A safety detection method for cast-in-place high piers in mountainous areas based on multi-source data fusion

Abstract

The present invention discloses a method for safety detection of cast-in-place high piers in mountainous areas based on multi-source data fusion, which relates to the field of intelligent building safety monitoring technology, including deploying sensor networks, multi-angle pan-tilt video monitoring and drones, covering high piers in mountainous areas for comprehensive monitoring, and collecting multi-source data; transmitting multi-source data to edge computing nodes in real time via wireless networks, and preprocessing to form a standardized multi-source data set; applying spatiotemporal data fusion methods, combining historical data with comprehensive data models to predict safety risks and generate risk factors; performing threshold and rule anomaly identification on risk factors, and evaluating safety conditions in real time to immediately trigger warnings and provide disposal suggestions. The present invention significantly improves the safety detection efficiency and accuracy in the process of cast-in-place high piers in mountainous areas through multi-source data fusion and quantum enhanced multimodal algorithms, effectively speeds up the triggering speed of safety warnings, and significantly enhances the safety of construction sites.

Description

Technical Field

[0001] The invention relates to the technical field of intelligent building safety monitoring, in particular to a cast-in-place safety detection method for high piers in mountainous areas based on multi-source data fusion. Background Art

[0002] With the continuous advancement of infrastructure construction, the construction of high-pier bridges in mountainous areas has become an important part of transportation engineering. However, due to the complex terrain, changeable climatic conditions and the particularity of the construction environment in mountainous areas, the cast-in-place construction process of high-pier bridges in mountainous areas has high technical difficulty and risk. Therefore, how to effectively monitor and ensure the safety of cast-in-place construction of high-piers in mountainous areas has become the focus of attention of engineering and technical personnel. Traditional safety detection methods mainly rely on manual inspection and simple sensor monitoring. Manual inspection not only consumes a lot of manpower and time, but also is prone to omissions or misjudgments due to the influence of human factors; and the traditional sensor monitoring method, due to the limitations of a single data source, cannot fully and real-time reflect the complex state of the construction site, resulting in safety hazards that are difficult to detect and deal with early.

[0003] Existing safety detection technologies for cast-in-place high piers in mountainous areas usually rely on simple and single-type sensors, such as anemometers or equipment-provided monitoring videos. Although these technical means can monitor the status of the local environment to a certain extent, they often seem to be powerless when dealing with complex construction environments. First, due to the structural complexity and environmental diversity of high piers in mountainous areas, the data from existing single sensors is difficult to fully reflect the actual situation of construction, especially in a changing environment, where a single data source cannot effectively reflect the degree of danger and urgency of the site. Secondly, traditional monitoring methods lack effective fusion and analysis of multi-source data, and cannot form a comprehensive understanding of the overall status of the construction site, resulting in difficulty in early warning of safety hazards. In addition, most of the current safety detection methods rely on fixed threshold judgments, lack flexible dynamic adjustment mechanisms, and cannot adapt to the changing environment and structural status during the construction process. Therefore, how to combine multi-source data and use advanced algorithms and technical means to evaluate construction safety in real time and dynamically has become an urgent problem to be solved. Summary of the invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a method for cast-in-place safety detection of high piers in mountainous areas based on multi-source data fusion to solve the problem of incomplete monitoring caused by the limitation of a single data source in the prior art.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In the first aspect, an embodiment of the present invention provides a method for safety detection of cast-in-place high piers in mountainous areas based on multi-source data fusion, which includes deploying a sensor network, multi-angle pan-tilt video surveillance and drones to cover high piers in mountainous areas for comprehensive monitoring and collect multi-source data; transmitting the multi-source data to edge computing nodes in real time via a wireless network, and preprocessing it to form a standardized multi-source data set; inputting the standardized multi-source data set into a quantum enhanced multimodal fusion algorithm, using a deep learning method to extract features and fuse them to generate a comprehensive data model; based on a dynamic weight adjustment mechanism of autonomous learning, dynamically adjusting the weight coefficients of different modes in the comprehensive data model through real-time construction environment and structural status; applying a spatiotemporal data fusion method, combining historical data with the comprehensive data model to predict safety risks and generate risk factors; performing threshold and rule anomaly identification on risk factors, real-time assessment of safety conditions, immediate triggering of warnings and providing disposal suggestions.

[0008] As a preferred solution of the method for safety detection of cast-in-place high piers in mountainous areas based on multi-source data fusion described in the present invention, the sensor network, multi-angle pan-tilt video monitoring and drones are deployed to cover the high piers in mountainous areas for comprehensive monitoring and collect multi-source data. The specific steps are as follows:

[0009] Select vibration sensor, temperature and humidity sensor, displacement sensor, and ambient light sensor;

[0010] A grid layout is adopted, with at least one sensor node deployed near each pier to cover the area around the high pier;

[0011] Use the LoRaWAN standard to establish a wireless data collection framework and perform wireless transmission for each sensor node;

[0012] The transmitted data is encrypted with AES-256 before being sent to the edge computing node;

[0013] Choose a PTZ camera with 360° horizontal rotation and 180° vertical rotation capabilities;

[0014] The H.265 video compression standard is used for real-time transmission over a dedicated 5G network, and video data is sent to edge computing nodes using end-to-end encryption during transmission;

[0015] Choose a professional-grade quadcopter drone with autonomous flight capabilities and high-precision GPS positioning;

[0016] Use the A* search algorithm to plan the flight path and transmit the acquired video stream and image data back to the edge computing node in real time;

[0017] The edge computing nodes receive sensor, video surveillance and autopilot data and store them using RAID 5 arrays.

[0018] As a preferred solution of the method for detecting safety of cast-in-place high piers in mountainous areas based on multi-source data fusion described in the present invention, the preprocessing is performed to form a standardized multi-source data set, and the specific steps are as follows:

[0019] Use pre-trained AI image recognition models to perform image recognition on image data received in real time by edge computing nodes;

[0020] Categorize the identified unsafe behaviors and assign corresponding labels to each behavior;

[0021] Archive the annotated image data by category, and gradually build and enrich the standardized data set;

[0022] Continuously train AI image recognition models using standardized datasets;

[0023] The standardized multi-source data sets will be stored in the distributed database and transmitted to the cloud in real time through the API interface.

[0024] As a preferred solution of the cast-in-place safety detection method for high piers in mountainous areas based on multi-source data fusion described in the present invention, wherein: the standardized multi-source data set is input into the quantum enhanced multimodal fusion algorithm, and the deep learning method is used to extract features and fuse them to generate a comprehensive data model. The specific steps are as follows:

[0025] For each data in the standardized multi-source data set, the feature extraction algorithm is applied to extract feature information to form a preliminary feature vector expression: X = [x1, x2, ..., x n ];

[0026] Introducing the concept of quantum computing, using the characteristics of quantum bits to enhance features and form quantum enhancement factors

[0027] The eigenvector X and the quantum enhancement factor Combined, the real-time comprehensive feature vector Z expression is generated as:

[0028]

[0029] Among them, x i is the time series eigenvalue of the i-th mode in the eigenvector set, t is the time variable, x i (t) represents the eigenvalue of the i-th mode at time t, n is the number of modes, e -αt represents the exponential decay function, e is the base of the natural logarithm, α is the decay coefficient, is the information filtering function, dt is the time increment;

[0030] when When it shows a gradually increasing trend, it means that the characteristics of mode i are more significant at a given time t;

[0031] when When it shows a decreasing trend, it means that the feature importance of mode i at a given time t decreases;

[0032] When time t and e -αt When an exponential decay relationship is presented, it means that the feature has a time decay relationship with the final result;

[0033] The information filtering function expression is:

[0034]

[0035] Among them, x j is the eigenvalue of the jth mode in the eigenvector set, μ is the eigenmean, σ 2 is the variance of the feature;

[0036] Perform data cleaning, feature engineering and feature normalization on the comprehensive feature vector;

[0037] Based on the processed comprehensive feature vector, the model is trained through the convolutional neural network deep learning model;

[0038] After training the model, the cross-validation method is used to optimize the model parameters and generate the final comprehensive data model.

[0039] As a preferred solution of the cast-in-place safety detection method for high piers in mountainous areas based on multi-source data fusion described in the present invention, the specific steps of dynamically adjusting the weight coefficients of different modes in the comprehensive data model are as follows:

[0040] Based on the comprehensive feature vector, an initial weighted comprehensive feature vector is constructed using the initial weights;

[0041] For the weighted comprehensive feature vector, the mean square error is used as the loss function to measure the deviation between the model prediction and the actual value;

[0042] Use the gradient descent method to gradually adjust the weights of each mode by minimizing the loss function;

[0043] Introducing dynamically adjusted learning rate and time decay factor to adjust the convergence speed and adapt to changes in the construction environment in real time;

[0044] After each round of weight adjustment, the accuracy of the comprehensive data model prediction is evaluated based on the current construction environment and structural status, and the frequency of weight adjustment and learning rate are dynamically adjusted based on the evaluation results;

[0045] If the prediction deviation of the comprehensive data model is large, the pace of weight adjustment will be accelerated. If the prediction result is accurate, the frequency will be reduced.

[0046] As a preferred solution of the cast-in-place safety detection method for high piers in mountainous areas based on multi-source data fusion described in the present invention, the spatiotemporal data fusion method is applied to combine historical data with a comprehensive data model to predict safety risks and generate risk factors. The specific steps are as follows:

[0047] Extract sensor data, video surveillance data, and drone inspection data from historical construction records as historical data;

[0048] Use wavelet transform to denoise historical sensor data, Gaussian filtering to clean video data, and interpolation to fill in missing data;

[0049] Align the timestamps of historical multi-source datasets and standardize the data using the Min-Max normalization method;

[0050] Apply feature extraction algorithms to the standardized historical data to extract key features;

[0051] The extracted features are represented as the historical feature vector L ht The expression is:

[0052]

[0053] i=1,2...n;

[0054] in, represents the eigenvalue of the i-th mode in the historical data, n is the number of modes, and i is the mode index variable;

[0055] The real-time comprehensive feature vector Z and the historical feature vector L ht Perform weighted fusion to form weighted comprehensive features The vector expression is:

[0056]

[0057] Among them, λ is the fusion coefficient;

[0058] For the comprehensive feature vector For each feature, use the dynamic weighting method to adjust the weight coefficient of each feature;

[0059] The weighted comprehensive feature vector Input deep learning model for training;

[0060] The mean square error is used as the loss function of the deep learning model to measure the deviation between the output prediction value and the true value;

[0061] Based on the prediction results of the deep learning model after training, the risk factor of each potential security risk is calculated.

[0062] As a preferred solution of the cast-in-place safety detection method for high piers in mountainous areas based on multi-source data fusion described in the present invention, the specific steps of calculating the risk factor of each potential safety risk are as follows:

[0063] For each prediction result P at time point t t and the true value R t , calculate the prediction error E t , the expression is: E t =|P t -R t |;

[0064] Based on the prediction error E t , define the risk factor F for each potential security risk t The expression is:

[0065] F t =θ·E t +β·P t +γ·R t ;

[0066] Among them, θ, β and γ are weight coefficients, which are used to balance the impact of prediction error, predicted value and true value on risk factors respectively;

[0067] For the obtained risk factor F t Perform normalization.

[0068] As a preferred solution of the safety detection method for cast-in-place high piers in mountainous areas based on multi-source data fusion described in the present invention, the specific steps are as follows: the threshold and rule anomaly identification of risk factors, real-time assessment of safety conditions, immediate triggering of warnings and provision of disposal suggestions are performed,

[0069] Use statistical methods to set the initial threshold T t The expression is:

[0070] T t =μ F +k·σ F

[0071] Among them, μ F The risk factor F t The mean value, σ F is the standard deviation of the risk factor, k is the adjustment coefficient;

[0072] After setting the initial threshold T tAfter that, the risk factor is monitored in real time using the rule anomaly recognition method. For each time t, the risk factor F is calculated. t With its threshold T t The deviation between t The expression is: ΔF t =F t -T t ;

[0073] When ΔF t >0, and the risk factor F t At a certain time t, the threshold T is exceeded t , marked as a first-level abnormality, identified as a minor abnormality, pay attention to observation;

[0074] When ΔF t >0, and the risk factor F t Exceeding the threshold T at multiple consecutive times t t , marked as a secondary abnormality, identified as a moderate abnormality, and preventive measures were taken;

[0075] When ΔF t >0, and the risk factor F t If it rises sharply in a short period of time and exceeds the threshold, it will be marked as a Level 3 abnormality, considered a serious abnormality, and immediate action will be taken.

[0076] In a second aspect, an embodiment of the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the method for safety detection of cast-in-place high piers in mountainous areas based on multi-source data fusion as described in the first aspect of the present invention is implemented.

[0077] In a third aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, any step of the method for safety detection of cast-in-place high piers in mountainous areas based on multi-source data fusion as described in the first aspect of the present invention is implemented.

[0078] The beneficial effects of the present invention are as follows: the present invention achieves comprehensive coverage of the construction site by deploying sensor networks, multi-angle PTZ video monitoring and drone collaborative monitoring, overcomes the shortcomings of traditional single sensor monitoring of local status, and thus enhances the comprehensiveness and accuracy of data; adopts quantum enhanced multimodal fusion algorithm, combined with deep learning technology, to fuse and extract features of multi-source data, generate a comprehensive data model, which can more accurately reflect the overall status of construction and significantly improve the accuracy of safety risk prediction; through the dynamic weight adjustment mechanism of autonomous learning, it can adapt to changes in construction environment and structural status in real time, provide flexible safety assessment and early warning, and ensure the timeliness and accuracy of early warning. The present invention not only solves the problem of incomplete monitoring in the prior art, but also greatly improves the intelligent level of safety detection, and is suitable for complex mountain construction environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0080] Figure 1 This is a flow chart of the cast-in-place safety detection method for high piers in mountainous areas based on multi-source data fusion in Example 1.

[0081] Figure 2 This is a flow chart for forming a standardized multi-source data set in Example 1. DETAILED DESCRIPTION

[0082] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0083] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0084] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0085] Example 1, reference Figure 1 and Figure 2, which is the first embodiment of the present invention, and provides a method for safety detection of high pier cast-in-place in mountainous areas based on multi-source data fusion, comprising the following steps:

[0086] S1. Deploy sensor networks, multi-angle PTZ video surveillance and drones to cover high piers in mountainous areas for comprehensive monitoring and collect multi-source data.

[0087] Furthermore, vibration sensors, temperature and humidity sensors, displacement sensors, and ambient light sensors are selected to ensure that the environment and structural status of the high pier and its surroundings can be fully monitored;

[0088] A grid layout is adopted, with at least one sensor node deployed near each pier to cover the area around the high pier;

[0089] Use the LoRaWAN standard to establish a wireless data collection framework and perform wireless transmission for each sensor node;

[0090] Preferably, the wireless data acquisition framework is established using the LoRaWAN standard, which can ensure that the data of each sensor node can be transmitted in a low-power and long-distance manner.

[0091] The transmitted data is encrypted with AES-256 before being sent to the edge computing node;

[0092] Choose a PTZ camera with 360° horizontal rotation and 180° vertical rotation capabilities;

[0093] Adopting the H.265 video compression standard, real-time transmission is carried out through a dedicated 5G network to ensure the transmission efficiency of high-quality video streams. At the same time, video data uses end-to-end encryption during transmission to protect the integrity and security of the data and send it to the edge computing node;

[0094] Choose a professional-grade quadcopter drone with autonomous flight capabilities and high-precision GPS positioning;

[0095] Use the A* search algorithm to plan the flight path and transmit the acquired video stream and image data back to the edge computing node in real time;

[0096] Preferably, the A search algorithm is an existing and widely used algorithm. It is a heuristic search algorithm used to find the shortest path from a starting point to an end point or a solution to a multi-objective optimization problem. The A search algorithm combines the advantages of the Dijkstra algorithm and the heuristic best-first search, which not only ensures that the optimal solution is found, but also speeds up the search speed through the heuristic function.

[0097] It should be noted that the A search algorithm can effectively reduce the search space through heuristic functions while ensuring that the optimal solution is found, thereby improving the search efficiency. The A search algorithm guides the search process by comprehensively considering the actual path cost and the heuristic estimated cost to ensure that the shortest path from the starting point to the end point is found within a limited time. This algorithm is particularly suitable for application scenarios that require finding the optimal path in a complex environment, such as automatic cruise mission planning for drones. In this solution, the A search algorithm is used to plan the flight path of the drone because compared with other algorithms, the A search algorithm can generate the optimal path more efficiently while ensuring the feasibility of the path, which is crucial for drones that need to perform automatic cruise missions in complex terrains such as mountainous high piers.

[0098] The A* search algorithm expression is:

[0099] f(n)=g(n)+h(n);

[0100] Where f(n) represents the total cost estimate of node n, g(n) represents the actual cost from the starting point to node n, and h(n) represents the heuristic estimate of the best path from node n to the end point;

[0101] The edge computing node receives sensor, video surveillance and autopilot data and stores it using a RAID 5 array;

[0102] Preferably, a "RAID 5 array" is a commonly used storage configuration that combines the advantages of data distribution and redundancy. RAID 5 achieves efficient storage utilization and data protection by striping data across multiple hard disks and storing a parity bit on each hard disk.

[0103] At the same time, "RAID 5 array" also has the following effects: RAID 5 can rely on the parity bits on other hard disks to rebuild the lost data when a hard disk fails, thereby providing fault tolerance, improving system reliability and data security, and ensuring that data will not be lost in the event of a single hard disk failure; compared to RAID 1 (mirror array), RAID 5 has higher storage efficiency because it only requires the capacity of one hard disk to store parity information instead of simply copying data. This allows RAID 5 to use storage space more efficiently while providing redundant protection. RAID 5 can improve read performance by distributing data across multiple hard disks in a striped manner. Although write operations will be slightly slowed down due to the need to calculate and write parity information, the overall performance can still meet the needs of most application scenarios.

[0104] The above advantages show that when RAID 5 arrays are used to store sensor, video surveillance and drone cruise data, they can not only ensure data reliability and security, but also maintain data integrity when hard disk fails, thus providing a reliable storage solution.

[0105] S2: Transmit multi-source data to edge computing nodes in real time via wireless networks and perform preprocessing to form a standardized multi-source data set.

[0106] Furthermore, a pre-trained AI image recognition model is used to perform image recognition on the image data received in real time by the edge computing node to identify unsafe behaviors;

[0107] It should be noted that the pre-trained AI image recognition model is based on the network structure of the Convolutional Neural Network (CNN) and is designed specifically for image processing. The model was initially trained on a large-scale general image dataset (such as ImageNet) and learned many common image features, such as edges, textures, and shapes.

[0108] It should also be noted that through pre-training on large-scale data sets, AI image recognition models have basic image recognition capabilities. These common features have good transferability for different tasks (such as detecting unsafe behaviors). Using these pre-trained AI image recognition models and fine-tuning them can better adapt to specific construction safety detection tasks.

[0109] Classify the identified unsafe behaviors and assign corresponding labels to each behavior, such as "not wearing a safety helmet", "working without a safety rope", "standing in the lifting area", etc.

[0110] Archive the annotated image data by category, and gradually build and enrich the standardized data set;

[0111] Use standardized data sets to continuously train AI image recognition models to improve the model's ability and accuracy in identifying unsafe behaviors;

[0112] The standardized multi-source data sets will be stored in the distributed database and transmitted to the cloud in real time through the API interface.

[0113] S3. Input the standardized multi-source data set into the quantum enhanced multimodal fusion algorithm, use deep learning methods to extract features and fuse them to generate a comprehensive data model.

[0114] Furthermore, for each data item in the standardized multi-source data set, the feature extraction algorithm is applied to extract feature information to form a preliminary feature vector expression: X = [x1, x2, ..., x n ];

[0115] Introducing the concept of quantum computing, using the characteristics of quantum bits to enhance features and form quantum enhancement factors Used to adjust the importance or weight of features;

[0116] The eigenvector X and the quantum enhancement factor Combined, a comprehensive feature vector Z is generated as follows:

[0117]

[0118] Among them, x i is the time series eigenvalue of the i-th mode in the eigenvector set, t is the time variable, x i (t) represents the eigenvalue of the i-th mode at time t, representing the time point of the feature, n is the number of modes, that is, the total number of features in the feature vector X, e -αt represents an exponential decay function, e is the base of the natural logarithm, α is the decay coefficient, and is used to control the importance of features over time. is the information filtering function, dt is the time increment, which indicates the time step of integration;

[0119] It should be noted that the comprehensive eigenvector X is the result of combining different modal eigenvectors and quantum enhancement factors.

[0120] It should be noted that the range of the above formula depends on the eigenvalues ​​in X and the quantum enhancement factor The value range of is a function based on Gaussian distribution, and its range is (0,1]. Therefore, the range of the entire formula mainly depends on range.

[0121] when When it shows a gradually increasing trend, it means that the characteristics of mode i are more significant at a given time t.

[0122] when When it shows a decreasing trend, it means that the feature importance of mode i at a given time t decreases;

[0123] When time t and e -αt When an exponential decay relationship is presented, it means that the feature has a time decay relationship with the final result, that is, the earlier the feature occurs, the smaller the impact on the final result.

[0124] The information filtering function expression is:

[0125]

[0126] Among them, x j is the eigenvalue of the jth mode in the eigenvector set, μ is the eigenmean, which is used to calculate the deviation between each feature and the mean, σ 2 is the variance of the feature, which is the square of the standard deviation and is used to calculate the variance part of the Gaussian distribution;

[0127] Perform data cleaning, feature engineering and feature normalization on the comprehensive feature vector;

[0128] Based on the processed comprehensive feature vector, the convolutional neural network deep learning model is used to train the model and predict the safety status of high piers in mountainous areas.

[0129] After training the model, a cross-validation method (such as K-fold cross-validation) is used to optimize the model parameters to ensure the generalization ability and accuracy of the model and generate the final comprehensive data model.

[0130] It should be noted that during “model training and optimization of model parameters”, the processed comprehensive feature vector should be divided into training set, validation set and test set, with the common division ratio being 70% training set, 15% validation set and 15% test set (adjusted according to specific circumstances); the training set is used to train the model, the validation set is used to tune the model parameters and select the model, and the test set is used to evaluate the final performance of the model.

[0131] S4. A dynamic weight adjustment mechanism based on autonomous learning dynamically adjusts the weight coefficients of different modes in the comprehensive data model through real-time construction environment and structural status.

[0132] Furthermore, based on the comprehensive feature vector, an initial weighted comprehensive feature vector is constructed using the initial weights;

[0133] For the weighted comprehensive feature vector, the mean square error is used as the loss function to measure the deviation between the model prediction and the actual value;

[0134] Use the gradient descent method to gradually adjust the weights of each mode by minimizing the loss function;

[0135] Each iteration fine-tunes the weights based on the current loss.

[0136] Introducing dynamically adjusted learning rate and time decay factor to adjust the convergence speed and adapt to changes in the construction environment in real time;

[0137] Preferably, a dynamic learning rate enables the model to quickly adjust weights in the early stages of training. Especially in large-scale data or complex models, rapid adjustments in the initial stage can significantly shorten the training time. This is because a large learning rate can enable the model to quickly skip some local minima and move closer to the global optimal solution.

[0138] As the training progresses, the dynamic learning rate will gradually decrease, thereby preventing the weights from fluctuating greatly when approaching the optimal solution. A smaller learning rate allows the model to make more detailed adjustments when approaching the optimal solution, preventing the model from oscillating back and forth near the global optimum. This mechanism also helps prevent overfitting, because in the later stages of training, a smaller learning rate prevents the model from excessively following the noise in the training data.

[0139] It should also be noted that changes in the construction process can be short-term or long-term. The time decay factor allows the model to be more flexible in dealing with changes on these different time scales. Drastic changes in the short term may require a larger time decay, while longer-term trends may require a smaller decay.

[0140] Preferably, the construction environment and structural status are dynamically changing. The introduction of the time decay factor can make the model gradually reduce its dependence on earlier data and pay more attention to the most recent real-time data. This is especially important in a dynamic environment because the latest data usually better reflects the current real situation.

[0141] The time decay factor helps the model find a balance between utilizing historical data and not ignoring changes in the current environment. For example, when monitoring a high pier under construction, historical data may show a certain trend, but current real-time data may show new and important changes. The time decay factor enables the model to adapt to this rapidly changing environment.

[0142] It should also be noted that dynamic learning rates and time decay factors are highly adaptive and flexible, and can be automatically adjusted according to the data and different stages of training. In contrast, fixed learning rates or weight adjustment methods that do not consider time changes may lack this flexibility and easily lead to unstable performance of the model in dynamic environments.

[0143] Compared with other methods (such as fixed learning rate, weight adjustment without time decay, and Momentum and Adam optimization methods), the differences are as follows:

[0144] A fixed learning rate remains unchanged throughout the training process, which may lead to slow convergence in the early stage, or oscillation or skipping of the optimal solution when approaching the optimal solution. In contrast, a dynamic learning rate can be adjusted according to the training status, taking into account both fast convergence in the early stage and stable fine-tuning in the later stage.

[0145] Weight adjustment methods that do not consider time factors may rely too much on historical data and ignore changes in the real-time environment. The time decay factor enables the model to better adapt to the current dynamic environment by gradually reducing the importance of historical data.

[0146] Optimization methods like Momentum or Adam are also adaptive, but they mainly target the way gradients are calculated and parameters are updated. In contrast, dynamic learning rates and time decay factors more directly control the adjustment of learning rates and time weights, and can be optimized more precisely for specific problems.

[0147] After each round of weight adjustment, the accuracy of the comprehensive data model prediction is evaluated based on the current construction environment and structural status, and the frequency of weight adjustment and learning rate are dynamically adjusted based on the evaluation results;

[0148] If the prediction deviation of the comprehensive data model is large, the pace of weight adjustment will be accelerated. If the prediction result is accurate, the frequency will be reduced to prevent the model from overfitting or underfitting.

[0149] S5. Apply spatiotemporal data fusion methods, combine historical data with comprehensive data models to predict safety risks and generate risk factors.

[0150] Furthermore, historical data can be extracted from historical construction records, sensor data, video surveillance data, and drone inspection data as historical data;

[0151] Use wavelet transform to denoise historical sensor data, Gaussian filtering to clean video data, and interpolation to fill in missing data;

[0152] Align the timestamps of historical multi-source datasets and standardize the data using the Min-Max normalization method to facilitate fusion with real-time data;

[0153] Apply feature extraction algorithms to historical data to extract key features, such as vibration frequency, temperature change rate, displacement deformation, structural changes in the video, etc.

[0154] Apply feature extraction algorithms to the standardized historical data to extract key features;

[0155] The extracted features are represented as the historical feature vector L ht The expression is:

[0156]

[0157] i=1,2...n;

[0158] in, represents the eigenvalue of the i-th mode in the historical data, n is the number of modes, and i is the mode index variable;

[0159] The real-time comprehensive feature vector Z and the historical feature vector L ht Perform weighted fusion to form weighted comprehensive features The vector expression is:

[0160]

[0161] Among them, λ is the fusion coefficient, which is usually between 0 and 1 and is used to adjust the relative weight of real-time data and historical data;

[0162] For the comprehensive feature vector For each feature, a dynamic weighting method is used to adjust the weight coefficient of each feature to reflect the importance of each feature under different time and space conditions;

[0163] It should be noted that the "dynamic weighting method" refers to the real-time adjustment of the weights of each feature in the comprehensive feature vector according to changes in the current environment or conditions. This is different from static weighting (fixed weights). The dynamic weighting method can flexibly adjust the importance of features according to the actual situation of the input data so as to more accurately reflect the current risk situation.

[0164] It should also be noted that in the context of spatiotemporal data fusion, data at different time points and spatial locations have different influences on current decisions. For example, data closer to the current time and location are usually more important, so the weights of these features should be higher; at the same time, some features may be more critical to risk prediction under certain conditions (such as specific weather and construction environment), so the weights of these features need to be dynamically increased. Under other conditions, the impact of these features may be smaller, and the weights should be reduced accordingly.

[0165] The implementation of the dynamic weighting method is as follows:

[0166] First, you can use trained models (such as attention mechanism, weighted regression, etc.) to automatically adjust the weights of features. The model determines which features are more important in the current environment by learning the relationship between features and risks in historical data.

[0167] Secondly, the weights of various features can be dynamically calculated based on changes in real-time data according to preset rules or algorithms (such as Bayesian update, entropy weight method, etc.). These rules may be automatically determined based on experience or in a data-driven manner.

[0168] An example of the implementation of the "dynamic weighting method" is as follows: suppose a construction site experiences strong seismic activity during a period of time, and the vibration data captured by the sensors is unusually active. In this case, the system will dynamically increase the weight of vibration-related features because these features are critical to predicting geological risks in the current situation. Under normal conditions, the weight of these features may be lower.

[0169] The weighted comprehensive feature vector Input deep learning models (such as LSTM or GRU) for training to predict future security risks;

[0170] The mean square error (MSE) is used as the loss function of the deep learning model to measure the deviation between the output prediction value and the true value;

[0171] Based on the prediction results of the deep learning model after training, the risk factor of each potential security risk is calculated.

[0172] Furthermore, the risk factor of each potential security risk is calculated. The specific steps are as follows:

[0173] For each prediction result P at time point t t and the true value R t , calculate the prediction error E t , the expression is: E t =|P t -R t |;

[0174] Based on the prediction error E t , define the risk factor F for each potential security risk t The expression is:

[0175] F t =θ·E t +β·P t +γ·R t ;

[0176] Among them, θ, β and γ are weight coefficients, which are used to balance the impact of prediction error, predicted value and true value on risk factors respectively;

[0177] For the obtained risk factor F t The normalized expression is:

[0178]

[0179] Among them, F min and F max Represents the minimum and maximum risk factor values ​​at all time points in the historical data, normalized The range is between [0, 1], which facilitates subsequent risk classification and assessment;

[0180] It should be noted that, through normalization, the risk factor F t Being standardized to a relative scale eliminates differences in the magnitude and range of the original data, improves the efficiency and accuracy of data processing, and makes subsequent risk assessment and decision-making more intuitive and reliable.

[0181] S6. Identify risk factors based on thresholds and rule anomalies, evaluate safety conditions in real time, trigger warnings immediately, and provide disposal suggestions.

[0182] Furthermore, a statistical method (such as a 95% confidence interval) is used to set the initial threshold T t The expression is:

[0183] T t =μF+k·σ F

[0184] Among them, μ F The risk factor F t The mean value, σ F is the standard deviation of the risk factor, k is the adjustment factor (usually 2 or 3, depending on the risk tolerance);

[0185] After setting the threshold, the risk factor is monitored in real time using the rule anomaly recognition method. For each time t, the risk factor F is calculated. t With its threshold T t The deviation between t The expression is: ΔF t =F t -T t ;

[0186] When ΔF t >0, and the risk factor F t At a certain time t, the threshold T is exceeded t , marked as a first-level abnormality, identified as a minor abnormality, pay attention to observation;

[0187] It should be noted that when a level 1 anomaly is identified, the following steps should be performed:

[0188] Notify relevant persons in charge: The system notifies the person in charge of safety at the construction site via SMS, email or APP, stating that there are minor abnormalities at present and recommends close monitoring of relevant indicators.

[0189] Increase monitoring frequency: Automatically increase the monitoring frequency of abnormal data sources, for example, doubling the sampling frequency of sensors or increasing the frequency of drone inspections.

[0190] Record abnormal events: Record the detailed information of the first-level abnormality (time, location, abnormality type, data, etc.) in the system log for subsequent analysis.

[0191] Continuous monitoring: In the next monitoring cycle, the system will focus on the abnormal point. If the abnormal situation persists or intensifies, the system will upgrade to a level 2 abnormality.

[0192] When ΔF t >0, and the risk factor F t Exceeding the threshold T at multiple consecutive times t t , marked as a secondary abnormality, identified as a moderate abnormality, and preventive measures were taken;

[0193] It should be noted that when a secondary anomaly is identified, the following operational measures should be taken:

[0194] Immediate notification of multiple levels of stakeholders: The system notifies not only the site safety manager, but also the project manager and technical support team, providing a detailed exception report.

[0195] Suspend high-risk construction activities: The system automatically triggers a suspension command to temporarily stop high-risk construction activities related to abnormal areas to prevent potential accidents. The command to suspend construction can be automatically issued through the on-site broadcast system or the construction management platform.

[0196] Manual on-site inspection: The system recommends and arranges technicians or engineers to go to the site for manual inspection, confirm abnormal conditions, and take necessary preventive measures.

[0197] Generate detailed exception analysis report: The system automatically generates a detailed exception analysis report, including an analysis of the cause of the exception, an assessment of the scope of impact, and recommended countermeasures. The report can be automatically sent to management and technical teams.

[0198] Prepare emergency plans: If on-site manual inspection confirms that the anomaly is serious, the system will recommend preparing an emergency plan to ensure a rapid response when the anomaly is upgraded to level three.

[0199] When ΔF t >0, and the risk factor F t If it rises sharply in a short period of time and exceeds the threshold, it will be marked as a Level 3 abnormality, considered a serious abnormality, and immediate action will be taken.

[0200] It should be noted that when a level 3 abnormality is identified, the following response measures should be taken promptly:

[0201] Immediately trigger an emergency warning: Immediately notify all relevant personnel through all available communication means (SMS, phone, broadcast, dedicated APP), including all on-site construction personnel, project management team, senior management, and local emergency management departments.

[0202] Emergency stop of all construction activities: Automatically issue emergency stop instructions, suspend all construction activities, and initiate emergency evacuation procedures. Through broadcasting, direct on-site personnel to quickly evacuate to a safe area.

[0203] Activate the emergency plan: According to the preset emergency plan, activate the on-site emergency equipment and measures, including but not limited to support reinforcement, on-site isolation, emergency rescue, etc.

[0204] Real-time monitoring and adjustment of response measures: Continuously monitor the situation on site and adjust response measures based on real-time data. If the situation worsens, it will be recommended to expand the scope of the emergency or upgrade the emergency response level.

[0205] Comprehensive Recording and Evaluation: The entire process of the three-level anomaly is recorded in detail, including all data and operation steps from triggering the anomaly to emergency response. A complete post-event evaluation report will be automatically generated to analyze the cause of the anomaly and optimize future emergency plans.

[0206] It should also be noted that after the implementation of the response measures, an effective feedback mechanism should be established to ensure the effectiveness of the response measures and the continuous improvement of the system. The specific steps are as follows: First, after the implementation of the response measures, feedback from on-site personnel will be collected, including the effectiveness of the response measures, changes in on-site conditions, etc. These feedbacks will be collected through a dedicated APP or a feedback terminal directly connected to the system; second, the data of various sensors on site will be continuously monitored to evaluate whether the response measures are effective. For example, if the abnormal data gradually returns to the normal range, it means that the measures are effective; if the data fluctuations intensify, further adjustments to the measures may be required; finally, the collected feedback data will be analyzed to determine the effectiveness of the response measures, and the threshold or response strategy will be adjusted based on the feedback results. For example, by comparing the data differences before and after the response, the system can automatically adjust the threshold of the risk factor to optimize the subsequent abnormal identification and response.

[0207] This embodiment also provides a computer device, which is suitable for the case of a method for detecting the safety of cast-in-place high piers in mountainous areas based on multi-source data fusion, and includes: a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement the method for detecting the safety of cast-in-place high piers in mountainous areas based on multi-source data fusion as proposed in the above embodiment.

[0208] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0209] The present embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the method for safety detection of cast-in-place high piers in mountainous areas based on multi-source data fusion proposed in the above embodiment is implemented; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, referred to as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, referred to as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, referred to as EPROM), programmable read-only memory (Programmable Red-Only Memory, referred to as PROM), read-only memory (Read-Only Memory, referred to as ROM), magnetic storage, flash memory, disk or optical disk.

[0210] In summary, the present invention achieves comprehensive coverage of the construction site by deploying sensor networks, multi-angle PTZ video monitoring and drone collaborative monitoring, overcomes the shortcomings of traditional single sensor monitoring of local status, and thus enhances the comprehensiveness and accuracy of data; adopts quantum enhanced multimodal fusion algorithm, combined with deep learning technology, to fuse and extract features of multi-source data, generate a comprehensive data model, which can more accurately reflect the overall status of construction and significantly improve the accuracy of safety risk prediction; through the dynamic weight adjustment mechanism of autonomous learning, it can adapt to changes in construction environment and structural status in real time, provide flexible safety assessment and early warning, and ensure the timeliness and accuracy of early warning. The present invention not only solves the problem of incomplete monitoring in the prior art, but also greatly improves the intelligent level of safety detection, and is suitable for complex mountain construction environments.

[0211] Example 2, referring to Table 1, is the second example of the present invention. To further verify the advancement of the present invention, experimental simulation data of a cast-in-place safety detection method for high piers in mountainous areas based on multi-source data fusion are provided.

[0212] A typical mountain high pier was selected as the test object for comprehensive safety monitoring. The experimental steps are as follows:

[0213] First, different types of sensors were deployed around the high pier, including vibration sensors, temperature and humidity sensors, displacement sensors, and ambient light sensors. The sensor nodes were deployed in a grid at intervals of 5 meters to ensure comprehensive coverage of the environment around the high pier. The sensors transmitted data through the LoRaWAN wireless communication standard, and the data was encrypted with AES-256 before transmission to ensure data security.

[0214] Secondly, a multi-angle pan-tilt video surveillance system was deployed, using a camera with 360° horizontal rotation and 180° vertical rotation capabilities. Video data with H.265 compression standard was transmitted to edge computing nodes in real time through a dedicated 5G network. In addition, a four-axis drone with autonomous flight function and high-precision GPS positioning was used to conduct regular inspections of high piers and transmit video streams and image data in real time.

[0215] Next, for the multi-source data obtained from sensors, video surveillance and drones, wavelet transform is used to denoise the sensor data, and Gaussian filtering is used to clean the video data. The cleaned data is timestamped and aligned using the NTP protocol, and Min-Max normalization is performed to form a standardized multi-source data set.

[0216] In terms of data fusion, the standardized multi-source data set is input into the quantum enhanced multimodal fusion algorithm to extract features and generate comprehensive feature vectors. The feature expression ability is improved through the quantum enhancement factor. The comprehensive data model is trained in combination with the deep learning model, and the cross-validation method is used to optimize the model parameters.

[0217] Finally, through the dynamic weight adjustment mechanism, the weight coefficients of different modes in the comprehensive data model are adjusted according to the real-time construction environment and structural status, the historical data and real-time data are fused in time and space to generate risk factors, and the safety status of high piers is evaluated in real time. When the risk factor exceeds the preset threshold, an early warning is triggered and corresponding disposal suggestions are provided.

[0218] The details are shown in Table 1 below:

[0219] Table 1 Performance comparison of cast-in-place safety detection methods for high piers in mountainous areas

[0220]

[0221] According to experimental data, the method of the present invention shows significant advantages in multiple key parameters.

[0222] First, in terms of the response time of the vibration sensor, the sensor data was denoised in real time through wavelet transform, which improved the signal quality and ultimately shortened the response time by 37.78%, from 0.45 seconds to 0.28 seconds. This improvement is crucial for real-time monitoring of high pier structures, because faster response time means that potential structural problems can be discovered earlier, so that timely measures can be taken to avoid accidents.

[0223] Secondly, in terms of the accuracy of temperature and humidity monitoring, through the multimodal data fusion algorithm combined with the feature extraction capability of the deep learning model, the monitoring accuracy has been improved from 0.85 to 0.93, an increase of 9.41%. This improvement in accuracy ensures the accurate monitoring of environmental parameters and helps to better evaluate the impact of the construction environment on high pier structures.

[0224] In terms of video image processing delay, Gaussian filtering and 5G network transmission technology are used to greatly reduce the delay, from 1.5 seconds of traditional methods to 0.75 seconds, an improvement of 50.00%. Low-latency video processing greatly improves the real-time performance of drone inspections, which is especially important for complex construction scenarios in mountainous environments.

[0225] In terms of risk prediction accuracy, the quantum-enhanced multimodal fusion algorithm, combined with the spatiotemporal fusion of historical data and real-time data, ultimately improved the prediction accuracy by 16.67%, from 0.78 to 0.91. This means that in actual construction environments, potential risks can be predicted more accurately, thereby improving the reliability of safety warnings.

[0226] Finally, the improvement in the safety warning trigger time is also very significant. Through the dynamic weight adjustment mechanism, combined with the fusion of real-time data and historical data, the warning trigger time has been reduced by 50.00%, from 2.2 seconds to 1.1 seconds. Faster warning trigger time means that when dangerous situations occur, response measures can be taken more promptly to avoid the expansion of accidents.

[0227] In summary, the method of the present invention has shown obvious advantages in multiple key technical indicators, especially in response time, monitoring accuracy, processing delay, prediction accuracy and warning trigger time. These improvements have significantly improved the safety and reliability of cast-in-place construction of high piers in mountainous areas.

[0228] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for safety detection of high cast-in-place piers in mountainous areas based on multi-source data fusion, characterized by: include, Comprehensively monitor high piers in mountainous areas through sensor networks, multi-angle PTZ video surveillance and drones to collect multi-source data; Transmit multi-source data to edge computing nodes in real time via wireless networks and perform preprocessing to form standardized multi-source data sets; The standardized multi-source data set is input into the quantum enhanced multimodal fusion algorithm, and the features are extracted and fused using deep learning methods to generate a comprehensive data model. A dynamic weight adjustment mechanism based on autonomous learning dynamically adjusts the weight coefficients of different modes in the comprehensive data model through real-time construction environment and structural status; Apply spatiotemporal data fusion methods, combine historical data with comprehensive data models to predict safety risks and generate risk factors; By identifying the thresholds and rule anomalies of risk factors, the safety status can be evaluated in real time, warnings can be triggered in time, and disposal suggestions can be provided; The standardized multi-source data set is input into the quantum enhanced multimodal fusion algorithm, and the features are extracted and fused using a deep learning method to generate a comprehensive data model. The specific steps are: For each data in the standardized multi-source data set, the feature extraction algorithm is applied to extract feature information to form a preliminary feature vector expression: X = [x1, x2, ..., x n ]; Using the properties of quantum bits to enhance features and form quantum enhancement factors The eigenvector X and the quantum enhancement factor Combined, a real-time comprehensive feature vector Z is generated. The expression of the comprehensive feature vector Z is: Among them, x i is the time series eigenvalue of the i-th mode in the eigenvector set, t is the time variable, x i (t) represents the eigenvalue of the i-th mode at time t, n is the number of modes, e -αt represents the exponential decay function, e is the base of the natural logarithm, α is the decay coefficient, is the information filtering function, dt is the time increment; Perform data cleaning, feature engineering and feature normalization on the comprehensive feature vector; Based on the processed comprehensive feature vector, the model is trained through the convolutional neural network deep learning model; After training the model, the cross-validation method is used to optimize the model parameters and generate the final comprehensive data model.

2. The method for detecting safety of cast-in-place high piers in mountainous areas based on multi-source data fusion according to claim 1, characterized in that: The method uses sensor networks, multi-angle pan-tilt video monitoring and drones to conduct comprehensive monitoring of high piers in mountainous areas and collect multi-source data. The specific steps are: Select vibration sensors, temperature and humidity sensors, displacement sensors, and ambient light sensors; A grid layout is adopted, with at least one sensor node deployed near each pier to cover the surrounding area of ​​the high pier; Use the LoRaWAN standard to establish a wireless data collection framework and perform wireless transmission for each sensor node; The transmitted data is encrypted with AES-256 before being sent to the edge computing node; Choose a PTZ camera with 360° horizontal rotation and 180° vertical rotation capabilities; The H.265 video compression standard is used for real-time transmission over a dedicated 5G network, and video data is sent to edge computing nodes using end-to-end encryption during transmission; Choose a professional-grade quadcopter drone with autonomous flight capabilities and high-precision GPS positioning; Use the A* search algorithm to plan the flight path and transmit the acquired video stream and image data back to the edge computing node in real time; Sensor, video surveillance, and autopilot data are received through edge computing nodes and stored using RAID 5 arrays.

3. The method for detecting safety of cast-in-place high piers in mountainous areas based on multi-source data fusion according to claim 2, characterized in that: The preprocessing is performed to form a standardized multi-source data set, and the specific steps are: Use pre-trained AI image recognition models to perform image recognition on image data received in real time by edge computing nodes; Categorize the identified unsafe behaviors and assign corresponding labels to each behavior; Archive the annotated image data by category, and gradually build and enrich the standardized data set; Continuously train AI image recognition models using standardized datasets; The standardized multi-source data sets will be stored in the distributed database and transmitted to the cloud in real time through the API interface.

4. The method for detecting safety of cast-in-place high piers in mountainous areas based on multi-source data fusion according to claim 3 is characterized in that: when When it shows a gradually increasing trend, it means that the characteristics of mode i are more significant at a given time t; when When it shows a decreasing trend, it means that the feature importance of mode i at a given time t decreases; When time t and e -αt When an exponential decay relationship is presented, it means that the feature has a time decay relationship with the final result; The information filtering function expression is: Among them, x j is the eigenvalue of the jth mode in the eigenvector set, μ is the eigenmean, σ 2 is the variance of the feature.

5. The method for detecting safety of cast-in-place high piers in mountainous areas based on multi-source data fusion according to claim 4, characterized in that: The specific steps of dynamically adjusting the weight coefficients of different modes in the comprehensive data model are as follows: Based on the comprehensive feature vector, an initial weighted comprehensive feature vector is constructed using the initial weights; For the weighted comprehensive feature vector, the mean square error is used as the loss function to measure the deviation between the model prediction and the actual value; Use the gradient descent method to gradually adjust the weights of each mode by minimizing the loss function; Introducing dynamically adjusted learning rate and time decay factor to adjust the convergence speed and adapt to changes in the construction environment in real time; After each round of weight adjustment, the accuracy of the comprehensive data model prediction is evaluated based on the current construction environment and structural status, and the frequency of weight adjustment and learning rate are dynamically adjusted based on the evaluation results; If the prediction deviation of the comprehensive data model is greater than the preset deviation threshold, the pace of weight adjustment will be accelerated.

6. The method for detecting safety of cast-in-place high piers in mountainous areas based on multi-source data fusion according to claim 5, characterized in that: The application of the spatiotemporal data fusion method combines historical data with a comprehensive data model to predict safety risks and generate risk factors. The specific steps are: Extract sensor data, video surveillance data, and drone inspection data from historical construction records as historical data; Use wavelet transform to denoise historical sensor data, Gaussian filtering to clean video data, and interpolation to fill in missing data; Align the timestamps of historical multi-source datasets and standardize the data using the Min-Max normalization method; Apply feature extraction algorithms to the standardized historical data to extract key features; The extracted features are represented as the historical feature vector L ht , historical feature vector L ht The expression is: in, represents the eigenvalue of the i-th mode in the historical data, n is the number of modes, and i is the mode index variable; The real-time comprehensive feature vector Z and the historical feature vector L ht Perform weighted fusion to form weighted comprehensive feature C fd Vector, weighted comprehensive feature C fd The vector expression is: C fd =λZ+(1-λ)L ht ; Among them, λ is the fusion coefficient; For the comprehensive feature vector C fd For each feature, use the dynamic weighting method to adjust the weight coefficient of each feature; The weighted comprehensive feature vector C fd Input deep learning model for training; The mean square error is used as the loss function of the deep learning model to measure the deviation between the output prediction value and the true value; Based on the prediction results of the deep learning model after training, the risk factor of each potential security risk is calculated.

7. The method for detecting safety of cast-in-place high piers in mountainous areas based on multi-source data fusion according to claim 6, characterized in that: The specific steps of calculating the risk factor of each potential safety risk are as follows: For each prediction result P at time point t t and the true value R t , calculate the prediction error E t , the expression is: E t =|P t -R t |; Based on the prediction error E t , define the risk factor F for each potential security risk t The expression is: F t =θ·E t +β·P t +γ·R t ; Among them, θ, β and γ are weight coefficients, which are used to balance the impact of prediction error, predicted value and true value on risk factors respectively; For the obtained risk factor F t Perform normalization.

8. The method for detecting safety of cast-in-place high piers in mountainous areas based on multi-source data fusion according to claim 7, characterized in that: By identifying the threshold and rule anomalies of risk factors, the safety status is evaluated in real time, early warnings are triggered in time, and disposal suggestions are provided. The specific steps are as follows: Use statistical methods to set the initial threshold T t , initial threshold T t The expression is: T t =μ F +k·s F Among them, μ F The risk factor F t The mean value, σ F is the standard deviation of the risk factor, k is the adjustment coefficient; After setting the initial threshold T t After that, the risk factor is monitored in real time using the rule anomaly recognition method. For each time t, the risk factor F is calculated. t With its threshold T t The deviation between t , deviation ΔF t The expression is: ΔF t =F t -T t ; When ΔF t >0, and the risk factor F t At a certain time t, the threshold T is exceeded t , marked as a first-level abnormality, identified as a minor abnormality, pay attention to observation; When ΔF t >0, and the risk factor F t Exceeding the threshold T at multiple consecutive times t t , marked as a secondary abnormality, identified as a moderate abnormality, and preventive measures were taken; When ΔF t >0, and the risk factor F t It rises sharply in a short time and exceeds the threshold T t , marked as a level 3 anomaly, identified as a serious anomaly, and took immediate action.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the method for safety detection of cast-in-place high piers in mountainous areas based on multi-source data fusion as described in any one of claims 1 to 8 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for safety detection of cast-in-place high piers in mountainous areas based on multi-source data fusion as described in any one of claims 1 to 8 is implemented.

Images