CN-120781194-B - Suspension type carrier state monitoring method and device based on dynamic environment sensing

CN120781194BCN 120781194 BCN120781194 BCN 120781194BCN-120781194-B

Abstract

The embodiment of the disclosure provides a suspension type carrier state monitoring method and device based on dynamic environment sensing, wherein the method comprises the steps of collecting sensor data related to a suspension type carrier, wherein the sensor data comprises acceleration, air pressure and gesture data; the method comprises the steps of respectively extracting the characteristics of acceleration, air pressure and gesture data, fusing the extracted characteristics through a linear fusion layer to obtain fusion characteristics, inputting the fusion characteristics into a strategy network, outputting probability distribution of three actions, training the strategy network through rewarding calculation and a double experience pool mechanism, updating parameters of the strategy network, and displaying dynamic curves of the acceleration, air pressure and gesture data and the action probability distribution output by the strategy network in a real-time visualization mode.

Inventors

HE YONGQI
YU XIAOYU
Diao Yanwei
LIU JINGJING
YAO XIAOQIANG

Assignees

凯通科技股份有限公司

Dates

Publication Date: 20260508
Application Date: 20250530

Claims (9)

1. A method for monitoring the status of a suspended carrier based on dynamic environment sensing, comprising the steps of: Collecting sensor data related to a suspension type carrier, wherein the sensor data comprise acceleration, air pressure and gesture data; respectively extracting the characteristics of the acceleration, the air pressure and the gesture data, and fusing the extracted characteristics through a linear fusion layer to obtain fusion characteristics; Inputting the fusion characteristics into a strategy network, outputting probability distribution of three actions, training the strategy network through reward calculation and a double experience pool mechanism, and updating parameters of the strategy network, wherein the method comprises the steps of inputting the fusion characteristics into the strategy network, outputting preference values of the three actions, namely normal, dragging and dropping, mapping the preference values into the probability distribution of the actions through a softmax function, randomly selecting one action to execute according to the probability distribution, calculating instant rewards and discount accumulated rewards based on sensor data of the current state, storing experience combination data of the state, the actions and the rewards executed by each action in an experience pool, wherein the experience pool comprises a common pool and an important pool, sampling the experience combination data from the experience pool if the experience quantity in the experience pool is larger than the sample quantity required by one training batch, calculating strategy gradient loss and entropy loss according to the experience combination data, and updating parameters of the strategy network through back propagation; And displaying dynamic curves of acceleration, air pressure and attitude data in real time and outputting action probability distribution by the strategy network.
2. The method for monitoring the status of a suspension type carrier based on dynamic environment sensing according to claim 1, wherein the steps of extracting the characteristics of the acceleration, the air pressure and the gesture data respectively, and fusing the extracted characteristics through a linear fusion layer, and obtaining the fused characteristics comprise: mapping three-dimensional acceleration data acquired by an accelerometer into eight-dimensional feature vectors through a full connection layer, and adding nonlinear expression by using Swish activation functions; After the air pressure data collected by the air pressure meter are normalized, the air pressure data are mapped into four-dimensional feature vectors through a linear layer, and a Swish activation function is used for adding nonlinear expression; mapping gesture data acquired by a gyroscope into eight-dimensional feature vectors through a full connection layer, and adding nonlinear expression by using Swish activation functions; and splicing the characteristic vectors extracted from the acceleration data, the air pressure data and the gesture data, and inputting the spliced characteristics into a linear fusion layer to obtain fusion characteristics.
3. The method for monitoring the status of a suspension type carrier based on dynamic environment sensing according to claim 2, wherein the splicing the feature vectors extracted from the acceleration data, the air pressure data and the gesture data, inputting the spliced features into a linear fusion layer, and obtaining the fusion features comprises: Splicing the feature vectors extracted from the acceleration data, the air pressure data and the gesture data to generate a joint feature: Wherein F is a joint characteristic, The characteristic vector of the acceleration is represented, The air pressure characteristic vector is represented by a model, Representing a gesture feature vector; feature fusion is carried out on the combined features through a linear fusion layer, and fusion features are generated: Wherein, the In order to fuse the features of the features, As a matrix of weights, the weight matrix, Is the first of the linear fusion layers And the bias term of the output neuron.
4. The method for monitoring a suspended carrier state based on dynamic environment sensing according to claim 3, wherein the feature fusion of the joint features by the linear fusion layer, generating fusion features comprises: And dynamically adjusting the weight matrix according to the real-time sensor data to strengthen key features, wherein the key features comprise abrupt change momentum, instantaneous change rate and frequency domain features.
5. The dynamic environment awareness based suspended carrier state monitoring method of claim 1 wherein the discount-jackpot is calculated by the formula: Wherein, the Expressed in time steps Is a discount-accumulating prize of (1), Indicating an instant prize is provided, Representing a discount factor; calculating a strategic gradient loss function by the following formula : Wherein, the Is the policy network in state Downward movement Is a function of the probability of (1), The natural logarithm of the probability of an action is represented, Accumulating rewards for the normalized discounts; Calculating the entropy loss function by the following formula : Wherein, the Is the policy network in state Down selection action Is a function of the probability of (1), Representing the natural logarithm of the probability of an action.
6. The method of claim 5, wherein calculating the policy gradient loss and the entropy loss from the empirical combined data, updating parameters of the policy network by back propagation comprises: Combining the strategy gradient loss and the entropy loss into total loss in a weighting way; the gradient of each parameter relative to the total loss is calculated, and the gradient is limited to a set threshold value through gradient clipping, and the parameters of the strategy network are updated by using an optimizer.
7. The method of claim 1, wherein the real-time visual display of dynamic curves of acceleration, barometric pressure and attitude data and the action probability distribution of the strategic network output comprises: selecting an action corresponding to the maximum probability from the action probability distribution output by the strategy network as a current predicted action; after each time of calculation and execution of the motion, recording current sensor data, current predicted motion and probability value of the motion, and updating drawing of an acceleration curve, a gyroscope curve and an air pressure curve through a visual interface.
8. A condition monitoring device for a suspended carrier, the device comprising: at least one processor and at least one memory storing a computer program; wherein the computer program, when executed by the at least one processor, causes the apparatus to perform the steps of the dynamic environment awareness based suspended carrier state monitoring method according to any one of claims 1 to 7.
9. A computer-readable storage medium storing a computer program, characterized in that the computer program, when executed by a processor, implements the steps of the dynamic context awareness based suspended carrier state monitoring method according to any one of claims 1 to 7.

Description

Suspension type carrier state monitoring method and device based on dynamic environment sensing Technical Field The embodiment of the disclosure relates to the technical field of intelligent monitoring, in particular to a method and a device for monitoring a suspension type carrier state based on dynamic environment sensing. Background Suspended carriers are engineering facilities deployed through flexible support structures such as cables, booms or catenaries that rely on tension to maintain the overall morphology, which gives the suspended carrier significant flexible structural characteristics. Suspended carriers are more susceptible to external factors such as wind loading, mechanical shock, vibration, etc. during operation than are rigid structures, which can cause dynamic deformation or deflection of the carrier. Thus, the stability and safety of such carriers often need to be dependent on highly accurate design and real-time monitoring. As an important component of the key infrastructure, the suspended carrier is widely applied to high-altitude and complex-environment operation scenes, such as maintenance pulleys, high-altitude cables, tourist cable cars and the like of the electric power transmission line. However, due to the extremely complex environment in which suspended carriers are used, these carriers typically span large spans of terrain ranging from hundreds to thousands of meters and are exposed to a variety of harsh external environments for extended periods of time, such as extreme temperatures, salt spray corrosion, high humidity, mechanical wear, and the like. Over time, these environmental factors can cause varying degrees of damage to the carrier. For example, wind and temperature variations may cause stretching or shrinking of the cable, salt spray corrosion may weaken the cable strength, and mechanical wear may lead to increased friction, thereby affecting the overall load carrying capacity. If the carrier is damaged locally, cascading failures may be caused, resulting in serious economic loss or safety accidents. Based on the characteristics, it is important to build a perfect real-time monitoring and intelligent early warning system. Currently, fault monitoring based on a suspension type carrier is not limited to traditional sensor acquisition and manual judgment, but gradually evolves towards an intelligent and data-driven direction, and the accuracy, the instantaneity and the automation degree of monitoring are remarkably improved. Suspension type carrier monitoring techniques can be divided into two categories, traditional optimization algorithm learning and deep learning techniques. The optimization algorithm learning is to use genetic algorithm and other technologies to perform optimization processing on simulation data (temperature, humidity, pressure and the like) of the suspension type carrier so as to judge whether the suspension type carrier is in a fault state or not. In the deep learning technology, the visual data such as the appearance, the running state and the like of the carrier are monitored to identify fault characteristics such as structural abnormality, displacement deviation and the like. The technology relies on a large-scale data set for training, continuously optimizes a neural network model, improves the recognition accuracy and the response speed, and realizes real-time dynamic monitoring and intelligent judgment of the suspension type carrier state. However, although the prior art greatly improves the shortcomings of the traditional mode, challenges and limitations are still unavoidable during practical application, and are mainly reflected in: (1) Because the suspension type carrier can continuously change along with the environment, the existing genetic algorithm lacks the online learning capability, the motion state change of the suspension type carrier in the three-dimensional space cannot be captured, the genetic algorithm excessively depends on an initial population, and the initial population can be converged to an optimal solution too early due to poor selection, so that the complex detection omission situation is caused. (2) Deep learning methods are highly dependent on large-scale and high-quality labeled datasets. The process of collecting, cleaning and labeling data is time-consuming and labor-consuming, and especially the difficulty of acquiring the data of the real scene in a high altitude or extreme environment is higher. Furthermore, deep learning models typically require a large amount of computational resources and a long iteration period during the training phase, resulting in a high overall cost. More importantly, when the model faces to unseen environmental changes, abnormal working conditions or edge cases, the model often lacks good generalization capability, and misjudgment or recognition failure can occur, so that the stability and reliability of the system under a practical complex scene are affected. Disclosure of Invention In