CN117035456A - Intelligent building site monitoring and management method and system - Google Patents

Abstract

The invention provides an intelligent building site monitoring and managing method and system, which relate to the technical field of building site management and comprise the following steps: the method comprises the steps of data acquisition, data processing, data anomaly detection, processing suggestion recommendation, monitoring result generation log, monitoring result transmission and the like, wherein anomaly detection of multiple iterations is carried out on monitoring data through combination of reinforcement learning and a knowledge base, anomaly data and anomaly degree are obtained, and processing suggestions are obtained according to a set recommendation flow. The system of the invention ensures the safety of the construction site to a great extent, utilizes the characteristic of reinforcement learning, is continuously optimized in the operation process, and monitors the construction site more and more comprehensively and accurately.

Description

Intelligent building site monitoring and management method and system

Technical Field

The invention relates to the technical field of building site management, in particular to an intelligent building site monitoring and management method and system.

Background

At present, in the construction and construction process of various construction sites, the related management modules are quite complicated, such as project management, safety management, equipment management and the like, the cost of manual management is too high, although some management systems appear, the performance of the management systems cannot meet the requirement of the comprehensive real-time management of the construction sites, the management efficiency is low, the system is easy to fail, and the cost is possibly increased additionally.

The invention patent with the Chinese application number of 202210138195.0 discloses a construction site monitoring method and a construction site monitoring system based on edge calculation, wherein edge equipment and edge calculation nodes are arranged in different areas of a construction site, construction site data of corresponding areas are collected according to the edge equipment, the edge calculation nodes process the construction site data and send processing results to a server, and the server generates control instructions according to the processing results and sends the control instructions to alarm modules in the corresponding construction site, so that the construction site is monitored; by arranging the edge equipment and the edge computing nodes in different areas of the construction site, the construction site data acquired by the edge equipment can be processed through the edge computing nodes, and the server only needs to generate a control instruction according to the processing result of the edge computing nodes, so that the operation of the server is greatly reduced, and the processing efficiency of the construction site data is greatly improved. However, although the technology improves the processing efficiency, the accuracy and the integrity of monitoring are correspondingly reduced, and the technology has a good effect on small-sized construction sites, but cannot meet the management requirement of the construction sites for medium and large-sized construction sites.

Disclosure of Invention

In view of the above, the invention provides an intelligent construction site monitoring and managing method and system, which monitor the construction site in real time by introducing technologies such as reinforcement learning, knowledge graph and the like, so that the safety of the construction site is greatly ensured, and the system is continuously optimized in the operation process by utilizing the characteristic of reinforcement learning, so that the monitoring of the construction site is more and more comprehensive and accurate.

The technical scheme of the invention is realized as follows:

in one aspect, the invention provides a method for intelligent building site monitoring and management, comprising the following steps:

s1, acquiring monitoring data of each management sub-module in a preset time period, and processing the monitoring data to obtain data to be analyzed;

s2, constructing a reinforcement learning module, wherein the reinforcement learning module comprises a reinforcement learning model and a knowledge base;

s3, acquiring environment data, inputting the data to be analyzed into a reinforcement learning model, generating an analysis strategy according to the environment data, analyzing the data to be analyzed by utilizing the analysis strategy, and confirming and implementing actions based on analysis results to obtain abnormal data and the degree of abnormality thereof;

s4, inputting the abnormal data and the degree of abnormality thereof into a knowledge base for inquiring to obtain feedback, determining the weight of the abnormal data according to a weight setting rule, and constructing a reward function based on the weight of the abnormal data and the feedback;

S5, acquiring a current rewarding value according to the rewarding function, and updating an analysis strategy according to the current rewarding value so as to obtain updated abnormal data and the abnormal degree thereof;

s6, feeding back updated abnormal data and the abnormal degree thereof by utilizing a knowledge base, and updating the rewarding function according to new feedback;

s7, constructing a return function based on the rewarding value, and repeatedly executing the steps S5-S6 until the return function reaches the maximum value, and outputting final abnormal data and the degree of abnormality thereof;

s8, inputting the final abnormal data and the degree of abnormality thereof into a recommendation module, obtaining a processing suggestion according to a recommendation algorithm, and taking the final abnormal data and the degree of abnormality thereof and the processing suggestion as monitoring results;

s9, generating a log of the monitoring result, transmitting the monitoring log to a display module, and simultaneously transmitting the monitoring log to an administrator of each management sub-module according to membership of each management sub-module corresponding to the monitoring log for timely processing by the administrator.

Further preferably, step S1 includes:

s11, dividing the monitoring data into table data, picture data and video data according to a data format;

s12, processing the form data by using a text processing method to obtain keywords to be analyzed;

S13, extracting frames of the video data, converting the video data into picture data by taking the frames as units, extracting target features of the picture data by utilizing a target detection model, and converting the target features into gray scales to obtain gray scale features to be analyzed;

s14, the keywords to be analyzed and the gray features to be analyzed form data to be analyzed.

Further preferably, in step S2, the knowledge base construction process is as follows:

constructing a domain knowledge graph and a Bayesian network, pre-training, vectorizing the pre-trained domain knowledge graph and Bayesian network, inputting the pre-trained domain knowledge graph and Bayesian network into a graph neural network, transmitting information in the Bayesian network to the domain knowledge graph by adopting a message transmission mechanism, and fusing the Bayesian network and the domain knowledge graph to obtain a knowledge base;

wherein the messaging mechanism includes a message function having the expression:

wherein l and l-1 refer to the present transfer and the previous transfer, M _l-1 For message functions, Σ represents aggregation, U _l-1 Representing an update function, v representing a target node, a representing a neighbor node of v, N (v) representing a set of neighbor nodes of v,message representing the current transfer->And->Hidden states of the target node and the neighboring node, e _va Representing a unit vector.

Further preferably, step S3 includes:

s31, the environment data comprise historical data of each management sub-module, and the historical data comprise historical keywords and historical gray scale features;

s32, generating an analysis strategy according to the environmental data, and respectively analyzing keywords to be analyzed and gray features to be analyzed in the data to be analyzed by utilizing the analysis strategy to obtain an analysis result;

s33, confirming and implementing actions based on the analysis result, wherein the actions comprise carrying out abnormality judgment on the keywords to be analyzed and carrying out abnormality judgment on the gray scale characteristics to be analyzed, so as to obtain abnormal data and the abnormality degree thereof.

Further preferably, step S32 includes:

coding the historical keywords and the historical gray scale characteristics to obtain a coding value;

constructing an index table according to the historical keywords and the coding values thereof, the historical gray features and the coding values thereof, wherein the index of the index table is the coding value;

respectively encoding the keywords to be analyzed and the gray features to be analyzed to obtain encoded values of the keywords to be analyzed and the gray features to be analyzed;

searching in an index table by utilizing the key words to be analyzed and the coding values of the gray features to be analyzed, and matching to obtain the association history key words related to the key words to be analyzed and the association history gray features related to the gray features to be analyzed as analysis results;

Accordingly, step S33 includes:

judging whether the keywords to be analyzed are abnormal or not according to the correlation degree of the correlation history keywords and the keywords to be analyzed, classifying the keywords to be analyzed into abnormal data if the keywords to be analyzed are abnormal, carrying out semantic description on the keywords to be analyzed according to a semantic model after judging that the keywords to be analyzed are abnormal data, and judging the abnormal degree of the keywords to be analyzed based on the content of the semantic description;

judging whether the gray scale feature to be analyzed is abnormal or not according to the correlation degree of the correlation history gray scale feature and the gray scale feature to be analyzed, classifying the gray scale feature to be analyzed into abnormal data if the gray scale feature to be analyzed is abnormal, and calculating the abnormal degree of the gray scale feature to be analyzed according to the information entropy after judging that the gray scale feature to be analyzed is the abnormal data.

Further preferably, step S4 includes:

s41, sending the abnormal data and the abnormal degree thereof to a knowledge base, and evaluating the abnormal data and the abnormal degree thereof by the knowledge base to obtain feedback;

s42, quantifying corresponding abnormal data according to the importance degree of each management sub-module to obtain a weight, and judging the influence degree of the abnormal data and the abnormality degree thereof based on the influence judgment rule according to feedback of the abnormal data and the abnormality degree thereof to obtain an influence value;

S43, constructing a reward function according to the weight and the influence value, wherein the formula is as follows:

f(x _i )＝α _i f ₁ (x _i )+β _i f ₂ (x _i )

wherein R is a prize value, n is the number of abnormal data, x _i Represents the ith anomaly data, ω _i Weight value representing ith abnormal data, f (x _i ) Feedback representing ith anomaly data, f ₁ (x _i ) Is the influence value of the ith abnormal data, f ₂ (x _i ) The influence value of the degree of abnormality corresponding to the ith abnormal data, and α and β are influence coefficients.

Further preferably, step S4 further includes:

the knowledge base predicts the current abnormal data, if the influence degree of the current abnormal data on the abnormal data subordinate to other management submodules exceeds a preset threshold, a deviation function is formed according to the prediction information to correct the reward function, and the corrected reward function is as follows:

wherein R' is the corrected prize value, x _j～i Representing abnormal data x _j For abnormal data x _i The degree of influence of (2) exceeds a preset threshold, W represents a deviation function, lambda represents a deviation coefficient, E represents abnormal data x _j For abnormal data x _i Is not limited to the above-described ones.

Further preferably, f ₁ (x _i ) = { -3, if x _i Is erroneous exception data; +2, if x _i Correct exception data };

f ₂ (x _i ) = { -2, if x _i Is greater than actual; -3, if x _i Is less abnormal than actual; +2, if x _i Correct degree of abnormality of (c).

Further preferably, step S8 includes:

s81, dividing the final abnormal data into keywords to be recommended and gray scale features to be recommended according to the keywords and the gray scale features;

s82, the recommendation module comprises a keyword processing suggestion library and a gray feature processing suggestion library, wherein the keyword processing suggestion library is constructed by taking keywords as indexes, the indexes of the keyword processing suggestion library are linked into keyword-processing suggestions, and the gray feature processing suggestion library is constructed by taking gray features as identifiers and binding the gray features with corresponding processing suggestions into groups;

s82, quantifying the degree of abnormality of the keywords to be recommended according to the content of semantic description of the keywords to be recommended to form abnormal values of the keywords to be recommended, searching in a recommendation module according to the keywords to be recommended to obtain a group of first candidate processing suggestions, sorting the first candidate processing suggestions according to the abnormal values of the keywords to be recommended, and selecting the first candidate processing suggestion with the highest matching degree as the processing suggestion of the keywords to be recommended;

s83, normalizing the abnormal degree of the gray feature to be recommended to obtain an abnormal value of the gray feature to be recommended, screening the gray feature with the feature distance within a preset range from the gray feature to be recommended in a recommendation module, obtaining a corresponding second candidate processing suggestion according to the screened gray feature, sorting the second candidate processing suggestion based on the abnormal value of the gray feature to be recommended, and selecting the second candidate processing suggestion with the highest matching degree as the processing suggestion of the gray feature to be recommended;

S84, the final abnormal data, the degree of abnormality and the processing advice are respectively formed into monitoring results of all the management sub-modules according to the attribution of the management sub-modules, and the monitoring results of all the management sub-modules are summarized to obtain the final monitoring results.

On the other hand, the invention also provides an intelligent building site monitoring and managing system, which comprises:

the management submodules are configured to monitor the respective task target areas in all weather according to preset instructions and store monitoring contents in corresponding formats in time units;

the data acquisition module is configured to simultaneously acquire monitoring data of each management sub-module according to a preset time period;

the data processing module is configured to respectively process the monitoring data according to a data format to form data to be analyzed, wherein the data to be analyzed comprises keywords to be analyzed and gray features to be analyzed;

an environmental data module configured to obtain environmental data, the environmental data including historical data of each management sub-module, the historical data including historical keywords and historical gray scale features;

the reinforcement learning module is configured to generate an analysis strategy according to the environmental data, analyze the data to be analyzed by utilizing the analysis strategy, confirm and implement actions based on analysis results to obtain abnormal data and abnormal degrees thereof, inquire and feed back the abnormal data and the abnormal degrees thereof according to a knowledge base, obtain a reward value according to feedback, perform multiple iterative updating on the analysis strategy by utilizing the reward value, perform multiple iterative updating on the abnormal data and the abnormal degrees thereof by utilizing the updated analysis strategy, construct a return function based on the reward value, and finish the iterative updating when the return function reaches the maximum value to obtain final abnormal data and the abnormal degrees thereof;

The recommendation module is configured to recommend the final abnormal data and the abnormal degree thereof according to a recommendation algorithm to obtain corresponding processing suggestions, and the final abnormal data, the abnormal degree thereof and the processing suggestions respectively form monitoring results of all the management sub-modules according to attribution of the management sub-modules, and the monitoring results of all the management sub-modules are summarized to obtain final monitoring results;

the log module is configured to receive the monitoring results and respectively generate monitoring logs for the monitoring results according to the attribution of the management sub-module;

the display module is configured to receive the monitoring log and display the monitoring log according to the visualization tool;

the data sending module is configured to receive the monitoring logs and send the monitoring logs to the manager of the corresponding management sub-module according to membership of the management sub-module of each monitoring log.

Compared with the prior art, the method has the following beneficial effects:

(1) The method introduces reinforcement learning in site monitoring, realizes that the system is more and more intelligent in interaction, continuously improves the monitoring capability of the site, and has substantial significance for the management and monitoring of intelligent sites;

(2) According to the invention, the knowledge base is constructed by fusing the Bayesian network and the knowledge map so as to take on the role of the other party of interaction, and the knowledge base combines the advantages of the Bayesian network, the knowledge map and the graph neural network, has excellent performance, can realize a better interaction process with the reinforcement learning model, and reduces the labor cost;

(3) According to the invention, through detecting the abnormality of the data, not only is the abnormal data identified, but also a judgment rule of the abnormality degree is provided, so that the data can be monitored more deeply and comprehensively;

(4) The invention combines the importance degree, the abnormal data and the influence rule of the abnormal degree of the management submodule to construct the reward function so as to enable the reinforcement learning model to be more adaptive to the monitoring of the construction site;

(5) According to the invention, the influence possibly associated with the occurrence of the abnormality before the management sub-module is considered, and the reward function is corrected by introducing a deviation function by utilizing the strong reasoning performance of the knowledge base, so that the performance of the reinforcement learning model is improved;

(6) The invention also provides a recommendation of the processing proposal, and provides a recommendation flow with specific settings so as to save the time of an administrator for processing the abnormal situation, and increase the function of the reinforcement learning model to enhance the generalization capability of the reinforcement learning model.

Drawings

In order to more clearly illustrate the embodiments of the invention or the technical solutions in the prior art, the drawings that are required in the embodiments or the description of the prior art will be briefly described, it being obvious that the drawings in the following description are only some embodiments of the invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.

FIG. 1 is a flow chart of a method according to an embodiment of the invention

Fig. 2 is a system configuration diagram of an embodiment of the present invention.

Detailed Description

The following description of the embodiments of the present invention will clearly and fully describe the technical aspects of the embodiments of the present invention, and it is apparent that the described embodiments are only some embodiments of the present invention, not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the present invention without making any inventive effort, are intended to fall within the scope of the present invention.

As shown in fig. 1, the present invention provides an intelligent building site monitoring and managing method, which includes:

s1, acquiring monitoring data of each management sub-module in a preset time period, and processing the monitoring data to obtain data to be analyzed;

s2, constructing a reinforcement learning module, wherein the reinforcement learning module comprises a reinforcement learning model and a knowledge base;

s3, acquiring environment data, inputting the data to be analyzed into a reinforcement learning model, generating an analysis strategy according to the environment data, analyzing the data to be analyzed by utilizing the analysis strategy, and confirming and implementing actions based on analysis results to obtain abnormal data and the degree of abnormality thereof;

S4, inputting the abnormal data and the degree of abnormality thereof into a knowledge base for inquiring to obtain feedback, determining the weight of the abnormal data according to a weight setting rule, and constructing a reward function based on the weight of the abnormal data and the feedback;

s5, acquiring a current rewarding value according to the rewarding function, and updating an analysis strategy according to the current rewarding value so as to obtain updated abnormal data and the abnormal degree thereof;

s6, feeding back updated abnormal data and the abnormal degree thereof by utilizing a knowledge base, and updating the rewarding function according to new feedback;

s7, constructing a return function based on the rewarding value, and repeatedly executing the steps S5-S6 until the return function reaches the maximum value, and outputting final abnormal data and the degree of abnormality thereof;

s8, inputting the final abnormal data and the degree of abnormality thereof into a recommendation module, obtaining a processing suggestion according to a recommendation algorithm, and taking the final abnormal data and the degree of abnormality thereof and the processing suggestion as monitoring results;

s9, generating a log of the monitoring result, transmitting the monitoring log to a display module, and simultaneously transmitting the monitoring log to an administrator of each management sub-module according to membership of each management sub-module corresponding to the monitoring log for timely processing by the administrator.

The technical implementation flow of the invention is as follows: 1) Corresponding management sub-modules are set according to different management contents to monitor task targets, and monitoring data of each management sub-module are obtained according to a preset time period; 2) Processing the monitoring data according to a data format to respectively obtain keywords to be analyzed and gray features to be analyzed; 3) According to the content and the result of the history monitoring, environmental data are formed, an analysis strategy is generated by utilizing a reinforcement learning module based on the environmental data, keywords to be analyzed and gray features to be analyzed are respectively analyzed, actions are confirmed and implemented based on the analysis result, then the analysis strategy is updated through feedback interaction between a knowledge base and the reinforcement learning model, and final abnormal data and abnormal degree are obtained after multiple iterations; 4) The recommendation module gives candidate processing suggestions according to the content of the abnormal data, then determines the most suitable processing suggestions according to the degree of abnormality, binds the abnormal data, the degree of abnormality and the processing suggestions into groups one by one, and respectively obtains the monitoring results of all the management sub-modules according to the management sub-modules to which the group of data belongs; 5) The monitoring results are formed into monitoring logs which are respectively stored according to membership of the management sub-module, the monitoring logs are sent to the display module so as to be conveniently referred and traced at any time, and meanwhile, the monitoring logs are sent to an administrator corresponding to the management sub-module so as to be conveniently and timely processed on abnormal conditions.

Specifically, in an embodiment of the present invention, step S1 includes:

s11, dividing the monitoring data into table data, picture data and video data according to a data format;

s12, processing the form data by using a text processing method to obtain keywords to be analyzed;

s13, extracting frames of the video data, converting the video data into picture data by taking the frames as units, extracting target features of the picture data by utilizing a target detection model, and converting the target features into gray scales to obtain gray scale features to be analyzed;

s14, the keywords to be analyzed and the gray features to be analyzed form data to be analyzed.

In this embodiment, the management submodule may be specifically divided into a project management submodule, a security management submodule, and a field management submodule, and according to the function of each management submodule, the format of the corresponding monitoring data is specifically: the monitoring data of the project management sub-module mainly comprise report types, the monitoring data of the safety management sub-module mainly comprise report types, video types and picture types, and the monitoring data of the field management sub-module mainly comprise video types and picture types. Accordingly, the monitoring data as a whole can be divided into form data, picture data, and video data according to the data format.

Specifically, the process of processing the table data includes: extracting text information from the table data, segmenting the text information, removing stop words, associating redundant words and mapping synonyms to obtain keywords which can express important meanings in the table data and serve as keywords to be analyzed.

Specifically, the target detection model may adopt a Yolov5 model, the image data is input into the Yolov5 model, the Yolov5 model is pre-trained, the target object in the image data can be detected according to the model, the target object includes but is not limited to constructors, towers, safety helmets, lighters, various construction devices and the like, after the target object is identified, the feature map of the target object is calculated by a gray level co-occurrence matrix, and the gray level co-occurrence matrix is converted into gray level features to be used as gray level features to be analyzed.

Specifically, in an embodiment of the present invention, the knowledge base in step S2 is a knowledge graph model, the model adopts a bayesian network, a domain knowledge graph is built in, and the construction process of the domain knowledge graph is as follows:

(1) Data acquisition

Firstly, all data of a relevant construction site are acquired, wherein the data comprise historical data, historical construction data, historical project data, historical planning data, climate data, geographic data, urban area data, personnel management data, historical event data, historical decision data and the like of each management sub-module, and the historical event data and the historical decision data refer to emergency conditions, management loopholes, unexpected events and the like which occur in the past construction site, and decisions and measures made for the conditions. The historical planning data refers to the whole planning content of the past construction site, the deviation between the actual implementation condition and the planning, the reason for generating the deviation, the actions made and the like.

(2) Body learning

Including term extraction, synonym extraction, concept extraction, classification relation extraction, axiom and rule extraction.

The term refers to a linguistic representation form of a concept, entity or attribute in a knowledge graph, the term extraction aims at finding a marked set for representing the concept, entity or attribute, and the term extraction can be based on a dictionary method, namely, a dictionary of related words of a construction site is predefined, and the terms defined in the dictionary are searched in summarized data materials.

The synonymous relation refers to entities which are the same or similar on the concept level, the purpose of synonymous relation extraction is to search terms which are literally different but have substantial meaning to refer to the same concept, entity or attribute, the synonymous relation extraction mode can be a machine learning algorithm, and particularly can be to adopt a classification model, and the text can be classified more accurately after training.

Concepts have a meaning similar to entities, but concepts are a more abstract meaning that is an idea, concept, or category or class of entities that serve as named entities, events, or relationships. The concept extraction can adopt a recognition method combining linguistics and statistics so as to achieve better extraction effect from the data materials.

The classification relation is a concept hierarchical relation, namely, association of upper and lower levels of concepts, leading cause and consequence and the like, specifically, the classification relation extraction can adopt a co-occurrence analysis method, firstly, data materials are converted into digitally represented information, various analysis methods in the mathematical field are adopted to quantitatively calculate the digital information, and then qualitative analysis and quantitative calculation are combined to obtain the classification relation.

Axiom and rule learning refers to a process of learning general sentence patterns or template rules containing certain entities and attributes, which can be simply understood as constraint conditions for matching concepts, terms and relations, the constraint conditions can be continuously perfected in learning, and enough examples and sentence patterns can be extracted and matched by using the constraint conditions.

(3) Entity learning

Namely, entity recognition, namely, an entity recognition method based on semantic and statistical analysis can be adopted, and recognition results of the entity are continuously optimized through three steps of text rough matching, semantic analysis and grouping statistics.

(4) Map construction

After the entity is identified, the relation and the attribute between the entities can be matched according to the results of the previous synonymous relation extraction, the classification relation extraction and the axiom and rule extraction, the entity-relation-entity is formed into triples, and the triples are linked according to the relation, so that the domain knowledge graph is obtained.

After the domain knowledge graph is built, the Bayesian network is pre-trained, and is commonly used for fault identification, specifically, a probability directed acyclic graph is formed to represent the occurrence probability, occurrence reason, relationship among faults and the like. Therefore, the embodiment firstly trains the Bayesian network by using some abnormal data and abnormal degrees marked by manpower, so that the Bayesian network has certain abnormal recognition capability, then stores the domain knowledge graph into the Bayesian network, and perfects the domain knowledge graph by using the Bayesian network after pre-training, wherein the process is as follows:

since the bayesian network is also a graph model, the bayesian network is marked as graph a, wherein vertices in the graph a are abnormal data, abnormal reasons or abnormal objects, and edges in the graph a are influences among the abnormal data or causal relations among the abnormal data. Each vertex and each edge in a contains a corresponding attribute, e.g., the anomaly data includes its scope of anomalies, the anomaly cause includes its probability value, the anomaly object includes its operating parameters, the impact between the anomaly data includes its impact probability and extent, and the causal relationship between the anomaly data includes its causal description.

The domain knowledge graph is marked as a graph B, each vertex in A is regarded as an entity, entity alignment is carried out on A and B, the vertex is defined as V, the edge is defined as E, the attribute is u, and then A is expressed as A= (u) ^A ,V ^A ,E ^A ) B is represented as b= (u) ^B ,V ^B ,E ^B ) Where u refers to a global attribute, e.g., u may represent a worksite type. Set of V and EIncluding its corresponding set of attributes.

A message transmission mechanism is adopted to transmit various information in A to B and merge, and in order to realize the step, a graph neural network is utilized to assist in message transmission.

Firstly, vectorizing various parameters in the graph A and the graph B, inputting the parameters into a graph neural network, randomly selecting a vertex as a target node, aggregating neighbor nodes of the target node by using a message function, and transmitting the aggregated neighbor node message to the target node. The process is repeated, with aggregation and messaging for each vertex, and similarly, for each edge.

The message function employed in message delivery is as follows:

wherein l and l-1 refer to the present transfer and the previous transfer, M _l-1 For message functions, Σ represents aggregation, U _l-1 Representing an update function, v representing a target node, a representing a neighbor node of v, N (v) representing a set of neighbor nodes of v, Message representing the current transfer->And->Hidden states of the target node and the neighboring node, e _va Representing a unit vector.

Specifically, in an embodiment of the present invention, step S3 includes:

s31, the environment data comprise historical data of each management sub-module, and the historical data comprise historical keywords and historical gray scale features;

s32, generating an analysis strategy according to the environmental data, and respectively analyzing keywords to be analyzed and gray features to be analyzed in the data to be analyzed by utilizing the analysis strategy to obtain an analysis result;

s33, confirming and implementing actions based on the analysis result, wherein the actions comprise carrying out abnormality judgment on the keywords to be analyzed and carrying out abnormality judgment on the gray scale characteristics to be analyzed, so as to obtain abnormal data and the abnormality degree thereof.

Specifically, in this embodiment, the overall reinforcement learning strategy may employ a Q-learning algorithm, and the analysis strategy may be represented by a strategy function:

Q(s,a)＝Q(s,a)+δ(R+γQ(s',a')-Q(s,a))

where s represents the current state, a is the current action, Q is the state action function, δ is the learning rate, γ is the rewarding decay factor, s ' is the state reached after action a is taken, a ' is the action that may be taken in state s '.

In this embodiment, step S32 includes:

coding the historical keywords and the historical gray scale characteristics to obtain a coding value;

Constructing an index table according to the historical keywords and the coding values thereof, the historical gray features and the coding values thereof, wherein the index of the index table is the coding value;

respectively encoding the keywords to be analyzed and the gray features to be analyzed to obtain encoded values of the keywords to be analyzed and the gray features to be analyzed;

and searching in an index table by utilizing the key words to be analyzed and the coding values of the gray features to be analyzed, and matching to obtain the association history key words related to the key words to be analyzed and the association history gray features related to the gray features to be analyzed as analysis results.

In this embodiment, the encoding mode may be hash encoding, after hash encoding is performed on the historical keyword and the historical gray feature, the historical keyword and the historical gray feature generate a unique encoding value, and then the unique encoding value is used to construct an index table, and it should be noted that in this embodiment, two index tables are constructed, one is the keyword index table, the other is the gray feature index table, and the header in the index table is the unique encoding value, the keyword or the gray feature. After the keywords to be analyzed and the gray features to be analyzed are obtained, hash encoding is carried out on the keywords to be analyzed and the gray features to be analyzed, unique encoding values are obtained, the unique encoding values of the keywords to be analyzed and the gray features to be analyzed are respectively searched in corresponding index tables, and historical keywords and historical gray features with the correlation degree exceeding 70% are searched and used as correlation historical keywords and correlation historical gray features. Specifically, the correlation degree may be obtained by measuring the degree of difference between two encoded values using a hamming distance.

Accordingly, step S33 includes:

judging whether the keywords to be analyzed are abnormal or not according to the correlation degree of the correlation history keywords and the keywords to be analyzed, classifying the keywords to be analyzed into abnormal data if the keywords to be analyzed are abnormal, carrying out semantic description on the keywords to be analyzed according to a semantic model after judging that the keywords to be analyzed are abnormal data, and judging the abnormal degree of the keywords to be analyzed based on the content of the semantic description;

judging whether the gray scale feature to be analyzed is abnormal or not according to the correlation degree of the correlation history gray scale feature and the gray scale feature to be analyzed, classifying the gray scale feature to be analyzed into abnormal data if the gray scale feature to be analyzed is abnormal, and calculating the abnormal degree of the gray scale feature to be analyzed according to the information entropy after judging that the gray scale feature to be analyzed is the abnormal data.

In this embodiment, the calculation method for the correlation degree between the correlation history keyword and the keyword to be analyzed may be: and vectorizing the associated historical keywords and the keywords to be analyzed respectively to obtain associated historical word vectors and word vectors to be analyzed, and calculating cosine similarity between the two word vectors to be used as correlation. And when judging the keywords to be analyzed, if the cosine similarity is greater than 0.7, classifying the keywords to be analyzed into abnormal data.

In addition, if the correlation history keyword of one keyword to be analyzed is more than one, the cosine similarity between the word vector to be analyzed of the keyword to be analyzed and all the corresponding correlation history word vectors is calculated, the cosine similarity is averaged, and if the average value is more than 0.7, the keyword to be analyzed is classified into abnormal data.

In this embodiment, the semantic model is also pre-trained, and the model adopts a light model to reduce the complexity of the model, specifically may be an ALBERT model, which is a model version after the light modification of the BERT model, and the semantic extraction is performed on the keywords to be analyzed by using the ALBERT model, so that the keywords to be analyzed can be better filled in sentences, accurate semantic description can be performed on the keywords to be analyzed by learning hidden part-of-speech features, and the anomaly degree of the keywords to be analyzed can be obtained based on a classification layer through the content of the semantic description. Specifically, the degree of abnormality is classified into four cases of very serious, relatively serious, slightly serious, and controllable degree. When the semantic model is pre-trained, the used labeling samples contain the abnormal degree of abnormal data, and the labels of the abnormal degree are sequentially set to be 4, 3, 2 and 1 according to the four conditions of very serious, relatively serious, slightly serious and controllable degree, so that the model is conveniently trained.

In this embodiment, the correlation calculation mode between the correlation gray scale features and the gray scale features to be analyzed may be euclidean distance, and the gray scale features to be analyzed are determined according to whether the euclidean distance is greater than 0.7, meanwhile, if one correlation gray scale feature of the gray scale features to be analyzed is more than one, the average value of all euclidean distances is calculated, and if the average value is greater than 0.7, the gray scale features to be analyzed are abnormal data.

In this embodiment, the gray scale feature to be analyzed is a feature obtained by gray scale conversion after target detection, and is a pixel region, and each gray scale feature to be analyzed should include a plurality of pixel points, and since the degree of abnormality is divided into four cases of very serious, relatively serious, slightly serious, and controllable, the four cases are used as indexes, the calculation result of the information entropy is divided into regions, the value range of the information entropy is [0,1], and the larger the information entropy is, the more abnormal information in the pixel region is represented, the greater the degree of abnormality is, so that the embodiment divides the value range of the information entropy into four cases of [0,0.2 ], [0.2,0.5 ], [0.5,0.8 ], and [0.8,1] respectively controllable, slightly serious, relatively serious, and very serious. The corresponding information entropy formula is as follows:

Wherein X represents the gray scale characteristic to be analyzed, D is the number of pixels, H is the information entropy, D is the D-th pixel, and p (D) represents the number of times the gray scale value of the pixel D appears in the gray scale values of all the pixels.

Specifically, in an embodiment of the present invention, step S4 includes:

s41, sending the abnormal data and the abnormal degree thereof to a knowledge base, and evaluating the abnormal data and the abnormal degree thereof by the knowledge base to obtain feedback;

s42, quantifying corresponding abnormal data according to the importance degree of each management sub-module to obtain a weight, and judging the influence degree of the abnormal data and the abnormality degree thereof based on the influence judgment rule according to feedback of the abnormal data and the abnormality degree thereof to obtain an influence value;

s43, constructing a reward function according to the weight and the influence value, wherein the formula is as follows:

f(x _i )＝α _i f ₁ (x _i )+β _i f ₂ (x _i )

wherein R is a prize value, n is the number of abnormal data, x _i Represents the ith anomaly data, ω _i Weight value representing ith abnormal data, f (x _i ) Feedback representing ith anomaly data, f ₁ (x _i ) Is the influence value of the ith abnormal data, f ₂ (x _i ) The influence value of the degree of abnormality corresponding to the ith abnormal data, and α and β are influence coefficients.

Wherein:

f ₁ (x _i ) = { -3, if x _i Is erroneous exception data; +2, if x _i Correct exception data };

f ₂ (x _i ) = { -2, if x _i Is greater than actual; -3, if x _i Is less abnormal than actual; +2, if x _i Correct degree of abnormality of (c).

Step S4 further includes:

the knowledge base predicts the current abnormal data, if the influence degree of the current abnormal data on the abnormal data subordinate to other management submodules exceeds a preset threshold, a deviation function is formed according to the prediction information to correct the reward function, and the corrected reward function is as follows:

wherein R' is the corrected prize value, x _j～i Representing abnormal data x _j For abnormal data x _i The degree of influence of (2) exceeds a preset threshold, W represents a deviation function, lambda represents a deviation coefficient, E represents abnormal data x _j For abnormal data x _i Is not limited to the above-described ones.

In reinforcement learning models, the following concepts are generally included:

the agent, which is the subject of execution, is referred to in this embodiment as a reinforcement learning module, which is a virtual "person" that performs the operation.

The environment refers to the environment in the whole reinforcement learning process, and in this embodiment, historical abnormal data, such as historical keywords, historical gray scale characteristics, historical abnormal degrees, and the like, are combined into the reinforcement learning model.

The state generally refers to the state in which the environment and the agent are located, and in this embodiment, the state is slightly changed, and specifically, the analysis result represents the state.

The action refers to an action that the intelligent agent can take based on the current state, and in this embodiment, the action includes performing abnormality determination on a keyword to be analyzed and performing abnormality determination on a gray feature to be analyzed.

Rewarding, namely, after the intelligent agent takes a specific action in the current state, feedback is obtained, a rewarding function can be constructed based on the feedback, and a rewarding value can be calculated by using the rewarding function.

In the step of obtaining feedback, the above-mentioned built knowledge base is introduced, and the knowledge base merges the performance of the Bayesian network and the knowledge map, and at the same time, the knowledge base is trained by the transmission of the graph neural network, so that the knowledge base has the functions of query classification and inference prediction.

Specifically, the abnormal data and the degree of abnormality obtained by the preliminary detection of the reinforcement learning model are input into a knowledge base, the knowledge base is queried by taking the keyword or the gray average value of gray features in the abnormal data as a query condition, the entity, the attribute and the relation related to the current abnormal data can be obtained, and the knowledge base is utilized to further infer the query condition by a learning expression method, so that the extended entity, the attribute and the relation are obtained. Summarizing query and reasoning results to obtain an entity set associated with the current abnormal data, and feeding back the current abnormal data according to the abnormal data duty ratio and the abnormal degree description in the entity set, wherein the feedback contents are as follows: whether the abnormality determination result of the current abnormal data is correct or not, and whether the abnormality degree of the current abnormal data is correct or not are determined.

After feedback is obtained, the feedback needs to be quantified according to an influence judgment rule to form an influence value, and the influence judgment rule is specifically set according to the influence of abnormal data missing influence and the influence of abnormal degree judgment errors. In other cases, for example, if the abnormal data is correctly determined and the degree of abnormality is correctly determined, the initial determination result of the model is more accurate, a proper positive score is given to encourage the model to continue learning, and if the degree of abnormality is greater than the actual value, the accuracy of the model is still to be improved, but the accuracy of the model is not greatly affected by the subsequent processing of the manager, so that the negative score is higher than if the degree of abnormality is smaller than the actual value.

Because the importance degree of each management sub-module in construction site is different, which is specifically set according to the actual situation, but the sum of the weights of the abnormal data corresponding to all the management sub-modules is 1.

After the weight and the influence value are obtained, the reward value of the model judgment result can be calculated according to the reward function, in addition, the embodiment also provides a solution for special situations, namely, the knowledge base finds that abnormal data of a certain management sub-module should influence abnormal data of another management sub-module during reasoning, when the influence degree exceeds a preset threshold, the influence degree is used as prediction information, the influence degree is quantized into influence degree, the influence degree is expected to form a deviation function, and the reward function is modified by the deviation function. In this embodiment, the influence degree may be evaluated by how much relationship there is between two pieces of abnormal data at the time of reasoning, and the preset threshold may be set so that at most one relationship should be included between two pieces of abnormal data.

After the reward value of the iteration is calculated, the reward value is input into a strategy function for updating, and meanwhile, a return function is constructed, wherein the expression is as follows:

where J represents the reward function, E represents the expectation of the policy function, which may be calculated from the summation, and ΣR represents the sum of the prize values in all iterations.

When the return function reaches the maximum value, the whole reinforcement learning process is finished, and final abnormal data and the degree of abnormality are output.

Specifically, in an embodiment of the present invention, step S8 includes:

s81, dividing the final abnormal data into keywords to be recommended and gray scale features to be recommended according to the keywords and the gray scale features;

s82, the recommendation module comprises a keyword processing suggestion library and a gray feature processing suggestion library, wherein the keyword processing suggestion library is constructed by taking keywords as indexes, the indexes of the keyword processing suggestion library are linked into keyword-processing suggestions, and the gray feature processing suggestion library is constructed by taking gray features as identifiers and binding the gray features with corresponding processing suggestions into groups;

s82, quantifying the degree of abnormality of the keywords to be recommended according to the content of semantic description of the keywords to be recommended to form abnormal values of the keywords to be recommended, searching in a recommendation module according to the keywords to be recommended to obtain a group of first candidate processing suggestions, sorting the first candidate processing suggestions according to the abnormal values of the keywords to be recommended, and selecting the first candidate processing suggestion with the highest matching degree as the processing suggestion of the keywords to be recommended;

s83, normalizing the abnormal degree of the gray feature to be recommended to obtain an abnormal value of the gray feature to be recommended, screening the gray feature with the feature distance within a preset range from the gray feature to be recommended in a recommendation module, obtaining a corresponding second candidate processing suggestion according to the screened gray feature, sorting the second candidate processing suggestion based on the abnormal value of the gray feature to be recommended, and selecting the second candidate processing suggestion with the highest matching degree as the processing suggestion of the gray feature to be recommended;

S84, the final abnormal data, the degree of abnormality and the processing advice are respectively formed into monitoring results of all the management sub-modules according to the attribution of the management sub-modules, and the monitoring results of all the management sub-modules are summarized to obtain the final monitoring results.

It should be noted that, this embodiment also provides a recommendation method of processing advice, two processing advice libraries are created according to the keywords and the gray scale features, and the specific creation method is similar to the above-mentioned index table, except that the table head in the processing advice library is the processing advice, the keywords or the gray scale features and the unique code value. When recommendation is performed, cosine similarity can be used for comparing keywords to be recommended with keywords in a processing suggestion library, euclidean distance is used for comparing gray scale features to be recommended with gray scale features in the processing suggestion library, processing suggestions are arranged according to the cosine similarity and Euclidean distance results, then the first 30% of the processing suggestions are selected as first candidate processing suggestions and second candidate processing suggestions, abnormal values are formed by retrieving abnormal degrees of the keywords to be recommended and the gray scale features to be recommended, and the abnormal values can be set to 4, 3, 2 and 1 in sequence according to four conditions of very serious, relatively serious, slightly serious and controllable degrees, and then proper processing suggestions are selected according to the matching degree of the abnormal values. And taking the final abnormal data and the degree of abnormality and the processing suggestion as monitoring results.

And then, converting the monitoring result into a log by using a log generating tool, wherein the log contains all the monitored contents of each management sub-module, transmitting the monitoring log to a display module so as to be convenient for looking up and tracing at any time, and simultaneously transmitting the monitoring log to an administrator corresponding to the management sub-module, wherein for the monitoring log with extremely serious abnormal degree, a warning mark is added when the system is transmitted so as to remind the administrator of urgent processing.

In addition, referring to fig. 2, the present invention further provides an intelligent building site monitoring and managing system, which includes:

the management submodules are configured to monitor the respective task target areas in all weather according to preset instructions and store monitoring contents in corresponding formats in time units;

the data acquisition module is configured to simultaneously acquire monitoring data of each management sub-module according to a preset time period;

the data processing module is configured to respectively process the monitoring data according to a data format to form data to be analyzed, wherein the data to be analyzed comprises keywords to be analyzed and gray features to be analyzed;

an environmental data module configured to obtain environmental data, the environmental data including historical data of each management sub-module, the historical data including historical keywords and historical gray scale features;

The reinforcement learning module is configured to generate an analysis strategy according to the environmental data, analyze the data to be analyzed by utilizing the analysis strategy, confirm and implement actions based on analysis results to obtain abnormal data and abnormal degrees thereof, inquire and feed back the abnormal data and the abnormal degrees thereof according to a knowledge base, obtain a reward value according to feedback, perform multiple iterative updating on the analysis strategy by utilizing the reward value, perform multiple iterative updating on the abnormal data and the abnormal degrees thereof by utilizing the updated analysis strategy, construct a return function based on the reward value, and finish the iterative updating when the return function reaches the maximum value to obtain final abnormal data and the abnormal degrees thereof;

the recommendation module is configured to recommend the final abnormal data and the abnormal degree thereof according to a recommendation algorithm to obtain corresponding processing suggestions, and the final abnormal data, the abnormal degree thereof and the processing suggestions respectively form monitoring results of all the management sub-modules according to attribution of the management sub-modules, and the monitoring results of all the management sub-modules are summarized to obtain final monitoring results;

the log module is configured to receive the monitoring results and respectively generate monitoring logs for the monitoring results according to the attribution of the management sub-module;

The display module is configured to receive the monitoring log and display the monitoring log according to the visualization tool;

the data sending module is configured to receive the monitoring logs and send the monitoring logs to the manager of the corresponding management sub-module according to membership of the management sub-module of each monitoring log.

The construction site is monitored in real time through collaborative operation before each module, embedded reinforcement learning models, knowledge maps and other technologies, emergency situations can be timely dealt with, the dangerous degree of the construction site is reduced, and the safety of the construction site is guaranteed.

The foregoing description of the preferred embodiments of the invention is not intended to be limiting, but rather is intended to cover all modifications, equivalents, alternatives, and improvements that fall within the spirit and scope of the invention.

Claims (10)

1. An intelligent building site monitoring and management method is characterized by comprising the following steps:

s1, acquiring monitoring data of each management sub-module in a preset time period, and processing the monitoring data to obtain data to be analyzed;

s2, constructing a reinforcement learning module, wherein the reinforcement learning module comprises a reinforcement learning model and a knowledge base;

s3, acquiring environment data, inputting the data to be analyzed into a reinforcement learning model, generating an analysis strategy according to the environment data, analyzing the data to be analyzed by utilizing the analysis strategy, and confirming and implementing actions based on analysis results to obtain abnormal data and the degree of abnormality thereof;

S4, inputting the abnormal data and the degree of abnormality thereof into a knowledge base for inquiring to obtain feedback, determining the weight of the abnormal data according to a weight setting rule, and constructing a reward function based on the weight of the abnormal data and the feedback;

s5, acquiring a current rewarding value according to the rewarding function, and updating an analysis strategy according to the current rewarding value so as to obtain updated abnormal data and the abnormal degree thereof;

s6, feeding back updated abnormal data and the abnormal degree thereof by utilizing a knowledge base, and updating the rewarding function according to new feedback;

s7, constructing a return function based on the rewarding value, and repeatedly executing the steps S5-S6 until the return function reaches the maximum value, and outputting final abnormal data and the degree of abnormality thereof;

s8, inputting the final abnormal data and the degree of abnormality thereof into a recommendation module, obtaining a processing suggestion according to a recommendation algorithm, and taking the final abnormal data and the degree of abnormality thereof and the processing suggestion as monitoring results;

s9, generating a log of the monitoring result, transmitting the monitoring log to a display module, and simultaneously transmitting the monitoring log to an administrator of each management sub-module according to membership of each management sub-module corresponding to the monitoring log for timely processing by the administrator.

2. The intelligent worksite monitoring and management method as set forth in claim 1, wherein the step S1 includes:

s11, dividing the monitoring data into table data, picture data and video data according to a data format;

s12, processing the form data by using a text processing method to obtain keywords to be analyzed;

s13, extracting frames of the video data, converting the video data into picture data by taking the frames as units, extracting target features of the picture data by utilizing a target detection model, and converting the target features into gray scales to obtain gray scale features to be analyzed;

s14, the keywords to be analyzed and the gray features to be analyzed form data to be analyzed.

3. The intelligent building site monitoring and managing method according to claim 2, wherein in step S2, the knowledge base is constructed by:

constructing a domain knowledge graph and a Bayesian network, pre-training, vectorizing the pre-trained domain knowledge graph and Bayesian network, inputting the pre-trained domain knowledge graph and Bayesian network into a graph neural network, transmitting information in the Bayesian network to the domain knowledge graph by adopting a message transmission mechanism, and fusing the Bayesian network and the domain knowledge graph to obtain a knowledge base;

wherein the messaging mechanism includes a message function having the expression:

Wherein l and l-1 refer to the present transfer and the previous transfer, M _l-1 For message functions, Σ represents aggregation, U _l-1 Representing an update function, v representing a target node, a representing a neighbor node of v, N (v) representing a set of neighbor nodes of v,message representing the current transfer->And->Hidden states of the target node and the neighboring node, e _va Representing a unit vector.

4. The intelligent worksite monitoring and management method as set forth in claim 2, wherein the step S3 includes:

s31, the environment data comprise historical data of each management sub-module, and the historical data comprise historical keywords and historical gray scale features;

s32, generating an analysis strategy according to the environmental data, and respectively analyzing keywords to be analyzed and gray features to be analyzed in the data to be analyzed by utilizing the analysis strategy to obtain an analysis result;

s33, confirming and implementing actions based on the analysis result, wherein the actions comprise carrying out abnormality judgment on the keywords to be analyzed and carrying out abnormality judgment on the gray scale characteristics to be analyzed, so as to obtain abnormal data and the abnormality degree thereof.

5. The intelligent worksite monitoring and management method as set forth in claim 4, wherein the step S32 includes:

coding the historical keywords and the historical gray scale characteristics to obtain a coding value;

Constructing an index table according to the historical keywords and the coding values thereof, the historical gray features and the coding values thereof, wherein the index of the index table is the coding value;

respectively encoding the keywords to be analyzed and the gray features to be analyzed to obtain encoded values of the keywords to be analyzed and the gray features to be analyzed;

searching in an index table by utilizing the key words to be analyzed and the coding values of the gray features to be analyzed, and matching to obtain the association history key words related to the key words to be analyzed and the association history gray features related to the gray features to be analyzed as analysis results;

accordingly, step S33 includes:

judging whether the keywords to be analyzed are abnormal or not according to the correlation degree of the correlation history keywords and the keywords to be analyzed, classifying the keywords to be analyzed into abnormal data if the keywords to be analyzed are abnormal, carrying out semantic description on the keywords to be analyzed according to a semantic model after judging that the keywords to be analyzed are abnormal data, and judging the abnormal degree of the keywords to be analyzed based on the content of the semantic description;

judging whether the gray scale feature to be analyzed is abnormal or not according to the correlation degree of the correlation history gray scale feature and the gray scale feature to be analyzed, classifying the gray scale feature to be analyzed into abnormal data if the gray scale feature to be analyzed is abnormal, and calculating the abnormal degree of the gray scale feature to be analyzed according to the information entropy after judging that the gray scale feature to be analyzed is the abnormal data.

6. The intelligent worksite monitoring and management method as set forth in claim 2, wherein the step S4 includes:

s41, sending the abnormal data and the abnormal degree thereof to a knowledge base, and evaluating the abnormal data and the abnormal degree thereof by the knowledge base to obtain feedback;

s42, quantifying corresponding abnormal data according to the importance degree of each management sub-module to obtain a weight, and judging the influence degree of the abnormal data and the abnormality degree thereof based on the influence judgment rule according to feedback of the abnormal data and the abnormality degree thereof to obtain an influence value;

s43, constructing a reward function according to the weight and the influence value, wherein the formula is as follows:

f(x _i )＝α _i f ₁ (x _i )+β _i f ₂ (x _i )

wherein R is a prize value, n is the number of abnormal data, x _i Represents the ith anomaly data, ω _i Weight value representing ith abnormal data, f (x _i ) Feedback representing ith anomaly data, f ₁ (x _i ) Is the influence value of the ith abnormal data, f ₂ (x _i ) The influence value of the degree of abnormality corresponding to the ith abnormal data, and α and β are influence coefficients.

7. The intelligent worksite monitoring and management method as set forth in claim 6, wherein the step S4 further comprises:

the knowledge base predicts the current abnormal data, if the influence degree of the current abnormal data on the abnormal data subordinate to other management submodules exceeds a preset threshold, a deviation function is formed according to the prediction information to correct the reward function, and the corrected reward function is as follows:

Wherein R' is the corrected prize value, x _j～i Representing abnormal data x _j For abnormal data x _i The degree of influence of (2) exceeds a preset threshold, W represents a deviation function, lambda represents a deviation coefficient, E represents abnormal data x _j For abnormal data x _i Is not limited to the above-described ones.

8. The intelligent worksite monitoring and management method as set forth in claim 6, wherein:

f ₁ (x _i ) = { -3, if x _i Is erroneous exception data; +2, if x _i Correct exception data };

f ₂ (x _i ) = { -2, if x _i Is greater than actual; -3, if x _i Is less abnormal than actual; +2, if x _i Correct degree of abnormality of (c).

9. The intelligent worksite monitoring and management method according to claim 5, wherein step S8 comprises:

s81, dividing the final abnormal data into keywords to be recommended and gray scale features to be recommended according to the keywords and the gray scale features;

s82, the recommendation module comprises a keyword processing suggestion library and a gray feature processing suggestion library, wherein the keyword processing suggestion library is constructed by taking keywords as indexes, the indexes of the keyword processing suggestion library are linked into keyword-processing suggestions, and the gray feature processing suggestion library is constructed by taking gray features as identifiers and binding the gray features with corresponding processing suggestions into groups;

S82, quantifying the degree of abnormality of the keywords to be recommended according to the content of semantic description of the keywords to be recommended to form abnormal values of the keywords to be recommended, searching in a recommendation module according to the keywords to be recommended to obtain a group of first candidate processing suggestions, sorting the first candidate processing suggestions according to the abnormal values of the keywords to be recommended, and selecting the first candidate processing suggestion with the highest matching degree as the processing suggestion of the keywords to be recommended;

s83, normalizing the abnormal degree of the gray feature to be recommended to obtain an abnormal value of the gray feature to be recommended, screening the gray feature with the feature distance within a preset range from the gray feature to be recommended in a recommendation module, obtaining a corresponding second candidate processing suggestion according to the screened gray feature, sorting the second candidate processing suggestion based on the abnormal value of the gray feature to be recommended, and selecting the second candidate processing suggestion with the highest matching degree as the processing suggestion of the gray feature to be recommended;

s84, the final abnormal data, the degree of abnormality and the processing advice are respectively formed into monitoring results of all the management sub-modules according to the attribution of the management sub-modules, and the monitoring results of all the management sub-modules are summarized to obtain the final monitoring results.

10. An intelligent worksite monitoring and management system, characterized in that the system performs the method of any of claims 1-9, the system comprising:

the management submodules are configured to monitor the respective task target areas in all weather according to preset instructions and store monitoring contents in corresponding formats in time units;

the data acquisition module is configured to simultaneously acquire monitoring data of each management sub-module according to a preset time period;

the data processing module is configured to respectively process the monitoring data according to a data format to form data to be analyzed, wherein the data to be analyzed comprises keywords to be analyzed and gray features to be analyzed;

an environmental data module configured to obtain environmental data, the environmental data including historical data of each management sub-module, the historical data including historical keywords and historical gray scale features;

the reinforcement learning module is configured to generate an analysis strategy according to the environmental data, analyze the data to be analyzed by utilizing the analysis strategy, confirm and implement actions based on analysis results to obtain abnormal data and abnormal degrees thereof, inquire and feed back the abnormal data and the abnormal degrees thereof according to a knowledge base, obtain a reward value according to feedback, perform multiple iterative updating on the analysis strategy by utilizing the reward value, perform multiple iterative updating on the abnormal data and the abnormal degrees thereof by utilizing the updated analysis strategy, construct a return function based on the reward value, and finish the iterative updating when the return function reaches the maximum value to obtain final abnormal data and the abnormal degrees thereof;

The recommendation module is configured to recommend the final abnormal data and the abnormal degree thereof according to a recommendation algorithm to obtain corresponding processing suggestions, and the final abnormal data, the abnormal degree thereof and the processing suggestions respectively form monitoring results of all the management sub-modules according to attribution of the management sub-modules, and the monitoring results of all the management sub-modules are summarized to obtain final monitoring results;

the log module is configured to receive the monitoring results and respectively generate monitoring logs for the monitoring results according to the attribution of the management sub-module;

the display module is configured to receive the monitoring log and display the monitoring log according to the visualization tool;

the data sending module is configured to receive the monitoring logs and send the monitoring logs to the manager of the corresponding management sub-module according to membership of the management sub-module of each monitoring log.