Early warning method for dangerous conditions in the lower reaches of the Yellow River based on machine learning

Abstract

The present invention relates to the field of river management early warning technology, and specifically discloses a dangerous situation early warning method for dangerous works in the lower reaches of the Yellow River based on machine learning, including: S1, data collection, S2, inter-frame difference image processing, S3, convolutional neural network design, S4, data set construction, S5, deep neural network automatic deployment model, S6, data-driven engineering accident mechanism analysis, S7, Bayesian network model construction, S8, model optimization and integration, and S9, evaluation and early warning; the present invention can more accurately identify and predict dangerous situations and improve the accuracy of early warning through deep learning models and Bayesian network models. The data-driven analysis method can learn potential risk patterns and characteristics from a large amount of historical data, reduce human errors, combine multi-source data such as video monitoring, flow, water level, flow velocity, soil moisture and pore water pressure, comprehensively monitor the situation in dangerous work areas, conduct risk assessment from multiple angles, and provide more comprehensive and accurate early warning of dangerous situations.

Description

Technical Field

[0001] The present invention belongs to the technical field of river management early warning, and specifically relates to a method for early warning of dangerous conditions in the lower reaches of the Yellow River based on machine learning. Background Art

[0002] The dangerous works in the lower reaches of the Yellow River play a pivotal role in river flood control safety, and the evolution of its dangerous situation changes rapidly. Strengthen the "four precautions" of forecasting, early warning, rehearsal, and emergency plan to prevent disaster risks. Previous research has focused on manual inspections, post-event rescue, and analysis of hydrodynamic conditions, and there is still a lack of research on the integration of machine learning algorithms and analysis of dangerous situation mechanisms. It is urgent to build professional models of water conservancy projects, artificial intelligence models, and visualization models to support intelligent simulations such as safety monitoring.

[0003] Compared with the demand for high-quality development of water conservancy, the following problems still exist:

[0004] 1. Imperfect early warning measures for water conservancy projects

[0005] 2. The model algorithm is far from the high-fidelity goal

[0006] 3. The coverage and stability of intelligent model application scenarios are not high

[0007] 4. The integration of advanced technologies such as artificial intelligence with business is not in-depth.

[0008] After an in-depth investigation of the current research status at home and abroad, many scholars have conducted research on image recognition algorithms as an important branch in the field of computer vision. The requirements for the accuracy of target object recognition and adaptability in complex environments are constantly increasing. The target monitoring system combined with the deep learning framework has become a hot topic in current research. It is widely used in medicine, agriculture and other fields. In recent years, it has been promoted and applied in the water conservancy industry, which has improved the accuracy and precision of image recognition in the water conservancy industry. It has strong real-time performance, but it is still in its infancy in the industry and is not widely used for water project safety warning. Bayesian networks have been successfully used in many intelligent fields, such as artificial intelligence and data mining, with their powerful uncertainty reasoning methods. Due to the unique advantages of Bayesian methods in training, learning and probability prediction, many scholars at home and abroad have used this method for risk assessment in the field of disaster risk management. However, in the study of flood disasters, due to the uneven characteristics of submerged sample data and influencing factors, few researchers have proposed corresponding improvement methods specifically for this type of disaster.

[0009] This invention aims to introduce digital technology into the management of typical dangerous projects in the lower reaches of the Yellow River. Based on machine learning and mechanism analysis methods, it carries out research on key technologies for early warning of dangerous situations in typical dangerous projects in the lower reaches of the Yellow River, so as to improve the natural disaster prevention and control capabilities of the whole society.

[0010] In response to this, the inventor proposed a dangerous situation warning method for the lower reaches of the Yellow River based on machine learning to solve the above problems. Summary of the invention

[0011] The purpose of the present invention is to provide a method for early warning of dangerous conditions in the lower reaches of the Yellow River based on machine learning, so as to solve the problems raised in the above-mentioned background technology.

[0012] To achieve the above object, the present invention provides the following technical solutions:

[0013] The method for early warning of dangerous conditions in the lower reaches of the Yellow River based on machine learning includes the following steps:

[0014] S1. Data collection. Using the water project video monitoring system of the river management department, video data is collected, and measured data on levee types, measured flow, water level, and flow velocity are collected. Video data and measured data are preprocessed to eliminate noise, unify data formats, and combined with project monitoring data, including soil moisture and osmotic pressure, and preliminary anomaly detection is performed;

[0015] S2, inter-frame difference image processing, taking frames at intervals on the video data in step S1, and generating a difference map using an inter-frame difference method, extracting change information between video frames, setting a frame taking interval to ensure accurate capture of change information, and performing image enhancement processing on the generated difference map to highlight key features and reduce background interference;

[0016] S3, convolutional neural network design, design a convolutional neural network model, input the difference map processed in step S2 into the CNN model, perform feature learning and extraction, use labeled data for training, and use cross-validation method to evaluate model performance to prevent overfitting;

[0017] S4. Dataset construction: construct a dataset for detecting typical dangerous conditions in the lower reaches of the Yellow River, covering samples with diverse characteristics. Through data enhancement technology, the diversity of the dataset is increased and the robustness of the model is improved.

[0018] S5. Deep neural network automatic allocation model. According to the different single and multi-targets captured, a deep neural network automatic allocation model is proposed and trained. The model should have adaptive adjustment capabilities, automatically adjust parameters according to the characteristics of the input data, and use optimization algorithms, including Adam and RMSprop, to iteratively optimize network parameters to improve the detection accuracy of the model. An early stopping mechanism is set to prevent overfitting.

[0019] S6. Data-driven analysis of engineering accident mechanism: Based on the collected measured data and engineering monitoring data, a data-driven calculation model for engineering accident in the lower reaches of the Yellow River is constructed using statistical methods. The process of engineering accident under different conditions is simulated, and the abnormal correlation coefficient is used as the evaluation index to analyze the correlation of abnormal data, identify potential dangers, and study the accident mechanism under different conditions through simulation experiments to reveal the internal factors of engineering accident.

[0020] S7. Bayesian network model construction: Based on Bayesian theory, combined with expert knowledge and deterministic forecast results, a water project safety early warning model with a polymorphic Bayesian network method is constructed to handle a variety of uncertain factors, analyze the probability density function of various water project safety influencing factors, and calculate the probability of project accidents by combining the correlation between factors. By continuously optimizing the Bayesian network structure and parameters, the early warning accuracy of the model is improved;

[0021] S8. Model optimization and integration: by continuously optimizing convolutional neural networks, data-driven models, and Bayesian network models, the accuracy and real-time performance of the overall early warning system are improved. Hyperparameter optimization techniques, including grid search and random search, are used to adjust model parameters. The research results of the three methods are integrated to form a comprehensive early warning model. Model fusion techniques, including weighted average and voting methods, are used to improve the performance and stability of the comprehensive model.

[0022] S9. Assessment and early warning: Apply a comprehensive early warning model to conduct real-time monitoring and assessment of typical dangerous projects in the lower reaches of the Yellow River, identify potential dangerous situations, and use streaming data processing technology to achieve real-time data processing and analysis. According to the early warning results output by the model, timely issue early warning signals to guide engineering protection measures, set up a multi-level early warning mechanism, and distinguish early warning responses at different risk levels.

[0023] Preferably, the inter-frame difference image processing in step S2 adopts an inter-frame difference method, which is expressed as:

[0024] D i (x,y)=|I t (x,y)-I t-1 (x,y)|

[0025] Among them, Dt(x,y) represents the difference between the t-th frame and the t-1-th frame at the coordinate (x,y), It(x,y) and I t-1 (x, y) represents the pixel value at (x, y) in the t-th frame and the t-1-th frame respectively. By calculating the pixel difference between adjacent frames, the motion information of the target object in the video is extracted to provide basic data for subsequent feature extraction.

[0026] Preferably, the convolutional neural network model in S3 includes multiple convolutional layers, pooling layers, local response normalization layers and fully connected layers. The convolutional layers are used to extract image features, the pooling layers are used to reduce feature dimensionality and computational complexity, the local response normalization layers are used to improve the generalization ability of the model, and the fully connected layers map the extracted features to the output label space.

[0027] Preferably, the convolution operation in the convolutional neural network model is:

[0028]

[0029] Among them, f is the input image, g is the convolution kernel, and (i, j) is the coordinate of the output image;

[0030] The pooling operation is expressed as:

[0031]

[0032] Among them, P(i,j) is the value after pooling, and window is the pooling window;

[0033] The output of the fully connected layer is expressed as:

[0034] y=σ(Wx+b)

[0035] Where W is the weight matrix, x is the input feature vector, b is the bias vector, and σ is the activation function ReLU or Sigmoid;

[0036] Image features are extracted through convolution operations, dimensionality reduction and feature aggregation are performed through pooling, and finally the classification results are output through the fully connected layer.

[0037] Preferably, in step S5, the deep neural network automatic deployment model improves classification accuracy by optimizing the loss function and adjusting the model parameters, which is expressed as:

[0038]

[0039] L is the loss function, N is the number of samples, yi is the true label, is the predicted probability.

[0040] Preferably, in step S6, the potential danger is identified by calculating the abnormal correlation coefficient, which is expressed as:

[0041]

[0042] Among them, r is the correlation coefficient, Xi and Yi are the observed values ​​of the two variables, and is the mean of the variable.

[0043] Preferably, the Bayesian network in step S7 is:

[0044]

[0045] Among them, P(A|B) is the probability of event A occurring when event B occurs, P(B|A) is the probability of event B occurring when event A occurs, P(A) and P(B) are the prior probabilities of events A and B occurring, respectively. The probability of danger is calculated by Bayesian theorem, and a water project safety early warning model based on a polymorphic Bayesian network method is established.

[0046] Preferably, the model fusion technology in step S8 is expressed as:

[0047]

[0048] is the prediction result after fusion, is the prediction result of the i-th model, α is the corresponding weight, and the performance and stability of the comprehensive model are improved through hyperparameter optimization and model fusion.

[0049] Compared with the prior art, the present invention has the following beneficial effects:

[0050] (1) The present invention can more accurately identify and predict dangerous situations and improve the accuracy of early warnings through deep learning models and Bayesian network models. The data-driven analysis method can learn potential risk patterns and characteristics from a large amount of historical data, reduce human errors, and combine multi-source data such as video monitoring, flow, water level, flow velocity, soil moisture and pore water pressure to comprehensively monitor the situation in dangerous areas, conduct risk assessments from multiple angles, and provide more comprehensive and accurate early warnings of dangerous situations.

[0051] (2) In the present invention, from local image features to overall environmental factors, the model can comprehensively consider various influencing factors and conduct multi-level and multi-angle analysis. The Bayesian network can combine expert knowledge and data-driven analysis results to provide clear causal relationships and probabilistic explanations, helping relevant personnel understand the mechanism and cause of the dangerous situation and facilitate the adoption of targeted protective measures.

[0052] (3) This invention combines the pain points of dangerous engineering management in the lower reaches of the Yellow River, utilizes the fast response speed of deep learning models, and combines existing computer vision technology, Bayesian methods and other mature technologies to achieve accurate and efficient identification of multiple targets and scenarios, and constructs an engineering risk warning model based on machine learning and mechanism analysis to explore the uncertainty characteristics of water engineering risk changes during the flood season and identify influencing factors. Taking full account of the uncertainty of water engineering safety factors, based on historical experience, statistical data and normative standards, an improved Bayesian method based on data reconstruction is proposed to grasp the key safety hazards of danger on the basis of accurate identification, improve safety warning capabilities, and comprehensively improve the level of intelligent operation and management. Brief Description of the Drawings

[0053] Figure 1 This is the flowchart of the warning method for dangerous engineering situations in the lower reaches of the Yellow River based on machine learning according to the present invention. Detailed Embodiments

[0054] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0055] Embodiment 1:

[0056] Please refer to Figure 1 As shown, the warning method for dangerous engineering situations in the lower reaches of the Yellow River based on machine learning includes the following steps:

[0057] S1. Data collection: Collect video monitoring data of a certain dangerous engineering monitoring site in the lower reaches of the Yellow River, including measured flow rate, water level, flow velocity, soil humidity, and osmotic pressure data;

[0058] Video file name: 20240301_0800_A.mp4;

[0059] Shooting time: 08:00 on March 1, 2024;

[0060] Shooting location: A;

[0061] Video duration: 15 minutes;

[0062] Video content: A certain dangerous engineering area in the lower reaches of the Yellow River, monitoring the dike and water flow conditions.

[0063] Measured data: The following is the csv measured data from 08:00 to 09:00 on the morning of March 1, 2024:

[0064] Date, Time, Location, Dike Type, Flow Rate, Water Level, Flow Velocity, Soil Humidity, Osmotic Pressure, Label are as follows:

[0065] 2024-03-01, 08:00, A, TypeA, 300, 5.2, 2.5, 0.45, 120, Normal

[0066] 2024-03-01, 08:15, A, TypeA, 305, 5.25, 2.55, 0.46, 123, Normal

[0067] 2024-03-01,08:30,A,TypeA,310,5.3,2.6,0.47,125,Normal

[0068] 2024-03-01,08:45,A,TypeA,320,5.4,2.7,0.48,130,Abnormal

[0069] 2024-03-01,09:00,A,TypeA,330,5.5,2.8,0.49,135,Abnormal;

[0070] S2, inter-frame difference image processing, extracting one frame from the video every 5 minutes:

[0071] Extraction time points: 08:00, 08:05, 08:10, 08:15, 08:20, 08:25, 08:30, 08:35, 08:40, 08:45, 08:50, 08:55, 09:00;

[0072] Extract image filename:

[0073] frame_20240701_0800.png

[0074] frame_20240701_0805.png

[0075] frame_20240701_0810.png

[0076] frame_20240701_0815.png

[0077] frame_20240701_0820.png

[0078] frame_20240701_0825.png

[0079] frame_20240701_0830.png

[0080] frame_20240701_0835.png

[0081] frame_20240701_0840.png

[0082] frame_20240701_0845.png

[0083] frame_20240701_0850.png

[0084] frame_20240701_0855.png

[0085] frame_20240701_0900.png

[0086] Generate an inter-frame difference image:

[0087] Difference image file name:

[0088] diff_frame_20240701_0805_0800.png

[0089] diff_frame_20240701_0810_0805.png

[0090] diff_frame_20240701_0815_0810.png

[0091] diff_frame_20240701_0820_0815.png

[0092] diff_frame_20240701_0825_0820.png

[0093] diff_frame_20240701_0830_0825.png

[0094] diff_frame_20240701_0835_0830.png

[0095] diff_frame_20240701_0840_0835.png

[0096] diff_frame_20240701_0845_0840.png

[0097] diff_frame_20240701_0850_0845.png

[0098] diff_frame_20240701_0855_0850.pngdiff_frame_20240701_0900_0855.png;

[0099] S3. Convolutional neural network design, convolutional neural network architecture:

[0100] Input layer: difference image;

[0101] Convolutional layer 1: 32 3x3 convolution kernels, ReLU activation function;

[0102] Pooling layer 1: 2x2 max pooling;

[0103] Convolutional layer 2: 64 3x3 convolution kernels, ReLU activation function;

[0104] Pooling layer 2: 2x2 maximum pooling;

[0105] Fully connected layer 1: 128 neurons, ReLU activation function;

[0106] Fully connected layer 2: 2 neurons, Softmax activation function (output: normal and abnormal probabilities); model training:

[0107] Training dataset: features extracted from difference images and corresponding labels (normal / abnormal)

[0108] Loss function: cross entropy loss function;

[0109] Optimization algorithm: Adam optimizer;

[0110] Number of training rounds: 100;

[0111] Batch size: 32

[0112] S4, data set construction, to build a data set for the detection of typical dangerous conditions in the lower reaches of the Yellow River;

[0113] Use the extracted differential images and measured data, combined with the annotation labels, to construct training and test data sets; the training set: 70% of the data; the test set: 30% of the data;

[0114] S5. Automatically adjust the deep neural network model and use the validation set to adjust the model hyperparameters, such as learning rate, batch size, and convolution kernel size;

[0115] S6. Data-driven engineering accident mechanism analysis and abnormal correlation coefficient calculation: Taking the measured data from 08:00 to 09:00 on March 1, 2024 as an example, calculate the correlation coefficient between each variable and the label (normal / abnormal);

[0116] S7. Bayesian network model construction, using measured data and expert knowledge to build a Bayesian network model.

[0117] Nodes: flow, water level, flow velocity, soil moisture, osmotic pressure, probability of disaster;

[0118] Edge: dependency relationship between variables;

[0119] S8, model optimization and integration, weighted average of the outputs of the CNN model and the Bayesian network model to obtain the final accident probability prediction;

[0120] S9. Assessment and early warning: Apply comprehensive early warning models to conduct real-time monitoring and assessment of typical dangerous projects in the lower reaches of the Yellow River, identify potential dangerous situations, use streaming data processing technology to achieve real-time data processing and analysis, issue early warning signals in a timely manner based on the early warning results output by the model, guide project protection measures, set up a multi-level early warning mechanism, and distinguish early warning responses at different risk levels;

[0121] 2024-07-01,08:00,A,Normal,Normal,0.95

[0122] 2024-07-01,08:15,A,Normal,Normal,0.93

[0123] 2024-07-01,08:30,A,Normal,Normal,0.90

[0124] 2024-07-01,08:45,A,Abnormal,Normal,0.85

[0125] 2024-07-01,09:00,A location,abnormal,abnormal,0.92.

[0126] The above data and analysis show the entire process from data collection and preprocessing to model prediction and performance evaluation in the dangerous situation early warning system for the lower reaches of the Yellow River based on machine learning. By continuously optimizing and improving the model, the accuracy and reliability of the early warning system can be improved to ensure the safe operation of water projects in the lower reaches of the Yellow River.

[0127] As can be seen from the above, through deep learning models and Bayesian network models, dangerous situations can be more accurately identified and predicted, and the accuracy of early warnings can be improved. Data-driven analysis methods can learn potential risk patterns and characteristics from a large amount of historical data, reduce human errors, and combine multi-source data such as video surveillance, flow, water level, flow velocity, soil moisture and pore water pressure to comprehensively monitor the situation in dangerous areas, conduct risk assessments from multiple angles, and provide more comprehensive and accurate early warnings of dangerous situations.

[0128] From local image features to overall environmental factors, the model can comprehensively consider various influencing factors and conduct multi-level and multi-angle analysis. The Bayesian network can combine expert knowledge and data-driven analysis results to provide clear causal relationships and probabilistic explanations, helping relevant personnel understand the mechanisms and causes of dangerous situations and facilitate the adoption of targeted protective measures.

[0129] Combined with the pain points of dangerous engineering management in the lower reaches of the Yellow River, the fast response speed of deep learning models, and existing computer vision technology, Bayesian methods and other mature technologies, we can achieve accurate and efficient identification of multiple targets and scenarios, and build an engineering risk warning model based on machine learning and mechanism analysis to explore the uncertainty characteristics of water engineering risk changes during the flood season and identify influencing factors. Taking full account of the uncertainty of water engineering safety factors, based on historical experience, statistical data and normative standards, we propose an improved Bayesian method based on data reconstruction, grasp the key safety hazards of danger on the basis of accurate identification, improve safety warning capabilities, and comprehensively improve the level of intelligent operation and management.

[0130] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples" or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine different embodiments or examples described in this specification and the features of different embodiments or examples, unless they are contradictory.

[0131] In the drawings of the embodiments disclosed in the present invention, only the structures related to the embodiments disclosed in the present invention are involved, and other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of the present invention can be combined with each other.

[0132] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for early warning of dangerous conditions in the lower reaches of the Yellow River based on machine learning, characterized in that: The following steps are involved: S1. Data collection. Using the water project video monitoring system of the river management department, video data is collected, and measured data on levee types, measured flow, water level, and flow velocity are collected. Video data and measured data are preprocessed to eliminate noise, unify data formats, and combined with project monitoring data, including soil moisture and osmotic pressure, and preliminary anomaly detection is performed; S2, inter-frame difference image processing, taking frames at intervals on the video data in step S1, and generating a difference map using an inter-frame difference method, extracting change information between video frames, setting a frame taking interval to ensure accurate capture of change information, and performing image enhancement processing on the generated difference map to highlight key features and reduce background interference; S3, convolutional neural network design, design a convolutional neural network model, input the difference map processed in step S2 into the CNN model, perform feature learning and extraction, use labeled data for training, and use cross-validation method to evaluate model performance to prevent overfitting; S4. Dataset construction: construct a dataset for detecting typical dangerous conditions in the lower reaches of the Yellow River, covering samples with diverse characteristics. Through data enhancement technology, the diversity of the dataset is increased and the robustness of the model is improved. S5. Deep neural network automatic allocation model. According to the different single and multi-targets captured, a deep neural network automatic allocation model is proposed and trained. The model has adaptive adjustment capabilities and automatically adjusts parameters according to the characteristics of the input data. It uses optimization algorithms, including Adam and RMSprop, to iteratively optimize network parameters, improve the detection accuracy of the model, and set an early stopping mechanism to prevent overfitting. S6. Data-driven analysis of engineering accident mechanism: Based on the collected measured data and engineering monitoring data, a data-driven calculation model for engineering accident in the lower reaches of the Yellow River is constructed using statistical methods. The process of engineering accident under different conditions is simulated, and the abnormal correlation coefficient is used as the evaluation index to analyze the correlation of abnormal data, identify potential dangers, and study the accident mechanism under different conditions through simulation experiments to reveal the internal factors of engineering accident. S7. Bayesian network model construction: Based on Bayesian theory, combined with expert knowledge and deterministic forecast results, a water project safety early warning model with a polymorphic Bayesian network method is constructed to handle a variety of uncertain factors, analyze the probability density function of various water project safety influencing factors, and calculate the probability of project accidents by combining the correlation between factors. By continuously optimizing the Bayesian network structure and parameters, the early warning accuracy of the model is improved; S8. Model optimization and integration: by continuously optimizing convolutional neural networks, data-driven models, and Bayesian network models, the accuracy and real-time performance of the overall early warning system are improved. Hyperparameter optimization techniques, including grid search and random search, are used to adjust model parameters. The research results of the three methods are integrated to form a comprehensive early warning model. Model fusion techniques, including weighted average and voting methods, are used to improve the performance and stability of the comprehensive model. S9. Assessment and early warning: Apply comprehensive early warning models to conduct real-time monitoring and assessment of typical dangerous projects in the lower reaches of the Yellow River, identify potential dangerous situations, use streaming data processing technology to achieve real-time data processing and analysis, issue early warning signals in a timely manner based on the early warning results output by the model, guide project protection measures, set up a multi-level early warning mechanism, and distinguish early warning responses at different risk levels; Data-driven engineering accident mechanism analysis, abnormal correlation coefficient calculation: calculate the correlation coefficient between each variable and label, normal / abnormal; Bayesian network model construction, using measured data and expert knowledge to build a Bayesian network model; Nodes: flow, water level, flow velocity, soil moisture, osmotic pressure, probability of disaster; Edge: dependency relationship between variables; Model optimization and integration: weighted average of the outputs of the CNN model and the Bayesian network model to obtain the final prediction of the probability of accident.

2. The method for early warning of dangerous conditions in the lower reaches of the Yellow River based on machine learning according to claim 1 is characterized by: The inter-frame difference image processing in step S2 adopts the inter-frame difference method, which is expressed as: D i (x,y)=|I t (x,y)-I t-1 (x,y)| Among them, Dt(x,y) represents the difference between the t-th frame and the t-1-th frame at the coordinate (x,y), It(x,y) and I t-1 (x, y) represents the pixel value at (x, y) in the t-th frame and the t-1-th frame respectively. By calculating the pixel difference between adjacent frames, the motion information of the target object in the video is extracted to provide basic data for subsequent feature extraction.

3. The method for early warning of dangerous conditions in the lower reaches of the Yellow River based on machine learning according to claim 1 is characterized by: The convolutional neural network model in S3 contains multiple convolutional layers, pooling layers, local response normalization layers and fully connected layers. The convolutional layer is used to extract image features, the pooling layer is used to reduce feature dimensionality and computational complexity, the local response normalization layer is used to improve the generalization ability of the model, and the fully connected layer maps the extracted features to the output label space.

4. The method for early warning of dangerous conditions in the lower reaches of the Yellow River based on machine learning according to claim 3 is characterized by: The convolution operation in the convolutional neural network model is: Among them, f is the input image, g is the convolution kernel, and (i, j) is the coordinate of the output image; The pooling operation is expressed as: Among them, P(i,j) is the value after pooling, and window is the pooling window; The output of the fully connected layer is expressed as: y=σ(Wx+b) Where W is the weight matrix, x is the input feature vector, b is the bias vector, and σ is the activation function ReLU or Sigmoid; Image features are extracted through convolution operations, dimensionality reduction and feature aggregation are performed through pooling, and finally the classification results are output through the fully connected layer.

5. The method for early warning of dangerous conditions in the lower reaches of the Yellow River based on machine learning according to claim 1 is characterized by: In step S5, the deep neural network automatic deployment model optimizes the loss function and adjusts the model parameters to improve the classification accuracy, which is expressed as: L is the loss function, N is the number of samples, yi is the true label, is the predicted probability.

6. The method for early warning of dangerous conditions in the lower reaches of the Yellow River based on machine learning according to claim 1 is characterized by: In step S6, the potential danger is identified by calculating the abnormal correlation coefficient, which is expressed as: Among them, r is the correlation coefficient, Xi and Yi are the observed values ​​of the two variables, and is the mean of the variable.

7. The method for early warning of dangerous conditions in the lower reaches of the Yellow River based on machine learning according to claim 1 is characterized by: The Bayesian network in step S7 is: Among them, P(A|B) is the probability of event A occurring when event B occurs, P(B|A) is the probability of event B occurring when event A occurs, P(A) and P(B) are the prior probabilities of events A and B occurring, respectively. The probability of danger is calculated by Bayesian theorem, and a water project safety early warning model based on a polymorphic Bayesian network method is established.

8. The method for early warning of dangerous conditions in the lower reaches of the Yellow River based on machine learning according to claim 1 is characterized by: The model fusion technology in step S8 is expressed as: is the prediction result after fusion, is the prediction result of the i-th model, α is the corresponding weight, and the performance and stability of the comprehensive model are improved through hyperparameter optimization and model fusion.

Images