Construction project safety monitoring and early warning system based on multi-agent cooperation

Abstract

The invention provides a construction project safety monitoring and early warning system based on multi-agent cooperation. The construction project safety monitoring and early warning system comprises a plurality of movable edge processing units, a first data collection unit and a second data collection unit; the data storage unit is provided with a BIM compression module for compressing the BIM model and a distillation module for distilling the construction specification library; and the edge processing unit carries out risk analysis according to the sensor data or / and the image data to obtain a risk level and an influence area, and the edge processing unit goes to the influence area to carry out a monitoring and early warning task. By adopting the system, the dynamic change of a construction site can be responded in real time, and the problem of a monitoring blind area is solved; the method has an autonomous learning ability, and a weight grading algorithm can be continuously optimized; and all the edge processing units can cooperatively make decisions and act and jointly execute monitoring and early warning tasks.

Description

Technical Field

[0001] The present invention relates to a construction project safety monitoring and early warning system based on multi-agent collaboration. Background Art

[0002] Safety monitoring and early warning are crucial components of construction projects. As projects continue to grow in scale and complexity, construction sites often face numerous safety hazards, including high-altitude work, equipment operation, and environmental risks. Without timely warnings and action, these hazards can easily lead to casualties and property damage. Traditional approaches relying on manual inspections or localized sensor monitoring are no longer sufficient. Intelligent, real-time monitoring systems are urgently needed to improve warning accuracy and response speed. However, existing technologies generally utilize fixed data processing architectures, which are difficult to adapt to the complex and dynamic nature of construction sites.

[0003] Most existing construction safety monitoring systems are based on cloud or central server architectures. For example, some systems deploy fixed sensor arrays and cameras to collect data and transmit all information back to a central processing unit for analysis. At the same time, to assist in risk assessment, these systems integrate BIM models, but these models are typically large and computationally complex, consuming a large amount of bandwidth during transmission, resulting in significantly increased processing delays. In addition, existing technologies lack effective mechanisms to support intelligent collaboration between devices: when high risks are detected, warning information cannot be quickly disseminated, regional coverage is also limited, and it is impossible to achieve automatic movement of intelligent agents in response to changing areas.

[0004] Therefore, there is an urgent need for a multi-agent collaborative construction project safety monitoring and early warning system. Summary of the Invention

[0005] In order to overcome the existing technical problems, the present invention provides a multi-agent collaborative construction project safety monitoring and early warning system.

[0006] The present invention adopts the following technical solutions.

[0007] A construction project safety monitoring and early warning system based on multi-agent collaboration, including multiple movable edge processing units; A first data collection unit and a second data collection unit are configured to acquire a plurality of sensor data or a plurality of image data and transmit the data to an edge processing unit within a preset transmission range, wherein the second data collection unit is disposed on the edge processing unit; a data storage unit having a construction specification library and a BIM model of a construction project, a BIM compression module for compressing the BIM model and sending it to an edge processing unit, and a distillation module for distilling the construction specification library and sending it to the edge processing unit; The edge processing unit performs risk analysis based on sensor data and / or image data to obtain the risk level and affected area. If the risk level is greater than or equal to the warning threshold, the risk level and affected area are synchronized to other edge processing units, and the edge processing unit goes to the affected area to perform monitoring and early warning tasks.

[0008] As a further improvement of the present invention, the data storage unit is further provided with a historical database storing historical construction project data, and the historical database is provided with a data normal distribution model that is regularly updated to the first data collection unit and the second data collection unit; The first data collection unit and the second data collection unit are provided with a data classification module, an image mapping module, a sensitivity pulse module and a data transmission module; The image mapping module is used to map image data into image fuzzy data; The sensitivity pulse module is used to send a pulse to the data classification module when the sensor data or image blur data exceeds the corresponding classification warning threshold; The data classification module is used to perform weighted classification calculations on sensor data or image fuzzy data. The expression for weighted classification calculation is: , , in, is the calculation of the probability density function of the i-th data value in the normal distribution, is the continuous sensitivity of risk data, is the number of pulses in time T, is the maximum number of pulses within time T, is the pulse calibration factor, is the time calibration factor, t is the interval from the data timestamp to the calculation time; The weight ratio calculation is performed for each weight level. The expression for weight ratio calculation is: , Data transmission module according to weight ratio Send sensor data or image blur data in descending order to the edge processing unit within the preset transmission range. If the weight ratio of sensor data or image blur data is If the number is lower than the preset redundancy threshold, it will not be sent; When in a non-construction state, the first data collection unit and the second data collection unit send all sensor data or image blur data to the data storage unit.

[0009] As a further improvement of the present invention, the specific steps of mapping image data into image blurred data are: The first data collection unit and the second data collection unit include a plurality of camera modules for acquiring image data; Using a CNN model to identify construction disaster elements in image data, construction disaster elements include smoke, cracks, fire, and liquids; The construction disaster elements under the same type of camera module have a corresponding basic disaster threshold. , identify the proportion of construction disaster elements in the image data and obtain the disaster ratio , the image blur data is ,in, is the weight coefficient of the j-th construction disaster element.

[0010] As a further improvement of the present invention, the specific steps of compression include: Filter the BIM model through FilteredElementCollector to obtain multiple components, set the components as nodes, and calculate whether the contour lines between two components intersect. If they intersect, connect the nodes of the two components as edges; Use FilteredElementCollector combined with the built-in category BuiltInCategory to filter out the load-bearing components in the BIM model, integrate the filtered load-bearing components into a load-bearing structure set, and integrate the remaining components in the BIM model into a decorative component set. Check the Structural parameters of each load-bearing component in the load-bearing structure set. If there is a false value or no Structural parameter, move it to the decorative component set. The weight factor of the nodes located in the load-bearing structure is 0.9, and the weight factor of the nodes located in the decorative component is 0.2; Calculate the node importance of each node , , , in, is the number of nodes connected to node v, is the set of nodes connected to node v, is the weight factor of the i-th node connected to node v, It is to calculate the proportion of the shortest paths of all nodes passing through node v, are importance adjustment coefficients, is the total number of nodes, s and t are any two nodes that are not equal to each other and are not equal to node v, is the number of nodes v passed through by the shortest path, is the total number of shortest paths, the shortest paths are obtained by Yen's algorithm; Calculating significance thresholds , , in, is the importance of the entire graph node The average value of is the importance of the entire graph node The standard deviation of is the significance threshold coefficient; If the node importance Below the importance threshold Then delete the node, merge the attribute data of the node into the connected nodes in equal proportion, calculate the new node importance of the merged node, and recalculate the new average importance of the nodes in the entire graph to obtain a compressed topology graph, in which each node has the attribute data of the corresponding BIM model.

[0011] As a further improvement of the present invention, the BIM model is compressed to obtain a compressed topology graph, wherein each node in the compressed topology graph has attribute data corresponding to the BIM model; The specific steps of distillation include: The rule parser extracts the quadruple of the construction specification library and converts the quadruple into the main specification constraint vector , and the universal canonical constraint vector , ; Extract node feature vectors based on compressed topology graph , ; Extract the canonical constraint vector with the same subject encoding and node feature vector Make associations; The specific steps of sending to the edge processing unit are: compressing the topology map and the general specification constraint vector , and the subject specification constraint vector with associated relationship Sent to the edge processing unit.

[0012] As a further improvement of the present invention, both the sensor data and the image data are marked with detection time data, and position data corresponding to the nodes of the compressed topology graph; The specific steps for performing risk analysis based on sensor data and image data to obtain risk levels include: Convert sensor data or image blur data into real-time detection vectors , ; Calculating construction deviation scores , ; , , in, is the constraint weight coefficient corresponding to the subject's normative constraint vector, is the constraint weight coefficient corresponding to the universal specification constraint vector, and n is the number of universal specification constraint vectors; Calculating risk level If the risk level is greater than 0, the corresponding detection time data and location data are marked. , in, These are the risk grading thresholds from small to large.

[0013] As a further improvement of the present invention, the first data collection unit includes multiple data collection modules. If the edge processing unit does not receive sensor data or image data from the same data collection module within a preset interruption detection interval, and the edge processing unit is within a preset transmission range of the data collection module, it is considered that the data collection module has an interruption. When a flow interruption occurs, multiple edge processing units within a preset transmission range send interruption confirmations to each other. If all confirm the flow interruption, the edge processing unit closest to the data collection module is selected as the interruption master unit. The main disconnection unit sends connection confirmations to multiple data collection modules within a preset disconnection range, with the data collection module as the center. The data collection modules that do not respond are regarded as disconnection nodes. With the disconnection node as the center, the data collection modules within the preset disconnection range that have not sent connection confirmations send connection confirmations. The data collection modules marked as disconnection nodes are integrated into a disconnection set to obtain the affected range. If there are at least two data collection modules that are disconnected, the risk level is set to the highest level; otherwise, the risk level is set to the third highest level.

[0014] As a further improvement of the present invention, the attribute data includes material data, and the edge processing unit is preset with a material attenuation coefficient mapping table corresponding to the material data; The specific steps to obtain the impact area include: Extract nodes with risk level greater than 0, and extract the material data and material attenuation coefficient of the node and location data, integrated to form a risk node set ; Select the risk center node , , in, is the risk level of node i is the new node importance of node i, is the new average of the importance of nodes in the entire graph; From the risk center node First, perform iterative calculations based on the adjacent relationship. For the first-level adjacent nodes ,like , then the node is not calculated, and the node Adjacent secondary neighbor nodes Calculate, if , then calculate the node Adjacent third-level neighbor nodes Risk level And cover the original risk level of the node, , in, is the dynamic safety factor, is the height value, is the height scaling factor, is the dynamic material attenuation value, is a node The volume, is the rated attenuation volume, is the material attenuation coefficient; like Less than the preset safe propagation threshold , then the node is considered to have no propagation risk, the iterative calculation is stopped, and the previous level node is selected as the propagation node. If If it is greater than the preset safe propagation threshold, the iterative calculation will continue; in, , all risk levels greater than the safe propagation threshold The nodes are regarded as propagation nodes and integrated into a propagation node set to obtain the influence range.

[0015] As a further improvement of the present invention, the specific steps of the edge processing unit moving to the affected area include: The edge processing unit in the affected area is marked as a response unit. Other edge processing units calculate the closest distance between the current position and the affected area. If the closest distance is lower than the response threshold, it is recorded as a response unit and Search algorithms plan trips to the impact area; The response units in the affected area establish signal connections with each other, and the response unit closest to the risk center node is selected as the safety master unit. The safety master unit divides the affected area into multiple detection areas according to the number of response units, and allocates the nearest detection area according to the location data of each response unit. The response unit goes to the corresponding detection area.

[0016] As a further improvement of the present invention, the specific steps of performing the monitoring and early warning task include: The response unit is equipped with a broadcast module for broadcasting, and the response unit sends evacuation instructions to the construction workers through the broadcast module; The second data collection unit on the response unit collects image data of the current environment in real time, identifies whether there is any equipment in operation in the current environment, and if so, identifies whether there are construction workers. If there are construction workers, the response unit broadcasts to the construction workers to shut down the equipment. If the construction workers do not shut down the equipment within the shutdown broadcast interval, or if there are no construction workers, the response unit shuts down the equipment. After the detection and warning task is completed, the collected image data is uploaded to the historical database of the data storage unit.

[0017] The beneficial effects of the present invention are: 1. By deploying multiple mobile edge processing units for on-site data processing and risk analysis, and combining their autonomous mobility, they can respond to dynamic changes on the construction site in real time. Even if the first data collection unit is unable to continue working due to an accident, the edge processing unit can still go to the affected area to perform monitoring and early warning tasks, solving the problem of monitoring blind spots and improving the timeliness of early warnings and response speed.

[0018] 2. The BIM compression and distillation modules compress large BIM models and distill complex construction specification library knowledge, reducing the amount of data transmitted to the edge processing unit. When the first and second data collection units are not in construction mode, they can upload all data to the data storage unit to further optimize the weighted grading calculation, enabling autonomous learning and continuous optimization of the algorithm.

[0019] 3. By establishing an independent first data collection unit and a second data collection unit deployed on the edge processing unit, a data collection method combining fixed and mobile locations is implemented, expanding the monitoring range and improving the comprehensiveness of detection data acquisition and its adaptability to construction sites. Furthermore, the edge processing units can collaborate on decision-making and actions, jointly executing monitoring and early warning tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0021] Figure 1 It is a schematic diagram of the main modules of the present invention. DETAILED DESCRIPTION

[0022] The accompanying drawings are for illustrative purposes only and should not be construed as limiting this patent. In order to better illustrate this embodiment, certain components of the accompanying drawings may be omitted, enlarged, or reduced in size, and do not represent the actual size of the product.

[0023] It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings. The technical solution of the present invention will be further described below in conjunction with the drawings and embodiments.

[0024] Reference Figure 1 ,A construction project safety monitoring and early warning system based on multi-agent collaboration, including multiple mobile edge processing units; A first data collection unit and a second data collection unit are configured to acquire a plurality of sensor data or a plurality of image data and transmit the data to an edge processing unit within a preset transmission range, wherein the second data collection unit is disposed on the edge processing unit; By deploying multiple mobile edge processing units for on-site data processing and risk analysis, and combining their autonomous mobility capabilities, it is possible to respond to dynamic changes at the construction site in real time. Even if the first data collection unit is unable to continue working due to an accident, the edge processing unit can still go to the affected area to perform monitoring and early warning tasks, solving the problem of monitoring blind spots and improving the timeliness and response speed of early warnings. By setting up an independent first data collection unit and a second data collection unit deployed on the edge processing unit, a data collection method combining fixed and mobile points is realized, expanding the monitoring range and improving the comprehensiveness of detection data acquisition and adaptability to the construction site. In addition, the various edge processing units can coordinate decisions and actions to jointly perform monitoring and early warning tasks.

[0025] a data storage unit having a construction specification library and a BIM model of a construction project, a BIM compression module for compressing the BIM model and sending it to an edge processing unit, and a distillation module for distilling the construction specification library and sending it to the edge processing unit; The BIM compression and distillation modules compress large BIM models and distill complex construction specification library knowledge, reducing the amount of data transmitted to the edge processing unit. When the first and second data collection units are not in construction mode, they can upload all data to the data storage unit to further optimize the weighted grading calculation, enabling autonomous learning and continuous optimization of the algorithm.

[0026] As a specific embodiment of the present invention, data reception is built on the Flink dynamic DAG stream processing engine. Stream processing refers to processing data immediately after it is generated, without waiting for all data to be collected. Batch processing refers to the centralized processing of complete data sets collected within a fixed time window.

[0027] As a further improvement of the present invention, the data storage unit is further provided with a historical database storing historical construction project data, and the historical database is provided with a data normal distribution model that is regularly updated to the first data collection unit and the second data collection unit; Specifically, the scheduled update occurs after the first and second data collection units upload all sensor data or image blur data when they are not in a construction state. It is important to note that due to the varying scales of construction projects, the data within their historical construction project data will differ significantly. Therefore, this historical construction project data includes multiple standard historical construction data sets based on type, scale, and region. When updating the historical database to the first and second data collection units, the data normal distribution model of the most similar historical construction data set is updated.

[0028] The first data collection unit and the second data collection unit are provided with a data classification module, an image mapping module, a sensitivity pulse module and a data transmission module; The image mapping module is used to map image data into image fuzzy data; As a further improvement of the present invention, the specific steps of mapping image data into image blurred data are: The first data collection unit and the second data collection unit include a plurality of camera modules for acquiring image data; The CNN model is used to identify construction disaster elements in image data. Construction disaster elements include smoke, cracks, fire and liquid. Since the technology of using the CNN model to identify specific elements is very common, such as the invention patent with authorization number CN118279907B, it will not be described in detail in the present invention.

[0029] The construction disaster elements under the same type of camera module have a corresponding basic disaster threshold. , identify the proportion of construction disaster elements in the image data and obtain the disaster ratio , the image blur data is ,in, is the weight coefficient of the j-th construction disaster element.

[0030] Since the image blur data at this time will directly contribute to the construction deviation score in the future During the calculation process, the general specification constraint vector and the main specification constraint vector need to be compared and judged, so the basic disaster threshold of each disaster element needs to be adjusted. , thereby balancing the difference in image blur data. At this point, in order to facilitate the subsequent setting of the general norm constraint vector and the subject norm constraint vector of the image blur data, the upper limit of the constraint in the general norm constraint vector and the subject norm constraint vector is generally set to 1, and the lower limit of the constraint is set to 0.

[0031] As a specific embodiment of the present invention, smoke is generally caused by fires such as electrical short circuits or welding sparks. It develops rapidly and generally produces a huge amount of smoke in a short period of time. Therefore, its basic disaster threshold is relatively low and is set to 0.4. When smoke occupies 60% of the screen and the weight coefficient is When it is 1, the image blur data reaches 1. Of course, in order to achieve the effect of rapid warning, the corresponding weight coefficient can be increased If set to 2, the image blur value reaches 1 when smoke occupies 30% of the image. For cracks, because cracks pose a greater risk to engineering, the basic disaster threshold is set higher, set to 1. Generally, if the current image is judged to have cracks, an early warning should be issued. Regarding fire, fire should not occur in construction scenes, but the impact of fire is relatively low compared to cracks. Therefore, the basic disaster threshold is set to 0.85 to avoid misjudgments caused by construction workers smoking or small flames that could go out at any time. For liquids, because liquids are commonly used and stored during construction, the basic disaster threshold is set to 0.3, and warnings are only issued when large amounts of liquid are present.

[0032] As a further improvement, for different construction projects and the location of the camera module, the basic disaster threshold and the weight coefficient can be adjusted specifically. When the construction project is large and the camera module is placed far away, the basic disaster threshold and the weight coefficient are Both can be adjusted higher.

[0033] It's important to note that image blur data is generally used as a supplementary means for monitoring and evaluation, with primary monitoring performed through sensors. For example, when a fire occurs, the data from nearby temperature sensors will surge. Sensor data monitoring is typically much faster and more accurate than image data. Image blur data is primarily used in scenarios where, in the event of a sensor failure or power outage, image blur data reaching a certain level can signal an edge computing unit to the affected area. This edge computing unit is equipped with a second data collection unit, which includes a temperature sensor and camera module, enabling more precise monitoring, as well as monitoring, recording, and alarm functions.

[0034] The sensitivity pulse module is used to send a pulse to the data classification module when the sensor data or image blur data exceeds the corresponding classification warning threshold; The data classification module is used to perform weighted classification calculations on sensor data or image fuzzy data. The expression for weighted classification calculation is: , , in, is the calculation of the probability density function of the i-th data value in the normal distribution, is the continuous sensitivity of risk data, is the number of pulses in time T, is the maximum number of pulses within time T, is the pulse calibration factor, is the time calibration factor, t is the interval from the data timestamp to the calculation time; At this point, it is important to note that the normal distribution model has a range of divisions based on the mean and standard deviation. The calculation of the probability density function of the i-th data value in the normal distribution is specifically to calculate the probability of the range in which the value lies. Specifically, the division range starts with the mean and is divided into half a standard deviation. At this time, the more the data value deviates from the mean, the smaller its probability value will be. The larger the value, the higher the calculated value of the weight classification.

[0035] By calculating the continuous sensitivity, we can measure the proportion of sensor data or image blur data that continuously exceeds the graded warning threshold. If the data value of the current sensor continuously exceeds the graded warning threshold, the weighted graded calculation will be increased. For example, when a sensor data continuously slightly exceeds the graded warning threshold, although its The value is small, but because it continues to exceed the graded warning threshold, its pulse ratio within T time is 100%, which can increase its continuous sensitivity value, thereby making the weight grade calculation value larger, and the proportion in the subsequent weight ratio calculation is larger, avoiding the occurrence of missed reports. Furthermore, if you want to increase the warning sensitivity of continuous minor risks, you can increase the pulse calibration factor By detecting whether the most recent sensor data or image blur data is greater than the graded warning threshold, the weight of the most recent data value can be increased. This cross-time weighted grading method can comprehensively and effectively determine the urgency of the data value.

[0036] It should be noted that the thresholds for graded warnings are set one by one. For example, as mentioned above, for image blur data, in order to facilitate the subsequent construction deviation score Calculation, set 1 as the limit, so the graded warning threshold can be set to 0.8. The graded warning thresholds for other sensor data are calculated based on the construction safety value. For example, if the current environment requires the temperature to be below 50 degrees, the graded warning threshold is set to 45 degrees. In short, this graded warning threshold is slightly lower than the required value, thereby achieving the warning effect and facilitating the subsequent edge computing unit to move to the affected area.

[0037] The weight ratio calculation is performed for each weight level. The expression for weight ratio calculation is: , Data transmission module according to weight ratio Send sensor data or image blur data in descending order to the edge processing unit within the preset transmission range. If the weight ratio of sensor data or image blur data is If the number is lower than the preset redundancy threshold, it will not be sent; Due to the limited bandwidth of the data transmission module and the limited data processing speed of the edge computing unit, when there is a serious safety accident, a large amount of sensor data and image fuzzy data must be transmitted, and the data will also have different importance. For example, the data deviation is greater near the center point of the accident and the weight ratio is greater. Therefore, according to the weight ratio calculation, the edge computing unit can receive the most urgent data, thereby obtaining the most accurate risk level assessment and achieving rapid early warning.

[0038] When in a non-construction state, the first data collection unit and the second data collection unit send all sensor data or image blur data to the data storage unit.

[0039] Through asynchronous updates, the pressure of data transmission during construction can be reduced, and the data can be uploaded to the data storage unit for automatic optimization of weight grading calculations.

[0040] As a further improvement of the present invention, the specific steps of compression include: Filter the BIM model through FilteredElementCollector to obtain multiple components, set the components as nodes, and calculate whether the contour lines between two components intersect. If they intersect, connect the nodes of the two components as edges; Specifically, to calculate whether the contour lines intersect, the element intersection filter BoundingBoxIntersectsFilter can be used. Both FilteredElementCollector and BoundingBoxIntersectsFilter are set through the Revit API, and the BIM model is also opened through the Revit software. The Revit API is the application programming interface of the Revit software.

[0041] Use FilteredElementCollector combined with the built-in category BuiltInCategory to filter out the load-bearing components in the BIM model, integrate the filtered load-bearing components into a load-bearing structure set, and integrate the remaining components in the BIM model into a decorative component set. Check the Structural parameters of each load-bearing component in the load-bearing structure set. If there is a false value or no Structural parameter, move it to the decorative component set. According to the "Architectural Engineering Design Information Model Delivery Standard," the model's data parameters must be complete and accurate, allowing for direct selection of load-bearing components. Specifically, these components include columns, walls, beams, slabs, foundations, and trusses.

[0042] The weight factor of the nodes located in the load-bearing structure is 0.9, and the weight factor of the nodes located in the decorative component is 0.2; Calculate the node importance of each node , , , in, is the number of nodes connected to node v, is the set of nodes connected to node v, is the weight factor of the i-th node connected to node v, It is to calculate the proportion of the shortest paths of all nodes passing through node v, are importance adjustment coefficients, is the total number of nodes, s and t are any two nodes that are not equal to each other and are not equal to node v, is the number of nodes v passed through by the shortest path, is the total number of shortest paths, the shortest paths are obtained by Yen's algorithm; For the importance adjustment coefficient, when the components connected by the node are all load-bearing components, the node, as the intermediate node connecting them, should also have a higher importance. Its importance is much higher than the number of other nodes connected to the node, so To be greater than And because the fault propagates quickly along the path, calculating the proportion of the shortest paths of all nodes passing through node v can effectively determine the extent to which other nodes will be affected when the node fails. Greater than Greater than Specifically, The coefficient can be set to 0.4~0.6, which is between the weight factors of decorative components and load-bearing components. It can be set as 1 as a benchmark. Set it between 0.6 and 1.

[0043] Calculating significance thresholds , , in, is the importance of the entire graph node The average value of is the importance of the entire graph node The standard deviation of is the significance threshold coefficient; Set the importance threshold coefficient according to the proportion of nodes you want to delete. For example, if you want to delete 30% of the nodes, you can set Set to 0.524, If the node importance Below the importance threshold Then delete the node, merge the attribute data of the node into the connected nodes in equal proportion, calculate the new node importance of the merged node, and recalculate the new average importance of the nodes in the entire graph to obtain a compressed topology graph, in which each node has the attribute data of the corresponding BIM model.

[0044] The specific steps for node merging are to merge the deleted node's size attribute and node importance according to the ratio of the size attribute and node importance of the merged node to form a new node. For example, if the deleted node has two adjacent nodes and these two adjacent nodes do not need to be merged, they will be merged into the adjacent nodes with 50% of the size attribute and node importance respectively. The node importance of the merged node is calculated based on the ratio of the merged node's size attribute to its own size attribute.

[0045] As a further improvement of the present invention, the BIM model is compressed to obtain a compressed topology graph, wherein each node in the compressed topology graph has attribute data corresponding to the BIM model; The specific steps of distillation include: The rule parser extracts the quadruple of the construction specification library and converts the quadruple into the main specification constraint vector , and the universal canonical constraint vector , ; Extract node feature vectors based on compressed topology graph , ; Extract the canonical constraint vector with the same subject encoding and node feature vector Make associations; The specific steps of sending to the edge processing unit are: compressing the topology map and the general specification constraint vector , and the subject specification constraint vector with associated relationship Sent to the edge processing unit.

[0046] Specifically, the rule parser is Microsoft Azure Language Service, which is customized by uploading a construction term dictionary before use. For example, the temperature control index in 3.0.4 of the "Standard for Mass Concrete Construction" GB50496-2018, the cement entering the machine temperature in 4.2.1 (3), the concrete entering the mold temperature in 5.1.7, etc., and the provisions of the "Technical Specifications for Highway Bridge and Culvert Construction" JTGT3650-2020, etc.

[0047] It should be noted at this point that the present invention is an improvement based on the existing technology of using a fixed sensor array and a camera to collect data and transmitting all information back to the central processing unit for analysis. Therefore, the present invention does not elaborate on the specific placement of the sensors.

[0048] Furthermore, since the constraint value of the image blur data is difficult to determine, as mentioned above, the upper limit of the subject constraint is set to 1 and the lower limit of the subject constraint is set to 0.

[0049] The edge processing unit performs risk analysis based on sensor data and / or image data to obtain the risk level and affected area. If the risk level is greater than or equal to the warning threshold, the risk level and affected area are synchronized to other edge processing units, and the edge processing unit goes to the affected area to perform monitoring and early warning tasks.

[0050] The warning threshold is specifically 1.

[0051] As a further improvement of the present invention, both the sensor data and the image data are marked with detection time data, and position data corresponding to the nodes of the compressed topology graph; The specific steps for performing risk analysis based on sensor data and image data to obtain risk levels include: Convert sensor data or image blur data into real-time detection vectors , ; Calculating construction deviation scores , ; , , in, is the constraint weight coefficient corresponding to the subject's normative constraint vector, is the constraint weight coefficient corresponding to the universal specification constraint vector, and n is the number of universal specification constraint vectors; In this formula, the purpose of calculating the multiplication of the constraint weight coefficient is to calculate how much the detection value or image blur data exceeds the constraint value. It should be noted that since the image blur data is obtained through the basic disaster threshold The proportion of construction disaster elements in the image data , calculate and get , the image blur data will not exceed too high a proportion relative to the upper limit of the constraint, so there will not be a high construction deviation score As mentioned above, image fuzzy data is only an auxiliary means of monitoring. For example, when a fire occurs, the sensor data of the temperature sensor nearby will rise sharply, which will increase the construction deviation score. Image blur data is primarily used in scenarios where, due to a sensor failure or power outage, image blur reaches a certain level. This can be used to send an edge computing unit to the affected area. The edge computing unit is equipped with a second data collection unit, which includes a temperature sensor and camera module. This allows for more precise monitoring, recording, and alarm functions.

[0052] Specifically, if the vibration detection value is calculated, the main constraint upper limit is divided into four categories according to the level of the equipment. The first category is a small fixed machine with a power of less than 15KW. The second category is a medium-sized machine with a power of 15-75KW without a dedicated foundation, or a machine with a power of less than 300KW rigidly installed on a dedicated foundation. The third category is large rotating machinery on a rigid or heavy foundation. The fourth category is a large rotating machinery on a light structure foundation. The unit of the vibration detection value is mm / s, and the starting point is the vibration that can be felt by the human body, which is 0.07mm / s. For the first category, its vibration detection value is between 1.2 and 1.8, which is close to the limit, and the constraint upper limit is 1.8. If it is the second category, its vibration detection value is between 1.8 and 2.8, which is close to the limit, and the constraint upper limit is 2.8. For the fourth category, it is 7.1.

[0053] Calculating risk level If the risk level is greater than 0, the corresponding detection time data and location data are marked. , in, These are the risk grading thresholds from small to large.

[0054] In order to facilitate the processing of image fuzzy data, the constraint weight coefficient is and are all 1, the risk classification threshold is also set to 1. The calculation is based on the ratio of the detection value to the upper or lower constraint limit, so the risk classification threshold can be determined based on the ratio. For every 20% increase in the detection value, the severity of the incident increases by one level. However, since the increase in detection value is nonlinear with the urgency of the incident, 1.2 is set as Level 2, and 1.3 is set as Level 3.

[0055] As a further improvement of the present invention, the first data collection unit includes multiple data collection modules. If the edge processing unit does not receive sensor data or image data from the same data collection module within a preset interruption detection interval, and the edge processing unit is within a preset transmission range of the data collection module, it is considered that the data collection module has an interruption. When a flow interruption occurs, multiple edge processing units within a preset transmission range send interruption confirmations to each other. If all confirm the flow interruption, the edge processing unit closest to the data collection module is selected as the interruption master unit. The main disconnection unit sends connection confirmations to multiple data collection modules within a preset disconnection range, with the data collection module as the center. The data collection modules that do not respond are regarded as disconnection nodes. With the disconnection node as the center, the data collection modules within the preset disconnection range that have not sent connection confirmations send connection confirmations. The data collection modules marked as disconnection nodes are integrated into a disconnection set to obtain the affected range. If there are at least two data collection modules that are disconnected, the risk level is set to the highest level; otherwise, the risk level is set to the third highest level.

[0056] The main unit for interruption of traffic detects layer by layer from the central node instead of the polling mechanism, which can quickly spread and locate the affected area, and only starts the main unit's flaw detection process when multiple edge units confirm the interruption of traffic, so as to avoid defining the interruption caused by individual network fluctuations as an interruption event. If at least two data collection modules are interrupted, it is generally because of an emergency at the corresponding location, so it is set to the highest level and the edge processing unit can be mobilized to handle it. Specifically, if there are multiple affected areas, the number is divided according to the risk level, and the nearest edge computing unit is assigned to the affected area. If there are eight robots, one with a risk level of 3 and one with a risk level of 1, then six robots will be mobilized to the affected area with risk level 3, and two will be mobilized to the affected area with risk level 1.

[0057] As a further improvement of the present invention, the attribute data includes material data, and the edge processing unit is preset with a material attenuation coefficient mapping table corresponding to the material data; According to the "Architectural Engineering Design Information Model Delivery Standard", the model's attribute data should include material data, and the material attenuation coefficient mapping table includes multiple material attenuation coefficients. .

[0058] The specific steps to obtain the impact area include: Extract nodes with risk level greater than 0, and extract the material data and material attenuation coefficient of the node and location data, integrated to form a risk node set ; Select the risk center node , , in, is the risk level of node i, is the new node importance of node i, is the new average of the importance of nodes in the entire graph; By calculating the risk center node through this method, since the new node importance of each node is different, the unique node can be effectively obtained.

[0059] From the risk center node First, perform iterative calculations based on the adjacent relationship. For the first-level adjacent nodes ,like , then the node is not calculated, and the node Adjacent secondary neighbor nodes Calculate, if , then calculate the node Adjacent third-level neighbor nodes Risk level And cover the original risk level of the node, , in, is the dynamic safety factor, is the height value, is the height scaling factor, is the dynamic material attenuation value, is a node The volume, is the rated attenuation volume, is the material attenuation coefficient; Specifically, The value is accurate to 0.1, the rated attenuation volume The dynamic material attenuation value is 1. The dynamic material attenuation value is roughly determined based on the density or material stiffness per cubic meter. For example, the material attenuation coefficient of concrete C30 is 1.2, the material attenuation coefficient of steel Q345 is 1.5, and the material attenuation coefficient of pine wood is 0.6. The larger the material attenuation coefficient, the larger the dynamic material attenuation value, which means that when the risk is transmitted to the component, the component will absorb most of the risk, thereby reducing the risk of adjacent nodes. The dynamic material attenuation value is mixed with the node importance calculation because the node importance is based on the number of connected nodes, the number of load-bearing structures of the connected nodes, and the proportion of the shortest path passing through the node. The larger the values of these items, the greater the role that other nodes can provide in supporting the node. Therefore, increasing the calculation of node importance can effectively calculate the material attenuation value.

[0060] Dynamic safety factor At least 1. During construction, increasing the height will lead to a sharp increase in the kinetic energy of falling objects, increased difficulty in rescuing fallen personnel, and dense cross-operation stacking. That is, the higher the number of floors, the greater the danger and the more serious the consequences of accidents, which can easily lead to chain accidents. Therefore, a height value is set. It can prioritize the identification of high-rise construction hazards to avoid serious accidents. Specifically, the risk increases with each increase in the height value. For example, if the height value is 4 meters, the risk increases by 10% for each floor. The height scaling factor is Can be set to 0.25, each time a layer is raised, the dynamic safety factor The value increases by 0.1 It's important to note why adjacent nodes with the same risk level as the central node aren't calculated. This is because if the risk levels of two nodes are the same, they're both at the same high risk. Directly applying the dynamic material attenuation value would lower the risk level of the nodes with the same risk. Nodes on the periphery, however, may be relatively less important, resulting in a low sensitivity to numerical values and, consequently, an inaccurate risk level. Iterative risk level calculation can improve accuracy.

[0061] like Less than the preset safe propagation threshold , then the node is considered to have no propagation risk, the iterative calculation is stopped, and the previous level node is selected as the propagation node. If If it is greater than the preset safe propagation threshold, the iterative calculation will continue; in, , all risk levels greater than the safe propagation threshold The nodes are regarded as propagation nodes and integrated into a propagation node set to obtain the influence range.

[0062] As a further improvement of the present invention, the specific steps of the edge processing unit moving to the affected area include: The edge processing unit in the affected area is marked as a response unit. Other edge processing units calculate the closest distance between the current position and the affected area. If the closest distance is lower than the response threshold, it is recorded as a response unit and Search algorithms plan trips to the impact area; The response units in the affected area establish signal connections with each other, and the response unit closest to the risk center node is selected as the safety master unit. The safety master unit divides the affected area into multiple detection areas according to the number of response units, and allocates the nearest detection area according to the location data of each response unit. The response unit goes to the corresponding detection area.

[0063] As a further improvement of the present invention, the specific steps of performing the monitoring and early warning task include: The response unit is equipped with a broadcast module for broadcasting, and the response unit sends evacuation instructions to the construction workers through the broadcast module; The second data collection unit on the response unit collects image data of the current environment in real time, identifies whether there is any equipment in operation in the current environment, and if so, identifies whether there are construction workers. If there are construction workers, the response unit broadcasts to the construction workers to shut down the equipment. If the construction workers do not shut down the equipment within the shutdown broadcast interval, or if there are no construction workers, the response unit shuts down the equipment. The construction workers did not turn off the device during the shutdown broadcast interval. This was for two purposes: one was to allow the construction workers to turn off the device, and the other was that when the response unit had a false alarm, the construction workers could interact with the response unit to turn off the command.

[0064] After the detection and warning task is completed, the collected image data is uploaded to the historical database of the data storage unit.

[0065] Obviously, the above embodiments of the present invention are merely examples for the purpose of clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. Those skilled in the art will appreciate that other variations or modifications can be made based on the above description. It is not necessary and impossible to enumerate all embodiments here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the claims of the present invention.

Claims

1. A construction project safety monitoring and early warning system based on multi-agent collaboration, characterized by: including a plurality of movable edge processing units; A first data collection unit and a second data collection unit are configured to acquire a plurality of sensor data or a plurality of image data and transmit the data to an edge processing unit within a preset transmission range, wherein the second data collection unit is disposed on the edge processing unit; a data storage unit having a construction specification library and a BIM model of a construction project, a BIM compression module for compressing the BIM model and sending it to an edge processing unit, and a distillation module for distilling the construction specification library and sending it to the edge processing unit; The edge processing unit performs risk analysis based on sensor data and / or image data to obtain the risk level and affected area. If the risk level is greater than or equal to the warning threshold, the risk level and affected area are synchronized to other edge processing units, and the edge processing unit goes to the affected area to perform monitoring and early warning tasks.

2. A construction project safety monitoring and early warning system based on multi-agent collaboration according to claim 1, characterized in that: The data storage unit is further provided with a historical database storing historical construction project data, and the historical database is provided with a data normal distribution model that is regularly updated to the first data collection unit and the second data collection unit; The first data collection unit and the second data collection unit are provided with a data classification module, an image mapping module, a sensitivity pulse module and a data transmission module; The image mapping module is used to map image data into image fuzzy data; The sensitivity pulse module is used to send a pulse to the data classification module when the sensor data or image blur data exceeds the corresponding classification warning threshold; The data classification module is used to perform weighted classification calculations on sensor data or image fuzzy data. The expression for weighted classification calculation is: ， ， in, is the calculation of the probability density function of the i-th data value in the normal distribution, is the continuous sensitivity of risk data, is the number of pulses in time T, is the maximum number of pulses within time T, is the pulse calibration factor, is the time calibration factor, t is the interval from the data timestamp to the calculation time; The weight ratio calculation is performed for each weight level. The expression for weight ratio calculation is: ， Data transmission module according to weight ratio Send sensor data or image blur data in descending order to the edge processing unit within the preset transmission range. If the weight ratio of sensor data or image blur data is If the number is lower than the preset redundancy threshold, it will not be sent; When in a non-construction state, the first data collection unit and the second data collection unit send all sensor data or image blur data to the data storage unit.

3. A construction project safety monitoring and early warning system based on multi-agent collaboration according to claim 2, characterized in that: The specific steps of mapping image data into image blurred data are: The first data collection unit and the second data collection unit include a plurality of camera modules for acquiring image data; Using a CNN model to identify construction disaster elements in image data, construction disaster elements include smoke, cracks, fire, and liquids; The construction disaster elements under the same type of camera module have a corresponding basic disaster threshold. , identify the proportion of construction disaster elements in the image data and obtain the disaster ratio , the image blur data is ,in, is the weight coefficient of the j-th construction disaster element.

4. A construction project safety monitoring and early warning system based on multi-agent collaboration according to claim 1, characterized in that: The specific steps of compression include: Filter the BIM model through FilteredElementCollector to obtain multiple components, set the components as nodes, and calculate whether the contour lines between two components intersect. If they intersect, connect the nodes of the two components as edges; Use FilteredElementCollector combined with the built-in category BuiltInCategory to filter out the load-bearing components in the BIM model, integrate the filtered load-bearing components into a load-bearing structure set, and integrate the remaining components in the BIM model into a decorative component set. Check the Structural parameters of each load-bearing component in the load-bearing structure set. If there is a false value or no Structural parameter, move it to the decorative component set. The weight factor of the nodes located in the load-bearing structure is 0.9, and the weight factor of the nodes located in the decorative component is 0.2; Calculate the node importance of each node , ， ， in, is the number of nodes connected to node v, is the set of nodes connected to node v, is the weight factor of the i-th node connected to node v, It is to calculate the proportion of the shortest paths of all nodes passing through node v, are importance adjustment coefficients, is the total number of nodes, s and t are any two nodes that are not equal to each other and are not equal to node v, is the number of nodes v passed through by the shortest path, is the total number of shortest paths, the shortest paths are obtained by Yen's algorithm; Calculating significance thresholds , ， in, is the importance of the entire graph node The average value of is the importance of the entire graph node The standard deviation of is the significance threshold coefficient; If the node importance Below the importance threshold Then delete the node, merge the attribute data of the node into the connected nodes in equal proportion, calculate the new node importance of the merged node, and recalculate the new average importance of the nodes in the entire graph to obtain a compressed topology graph, in which each node has the attribute data of the corresponding BIM model.

5. A construction project safety monitoring and early warning system based on multi-agent collaboration according to claim 3, characterized in that: Compressing the BIM model to obtain a compressed topology graph, wherein each node in the compressed topology graph has attribute data corresponding to the BIM model; The specific steps of distillation include: The rule parser extracts the quadruple of the construction specification library and converts the quadruple into the main specification constraint vector , and the universal canonical constraint vector , ; Extract node feature vectors based on compressed topology graph , ; Extract the canonical constraint vector with the same subject encoding and node feature vector Make associations; The specific steps of sending to the edge processing unit are: compressing the topology map and the general specification constraint vector , and the subject specification constraint vector with associated relationship Sent to the edge processing unit.

6. A construction project safety monitoring and early warning system based on multi-agent collaboration according to claim 5, characterized in that: Both sensor data and image data are labeled with detection time data and location data corresponding to the nodes of the compressed topology graph; The specific steps for performing risk analysis based on sensor data and image data to obtain risk levels include: Convert sensor data or image blur data into real-time detection vectors , ; Calculating construction deviation scores , ； ， ， in, is the constraint weight coefficient corresponding to the subject's normative constraint vector, is the constraint weight coefficient corresponding to the universal specification constraint vector, and n is the number of universal specification constraint vectors; Calculating risk level If the risk level is greater than 0, the corresponding detection time data and location data are marked. ， in, These are the risk grading thresholds from small to large.

7. A construction project safety monitoring and early warning system based on multi-agent collaboration according to claim 1, characterized in that: The first data collection unit includes multiple data collection modules. If the edge processing unit does not receive sensor data or image data from the same data collection module within a preset interruption detection interval, and the edge processing unit is within a preset transmission range of the data collection module, it is considered that the data collection module has an interruption. When a flow interruption occurs, multiple edge processing units within a preset transmission range send interruption confirmations to each other. If all confirm the flow interruption, the edge processing unit closest to the data collection module is selected as the interruption master unit. The main disconnection unit sends connection confirmations to multiple data collection modules within a preset disconnection range, with the data collection module as the center. The data collection modules that do not respond are regarded as disconnection nodes. With the disconnection node as the center, the data collection modules within the preset disconnection range that have not sent connection confirmations send connection confirmations. The data collection modules marked as disconnection nodes are integrated into a disconnection set to obtain the affected range. If there are at least two data collection modules that are disconnected, the risk level is set to the highest level; otherwise, the risk level is set to the third highest level.

8. A construction project safety monitoring and early warning system based on multi-agent collaboration according to claim 6, characterized in that: The attribute data includes material data, and the edge processing unit is preset with a material attenuation coefficient mapping table corresponding to the material data; The specific steps to obtain the impact area include: Extract nodes with risk level greater than 0, and extract the material data and material attenuation coefficient of the node and location data, integrated to form a risk node set ; Select the risk center node , ， in, is the risk level of node i, is the new node importance of node i, is the new average of the importance of nodes in the entire graph; From the risk center node First, perform iterative calculations based on the adjacent relationship. For the first-level adjacent nodes ,like , then the node is not calculated, and the node Adjacent secondary neighbor nodes Calculate, if , then calculate the node Adjacent third-level neighbor nodes Risk level And cover the original risk level of the node, ， in, is the dynamic safety factor, is the height value, is the height scaling factor, is the dynamic material attenuation value, is a node The volume, is the rated attenuation volume, is the material attenuation coefficient; like Less than the preset safe propagation threshold , then the node is considered to have no propagation risk, the iterative calculation is stopped, and the previous level node is selected as the propagation node. If If it is greater than the preset safe propagation threshold, the iterative calculation will continue; in, , all risk levels greater than the safe propagation threshold The nodes are regarded as propagation nodes and integrated into a propagation node set to obtain the influence range.

9. A construction project safety monitoring and early warning system based on multi-agent collaboration according to claim 8, characterized in that: The specific steps for the edge processing unit to move to the affected area include: The edge processing unit in the affected area is marked as a response unit. Other edge processing units calculate the closest distance between the current position and the affected area. If the closest distance is lower than the response threshold, it is recorded as a response unit and Search algorithms plan trips to the impact area; The response units in the affected area establish signal connections with each other, and the response unit closest to the risk center node is selected as the safety master unit. The safety master unit divides the affected area into multiple detection areas according to the number of response units, and allocates the nearest detection area according to the location data of each response unit. The response unit goes to the corresponding detection area.

10. A construction project safety monitoring and early warning system based on multi-agent collaboration according to claim 9, characterized in that: The specific steps for carrying out monitoring and early warning tasks include: The response unit is equipped with a broadcast module for broadcasting, and the response unit sends evacuation instructions to the construction workers through the broadcast module; The second data collection unit on the response unit collects image data of the current environment in real time, identifies whether there is any equipment in operation in the current environment, and if so, identifies whether there are construction workers. If there are construction workers, the response unit broadcasts to the construction workers to shut down the equipment. If the construction workers do not shut down the equipment within the shutdown broadcast interval, or if there are no construction workers, the response unit shuts down the equipment. After the detection and warning task is completed, the collected image data is uploaded to the historical database of the data storage unit.

Images