CN-121982363-A - Transformer substation construction machinery anti-collision early warning method and system based on dynamic security domain

CN121982363ACN 121982363 ACN121982363 ACN 121982363ACN-121982363-A

Abstract

The invention relates to the technical field of substation construction safety early warning and discloses a substation construction machinery anti-collision early warning method and system based on a dynamic safety domain, wherein the method comprises the steps of synchronously acquiring multi-view images through a plurality of binocular cameras deployed in a substation construction area and reconstructing a three-dimensional semantic map of the construction area; the method comprises the steps of carrying out target identification on crane key components and peripheral electric facilities in a multi-view image by adopting an improved target detection network, extracting three-dimensional coordinates of the crane key components and the electric facilities based on identification results, outputting motion track time sequence data of the crane key components, decomposing the motion track time sequence data by utilizing a trend filtering decomposition algorithm, extracting trend components of a current motion track, obtaining predicted safe distances based on the trend components, and triggering early warning according to the predicted safe distances. The invention realizes real-time monitoring, dynamic risk prediction and early warning of the crane operation process.

Inventors

LIU JIANGE
DAI XIN
CAO LI
JIANG MENGNA
LIU XIA

Assignees

国网江苏省电力有限公司淮安供电分公司

Dates

Publication Date: 20260505
Application Date: 20251217

Claims (10)

1. The anti-collision early warning method for the transformer substation construction machinery based on the dynamic security domain is characterized by comprising the following steps of: Synchronously acquiring multi-view images through a plurality of binocular cameras deployed in a construction area of a transformer substation, and reconstructing a three-dimensional semantic map of the construction area; adopting an improved target detection network to identify targets of crane key components and peripheral electric power facilities in the multi-view image; Extracting three-dimensional coordinates of the crane key component and the electric power facility based on the identification result, calculating a real-time space distance between the crane key component and the electric power facility, and outputting motion track time sequence data of the crane key component; Decomposing the motion trail time sequence data by utilizing a trend filtering decomposition algorithm, and extracting trend components representing the overall motion direction of the current motion trail; And constructing Gaussian linear fuzzy information particles based on the trend components, judging the similarity between the current information particles and a sample in a pre-stored risk track set, if so, predicting the motion track of the crane key component in a period of time in the future, calculating the predicted safety distance between the crane key component and an electric facility, and triggering early warning according to the predicted safety distance.
2. The method of claim 1, wherein constructing gaussian linear fuzzy information particles based on the trend component, performing similarity judgment on the current information particles and samples in a pre-stored risk track set, predicting a motion track of the crane key component in a future period if the current information particles are similar, calculating a predicted safe distance between the crane key component and an electric power facility, and triggering early warning according to the predicted safe distance, wherein the method comprises the steps of: If the similarity between the current track and the risk track sample exceeds a preset threshold value, predicting the space coordinates of the crane key components in a period of time in the future by using a long-period memory network; Calculating a predicted safe distance between the crane critical component and the electric power facility based on the predicted coordinates; And triggering an early warning signal when the predicted safety distance is smaller than a first safety threshold value.
3. The method of claim 2, wherein said employing an improved target detection network for target identification of crane critical components and peripheral electrical facilities in said multi-perspective image comprises: An inverse convolution module is embedded in an FPN-PAN feature pyramid structure of the YOLO model to realize self-adaptive modeling and enhancement of cross-scale features, and a high-resolution detection head is added to perform target detection and semantic segmentation.
4. The method of claim 3, wherein the extracting three-dimensional coordinates of the crane critical component and the electric power facility based on the recognition result, and calculating a real-time spatial distance between the crane critical component and the electric power facility, and outputting the motion profile timing data of the crane critical component, comprises: in the bounding box regression of target detection, optimizing by adopting a comprehensive loss function, wherein the comprehensive loss function combines angle loss, distance loss and internal cross ratio loss based on an auxiliary box concept; The motion track time sequence data of the crane key components are data sequences which are generated by a visual perception system, are arranged in time sequence and are used for representing position information of the crane key components including the boom end points and the hooks in a three-dimensional space.
5. The method according to claim 4, wherein the method further comprises: Comparing the real-time space distance with a second safety threshold, and triggering a second-stage early warning if the real-time space distance is smaller than the second safety threshold; The comprehensive loss function optimizes the alignment of the prediction frame and the real frame in the direction through an angle loss term, corrects the center point distance error through combining the distance loss term with angle information, constructs an auxiliary frame through introducing scale factors, and calculates internal cross ratio loss so as to jointly restrict the regression process of the boundary frame.
6. The method according to claim 2, wherein the decomposing the motion trajectory time series data by using a trend filtering decomposition algorithm, extracting a trend component of the current motion trajectory representing the overall motion direction, includes: By using Trend filtering decomposition algorithm -TFD), decomposing the original motion trajectory sequence into a linear trend component and a residual component; The linear trend which can reflect the whole motion rule of the crane is obtained by constructing the indexes of intra-grain combination degree, intra-grain separation degree and inter-grain separation degree and carrying out self-adaptive optimization on the smooth parameters of the decomposition process by utilizing the effectiveness function.
7. The method according to claim 2, wherein the method further comprises: After the motion track time sequence data is decomposed, further performing sample entropy calculation on residual error components, and dividing the decomposed components into trend components for long-term prediction, periodic components for periodic motion capture and noise components to be filtered according to sample entropy values; Wherein the periodic component can be used to assist in assessing the smoothness of crane motion.
8. The method of claim 2, wherein said constructing gaussian linear fuzzy information particles based on said trend component comprises: gaussian linear fuzzy information grain G is constructed, which is expressed as g= (k, b, σ, T), where k is the linear slope of the trajectory, b is the initial value, σ is the trajectory fluctuation variance, and T is the granularity time length.
9. The method according to claim 2, wherein the similarity is obtained by calculating a custom distance function between two fuzzy information particles, the distance function taking into account the effect of the difference in slope, intercept, volatility and time length of the two particles.
10. A substation construction machine anti-collision early warning system based on a dynamic security domain, for implementing the method according to any one of claims 1 to 9, comprising: the three-dimensional perception module is used for acquiring images through a plurality of binocular cameras and reconstructing a three-dimensional semantic map of a construction site; The target recognition and positioning module is configured with an improved target detection network and is used for recognizing crane key components and peripheral electric power facilities in the image, calculating the spatial positions and real-time safety distances of the crane key components and peripheral electric power facilities, and outputting motion track time sequence data of the crane key components; the track processing and predicting module is used for carrying out trend filtering decomposition on the motion track time sequence data of the crane so as to extract the inherent trend components of the crane and obtain a predicted safe distance based on the trend components; And the advanced early warning decision module is used for triggering early warning according to the predicted safety distance.

Description

Transformer substation construction machinery anti-collision early warning method and system based on dynamic security domain Technical Field The invention relates to the technical field of substation construction safety early warning, in particular to a substation construction machinery anti-collision early warning method and system based on a dynamic safety domain. Background With the continuous expansion of the construction scale of the transformer substation and the improvement of the automation degree of construction machinery, large-scale equipment such as cranes, cranes and the like frequently operate in a high-voltage electrical equipment intensive environment, and the space dynamic safety problem of the large-scale equipment is more remarkable. The components such as the suspension arm and the suspension hook are very easy to approach high-voltage equipment such as a bus, a disconnecting switch and a framework due to visual shielding, inertial overshoot or operation errors in operation, and if collision prevention early warning is not timely, equipment damage, arc discharge and even casualties can be possibly caused. The traditional anti-collision system is mostly based on two-dimensional image recognition or ultrasonic ranging, and has the following limitations that three-dimensional space understanding capability is lacked, shielding, reflecting and shadow environment recognition of complex structures is unstable, safety distance judgment is mostly a static threshold value and cannot adapt to nonlinear change of a dynamic motion track of a crane, prediction capability of time dimension is lacked, and warning and early warning lag can be realized only when collision is about to happen. Therefore, an intelligent anti-collision system capable of combining high-precision visual perception, depth feature modeling and time sequence prediction analysis is needed, and dynamic security domain real-time perception and early warning of a crane under a transformer substation construction environment are realized. Disclosure of Invention The invention aims to solve the problems of poor adaptability to complex construction scenes and insufficient dynamic track prediction capability of the traditional anti-collision method, and provides a transformer substation construction machinery anti-collision early warning method and system based on a dynamic security domain, so as to realize real-time monitoring, dynamic risk prediction and early warning of a crane operation process. In a first aspect, an embodiment of the present invention provides a method for pre-warning a substation construction machine in an anti-collision manner based on a dynamic security domain, including the following steps: Synchronously acquiring multi-view images through a plurality of binocular cameras deployed in a construction area of a transformer substation, and reconstructing a three-dimensional semantic map of the construction area; adopting an improved target detection network to identify targets of crane key components and peripheral electric power facilities in the multi-view image; Extracting three-dimensional coordinates of the crane key component and the electric power facility based on the identification result, calculating a real-time space distance between the crane key component and the electric power facility, and outputting motion track time sequence data of the crane key component; Decomposing the motion trail time sequence data by utilizing a trend filtering decomposition algorithm, and extracting trend components representing the overall motion direction of the current motion trail; And constructing Gaussian linear fuzzy information particles based on the trend components, judging the similarity between the current information particles and a sample in a pre-stored risk track set, if so, predicting the motion track of the crane key component in a period of time in the future, calculating the predicted safety distance between the crane key component and an electric facility, and triggering early warning according to the predicted safety distance. Further, a Gaussian linear fuzzy information grain is constructed based on the trend component, similarity judgment is carried out on the current information grain and a sample in a pre-stored risk track set, if the current information grain is judged to be similar, the motion track of the crane key component in a future period is predicted, and the predicted safety distance between the crane key component and an electric power facility is calculated, and early warning is triggered according to the predicted safety distance, wherein the method comprises the following steps: based on the trend component, constructing Gaussian linear fuzzy information particles to represent the characteristics of the crane motion trail; Carrying out similarity judgment on fuzzy information particles constructed by the current crane motion track and a sample in a preset historical risk track set; If