Identification system and method for unsafe factors in new energy infrastructure construction process

Abstract

The invention discloses an identification system and method for unsafe factors in a new energy infrastructure construction process, and the specific technical scheme is as follows: constructing a feature database of unsafe factors in different construction stages and different accident types; carrying out similarity calculation on unsafe factor feature vectors collected in real time and unsafe factor feature vectors in a feature database, and when the calculated similarity exceeds a set similarity threshold value, preliminarily identifying a current unsafe factor; and analyzing a causal relationship, a time sequence and a spatial relationship between the current unsafe factor and other unsafe factors associated with the current unsafe factor, constructing a factor association network model, determining an unsafe factor chain, and mining potential unsafe factors. By analyzing the mutual influence between the current unsafe factors and the potential unsafe factors, the complex relationship between the data is deeply mined, so that workers can be helped to comprehensively understand the root and development trend of the unsafe factors, and more effective countermeasures can be formulated.

Description

Technical Field

[0001] The present invention relates to the field of identifying unsafe factors during infrastructure construction, and particularly to an identification system and method for unsafe factors during new energy infrastructure construction. Background Art

[0002] In the context of the global vigorous development of clean energy, the scale of new energy infrastructure continues to expand, covering multiple fields such as the construction of wind farms, photovoltaic power stations, and energy storage facilities. These infrastructure projects often have long construction periods, complex construction environments, and involve a large number of new equipment and technologies. During the construction process, they face various unsafe factors such as equipment failures, improper personnel operations, and environmental changes, which may not only lead to construction accidents but also affect the project progress and economic benefits. Therefore, effectively identifying unsafe factors during new energy infrastructure construction has become a key requirement for ensuring the healthy development of the new energy industry.

[0003] Currently, there are many deficiencies in the existing technologies for identifying unsafe factors in new energy infrastructure. On the one hand, traditional manual inspection methods rely on the experience and responsibility of staff, with problems such as large subjective judgment deviations, low efficiency, and inability to monitor in real time, making it difficult to comprehensively capture complex and changing unsafe factors. For example, in the construction of large wind farms, manual inspections cannot timely detect potential dangers in high-altitude operations or early fault hazards inside equipment. On the other hand, for some sensor-based monitoring technologies, the sensor layout lacks scientific planning, the data collection is incomplete, and the collected data is easily interfered by the environment, resulting in insufficient data accuracy. At the same time, existing identification methods are relatively single in the data processing and analysis links, mostly using simple threshold judgments or traditional statistical analyses, unable to deeply explore the complex relationships between data, difficult to identify potential chains of unsafe factors, and the assessment of the severity of unsafe factors is not accurate enough.

[0004] In addition, existing unsafe factor identification systems lack the ability of dynamic update and adaptability. With the continuous development of new energy infrastructure technologies, new types of unsafe factors are constantly emerging. However, the feature libraries of existing systems cannot be updated in time, and the identification models are also difficult to adapt to new situations, resulting in a gradual decline in the timeliness and accuracy of identification, unable to meet the growing safety requirements of new energy infrastructure. There is an urgent need for more advanced and intelligent identification systems and methods to solve these problems. Summary of the Invention

[0005] In order to overcome the disadvantages and deficiencies of the existing technology, the present invention provides an identification system and method for unsafe factors during new energy infrastructure construction.

[0006] A method for identifying unsafe factors in the process of new energy infrastructure construction, which constructs a characteristic database of unsafe factors under different construction stages and different accident types; collects on-site data of the infrastructure site in real time, and extracts corresponding characteristic vectors of unsafe factors; based on a matching algorithm based on similarity measurement, calculates the similarity between the collected characteristic vectors of unsafe factors and the characteristic vectors of unsafe factors in the characteristic database, and when the calculated similarity is greater than the set similarity threshold, determines the current unsafe factor; analyzes the causal relationship, time sequence and spatial relationship between the current unsafe factor and other factors associated with it through a factor association network model, determines the unsafe factor chain, thereby excavates potential unsafe factors, and comprehensively evaluates the severity of the current unsafe factor and potential unsafe factors.

[0007] Preferably, the identification method includes the following steps: Step S1: Analyze the unsafe factors under different construction stages and different accident types, construct the characteristic vectors of unsafe factors under different construction stages and different accident types, so as to construct a characteristic database of unsafe factors; Step S2: With the goal of maximizing the coverage of the infrastructure area and minimizing the number of sensors, arrange various sensors at the infrastructure site, collect on-site data of the infrastructure site in real time, and extract the characteristic vectors of unsafe factors; Step S3: Match the characteristic vectors of unsafe factors collected in real time with the characteristic vectors of unsafe factors in the characteristic database, use a matching algorithm based on similarity measurement to calculate the similarity between the two, and when the calculated similarity is greater than the set similarity threshold, preliminarily determine that there is a corresponding unsafe factor, which is recorded as the current unsafe factor; Step S4: Analyze other factors associated with the current unsafe factor, construct a factor association network model, consider the causal relationship, time sequence and spatial position relationship between different physical quantities, determine the unsafe factor chain, and excavate potential unsafe factors; Step S5: Use a model that combines fuzzy hierarchical analysis and grey relational analysis to comprehensively evaluate the severity of the current unsafe factor and potential unsafe factors.

[0008] Preferably, the step S1 includes: Step S11: Based on historical accident cases, industry standards and expert experience, analyze the unsafe factors under different construction stages and different accident types, and determine the corresponding relationship between the characteristics and the unsafe factors; Step S12: Use an improved hierarchical clustering algorithm to preliminarily cluster the characteristic vectors of the infrastructure site, and introduce the spatial correlation weight between the characteristics during the clustering process. The improved hierarchical clustering distance calculation formula is: , where and are two eigenvectors, and are respectively the and -th eigenvalue under the -th dimension of the vector ; is the spatial correlation weight of the -th dimension, is the dimension number of the eigenvector, is the number of features under each dimension; Step S13: Input the clustering result in Step S12 into the decision tree model for further classification. The decision tree model splits nodes according to the information gain ratio of features, classifies and arranges the features, and improves the feature database of unsafe factors; Step S14: Use the incremental learning algorithm to update the feature database of unsafe factors, and incorporate the newly emerged features of unsafe factors into the feature database in a timely manner. The weight update formula based on gradient descent is adopted during the incremental learning process: , wherein, is the weight vector of the feature library model at time, is the learning rate, is the loss function with respect to the weight vector gradient.

[0009] Preferably, the said Step S2 includes: Step S21: Arrange various sensors at the infrastructure construction site based on the spatial optimization layout model. The layout model formula is as follows: , Constraint condition: , wherein, indicates whether to arrange a sensor at the position , indicates the cost of arranging a sensor at the position , indicates the coverage ability of the sensor at the position to the area , indicates the set of areas that can be covered by the sensor at the position , indicates the set of all areas divided by the infrastructure construction site; Step S22: Determine the sampling frequencies of various sensors according to the sensor types and the dynamic characteristics of the infrastructure site, and collect data on the physical parameters and real-time scenarios in the infrastructure site at specific sampling frequencies; Step S23: Extract the feature vectors of unsafe factors in the infrastructure site in real time, where the feature vectors of unsafe factors include time-domain features, frequency-domain features, and abstract features.

[0010] Preferably, the similarity calculation formula in step S3 is: , where and are the real-time feature vector and the feature vector in the feature database respectively, and are the and th and th sub-vectors of the th and th time points of the vectors is the kernel function, and are the numbers of time points of the vectors and respectively.

[0011] Preferably, step S4 includes: Step S41: Analyze other factors associated with the current unsafe factor, and map the other factors as variables to the nodes of the Bayesian network; Step S42: Determine the directed edges between the nodes according to the causal relationship, time sequence, and spatial position relationship between different physical quantities, draw the factor association network, and adjust the positions of the nodes and the layout of the edges in combination with the actual logical relationship in the infrastructure process; Step S43: For each node, estimate its conditional probability table based on historical data and expert experience, and analyze the occurrence probabilities of various factors in different situations; Step S44: Sort the occurrence probabilities of various factors in different situations from large to small to mine potential unsafe factors.

[0012] Preferably, in step S5, the weights of different unsafe factors and their associated factors are determined by using fuzzy analytic hierarchy process. The elements of the fuzzy judgment matrix are calculated based on the fuzzy evaluation of the relative importance of factor and factor by experts. Then, the correlation degree between the real-time unsafe factor and the standard risk model is calculated through grey relational analysis. The grey relational degree calculation formula is: , in, For reference sequence Compare with sequence The grey correlation degree of and are the reference sequence and the ith comparison sequence in the The value under the indicator, is the resolution coefficient.

[0013] Correspondingly, a system for identifying unsafe factors in the process of new energy infrastructure construction includes a multi-source heterogeneous data aggregation and fusion unit, an adaptive data cleaning and feature extraction unit, a dynamic feature library construction and maintenance unit, a similarity matching and preliminary judgment unit, a multi-factor association mining and analysis unit, and a comprehensive evaluation and visualization presentation unit; the multi-source heterogeneous data aggregation and fusion unit is used to receive and integrate physical parameters and image multi-source data from different sensors; the adaptive data cleaning and feature extraction unit is used to clean the aggregated data and extract features based on a specific algorithm; the dynamic feature library construction and maintenance unit is responsible for constructing and dynamically updating the unsafe factor feature library based on historical cases; the similarity matching and preliminary judgment unit makes a preliminary judgment on the unsafe factors through a similarity algorithm; the multi-factor association mining and analysis unit uses a constructed factor association network model to analyze the associated factors of the unsafe factors; the comprehensive evaluation and visualization presentation unit uses a comprehensive evaluation algorithm to evaluate the severity of the unsafe factors.

[0014] Compared with the prior art, the present invention has the following beneficial effects: 1. By constructing a feature database of unsafe factors in different construction stages and accident types, the feature vectors of unsafe factors collected in real time are calculated for similarity with the feature vectors of unsafe factors in the feature database. When the calculated similarity exceeds the set similarity threshold, the current unsafe factor is preliminarily identified, and then the causal relationship, time sequence and spatial relationship between the current unsafe factor and other unsafe factors associated with it are analyzed, and a factor association network model is constructed to determine the unsafe factor chain, thereby mining potential unsafe factors. By analyzing the mutual influence between the current unsafe factors and potential unsafe factors and deeply mining the complex relationship between the data, the severity of the current unsafe factors and potential unsafe factors can be accurately assessed, which can help staff fully understand the root causes and development trends of unsafe factors and formulate more effective response measures.

[0015] 2. Through the coordinated operation of six major units such as the multi-source heterogeneous data aggregation and fusion unit and the adaptive data cleaning and feature refinement unit, the entire recognition system realizes the efficient processing of the full process of new energy infrastructure's unsafe factors from data collection, processing and analysis to comprehensive evaluation and output. It can timely and accurately identify potential risks, provide strong guarantees for the safe, stable and efficient construction of new energy infrastructure, effectively reduce the probability of accidents, reduce economic losses and the risk of casualties, and promote the healthy development of the new energy infrastructure industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a flowchart of the method steps of the present invention; Figure 2 It is a diagram showing the composition of the system units of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0017] It should be noted that, without conflict, the embodiments in the present application and the features in the embodiments can be combined with each other. The following further describes the present application in detail with reference to the drawings and specific embodiments.

[0018] In this embodiment, the construction and operation and maintenance scenarios of a wind farm are taken as an example to illustrate the specific implementation process of the recognition method in the present application. As Figure 1 shown, the present invention discloses a method for identifying unsafe factors in the process of new energy infrastructure, and the method includes: Step S1: Construct a feature database of unsafe factors: Based on historical accident cases, industry standards, and expert experience in the field of new energy infrastructure, analyze the unsafe factors in different construction stages and different accident types, construct feature vectors of unsafe factors in different construction stages and different accident types, and thus construct a feature database of unsafe factors. A clear mapping relationship is established between the features in the feature database and the unsafe factors, and a dynamic update mechanism is provided to adapt to the newly emerged features of unsafe factors.

[0019] Specifically, constructing a feature library of unsafe factors is an important foundation for the recognition system. Classify and organize the extracted feature vectors based on historical accident cases, industry standards, and expert experience in the field of new energy infrastructure. For example, in the construction of an energy storage power station, referring to past battery thermal runaway accident cases, classify features such as abnormal increase in battery temperature, excessive voltage fluctuation, and change in internal gas concentration; combine industry standards and incorporate relevant features such as operation specifications and safety distances of construction equipment into the feature library; and supplement some features that are difficult to quantify but have important indicative significance, such as the fatigue state of construction workers, through expert experience.

[0020] With the development of new energy infrastructure technologies and the emergence of new security risks, such as the new risks brought about by the application of new energy storage materials, the feature library can be updated in a timely manner to incorporate new features and security risks, ensuring that the feature library always meets the actual needs of new energy infrastructure and providing a reliable basis for accurately identifying security risks.

[0021] Step S2: Collect on-site data in real time and extract the feature vectors of security risks. Use various sensors distributed at the new energy infrastructure site to collect data on physical parameters and real-time scenarios in the infrastructure environment at a specific sampling frequency, and perform feature extraction of security risks. The sampling frequency is adaptively adjusted according to different sensor types and the dynamic characteristics of the infrastructure scenario. The collected data includes time, space dimensions, and multi-physical quantity features.

[0022] Specifically, in the construction and operation and maintenance projects of wind farms, the construction environment is complex and changeable, covering various scenarios such as high altitude, underground, high temperature, and high humidity, and there are many potential security risks. To comprehensively capture these risk information, it is necessary to use various sensors distributed at the infrastructure site for data collection, including temperature sensors, pressure sensors, vibration sensors, and image acquisition devices. For example, temperature sensors can monitor the temperature of welding parts in real time to prevent the degradation of material properties due to overheating; pressure sensors are installed on lifting equipment to monitor the stress during the lifting process and avoid accidents caused by overloading; vibration sensors are arranged in the foundation construction area to detect the vibration state of the foundation and prevent problems such as foundation settlement; image acquisition devices are used to monitor the operation specifications of construction workers and changes in the on-site environment.

[0023] It should be noted that in the specific implementation process, the image data collected by the image acquisition device can be preprocessed using an object detection model based on the fusion of multi-scale convolutional neural networks and attention mechanisms. This model performs convolutional operations on the image at different scales to extract multi-scale features, and at the same time uses the attention mechanism to enhance the attention to key regions. The calculation formula of the attention mechanism is: ; where is the input image is the attention weight matrix of is the activation function, and are the weight matrix and bias vector, is the number of network layers. Through this model, it is possible to more effectively extract the target information related to security risks from the image data and assist in the subsequent identification of security risks.

[0024] Step S3, Preliminary determination of unsafe factors: Match the feature vectors of the unsafe factors collected in real time with those in the feature database, and use a matching algorithm based on similarity measurement to calculate the similarity between the two. When the calculated similarity is greater than the set similarity threshold, it is preliminarily determined that there are corresponding unsafe factors, which are recorded as the current unsafe factors.

[0025] Specifically, a dynamic time warping similarity model based on kernel function is adopted. The model takes into account the scalability differences of feature vectors of different types of unsafe factors on the time scale, and uses the kernel function to map the feature vectors into a high-dimensional space to enhance the distinguishability of features. The similarity calculation formula is: ; where, and are the real-time feature vector and the feature vector in the feature library respectively, and are the sub-vectors of the and vectors at the th and th time points respectively, is the kernel function, is the time alignment factor determined according to the dynamic time warping algorithm, and are the number of time points of the and vectors respectively. Through this model, the similarity between feature vectors can be calculated more accurately, improving the accuracy of the preliminary identification of unsafe factors.

[0026] Specifically, in practical applications, the set similarity threshold is not fixed, but is optimized according to the risk levels of different types of unsafe factors and the misjudgment costs. For high-risk unsafe factors, the threshold will be set relatively low to ensure that potential risks can be detected in time; while for some low-risk and easily misjudged factors, the threshold will be appropriately increased to reduce false alarms.

[0027] Step S4, Mining potential unsafe factors: Analyze other factors associated with the current unsafe factors, construct a factor association network model, consider the causal relationship, time sequence, and spatial location relationship between different physical quantities, determine the unsafe factor chain, and mine potential unsafe factors. The factor association network model is constructed based on the Bayesian network theory and the parameter adjustment is carried out in combination with the actual logical relationship in the infrastructure construction process.

[0028] Specifically, for the initially identified unsafe factors, further analyze other factors associated with them. For example, in the operation and maintenance infrastructure of a new energy wind farm, if abnormal vibration of the wind turbine blades is initially identified, by constructing a factor correlation network model, considering the causal relationship, time sequence, and spatial position relationship between different physical quantities, potential unsafe factor chains can be mined. It may be found that the abnormal blade vibration is caused by the gearbox failure leading to unstable rotation speed, which in turn causes uneven stress on the blades. At the same time, the drastic change in environmental wind speed also exacerbates this vibration. This factor correlation network model is constructed based on the Bayesian network theory and the parameter adjustment is combined with the actual logical relationship in the infrastructure process, which can accurately analyze the mutual influence between various factors, help the staff comprehensively understand the root cause and development trend of unsafe factors, and formulate more effective countermeasures.

[0029] Step S5: Evaluate the severity of unsafe factors: Use a model that combines fuzzy analytic hierarchy process and grey relational analysis to comprehensively evaluate the severity of current unsafe factors and potential unsafe factors. The evaluation results are output in an intuitive visual form, including risk level identification, location marking of unsafe factors, and indication of the possible influence range. At the same time, a detailed evaluation report is generated, which contains key data and analysis basis during the identification process.

[0030] The model that combines fuzzy analytic hierarchy process and grey relational analysis is specifically as follows: First, use fuzzy analytic hierarchy process to determine the weights of different unsafe factors and their associated factors. The calculation of the fuzzy judgment matrix elements is based on the fuzzy evaluation of experts on the relative importance of factor and factor . Then, calculate the grey relational degree between the real-time unsafe factors and the standard risk mode through grey relational analysis. The grey relational degree calculation formula is: , where is the grey relational degree between the reference sequence and the comparison sequence , and are the values of the reference sequence and the i-th comparison sequence under the j-th index respectively, , and is the resolution coefficient. Through this fusion model, the severity of unsafe factors can be evaluated more comprehensively and accurately.

[0031] Specifically, considering the preliminary recognition results and the results of multi-factor correlation analysis comprehensively, a comprehensive evaluation algorithm is used to evaluate the severity of unsafe factors during the infrastructure construction process. Taking the construction of a large-scale solar photovoltaic power station as an example, if the deviation of the installation angle of the photovoltaic panels is identified, and the correlation analysis finds that this will lead to problems such as reduced power generation efficiency and accelerated equipment aging due to local overheating, the comprehensive evaluation algorithm will combine these factors and evaluate from multiple dimensions such as safety risks, economic losses, and impacts on the project schedule to determine the severity of the unsafe factor.

[0032] The evaluation results are output in an intuitive visual form. For example, on the monitoring system interface, the risk levels are identified by different colors (red indicates high risk, yellow indicates medium risk, and blue indicates low risk), the specific locations of unsafe factors are marked on the infrastructure map, and the possible influence ranges are shown graphically. At the same time, a detailed evaluation report is generated, which includes the key data during the recognition process (such as the original data collected, the processed feature vectors) and the analysis basis (such as the similarity calculation results, the process of factor correlation analysis), facilitating the staff to understand the specific situation of unsafe factors and take corresponding treatment measures in a timely manner to ensure the safe and smooth progress of new energy infrastructure projects.

[0033] A further implementation method is that in step S1, the feature classification adopts a fusion model based on improved hierarchical clustering and decision tree, including the following process: Step S11: Analyze the unsafe factors in different construction stages and different accident types based on historical accident cases, industry standards, and expert experience, and determine the corresponding relationship between features and unsafe factors; Step S12: Use the improved hierarchical clustering algorithm to preliminarily cluster the feature vectors at the infrastructure construction site. During the clustering process, the spatial correlation weight between features is introduced. The improved hierarchical clustering distance calculation formula is: , where, and are two feature vectors, and are respectively the and th th feature values under the th dimension of the vectors is the spatial correlation weight of the th dimension, is the dimension number of the feature vector, is the number of features under each dimension; Step S13: Input the clustering results in Step S12 into a decision tree model for further classification. The decision tree model splits nodes according to the information gain ratio of features, classifies and organizes the features, so as to more accurately classify and organize the features and improve the feature database of unsafe factors; Step S14: Use the incremental learning algorithm to update the feature database of unsafe factors, and incorporate the newly emerged features of unsafe factors into the feature database in a timely manner. In the incremental learning process, the weight update formula based on gradient descent is adopted: , where, is the weight vector of the feature library model at time is the learning rate, is the loss function with respect to the weight vector gradient.

[0034] Further implementation method, the said Step S2 includes the following process: Step S21: Arrange various sensors at the infrastructure construction site based on the spatial optimization layout model. This model aims to maximize the coverage of the infrastructure area and minimize the number of sensors, taking into account factors such as the topography and geomorphology of the infrastructure site, equipment distribution, and high-incidence areas of potential unsafe factors. The sensor layout determined by this model can collect comprehensive and targeted data more efficiently; the layout model formula is as follows: , Constraint conditions: , where, indicates whether to arrange a sensor at location , represents the cost of arranging a sensor at location , represents the coverage ability of the sensor at location to area , represents the set of areas that the sensor at location can cover, represents the set of all areas divided by the infrastructure site; Step S22: Determine the sampling frequencies of various sensors according to the sensor types and the dynamic characteristics of the infrastructure construction site, and collect data on physical parameters and real-time scenarios in the infrastructure construction site at specific sampling frequencies. For example, in areas where the temperature changes slowly, the sampling frequency of the temperature sensor can be set relatively low; while at the critical parts of equipment operation, the sampling frequency of the vibration sensor needs to be significantly increased to capture the subtle changes in abnormal vibrations of the equipment, ensuring that the collected data can accurately reflect the actual situation of the infrastructure construction site and providing rich and effective raw data for subsequent analysis; Step S23: Pass the collected raw data through a data cleaning algorithm to remove outliers caused by factors such as sensor failures and interference, and then adopt a feature extraction algorithm that combines time-frequency analysis and deep learning to extract feature vectors from the cleaned data that can characterize potential unsafe factors during the infrastructure construction process. The feature vectors include time-domain features, frequency-domain features, and abstract features learned through deep neural networks.

[0035] In the specific implementation process, the cleaning of raw data can adopt a data anomaly detection model based on dynamic threshold correction, and dynamically adjust the anomaly detection threshold according to the time series characteristics of the data and the change trend of adjacent data points. The model formula is: , where, is the anomaly detection threshold of the th data point, is the mean value of the first data points, is the standard deviation of the first data points, is the coefficient adaptively adjusted according to the data fluctuation situation. This model can more accurately identify and remove outliers caused by sensor failures or environmental interference, providing a more reliable data basis for subsequent feature extraction.

[0036] In some implementation manners, after removing outliers, a feature extraction algorithm that combines time-frequency analysis and deep learning is adopted to extract feature vectors from the cleaned data that can characterize potential unsafe factors during the infrastructure construction process. During the manufacturing process of wind turbine blades, time-frequency analysis can transform vibration data from the time domain to the frequency domain, clearly presenting the vibration characteristics at different frequencies, while deep learning algorithms (such as convolutional neural networks) can automatically learn the complex patterns and abstract features in the data, and extract key features of unsafe factors such as blade cracks and imbalances from a large amount of data. These feature vectors include time-domain features (such as mean value, variance), frequency-domain features (such as frequency peak), and abstract features learned through deep neural networks (such as the feature pattern of blade structure damage), laying a solid foundation for subsequent identification of unsafe factors.

[0037] Further embodiments, the step S4 includes the following process: Step S41, analyze other factors associated with the current unsafe factor according to the current unsafe factor, and map the other factors to the Bayesian network nodes as variables; Step S42, determine the directed edges between the nodes according to the causal relationship, time sequence, and spatial position relationship between different physical quantities, draw the factor association network, and adjust the positions of the nodes and the layout of the edges in combination with the actual logical relationship in the infrastructure construction process; Step S43, for each node, estimate its conditional probability table according to historical data and expert experience, and analyze the occurrence probabilities of various factors in different situations; Step S44, sort the occurrence probabilities of various factors in different situations from large to small to mine potential unsafe factors.

[0038] It should be noted that in some embodiments, it is recommended to calculate the causal relationship strength in the factor association network model by using a model based on the combination of Granger causality test and conditional entropy. This model determines the causal direction between factors through the Granger causality test, and then uses conditional entropy to measure the strength of the causal relationship. The causal relationship strength calculation formula is:

[0039] Among them, represents the causal relationship strength of factor to factor , is the conditional entropy of factor under the condition of knowing factor , is the conditional entropy of factor without considering the historical information of factor . Through this model, the causal relationship between different physical quantities can be mined more precisely, and the mining of the unsafe factor chain can be improved.

[0040] Such as Figure 2As shown in the figure, the present invention also discloses an identification system for unsafe factors in the process of new energy infrastructure, which includes the following six units: a multi-source heterogeneous data aggregation and fusion unit, which is communicatively connected to various sensors distributed at the new energy infrastructure site and is used to receive and integrate multi-source data such as physical parameters and images from different sensors; an adaptive data cleaning and feature refinement unit, whose input is connected to the output of the multi-source heterogeneous data aggregation and fusion unit and is used to clean the aggregated data and extract features based on specific algorithms; a dynamic feature library construction and maintenance unit, whose input is connected to the output of the adaptive data cleaning and feature refinement unit and is responsible for constructing and dynamically updating the unsafe factor feature library based on historical cases, etc.; a similarity matching and preliminary judgment unit, whose input is connected to the output of the adaptive data cleaning and feature refinement unit and the dynamic feature library construction and maintenance unit, and is used to make a preliminary judgment on unsafe factors through a specific similarity algorithm; a multi-factor association mining and analysis unit, whose input is connected to the output of the similarity matching and preliminary judgment unit and is used to analyze the associated factors of unsafe factors by using the constructed factor association network model; a comprehensive evaluation and visualization presentation unit, whose input is connected to the output of the similarity matching and preliminary judgment unit and the multi-factor association mining and analysis unit, and is used to evaluate the severity of unsafe factors by using a specific comprehensive evaluation algorithm and output the results in a visual form and generate a report.

[0041] The identification of unsafe factors in traditional new energy infrastructure relies on manual inspections or single-sensor monitoring, which has problems such as incomplete coverage and poor real-time performance. Through the multi-source data collection step, this system uses temperature, pressure, vibration sensors and image acquisition devices to achieve all-round monitoring of the construction environment. Taking the tower construction of a wind farm as an example, the vibration sensor samples at a high frequency to capture subtle anomalies of the equipment, and the image acquisition device monitors the personnel operations in real time, solving the problems of low efficiency and easy omission in manual inspections. At the same time, the sampling frequency is dynamically adjusted to ensure accurate data collection in key areas, avoiding data loss caused by unreasonable traditional sensor layouts and overcoming the defects of incomplete monitoring and low data quality in the existing technology.

[0042] In the data processing and analysis link, the existing technology mostly uses simple threshold judgment or traditional statistical methods, which cannot deeply explore complex data relationships and have poor adaptability to newly emerging unsafe factors. The dynamic feature library constructed by this method is based on historical cases, industry standards and expert experience, and combines an incremental learning update mechanism, which can timely incorporate the new risk features brought by the application of new energy storage materials. Compared with the traditional static feature library, it has stronger timeliness. In addition, the multi-factor association analysis model based on the Bayesian network can accurately sort out the causal relationships of various factors. For example, in the operation and maintenance of a wind farm, it can clearly analyze the association between abnormal blade vibration, gearbox failure and wind speed change, changing the limitation of the traditional method of analyzing data in isolation and greatly improving the accuracy and comprehensiveness of unsafe factor identification.

[0043] In terms of result output and application, the prior art lacks systematic evaluation of the severity of unsafe factors, and the result display form is single, making it difficult to assist in decision-making. This system adopts a comprehensive evaluation algorithm that combines fuzzy analytic hierarchy process and grey relational analysis to quantify the severity of unsafe factors from multiple dimensions such as safety risks, economic losses, and project progress. At the same time, it displays the risk level, location, and influence range through a visual interface and generates a report containing detailed data and analysis basis. Compared with the traditional method that only provides simple alarm information, it can help staff quickly grasp the overall situation of risks, formulate scientific response strategies, effectively make up for the deficiencies of the prior art in risk assessment and decision support, and provide strong guarantee for the safe and efficient construction of new energy infrastructure.

[0044] Although the embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various equivalent changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principle and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalent scope.

Claims

1. A method for identifying unsafe factors in the process of new energy infrastructure construction, characterized by: Construct a feature database of unsafe factors in different construction stages and accident types; collect field data of infrastructure sites in real time and extract corresponding unsafe factor feature vectors; calculate the similarity between the collected unsafe factor feature vectors and the unsafe factor feature vectors in the feature database based on a matching algorithm of similarity measurement, and determine the current unsafe factor when the calculated similarity is greater than the set similarity threshold; analyze the causal relationship, time sequence and spatial relationship between the current unsafe factor and other factors associated with it through a factor association network model, determine the unsafe factor chain, thereby exploring potential unsafe factors and comprehensively evaluating the severity of current unsafe factors and potential unsafe factors.

2. A method for identifying unsafe factors in the process of new energy infrastructure construction according to claim 1, characterized in that: The steps include: Step S1, analyzing unsafe factors in different construction stages and different accident types, constructing feature vectors of unsafe factors in different construction stages and different accident types, and thus constructing a feature database of unsafe factors; Step S2: with the goal of maximizing the coverage of the infrastructure area and minimizing the number of sensors, various sensors are arranged at the infrastructure site to collect field data of the infrastructure site in real time and extract feature vectors of unsafe factors; Step S3, matching the unsafe factor feature vector collected in real time with the unsafe factor feature vector in the feature database, and using a matching algorithm based on similarity measurement to calculate the similarity between the two. When the calculated similarity is greater than a set similarity threshold, it is preliminarily determined that the corresponding unsafe factor exists and is recorded as the current unsafe factor; Step S4: Analyze other factors associated with the current unsafe factor, build a factor association network model, consider the causal relationship, time sequence and spatial position relationship between different physical quantities, determine the unsafe factor chain, and mine potential unsafe factors; Step S5: Use the fusion model of fuzzy hierarchical analysis and grey relational analysis to comprehensively evaluate the severity of current unsafe factors and potential unsafe factors.

3. A method for identifying unsafe factors in the process of new energy infrastructure construction according to claim 2, characterized in that: The step S1 comprises: Step S11: Analyze the unsafe factors in different construction stages and accident types based on historical accident cases, industry standards and expert experience, and determine the corresponding relationship between the characteristics and the unsafe factors; Step S12: Use the improved hierarchical clustering algorithm to preliminarily cluster the feature vectors of the infrastructure site. In the clustering process, the spatial correlation weights between the features are introduced. The improved hierarchical clustering distance calculation formula is: ， in, and are two eigenvectors, and The vectors are and No. The first dimension eigenvalues, For the The spatial correlation weight of the dimension, is the dimension of the feature vector, is the number of features under each dimension; Step S13: Input the clustering result in step S12 into the decision tree model for further classification. The decision tree model splits the nodes according to the information gain ratio of the features, classifies and organizes the features, and improves the feature database of unsafe factors.

4. A method for identifying unsafe factors in the process of new energy infrastructure construction according to claim 3, characterized in that: The method further includes step S14, using an incremental learning algorithm to update the feature database of unsafe factors, and timely incorporate the features of newly emerged unsafe factors into the feature database. The weight update formula based on gradient descent is used in the incremental learning process: ， in, for The weight vector of the moment feature library model, is the learning rate, is the loss function About weight vector gradient.

5. The method for identifying unsafe factors in the process of new energy infrastructure construction according to claim 2 is characterized by: The step S2 comprises: Step S21: Arrange various sensors at the infrastructure site based on the spatial optimization layout model. The layout model formula is as follows: ， Constraints: , in, Indicates whether it is at the location Arrange sensors, Indicates at location The cost of deploying sensors, Indicates location Sensor pair area The coverage capability, Indicates location The set of areas that can be covered by the sensor, Represents the set of all areas where the infrastructure site is divided; Step S22: Determine the sampling frequency of various sensors according to the sensor type and the dynamic characteristics of the infrastructure site, and collect data on the physical parameters and real-time scenes in the infrastructure site at a specific sampling frequency; Step S23: extracting the unsafe factor feature vectors in the infrastructure construction site in real time.

6. A method for identifying unsafe factors in the process of new energy infrastructure construction according to claim 5, characterized in that: The unsafe factor feature vector includes time domain features, frequency domain features and abstract features.

7. The method for identifying unsafe factors in the process of new energy infrastructure construction according to claim 2 is characterized by: The similarity calculation formula in step S3 is: ， in, and are the real-time feature vector and the feature vector in the feature database respectively, and The vectors are and No. and subvectors at each time point, is the kernel function, is the time alignment factor determined by the dynamic time warping algorithm, and The vectors are and The number of time points.

8. The method for identifying unsafe factors in the process of new energy infrastructure construction according to claim 2 is characterized by: The step S4 comprises: Step S41, analyzing other factors associated with the current unsafe factor, and mapping the other factors as variables to the Bayesian network nodes; Step S42: According to the causal relationship, time sequence and spatial position relationship between different physical quantities, the directed edges between nodes are determined, a factor association network is drawn, and the positions of nodes and the layout of edges are adjusted in combination with the actual logical relationship in the infrastructure construction process; Step S43: for each node, estimate its conditional probability table based on historical data and expert experience, and analyze the probability of occurrence of each factor under different circumstances; Step S44: sort the occurrence probability of each factor in different situations from large to small to explore potential unsafe factors.

9. The method for identifying unsafe factors in the process of new energy infrastructure construction according to claim 2 is characterized by: In step S5, the weights of different unsafe factors and their associated factors are determined by fuzzy hierarchy analysis, and the fuzzy judgment matrix elements The calculation is based on expert evaluation of factors and factors The fuzzy evaluation of relative importance is then performed to calculate the correlation between the real-time unsafe factors and the standard risk model through grey correlation analysis. The grey correlation calculation formula is: ， in, For reference sequence Compare with sequence The grey correlation degree of and are the reference sequence and the ith comparison sequence in the The value under the indicator, is the resolution coefficient.

10. A system for identifying unsafe factors in the process of new energy infrastructure construction, characterized by: It includes a multi-source heterogeneous data aggregation and fusion unit, an adaptive data cleaning and feature extraction unit, a dynamic feature library construction and maintenance unit, a similarity matching and preliminary judgment unit, a multi-factor association mining and analysis unit, and a comprehensive evaluation and visualization presentation unit; the multi-source heterogeneous data aggregation and fusion unit is used to receive and integrate physical parameters and image multi-source data from different sensors; the adaptive data cleaning and feature extraction unit is used to clean the aggregated data and extract features based on specific algorithms; the dynamic feature library construction and maintenance unit is responsible for constructing and dynamically updating the unsafe factor feature library based on historical cases; the similarity matching and preliminary judgment unit makes a preliminary judgment on the unsafe factors through the similarity algorithm; the multi-factor association mining and analysis unit uses the constructed factor association network model to analyze the associated factors of the unsafe factors; the comprehensive evaluation and visualization presentation unit uses a comprehensive evaluation algorithm to evaluate the severity of the unsafe factors.

Images