error, mean absolute error, and determination coefficient.

4. The lake water regime monitoring method based on deep learning according to claim 1, characterized in that: The step of calculating the similarity between the predicted change curve subgraph and the corresponding historical change curve graph specifically includes: Determine the curve color according to the water regime parameter type, and color all the change curves in the predicted change curve subgraph and the historical change curve graph; Perform DTW distance calculation on the predicted change curve subgraph and the historical change curve graph, and normalize the distance value calculated by DTW to obtain the similarity value; All similarity values ​​obtained through the predicted change curve subgraph are summarized, and the largest one is taken as the final similarity value.

5. The lake water regime monitoring method based on deep learning according to claim 4 is characterized in that: The step of determining the potential anomaly type based on the similarity value specifically includes: Determine whether the similarity value of each predicted change curve subgraph is greater than the similarity threshold. If it is greater than, the corresponding water abnormality type is determined as a potential abnormality type; The potential anomaly types determined by a plurality of the predicted change curve subgraphs are aggregated to obtain an anomaly type set.

6. The lake water regime monitoring method based on deep learning according to claim 1, characterized in that: The step of generating early warning information and monitoring plans according to potential abnormality types specifically includes: Determine the anomaly level based on the similarity value, and generate water situation warning information based on the potential anomaly type and anomaly level; Determine the monitoring frequency, monitoring range, and newly added monitoring points for the corresponding strongly correlated water parameters based on the anomaly level; Determine the water situation treatment plan based on the potential abnormality type and abnormality level.

7. A lake water condition monitoring system based on deep learning, characterized in that: The system comprises: The historical water regime classification module is used to classify historical water regime data based on the type of water regime anomaly. Each historical water regime data contains time series data of various water regime parameters. The strongly correlated parameter module is used to analyze the historical water regime data of each category, determine the strongly correlated water regime parameters corresponding to each water regime anomaly type, and determine the historical change curve graph, which only contains the change curves of all strongly correlated water regime parameters; The real-time water regime prediction module is used to collect real-time water regime data, predict the real-time water regime data based on deep learning and climate conditions, and obtain a predicted change curve chart, which contains the change curves of all water regime parameters; The prediction graph splitting module is used to split the prediction change curve graph based on the strongly correlated water regime parameters corresponding to each water regime anomaly type to obtain several prediction change curve subgraphs; The potential anomaly type module is used to calculate the similarity between the predicted change curve subgraph and the corresponding historical change curve graph, and determine the potential anomaly type based on the similarity value; The abnormal warning monitoring module is used to generate early warning information and monitoring plans based on potential abnormality types.

8. The lake water condition monitoring system based on deep learning according to claim 7 is characterized in that: The strongly correlated parameter module includes: The water regime parameter grouping unit is used to extract and group all historical water regime data of the same category according to the water regime parameters, so that each group of data is time series data of the same water regime parameter; A similarity mean determination unit is used to draw a curve graph corresponding to the water condition parameter based on the time series data, and sequentially calculate the similarity between any two curve graphs in each group to obtain a similarity mean; The water regime parameter determination unit is used to determine whether the water regime parameter is a strongly correlated water regime parameter corresponding to the water regime anomaly type based on the mean similarity value of the water regime parameter.

9. The lake water condition monitoring system based on deep learning according to claim 7, characterized in that: The real-time water regime prediction module includes: The input window definition unit is used to define the input window. Each sample is all the characteristic data of a certain period of time in the past. The characteristic data includes water parameter values, climate data and statistical characteristics. The prediction window definition unit is used to define the prediction window. The label corresponding to each sample is the water condition parameter after a certain period of time in the past N hours. The collection classification unit is used to divide all samples into training set, validation set and test set in chronological order; The model learning unit is used to input the training set data into the LSTM model, minimize the loss function through the backpropagation algorithm and optimizer, and iteratively update the model parameters; The model optimization unit is used to perform early stopping on the validation set, monitor the performance of the model on the validation set, stop training when the validation set loss no longer decreases, and restore the model parameters of the group with the best performance on the validation set; The model evaluation unit is used to evaluate the final performance of the LSTM model based on the test set. The evaluation indicators include root mean square error, mean absolute error, and determination coefficient.

10. The lake water condition monitoring system based on deep learning according to claim 7, characterized in that: The potential exception type module includes: The curve color adding unit is used to determine the curve color according to the water condition parameter type and color all the change curves in the predicted change curve sub-graph and the historical change curve graph; A similarity value calculation unit is used to perform DTW distance calculation on the predicted change curve subgraph and the historical change curve graph, and normalize the distance value calculated by DTW to obtain a similarity value; The similarity value determination unit is used to summarize all similarity values ​​obtained through the predicted change curve subgraph and take the largest one as the final similarity value.

Images