Field operation video monitoring risk prevention and control method, system, equipment and medium

Abstract

The invention discloses a field operation video monitoring risk prevention and control method, system and device and a medium, and the method comprises the steps: analyzing the reference operation risks of the whole network over the years, and constructing a reference operation risk library according to the analysis result; establishing a first risk model through the reference operation risk library, and identifying risk influence factors; constructing a second risk model through the risk influence factors, determining weight values of the risk influence factors, outputting risk values, and performing risk grade division according to the risk values; deploying video monitoring equipment to identify operating personnel, synchronously monitoring an operation site, and analyzing and judging an abnormal condition through a pixel difference value of front and back frame images; and when the abnormal situation is judged, determining an early warning level in combination with the risk level, and giving an alarm. According to the method, historical risk data analysis and an intelligent video monitoring technology are deeply fused, so that risk assessment is fundamentally converted from static experience judgment to dynamic quantitative prediction.

Description

Technical Field

[0001] This invention relates to the field of video surveillance technology, and in particular to a method, system, equipment and medium for risk prevention and control of on-site operation video surveillance. Background Technology

[0002] Our organization conducted a research project titled "Innovation and Application of Safety Supervision and Management Throughout the Entire Process Based on 'Internet+'". This research aimed to standardize mobile safety operations, focusing on the entire process of operation planning, implementation, supervision, and management. Through mobile terminals and mobile apps, it achieved real-time information sharing, real-time reporting, and timely monitoring of on-site surveys. It facilitated the location of construction sites, identification of workers, on-site personnel status at all levels, problem identification during inspections, and statistical analysis, making operation management more standardized and transparent. This research fully utilized "Internet+" and big data technologies to explore the application of mobile internet-based operation list management and risk control apps, improving the efficiency of on-site operation risk control through both management and technical means.

[0003] Most existing video surveillance systems only have basic recording and playback functions, lacking intelligent analysis capabilities and unable to automatically identify operational risks and issue early warnings of abnormal behavior. Furthermore, traditional risk assessment methods are largely based on static standards and subjective experience, making it difficult to achieve dynamic quantitative assessments. Moreover, the assessment results lack effective correlation with actual monitoring, failing to provide accurate risk prevention and control guidance for on-site operations. Summary of the Invention

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

[0005] Therefore, this invention provides a method, system, equipment, and medium for risk prevention and control of on-site operation video surveillance to solve the problem that existing video surveillance systems lack intelligent analysis capabilities, cannot automatically identify operational risks and issue early warnings of abnormal behaviors, and are difficult to achieve dynamic quantitative assessment.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a method for risk prevention and control of on-site operation video surveillance, comprising: analyzing the historical benchmark operation risks of the entire network and constructing a benchmark operation risk database based on the analysis results; establishing a first risk model through the benchmark operation risk database to identify risk influencing factors; constructing a second risk model through the risk influencing factors, determining the weight values ​​of the risk influencing factors, outputting risk values, and classifying risk levels based on the risk values; deploying video surveillance equipment to identify operators, synchronously monitoring the operation site, and judging abnormal situations through pixel difference analysis of consecutive frames; when an abnormal situation is judged, determining the warning level in conjunction with the risk level and issuing an alarm.

[0008] As a preferred embodiment of the on-site operation video surveillance risk prevention and control method of the present invention, the analysis of historical benchmark operation risks across the entire network includes: collecting historical benchmark operation risk data across the entire network; classifying and organizing the collected benchmark operation risk data; identifying the risk characteristics of various types of operations; analyzing the frequency and impact of various risks based on the risk characteristics of various operations; and determining the distribution pattern of benchmark operation risks based on the analysis results. As a preferred embodiment of the on-site operation video surveillance risk prevention and control method of the present invention, the step of constructing the benchmark operation risk database includes: establishing a risk classification system based on the distribution pattern of the benchmark operation risks; encoding and storing various types of operations according to risk type and severity through the risk classification system; establishing a risk data index relationship; and forming the benchmark operation risk database.

[0009] The beneficial effects of this preferred technical solution are as follows: by analyzing the risk data of benchmark operations across the entire network over the years and constructing a unified risk database, standardized management and retrieval of risk data are achieved, providing a reliable data foundation for subsequent risk assessment and avoiding the problems of scattered risk data and inconsistent standards in traditional methods.

[0010] As a preferred embodiment of the on-site operation video monitoring risk prevention and control method of the present invention, the step of identifying the risk influencing factors includes: constructing a first risk model based on the risk data in the benchmark operation risk library; extracting risk influencing factors from the set factors through the first risk model; running the first risk model to quantify the risk influencing factors of each dimension, and outputting the risk influencing factors and risk events.

[0011] The beneficial effects of this preferred technical solution are as follows: by constructing a specialized first risk model to identify risk influencing factors, the automatic extraction and quantification of risk factors are realized, which improves the accuracy and efficiency of risk identification and overcomes the defects of strong subjectivity and high omission rate of manual identification.

[0012] As a preferred embodiment of the on-site operation video monitoring risk prevention and control method described in this invention, determining the weight values ​​of risk influencing factors includes: calculating the historical occurrence frequency of each risk influencing factor, analyzing the contribution of each risk influencing factor to the risk event; using mathematical algorithms to determine the weight coefficient of each risk influencing factor, and outputting the weight values ​​of risk influencing factors in each dimension.

[0013] The beneficial effects of this preferred technical solution are as follows: by using mathematical algorithms to determine the weight values ​​of risk influencing factors, the objectivity and scientific nature of weight allocation are achieved, avoiding the subjectivity and arbitrariness of weight setting in traditional methods, and improving the credibility of risk assessment results.

[0014] As a preferred embodiment of the on-site operation video monitoring risk prevention and control method described in this invention, the step of classifying risk levels according to risk values ​​includes: setting risk value level classification intervals; classifying risk values ​​into high-risk, medium-risk, and low-risk levels according to the level classification intervals; and formulating corresponding monitoring strategies according to different risk levels.

[0015] The beneficial effects of this preferred technical solution are: establishing a risk level classification mechanism and formulating corresponding monitoring strategies, realizing differentiated monitoring management, which not only improves the monitoring intensity of high-risk operations, but also avoids excessive monitoring of low-risk operations.

[0016] As a preferred embodiment of the on-site operation video monitoring risk prevention and control method of the present invention, the step of analyzing and judging abnormal situations includes: acquiring continuous video frame images; calculating the pixel difference between adjacent frame images; setting an abnormal judgment threshold for the pixel difference, and when the pixel difference exceeds the abnormal judgment threshold, it is judged as an abnormal situation.

[0017] Secondly, the present invention provides a field operation video monitoring risk prevention and control system, comprising:

[0018] The data acquisition module is used to collect historical benchmark operation risk data from the entire network and acquire continuous video frame images;

[0019] The first risk model module is used to establish a first risk model based on the benchmark operation risk library and identify risk influencing factors.

[0020] The second risk model module is used to construct a second risk model, determine the weight values ​​of risk influencing factors, and output the risk value.

[0021] The anomaly detection module is used to analyze the pixel differences between consecutive frames to determine abnormal situations.

[0022] The early warning and alarm module is used to determine the early warning level and issue an alarm when an abnormal situation is identified, based on the risk level.

[0023] Thirdly, the present invention provides an electronic device, comprising:

[0024] Memory and processor;

[0025] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the on-site operation video monitoring risk prevention and control method are implemented.

[0026] Fourthly, the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of the on-site operation video monitoring risk prevention and control method.

[0027] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0028] By deeply integrating historical risk data analysis with intelligent video surveillance technology, a fundamental shift has been achieved in risk assessment, moving from static, experience-based judgment to dynamic, quantitative prediction. By constructing a benchmark operational risk database and establishing a dual-risk model system—the first model specifically responsible for identifying risk influencing factors, and the second model specifically responsible for weight calculation and risk value output—this collaborative approach not only improves the accuracy of risk identification but, more importantly, achieves precise matching between risk assessment results and on-site monitoring strategies. When the system detects anomalies at high-risk work sites, it can immediately activate the highest-level early warning response, while minor anomalies at low-risk levels are handled using standard procedures. This avoids the resource waste and inefficiency caused by the traditional "one-size-fits-all" monitoring model.

[0029] By calculating the pixel difference between consecutive frames, subtle changes at the work site can be keenly detected. Combined with pre-assessed risk level information, it is possible to determine whether such changes constitute a real safety threat. For example, the increased pixel difference caused by rapid personnel movement may indicate an emergency evacuation or accident at a high-risk work site, while it may simply be a normal work adjustment at a low-risk work site. Differentiated warning levels can be provided based on different risk levels, avoiding the frequent false alarms or missed alarms of traditional monitoring systems. Attached Figure Description

[0030] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0031] Figure 1 This is a schematic diagram of the overall process of the on-site operation video monitoring risk prevention and control method according to an embodiment of the present invention. Detailed Implementation

[0032] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0033] Example 1, referring to Figure 1 As an embodiment of the present invention, a method for risk prevention and control of on-site operation video surveillance is provided, comprising S1 to S5:

[0034] S1. By analyzing the baseline operation risks of the entire network over the years, a baseline operation risk database is constructed based on the analysis results.

[0035] S2. Establish a first risk model using the benchmark operation risk library to identify risk influencing factors.

[0036] S3. Construct a second risk model through risk influencing factors, determine the weight values ​​of the risk influencing factors, output risk values, and classify risk levels based on the risk values.

[0037] S4. Deploy video surveillance equipment to identify workers, monitor the work site synchronously, and analyze the pixel difference between consecutive frames to determine abnormal situations.

[0038] S5. When an abnormal situation is identified, the warning level is determined based on the aforementioned risk level, and an alarm is triggered.

[0039] It should be noted that traditional on-site safety supervision mainly relies on manual inspections and subjective experience, resulting in low supervision efficiency, limited coverage, and frequent human oversights. Most existing video surveillance systems only have basic recording playback functions, lacking intelligent analysis capabilities and unable to automatically identify operational risks and issue warnings of abnormal behavior. Furthermore, traditional risk assessment methods are largely based on static standards and experience-based judgments, making dynamic quantitative assessment difficult, and the assessment results are disconnected from actual monitoring, failing to provide accurate and effective risk prevention guidance for on-site operations. On-site operating environments are complex and changeable, worker behavior is unpredictable, and safety risks are dynamic and hidden, requiring high real-time and accurate risk identification and warning capabilities; therefore, intelligent risk prevention and control for on-site operations is also crucial.

[0040] Therefore, to address the aforementioned issues of regulatory efficiency and risk prediction, a risk assessment model based on historical data is constructed through steps S1-S5. This model enables the automatic identification and weight quantification of risk influencing factors, resulting in a scientifically accurate risk level classification. It also enables intelligent analysis of video surveillance, identifying abnormal behavior of workers and dynamically monitoring the safety status of the work site. Furthermore, based on the correlation analysis between risk levels and abnormal situations, it provides precise early warnings and timely alarms for on-site operational safety risks.

[0041] Example 2, refer to Figure 1 As an embodiment of the present invention, a method for risk prevention and control of on-site operation video monitoring is provided based on the above embodiment.

[0042] In this application embodiment, step S1 involves analyzing the historical baseline operational risks of the entire network, including A1 to A2:

[0043] A1. Collect historical benchmark operation risk data from the entire network, classify and organize the collected benchmark operation risk data, and identify the risk characteristics of various types of operations;

[0044] Historical benchmark operation risk data was collected from across the entire network. This data was then categorized and organized to identify the risk characteristics of various operation types. Benchmark operation risk data for the past five years was collected from municipal and county-level power bureaus, including data on different types of operations such as transmission line maintenance, substation equipment maintenance, and distribution network upgrades. Specifically, the benchmark operation risk data included information such as operation type, location, time, personnel involved, equipment used, environmental conditions, and any accidents that occurred. The collected data was categorized by professional field into three main categories: transmission, substation, and distribution. Each category was further subdivided into subcategories such as maintenance, operation, and construction. Data mining techniques were used to identify the risk characteristics of various operations and extract key risk elements such as risks associated with working at heights, working on live lines, and overlapping operations. The data collection coverage was set at all operating units across the entire network, with a continuous five-year time span to ensure data completeness and representativeness.

[0045] A2. Analyze the frequency and impact of various risks based on the risk characteristics of different types of operations, and determine the distribution pattern of risks in the benchmark operation based on the analysis results;

[0046] Based on the risk characteristics of various operations, the frequency and impact of each risk are analyzed. Based on the analysis results, the distribution pattern of risks in the benchmark operation is determined. A risk statistical analysis model is established, and the formula for calculating the frequency of each risk is as follows:

[0047] F (Ri) =N i / N total ;

[0048] Among them, F (Ri) Let N be the frequency of occurrence of the i-th type of risk. i Let N be the number of occurrences of the i-th type of risk event. total This represents the total number of assignments.

[0049] The impact assessment adopts a multi-dimensional evaluation system, including three dimensions: the degree of casualties, the degree of equipment damage, and the degree of impact of the power outage. The formula for calculating the impact is as follows:

[0050] I (Ri) =α×Hp+β×Hd+γ×Hs;

[0051] Among them, I (Ri) To represent the impact level of the i-th type of risk, Hp, Hd, and Hs are the standardized scores for personnel casualties, equipment losses, and power outage impacts, respectively, with α, β, and γ being weighting coefficients. The risk distribution pattern is determined through probability distribution fitting, and mathematical models such as Weibull distribution, normal distribution, and Poisson distribution are used to describe the distribution characteristics of different types of risks, providing a theoretical basis for subsequent risk prediction.

[0052] In an optional implementation, step S1, which analyzes the baseline operational risks across the entire network over the years, can also employ machine learning algorithms for in-depth data mining and pattern recognition. Specifically, clustering algorithms are used for unsupervised learning of risk data to automatically identify hidden risk patterns and correlations; association rule mining algorithms are used to discover the correlations between different risk factors and construct a risk factor association network; time series analysis methods are introduced to identify the temporal regularity and seasonal characteristics of risk occurrence; principal component analysis dimensionality reduction techniques are used to extract key risk features, reducing data redundancy and improving analysis efficiency; and anomaly detection algorithms are used to identify abnormal risk events, providing early signals for risk warning.

[0053] In another optional implementation, step S1 can be enhanced by combining external factors and a dynamic update mechanism to analyze the baseline operational risks of the entire network over the years. External factors such as meteorological data, geographic information, and equipment status are incorporated into risk data collection to construct a multi-source data fusion risk analysis system. A sliding window mechanism is used to dynamically update the risk analysis results, and a fixed time interval is set to recalculate the risk distribution pattern to ensure the timeliness of the analysis results. A risk data quality assessment system is established, evaluating data quality through indicators such as data integrity, accuracy, and consistency, and cleaning and correcting low-quality data. An expert knowledge base is introduced, combining professional experience with data analysis, and expert rules are used to verify and correct the statistical analysis results, improving the accuracy and practicality of the risk analysis.

[0054] In this embodiment of the application, step S1, the step of constructing the benchmark operation risk library, includes B1 to B2:

[0055] B1. Establish a risk classification system based on the distribution pattern of the risks in the benchmark operation;

[0056] A risk classification system is established based on the distribution pattern of the baseline operation risks. Based on the aforementioned risk analysis results, a two-dimensional risk classification matrix is ​​constructed according to the frequency and impact of risks, dividing risks into four quadrants: high-frequency high-risk, high-frequency low-risk, low-frequency high-risk, and low-frequency low-risk. Specifically, the risk classification system adopts a three-level classification structure. The first-level classification is divided into three categories according to professional fields: transmission risk, substation risk, and distribution risk. The second-level classification is divided into four subcategories according to the nature of the operation: maintenance risk, operation risk, construction risk, and emergency risk. The third-level classification is divided into four subcategories according to the characteristics of the risk source: personnel risk, equipment risk, environmental risk, and management risk. Risk coding rules are established, using a hierarchical coding method with the coding format XXX-YYY-ZZZ, where XXX represents the first-level classification code, YYY represents the second-level classification code, and ZZZ represents the third-level classification code. Risk levels are divided into five levels according to the degree of impact: extremely serious (Level I), serious (Level II), relatively serious (Level III), moderate (Level IV), and minor (Level V). The level determination threshold is determined based on the aforementioned impact degree calculation results.

[0057] B2. Using the aforementioned risk classification system, encode and store various operations according to risk type and severity, establish risk data index relationships, and form a benchmark operation risk database;

[0058] The risk classification system encodes and stores various operations according to risk type and severity, establishing a risk data index relationship to form a benchmark operation risk database. The database storage structure is designed using a relational database, primarily comprising four core data tables: a basic risk information table, a risk classification table, a risk level table, and a risk association table. The basic risk information table stores basic risk attributes, including fields such as risk code, risk name, risk description, frequency of occurrence, impact level, and historical cases. The risk classification table stores classification system information, including fields such as classification code, classification name, parent classification, and classification level. The risk level table stores level classification standards, including fields such as level code, level name, judgment threshold, and response measures. A multi-dimensional indexing mechanism is established, including risk type index, risk level index, operation type index, and time index. The index structure uses a B+ tree algorithm to ensure query efficiency. The risk data storage format adopts standardized JSON format, with each risk record containing risk identifier, risk attributes, association relationships, and metadata.

[0059] In an optional implementation, the construction of the baseline job risk database in step S1 can also employ a distributed storage architecture and intelligent retrieval technology. Specifically, a distributed database cluster is used to store large-scale risk data, and data reliability and scalability are ensured through data sharding and replication mechanisms; an intelligent retrieval engine is established to support multiple retrieval methods such as fuzzy search, semantic search, and similarity matching; a version control mechanism is introduced to record the change history of the risk database and support data rollback and version comparison; caching technology is used to optimize the performance of high-frequency queries, and frequently used risk data is preloaded into memory; and a data synchronization mechanism is established to ensure the consistency and real-time performance of data on each node.

[0060] In another optional implementation, the construction of the baseline operation risk base in step S1 can also combine knowledge graph technology and a dynamic update mechanism. Knowledge graph technology is used to construct a risk relationship network, storing risk entities, attributes, and relationships in a graph structure to support complex relational queries and reasoning analysis. A dynamic update mechanism for the risk base is established, setting two modes: periodic updates and real-time updates. The periodic update cycle is quarterly, and the real-time update is triggered by the addition of a major risk event. A data quality monitoring system is introduced to ensure the quality of the risk base data through integrity checks, consistency verification, and accuracy assessments. An access control mechanism is established to control the scope of data access based on user roles and permission levels, ensuring the security of sensitive risk information. A risk base query service is provided via an API interface, supporting both RESTful and GraphQL interface protocols to meet the integration needs of different application systems.

[0061] In the embodiments of this application, step S2, the step of identifying the risk influencing factors, includes C1 to C3:

[0062] C1. Construct a first risk model based on the risk data in the benchmark operation risk library;

[0063] A first risk model is constructed based on risk data from the benchmark operation risk database. This model employs a Bayesian network algorithm, extracting risk sample data from the database, including information on historical risk events, risk occurrence conditions, and risk consequences, as training data. Specifically, the first risk model uses a three-layer network structure. The input layer contains raw data such as operational environment parameters, personnel status parameters, and equipment status parameters. The hidden layer performs feature extraction and pattern recognition through neural network nodes. The output layer generates the identification results of risk influencing factors. Supervised learning is used for model training, employing historical risk events as label data and optimizing network parameters through backpropagation.

[0064] C2. Extract risk impact factors from the defined factors using the first risk model;

[0065] The first risk model extracts risk influencing factors from the defined factors. These factors include four dimensions: personnel, equipment, environment, and management. Each dimension contains several specific influencing factors. Personnel factors include worker skill level, safety awareness, physical condition, and work experience. Equipment factors include equipment operating status, equipment reliability, tool integrity, and safety protection facilities. Environmental factors include meteorological conditions, geographical environment, work space, and traffic conditions. Management factors include safety system implementation, on-site supervision, emergency plans, and training. The first risk model automatically extracts key risk influencing factors from the defined factors using a feature selection algorithm. Statistical methods such as information gain, chi-square test, and mutual information are used to calculate the importance score of each influencing factor. (Xi) The calculation formula is:

[0066] Score (Xi) =IG (Xi) +χ 2 (Xi) +MI (Xi) ;

[0067] Among them, IG (Xi) χ is the information gain value. 2 (Xi) MI is the chi-square test value. (Xi) This represents the mutual information value. An importance threshold of 0.6 is set, and factors scoring above this threshold are selected as key risk factors.

[0068] C3. Run the first risk model to quantify the risk impact factors of each dimension and output the risk impact factors and risk events.

[0069] The first risk model is run to quantify the risk impact factors in each dimension, outputting risk impact factors and risk events. A quantitative assessment system for risk impact factors is established, using fuzzy mathematics theory to quantify qualitative impact factors. The quantification process includes four steps: membership function construction, fuzzification, fuzzy inference, and defuzzification. The membership function uses a combination of triangular and trapezoidal functions to convert the verbal descriptions of impact factors into numerical values ​​within the [0,1] interval. The quantification formula is:

[0070]

[0071] Among them, Q (Xi) w is the quantified value of the i-th impact factor. j The weight of the j-th membership function. Let be the membership degree value of the influencing factor Xi under the j-th membership function. Through the first risk model calculation, the quantitative results of risk influencing factors in each dimension and the corresponding risk event correlation matrix are output. The correlation matrix adopts a sparse matrix storage format with a matrix dimension of M×N, where M is the number of risk influencing factors and N is the number of risk event types.

[0072] In an optional implementation, the identification of risk influencing factors in step S2 can also employ deep learning and ensemble learning techniques for model optimization. Specifically, a convolutional neural network (CNN) is used to extract spatial features of risk data, and a recurrent neural network (RNN) is used to capture the temporal features of risk evolution; an attention mechanism is introduced to automatically identify key risk influencing factors, and the model's sensitivity to important factors is improved through self-attention weight allocation; an ensemble learning method is used to fuse the prediction results of multiple base models, including algorithms such as random forests, support vector machines, and gradient boosting trees, and the identification accuracy is improved through a voting mechanism or weighted averaging; an online learning mechanism for the model is established to continuously optimize model parameters based on newly added risk data, ensuring the model's adaptability and timeliness.

[0073] In another optional implementation, the identification of risk influencing factors in step S2 can be enhanced by combining knowledge graphs and causal reasoning techniques. A risk knowledge graph is constructed, representing risk influencing factors, risk events, risk consequences, and the relationships between them in a graph structure, supporting complex association queries and reasoning analysis. Causal reasoning algorithms are used to identify causal relationships between risk influencing factors, analyzing direct and indirect influences between factors using structural equation modeling (SEM) and directed acyclic graphs (DAGs). An uncertainty reasoning mechanism is introduced, employing evidence theory and probabilistic reasoning methods to handle the uncertainty and ambiguity of risk information. A multi-source data fusion mechanism is established, integrating historical risk data, real-time monitoring data, expert knowledge, and other information sources, improving the comprehensiveness and accuracy of risk identification through data fusion algorithms. A dynamic threshold adjustment mechanism is set, adjusting the screening thresholds of influencing factors according to changes in the risk environment to ensure the adaptability of risk identification results.

[0074] In this embodiment of the application, in step S3, the weight values ​​of the risk influencing factors include D1 to D2:

[0075] D1. Calculate the historical frequency of each risk factor and analyze the contribution of each risk factor to the risk event.

[0076] Calculate the historical frequency of each risk factor and analyze its contribution to risk events: Extract risk event records from the benchmark operation risk database for the past 5 years, and statistically analyze the number and frequency of occurrence of each risk factor in historical risk events. Specifically, the formula for calculating historical frequency is:

[0077] F (Ri) =C (Ri) / N;

[0078] Among them, F (Ri) Let C be the historical frequency of occurrence of the i-th risk factor. (Ri) Let N be the number of times this factor occurs in historical risk events, and N be the total number of historical risk events. The contribution of each risk factor to a risk event is calculated using conditional probability analysis. The formula for calculating the contribution is:

[0079] G (Ri) =P (Event|Ri) -P (Event) ;

[0080] Among them, G (Ri) For the contribution of the i-th risk factor, P (Event|Ri) P represents the probability of a risk event occurring given the presence of this factor. (Event) This represents the basic probability of occurrence of a risk event. The Pearson correlation coefficient is used to analyze the correlation between risk influencing factors and the severity of risk events. The correlation coefficient is calculated using the following formula:

[0081]

[0082] Where r(Ri,S) is the correlation coefficient between the i-th risk influencing factor and the severity of the risk event, and S is the severity score of the risk event. A contribution threshold of 0.15 and a correlation coefficient threshold of 0.3 are set to screen out key factors that significantly influence the occurrence of risk events.

[0083] D2. Use mathematical algorithms to determine the weight coefficients of each risk influencing factor and output the weight values ​​of each dimension of risk influencing factor;

[0084] The weight coefficients of each risk influencing factor are determined using mathematical algorithms, and the weight values ​​of each dimension of risk influencing factor are output. The comprehensive weight of the risk influencing factors is determined using the Analytic Hierarchy Process (AHP) combined with the entropy weight method. First, an AHP matrix is ​​constructed. The relative importance of each factor is determined through expert scoring. The matrix element aij represents the importance of factor i relative to factor j, quantified using a 1-9 scale. The maximum eigenvalue λmax and the corresponding eigenvector of the judgment matrix are calculated. After normalization of the eigenvector, the AHP weight wAHP is obtained. The entropy weight calculation steps include: first, calculating the information entropy of each factor. The entropy calculation formula is:

[0085] E(Ri)=-k∑[p ij ln(p ij )];

[0086] Where E(Ri) is the information entropy of the i-th risk influencing factor, pij Let k be the probability of the i-th factor in the j-th evaluation object, k = 1 / ln(m), where m is the number of evaluation objects. Then calculate the entropy weight:

[0087] gotropy(Ri)=(1-E(Ri)) / ∑(1-E(Ri));

[0088] The overall weights are calculated using the multiplicative composition method:

[0089] w(Ri)=(wAHP(Ri)×wentropy(Ri)) / ∑(wAHP(Ri)×wentropy(Ri));

[0090] The final output is the weight allocation for the four dimensions: personnel factor weight 0.35, equipment factor weight 0.28, environmental factor weight 0.22, management factor weight 0.15, and the sum of the internal factor weights of each dimension is 1.0.

[0091] In an optional implementation, the determination of risk influencing factor weights in step S3 can also employ machine learning algorithms for adaptive weight calculation. Specifically, a random forest algorithm is used to calculate feature importance scores, and the importance of each risk influencing factor is assessed using out-of-bag error and Gini impurity indices; principal component analysis (PCA) is used for dimensionality reduction to extract principal components and calculate the loading coefficients of each factor in the principal components as weight references; a genetic algorithm is introduced to optimize weight allocation, using risk prediction accuracy as the fitness function, and the optimal weight combination is found through selection, crossover, and mutation operations; a weight adjustment mechanism is established to automatically adjust factor weights based on new risk data and prediction performance feedback, with the adjustment cycle set to quarterly.

[0092] In another optional implementation, the determination of risk influencing factor weights in step S3 can also be achieved by combining multi-criteria decision-making and fuzzy evaluation techniques for comprehensive weight determination. The TOPSIS multi-criteria decision-making method is used to construct a decision matrix and calculate the distances of each risk influencing factor to the ideal and negative ideal solutions. Factor weights are determined by relative proximity. Interval numbers and fuzzy numbers are introduced to represent the uncertainty of weights, and triangular fuzzy numbers are used to describe the fuzziness and subjectivity of expert evaluations. A weight consistency verification mechanism is established, calculating the consistency ratio CR using the consistency index CI and the random consistency index RI. When CR < 0.1, the weight allocation is considered reasonable. A weight sensitivity analysis module is set up to assess the impact of weight changes on risk assessment results through perturbation analysis and identify key sensitive factors. A multi-expert group decision-making mechanism is established, using the Delphi method to collect weight evaluation opinions from multiple experts, and integrating the judgment results of different experts through a weight aggregation algorithm to improve the objectivity and credibility of weight determination.

[0093] In this embodiment of the application, step S3, the step of classifying the risk level according to the risk value, includes E1 to E3:

[0094] E1. Define the risk value level range;

[0095] A risk value grading range is established: Based on historical risk event statistical analysis and industry safety standards, a quantitative risk value scoring system is established, with risk values ​​ranging from 0 to 100 points. Specifically, by analyzing the severity and probability of occurrence of historical risk events, the quantile method is used to determine the grading thresholds. Data from 10,000 risk events over the past 5 years are collected, and the distribution characteristics of the risk values ​​are calculated: the 25th percentile is 20 points, the 50th percentile is 45 points, the 75th percentile is 70 points, and the 90th percentile is 85 points. Combining the power industry's safety management requirements and expert evaluation standards, the risk level grading ranges are set as follows: low risk level 0-30 points, medium risk level 30-70 points, and high risk level 70-100 points.

[0096] E2. Divide the risk values ​​into high-risk, medium-risk, and low-risk levels according to the risk level classification.

[0097] The risk values ​​are categorized into high-risk, medium-risk, and low-risk levels based on their risk level ranges. An automated risk level determination system is established to automatically assign risk level labels based on the calculated risk values. Specifically, the low-risk level (green level) corresponds to a risk value of 0-30 points, including routine maintenance, low-voltage equipment repair, and indoor work—types of work with relatively low risk. The medium-risk level (yellow level) corresponds to a risk value of 30-70 points, including high-voltage equipment repair, outdoor line work, and cross-operations—types of work with a certain degree of danger. The high-risk level (red level) corresponds to a risk value of 70-100 points, including ultra-high-voltage equipment repair, live-line work, and high-altitude work—types of work with high risk.

[0098] A risk level classification system is established, with low-risk levels designated as L, medium-risk levels as M, and high-risk levels as H. The system automatically generates risk level reports, including level distribution statistics, level change trends, and abnormal level warnings, with reports generated daily.

[0099] E3. Develop corresponding monitoring strategies based on different risk levels;

[0100] Develop corresponding monitoring strategies based on different risk levels: Establish a differentiated monitoring strategy system, defining corresponding monitoring frequencies, methods, and response mechanisms for different risk levels. Low-risk levels employ a conventional monitoring strategy, with a monitoring frequency of once every 4 hours, primarily using automated monitoring supplemented by manual spot checks, configuring one video surveillance device, and a monitoring personnel ratio of 1:20. Medium-risk levels adopt an enhanced monitoring strategy, with a monitoring frequency of once every 2 hours, primarily using a combination of human and machine monitoring, configuring 2-3 video surveillance devices, adding mobile monitoring equipment, a monitoring personnel ratio of 1:10, and assigning dedicated safety officers for on-site supervision. High-risk levels adopt a focused monitoring strategy, with a continuous real-time monitoring frequency, primarily using manual focused monitoring, configuring four or more video surveillance devices, deploying comprehensive monitoring equipment, a monitoring personnel ratio of 1:5, and assigning professional supervisors for full-time supervision. Establish an early warning response mechanism: 30 minutes for low-risk levels, 15 minutes for medium-risk levels, and 5 minutes for high-risk levels. Develop emergency response procedures, including the issuance of early warning signals, on-site handling measures, personnel evacuation plans, and emergency resource allocation, with emergency response times of 60 minutes, 30 minutes, and 10 minutes, respectively.

[0101] In an optional implementation, the risk level classification in step S3 can also employ dynamic level adjustment and intelligent prediction technologies. Specifically, a dynamic risk level adjustment mechanism is established to adjust the risk level in real time based on factors such as work progress, changes in environmental conditions, and changes in personnel status; machine learning algorithms are used to predict the trend of risk level changes, and time series analysis and regression analysis methods are used to identify the risk level escalation risk in advance; a multi-factor comprehensive evaluation model is introduced to correct the risk level by combining external factors such as weather conditions, equipment status, and personnel capabilities; an automatic trigger mechanism for level escalation is established to automatically raise the risk level and activate corresponding monitoring strategies when abnormal situations are detected; and a level downgrade assessment process is set up to determine whether to reduce the risk level through the evaluation of the effectiveness of risk mitigation measures.

[0102] In another optional implementation, the risk level classification in step S3 can also be combined with geographic information systems (GIS) and visualization technology for spatial management. GIS technology is used to construct a spatial distribution map of risk levels, linking risk level information with the geographic coordinates of the work site to achieve spatial visualization of risks; a risk level heat map is established, using color depth to represent the risk level distribution density in different areas; 3D visualization technology is introduced to construct a 3D model of the work site and overlay risk level information, supporting multi-angle observation of the risk situation; an animation display function for risk level changes is established, using a timeline to control and display the historical changes in risk levels; a risk level statistical analysis module is set up to generate various statistical charts such as level distribution pie charts, level change trend charts, and level correlation analysis charts, providing data support for risk management decisions; a risk level early warning map is established, displaying the risk level status and early warning information of each work site in real time on an electronic map, supporting overall control of the risk situation and regional coordinated management.

[0103] In this embodiment of the application, step S4, the step of analyzing and judging abnormal situations, includes F1 to F3:

[0104] F1: Acquire consecutive video frame images;

[0105] Acquiring continuous video frame images: Video stream data of the operation process is acquired in real time through video surveillance equipment deployed at the work site. Specifically, a video data acquisition system is established, comprising three components: front-end cameras, a video transmission network, and a back-end storage server. The front-end cameras are high-definition network cameras, supporting autofocus, infrared night vision, and waterproof / dustproof functions. The installation height is set at 3-5 meters, and the monitoring angle covers more than 90% of the work area. Video transmission uses a combination of wired and wireless networks. The wired network bandwidth is no less than 100Mbps, and the wireless network uses 4G / 5G mobile communication technology to ensure real-time video data transmission. The back-end storage server adopts a distributed storage architecture with a storage capacity of no less than 100TB, supporting 24 / 7 continuous recording, and the recording retention period is 30 days. A video frame extraction algorithm is established to extract static image frames from the video stream at fixed time intervals. The extraction interval is set to 40 milliseconds (corresponding to 25fps), extracting 25 frames per second. Image preprocessing includes denoising, enhancement, and format conversion steps, with the output format being RGB three-channel 24-bit images.

[0106] F2. Calculate the pixel difference between adjacent frames;

[0107] Calculate the pixel difference between adjacent frames: An inter-frame difference algorithm is used to calculate the pixel change between consecutive video frames. Pixel difference analysis is then used to detect moving targets and abnormal changes at the work site. Specifically, the inter-frame difference calculation formula is:

[0108] D(x,y)=|I(t+1)(x,y)-I(t)(x,y)|;

[0109] Where D(x,y) is the pixel difference at coordinates (x,y), I(t)(x,y) is the pixel value at coordinates (x,y) of the t-th frame, and I(t+1)(x,y) is the pixel value at the same coordinates of the (t+1)-th frame. To improve the difference effect, a three-frame difference method is used, and the calculation formula is as follows:

[0110] D3(x,y)=|I(t+1)(x,y)-I(t)(x,y)|∩|I(t+1)(x,y)-I(t-1)(x,y)|;

[0111] Noise interference is reduced through logical AND operations to improve the accuracy of moving target detection. Morphological processing, including erosion, dilation, opening, and closing operations, is performed on the calculated difference image to remove small noise points and fill holes inside the target. A difference statistical analysis mechanism is established to calculate statistical indicators such as the total difference, average difference, and maximum difference of the entire image. The formula for calculating the total difference (TotalDiff) is:

[0112] TotalDiff = ∑∑D(x,y);

[0113] Processing speed optimization employs multi-threaded parallel computing and GPU acceleration technology, keeping the single-frame processing time within 10 milliseconds.

[0114] F3. Set an abnormal judgment threshold for pixel difference. When the pixel difference exceeds the abnormal judgment threshold, it is judged as an abnormal situation.

[0115] An anomaly detection threshold for pixel differences is set. When the pixel difference exceeds this threshold, it is considered an abnormal situation. Based on historical video data analysis and on-site measurement results, a method for determining the anomaly detection threshold is established. By analyzing 1000 hours of normal operation video and 500 videos of abnormal events, the distribution characteristics of pixel differences are statistically analyzed. Under normal operation conditions, the average difference is 15-25 pixels, while under abnormal conditions, the average difference is 60-120 pixels. The optimal threshold is determined using ROC curve analysis. With detection accuracy and false alarm rate as evaluation indicators, the anomaly detection threshold is determined to be 40 pixels after optimization calculation. A multi-level threshold judgment mechanism is established, setting three threshold levels: a minor anomaly threshold of 40 pixels, a moderate anomaly threshold of 80 pixels, and a severe anomaly threshold of 120 pixels.

[0116] An adaptive threshold adjustment mechanism is established to dynamically adjust the judgment threshold based on factors such as ambient lighting conditions, weather conditions, and job type, with an adjustment coefficient ranging from 0.8 to 1.2. An anomaly duration judgment condition is set, triggering an anomaly alarm only when an anomaly is detected for more than three consecutive frames, thus avoiding occasional false alarms.

[0117] In an optional implementation, the abnormal situation judgment in step S4 can also employ deep learning and object detection technologies for intelligent identification. Specifically, a convolutional neural network (CNN) is used to construct an abnormal behavior recognition model, which is trained with a large amount of labeled data to identify various abnormal types such as personnel violations, abnormal equipment states, and hazardous environmental conditions; the YOLO object detection algorithm is introduced to identify target objects such as workers, vehicles, and equipment, and abnormal behavior patterns are judged through target trajectory analysis; optical flow motion analysis is established to calculate the motion vector field of pixels in the image, and abnormal motion is detected by changes in motion amplitude and direction; a dynamic background model is established using background modeling technology to automatically adapt to changes in lighting and weather conditions, improving the robustness of anomaly detection; and a time series analysis method is introduced to predict potential abnormal situations by analyzing the time variation trend of pixel differences.

[0118] In another optional implementation, the anomaly judgment in step S4 can also be enhanced by combining multimodal data fusion and edge computing technologies. Multi-sensor data fusion technology is employed to integrate multiple data sources such as video images, audio signals, vibration sensors, and temperature sensors, improving the accuracy and reliability of anomaly detection through data fusion algorithms. Edge computing nodes are established, deploying lightweight anomaly detection algorithms on the monitoring equipment to achieve local real-time processing and rapid response, reducing network transmission latency. An attention mechanism is introduced to automatically identify key areas and important features in images, improving the accuracy and efficiency of anomaly detection. An anomaly behavior database is established to collect and label various anomaly behavior samples, continuously optimizing the anomaly detection algorithm. An anomaly verification mechanism is set up, using cross-verification from multi-angle cameras and manual review to confirm anomaly detection results, reducing the risk of false alarms and missed alarms. An automatic anomaly level escalation mechanism is established, automatically escalating the anomaly level and initiating corresponding emergency response procedures when persistent or aggravating anomalies are detected.

[0119] In summary, by deeply integrating historical risk data analysis with intelligent video surveillance technology, a fundamental shift has been achieved in risk assessment, moving from static, experience-based judgment to dynamic, quantitative prediction. By constructing a benchmark operational risk database and establishing a dual-risk model system—with the first model specifically responsible for identifying risk influencing factors and the second model specifically responsible for weight calculation and risk value output—this collaborative approach not only improves the accuracy of risk identification but, more importantly, achieves precise matching between risk assessment results and on-site monitoring strategies. When the system detects anomalies at high-risk work sites, it can immediately initiate the highest-level early warning response, while minor anomalies at low-risk levels are handled using standard procedures, avoiding the resource waste and inefficiency caused by the "one-size-fits-all" monitoring model of traditional methods.

[0120] By calculating the pixel difference between consecutive frames, subtle changes at the work site can be keenly detected. Combined with pre-assessed risk level information, it is possible to determine whether such changes constitute a real safety threat. For example, the increased pixel difference caused by rapid personnel movement may indicate an emergency evacuation or accident at a high-risk work site, while it may simply be a normal work adjustment at a low-risk work site. Differentiated warning levels can be provided based on different risk levels, avoiding the frequent false alarms or missed alarms of traditional monitoring systems.

[0121] Example 3 illustrates a method for risk prevention and control through on-site video surveillance. It should be noted that the technical solution of this on-site video surveillance risk prevention and control system is based on the same concept as the technical solution of the on-site video surveillance risk prevention and control method described above. Details not described in detail in this example can be found in the description of the technical solution of the on-site video surveillance risk prevention and control method described above.

[0122] This embodiment also provides a field operation video monitoring risk prevention and control system, including:

[0123] The data acquisition module is used to collect historical benchmark operation risk data from the entire network and acquire continuous video frame images;

[0124] The first risk model module is used to establish a first risk model based on the benchmark operation risk library and identify risk influencing factors.

[0125] The second risk model module is used to construct a second risk model, determine the weight values ​​of risk influencing factors, and output the risk value.

[0126] The anomaly detection module is used to analyze the pixel differences between consecutive frames to determine abnormal situations.

[0127] The early warning and alarm module is used to determine the early warning level and issue an alarm when an abnormal situation is identified, based on the risk level.

[0128] This embodiment also provides an electronic device suitable for on-site operation video surveillance risk prevention and control, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the on-site operation video surveillance risk prevention and control method proposed in the above embodiment.

[0129] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements the method for risk prevention and control of on-site operation video monitoring as proposed in the above embodiments.

[0130] The storage medium proposed in this embodiment and the method for risk prevention and control of on-site operation video monitoring proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0131] Based on the above description of the implementation methods, those skilled in the art will clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0132] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for risk prevention and control through on-site video surveillance, characterized in that, include: By analyzing the baseline operation risks across the entire network over the years, a baseline operation risk database is constructed based on the analysis results; A first risk model is established using the benchmark operation risk library to identify risk influencing factors; A second risk model is constructed by using risk influencing factors, the weight values ​​of the risk influencing factors are determined, the risk value is output, and the risk level is classified according to the risk value. Deploy video surveillance equipment to identify workers, monitor the work site simultaneously, and analyze the pixel differences between consecutive frames to determine abnormal situations. When an abnormal situation is identified, the warning level is determined based on the aforementioned risk level, and an alarm is triggered.

2. The on-site operation video monitoring risk prevention and control method as described in claim 1, characterized in that, The analysis of historical benchmark operational risks across the entire network includes: Collect historical benchmark operation risk data from across the entire network, classify and organize the collected benchmark operation risk data, and identify the risk characteristics of various types of operations; Based on the risk characteristics of various operations, analyze the frequency and impact of various risks, and determine the distribution pattern of baseline operation risks based on the analysis results.

3. The on-site operation video monitoring risk prevention and control method as described in claim 2, characterized in that, The steps for constructing the benchmark job risk library include: A risk classification system is established based on the distribution pattern of risks in the benchmark operations. The risk classification system is used to encode and store various operations according to risk type and severity, establish risk data index relationships, and form a benchmark operation risk library.

4. The on-site operation video monitoring risk prevention and control method as described in claim 3, characterized in that, The steps for identifying the risk factors include: A first risk model is constructed based on the risk data in the benchmark operation risk library; The risk impact factors are extracted from the defined factors using the first risk model; The first risk model is run to quantify the risk impact factors of each dimension, and output the risk impact factors and risk events.

5. The on-site operation video monitoring risk prevention and control method as described in claim 4, characterized in that, The weights for determining the risk influencing factors include: Calculate the historical frequency of each risk factor and analyze the contribution of each risk factor to the risk event; Mathematical algorithms are used to determine the weight coefficients of each risk factor, and the weight values ​​of each risk factor in each dimension are output.

6. The on-site operation video surveillance risk prevention and control method as described in claim 5, characterized in that, The steps for classifying risk levels based on risk values ​​include: Define risk value grading ranges; The risk values ​​are divided into high-risk, medium-risk, and low-risk levels according to the risk level classification. Develop corresponding monitoring strategies based on different risk levels.

7. The on-site operation video monitoring risk prevention and control method as described in claim 6, characterized in that, The steps for analyzing and judging abnormal situations include: Acquire consecutive video frame images; Calculate the pixel difference between adjacent frames; An abnormal judgment threshold is set for pixel difference. When the pixel difference exceeds the abnormal judgment threshold, it is judged as an abnormal situation.

8. A field operation video monitoring risk prevention and control system, using the method described in any one of claims 1-7, characterized in that, include: The data acquisition module is used to collect historical benchmark operation risk data from the entire network and acquire continuous video frame images; The first risk model module is used to establish a first risk model based on the benchmark operation risk library and identify risk influencing factors. The second risk model module is used to construct a second risk model, determine the weight values ​​of risk influencing factors, and output the risk value. The anomaly detection module is used to analyze the pixel differences between consecutive frames to determine abnormal situations. The early warning and alarm module is used to determine the early warning level and issue an alarm when an abnormal situation is identified, based on the risk level.

9. An electronic device, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, they implement the steps of the on-site operation video monitoring risk prevention and control method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the on-site operation video monitoring risk prevention and control method according to any one of claims 1 to 7.

Images