Large special-shaped steel structure support unloading analysis method based on data processing

Abstract

The invention discloses a large special-shaped steel structure support unloading analysis method based on data processing, and belongs to the field of support unloading, and the method comprises the steps: building a connection topological graph of a special-shaped steel structure by taking a component connection point as a node, dividing the connection topological graph into a plurality of unloading areas based on a division structure during the installation of the special-shaped steel structure, obtaining a connection topological graph marked with unloading area information; marking the position of the temporary support in the connection topological graph marked with the unloading area information, marking the position as a supporting point, and obtaining a support unloading topological graph; according to the support unloading topological graph, by taking the unloading area as a unit, respectively constructing a mapping relation between the support force change of each support point and the deformation of the nodes except the support points; and solving a support unloading strategy by using a deep reinforcement learning network model according to the mapping relation between the support force change of each support point and the deformation of the nodes except the support points. The method solves the problem that the existing unloading method cannot balance the deformation of each node and the total deformation.

Description

Technical Field

[0001] The invention belongs to the field of support unloading, and in particular relates to a large-scale special-shaped steel structure support unloading analysis method based on data processing. Background Art

[0002] During the construction of steel structures, the support system is a key temporary facility to ensure the stability and safety of the structure during the construction phase. When the main construction of the steel structure is completed, it is necessary to carry out support unloading operations to gradually transform the stress state of the structure from relying on temporary supports to relying on its own structural system to bear all the loads.

[0003] Traditional steel structure support unloading methods often rely on the experience and judgment of construction personnel, lacking accurate mechanical analysis and real-time monitoring methods. In the past small and simple steel structure projects, experienced construction personnel were able to make quick judgments and implement unloading operations based on their accumulated experience, without the need for complex equipment and preliminary preparations, and the cost was relatively low. For example, in the construction of small steel structures for self-built houses in rural areas, the construction team can directly carry out support unloading based on their experience in many similar projects, and the operation process is simple. However, it relies heavily on the personal ability of construction personnel and lacks scientific and accurate analysis. There are differences in the experience and judgment of different construction personnel. When facing large and complex structures, it is difficult to accurately grasp the mechanical changes of the structure during the unloading process. Once the judgment is wrong, it is very easy to cause local instability of the structure. For example, in the unloading of some small gymnasium steel structures with unique shapes, due to the misjudgment of experience, the local bending and deformation of the steel beam occurred. At the same time, this method cannot predict the potential risks of the structure in advance, and cannot form a systematic unloading plan.

[0004] Alternatively, through finite element analysis software, the entire unloading process can be simulated before construction, and the deformation and stress distribution of the structure at different unloading stages can be known in advance, so as to optimize the unloading plan and improve the safety and scientificity of unloading. For example, before unloading a super-high-rise steel structure building, the finite element simulation can accurately plan the unloading sequence and value. At the same time, with the help of high-precision sensors for real-time monitoring, structural anomalies can be discovered in time, providing accurate data for construction decisions and ensuring that the unloading process is safe and controllable. During the unloading of steel structures in large stadiums, real-time monitoring can ensure the safety of the structure under complex forces.

[0005] However, the initial investment cost is high, requiring professional finite element analysis software and technicians who are proficient in software operation. The cost of purchasing software licenses and personnel training is high. At the same time, the layout of high-precision sensors and the establishment of data acquisition systems also require a large investment. In addition, simulation analysis relies on accurate model establishment. If the model parameters are not set reasonably, the simulation results will deviate greatly from the actual situation, affecting the reliability of the unloading solution.

[0006] During the unloading process, if the deformation and stress changes of the structure are not accurately grasped, it is easy to cause local instability of the structure, damage to components, and even cause safety accidents of the overall structure. In addition, due to the great differences in the structural form, load distribution and construction conditions of different steel structure projects, the general unloading method is difficult to meet the needs of all projects, which urgently requires a more scientific, accurate and adaptable steel structure support unloading technology. Summary of the invention

[0007] In view of the above-mentioned deficiencies in the prior art, the present invention provides a large-scale special-shaped steel structure support unloading analysis method based on data processing, which solves the problem that the existing unloading method cannot balance the deformation of each node and the total deformation.

[0008] In order to achieve the above-mentioned purpose of the invention, the technical solution adopted by the present invention is: a large-scale special-shaped steel structure support unloading analysis method based on data processing, comprising: A connection topology diagram of the special-shaped steel structure is constructed with the component connection points as nodes, and based on the division structure when the special-shaped steel structure is installed, the connection topology diagram is divided into several unloading areas to obtain a connection topology diagram marked with unloading area information; Mark the position of the temporary support in the connection topology diagram marked with the unloading area information, mark it as a support point, and obtain a support unloading topology diagram; According to the support unloading topology diagram, the mapping relationship between the change of the support force of each support point and the deformation of the nodes other than the support point is constructed respectively with the unloading area as the unit; According to the mapping relationship between the change of supporting force of each supporting point and the deformation of nodes other than the supporting point, the support unloading strategy is solved by using the deep reinforcement learning network model.

[0009] Furthermore, according to the support unloading topology map, the mapping relationship between the change of the support force of each support point and the deformation of the node other than the support point is constructed in units of the unloading area, specifically: According to the support unloading topology diagram, obtain the connection information between each support point and each node in each unloading area:

[0010] in, For the The first support point and the Connection information of each node; For the The first support point and the The shortest path step length on the node connection line; For the The first support point and the The set of nodes that the shortest path on the node connection line passes through; Model the special-shaped steel structure, use simulation technology to simulate the working conditions of the supporting force changes of each supporting point, and obtain the response of each node under different working conditions; According to the connection information between each support point and each node and the response of each node under different working conditions, the mapping relationship between the change of the support force of each support point and the deformation of the nodes except the support point is obtained.

[0011] Furthermore, according to the connection information between each support point and each node and the response of each node under different working conditions, the mapping relationship between the change of the support force of each support point and the deformation of the node other than the support point is obtained, which is specifically: According to the response of each node under different working conditions, the deformation of each node is obtained when the support force of a single support point decreases; According to the connection information between each support point and each node and the deformation of each node when the support force of a single support point decreases, a mapping relationship between the change of the support force of each support point and the deformation of the nodes other than the support point is obtained.

[0012] Furthermore, the expression for the mapping relationship between the change of the supporting force of each supporting point and the deformation of the node other than the supporting point is:

[0013] in, For the The first support point and the The mapping relationship of the deformation of each node; For the When the support force of the first support point decreases, Support point to The shortest path through the nodes nodes; For the When the support force of the first support point decreases, The deformation of a node.

[0014] Furthermore, according to the mapping relationship between the change of the support force of each support point and the deformation of the nodes other than the support point, the support unloading strategy is solved by using the deep reinforcement learning network model, which is specifically: According to the mapping relationship between the change of the support force of each support point and the deformation of the nodes other than the support point, the joint influence of the change of the support force of each support point on the deformation of the nodes other than the support point is solved:

[0015] in, The change of the supporting force of each supporting point The combined influence of the deformation of each node; For the The influence reduction coefficient of the supporting point with changing supporting force; For the The support point of the support force change to the The single-step influence coefficient of each node; For the The supporting point of the first supporting force change is The shortest path step length on the node connection line; For the When the support force of the first support point decreases, The deformation of each node; For the Support point to The single-step influence coefficient of each node; For the When the support force of the first support point decreases, Support point to The node number on the shortest path of nodes; For the Support point to The error correction coefficient of each node; For the When the support force of the first support point decreases, The deformation of each node; According to the joint influence of the change of supporting force of each supporting point on the deformation of nodes other than the supporting point, the special-shaped steel structure is used as the intelligent agent in the deep reinforcement learning network model. The state space, candidate action space and reward of the special-shaped steel structure are constructed respectively. Based on the state space, candidate action space and reward of the special-shaped steel structure, the support unloading strategy is solved.

[0016] Furthermore, the expression of the state space is:

[0017] in, for The state space at the moment; is a set of support points to be unloaded; is the total deformation of the steel structure; for Action at all times The weight of influence; for The final execution action at the moment; for The change of the support force of each support point at the moment The combined influence of the deformation of each node; for The number of the support point unloaded in the final execution action at the moment; For the The influence reduction coefficient of the supporting point with changing supporting force; For the The support point of the support force change to the The single-step influence coefficient of each node; For the The supporting point of the first supporting force change is The shortest path step length on the node connection line.

[0018] Furthermore, the expression of the candidate action space is:

[0019] in, for The candidate action space at time , denoted and All possible combinations; is the number of support points for unloading; To be unloaded from the support point collection Selected Support points; Unload the upper limit for the support point; Unload the maximum value for the batch support point; is the number of support points to be unloaded; For the The deformation of each node; It is the node deformation warning value.

[0020] Furthermore, the method for solving the maximum value of batch support point unloading is: B1. Use dichotomy to select Support points for unloading; The upper limit of the number of support points, the initial value is the total number of support points in the current unloading area; B2. Model the special-shaped steel structure and use simulation technology to select different The support points are unloaded, and the total deformation of each node under each working condition is calculated to obtain the error between the maximum total deformation and the minimum total deformation; under each working condition, the deformation of each node is less than the node deformation warning value; B3. Determine whether the error is within the threshold. If so, As the batch support point unload the maximum value, otherwise, Updated to , and returns B1.

[0021] Furthermore, the reward is expressed as:

[0022]

[0023] in, For reward; is the normalization function; is the upper limit of deformation deviation; for After the final execution action is executed, The deformation of each node; for The mean deformation of each node after the final execution action is executed at the moment; for The total number of nodes at the moment; is the node deformation warning value; To support the objective function of the offloading task; is the first objective, which means minimizing the deformation fluctuation of each node; For the The deformation of each node; is the mean deformation of each node; is the second objective, which means minimizing the total deformation of the steel structure; is the deformation fluctuation weight; is the deformation warning weight; is the deformation weight.

[0024] Furthermore, the support unloading strategy is solved according to the state space, candidate action space and reward of the special-shaped steel structure, specifically: A1. Determine the current unloading area and initialize the Q network, target network, experience pool and state space of the deep reinforcement learning network model; A2. Obtain candidate action space based on the current state space; A3. Use the Q network to calculate the action values ​​of all candidate actions in the candidate action space; A4. According to the action values ​​of all candidate actions, use the greedy strategy to select the final execution action; A5. Execute the final execution action and feedback the reward to obtain the updated state space; A6. According to the current state space , final execution action ,award and the next state space , obtain the state transition sequence, and store the state transition sequence in the experience pool; A7. Select several groups of state transition sequences from the experience pool as training samples; A8. Update the Q network and target network using training samples; A9, return to step A2, until all the support points in the current unloading area are unloaded, and obtain the deep reinforcement learning network model corresponding to the current unloading area, that is, the support point unloading strategy obtaining model of the current unloading area; A10, return to step A1, until all the support points of the unloading area are unloaded, and obtain the unloading strategy model of the support points of each unloading area; A11. Use the support point unloading strategy of each unloading area to obtain the model, and obtain the support unloading strategy of each unloading area respectively.

[0025] The beneficial effects of the present invention are as follows: based on the structural division during installation, through the mapping relationship between the support points and the node deformation, relying on the deep reinforcement learning network, the unloading strategy of each unloading area is solved. The method of unloading by area takes into account the overall balance of the large steel structure. Secondly, the deformation balance of the node is taken into account to avoid significant instability of a certain part of the steel structure after the support is unloaded, thereby ensuring the overall stability of the steel structure. This stability is achieved while ensuring that the support unloading has the least impact on the steel structure, thus achieving a balance between the two. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 The figure is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0027] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0028] like Figure 1 As shown, in one embodiment of the present invention, a large-scale special-shaped steel structure support unloading analysis method based on data processing includes: A connection topology diagram of the special-shaped steel structure is constructed with the component connection points as nodes, and based on the division structure when the special-shaped steel structure is installed, the connection topology diagram is divided into several unloading areas to obtain a connection topology diagram marked with unloading area information; Mark the position of the temporary support in the connection topology diagram marked with the unloading area information, mark it as a support point, and obtain a support unloading topology diagram; According to the support unloading topology diagram, the mapping relationship between the change of the support force of each support point and the deformation of the nodes other than the support point is constructed respectively with the unloading area as the unit; According to the mapping relationship between the change of supporting force of each supporting point and the deformation of nodes other than the supporting point, the support unloading strategy is solved by using the deep reinforcement learning network model.

[0029] In this embodiment, the large-scale special-shaped steel structure is divided into several installation areas during installation and installed separately. When unloading, the unloading can be carried out in different areas based on the area division during installation.

[0030] According to the support unloading topology diagram, the mapping relationship between the change of the support force of each support point and the deformation of the node other than the support point is constructed in units of the unloading area, specifically: According to the support unloading topology diagram, obtain the connection information between each support point and each node in each unloading area:

[0031] in, For the The first support point and the Connection information of each node; For the The first support point and the The shortest path step length on the node connection line; For the The first support point and the The set of nodes that the shortest path on the node connection line passes through; Model the special-shaped steel structure, use simulation technology to simulate the working conditions of the supporting force changes of each supporting point, and obtain the response of each node under different working conditions; According to the connection information between each support point and each node and the response of each node under different working conditions, the mapping relationship between the change of the support force of each support point and the deformation of the nodes except the support point is obtained.

[0032] In this embodiment, the support point is a temporary support point, and the node is a point other than the support point. The shortest path from the support point to the node is the direct influence propagation path from the support point to the node.

[0033] According to the connection information between each support point and each node and the response of each node under different working conditions, the mapping relationship between the change of the support force of each support point and the deformation of the node other than the support point is obtained, which is specifically: According to the response of each node under different working conditions, the deformation of each node is obtained when the support force of a single support point decreases; According to the connection information between each support point and each node and the deformation of each node when the support force of a single support point decreases, a mapping relationship between the change of the support force of each support point and the deformation of the nodes other than the support point is obtained.

[0034] In this embodiment, the mapping relationship between the changes in the support force of each support point and the deformation of the nodes other than the support point is obtained, which describes the deformation of each node in the shortest path between the support point and the node after the unloading of a single support point. Based on this, the fitting formula for the deformation influence propagated from the support point to the node can be obtained.

[0035] The expression for the mapping relationship between the change of the supporting force of each supporting point and the deformation of the nodes other than the supporting point is:

[0036] in, For the The first support point and the The mapping relationship of the deformation of each node; For the When the support force of the first support point decreases, Support point to The shortest path through the nodes nodes; For the When the support force of the first support point decreases, The deformation of a node.

[0037] According to the mapping relationship between the change of the support force of each support point and the deformation of the nodes other than the support point, the support unloading strategy is solved by using the deep reinforcement learning network model, which is specifically: According to the mapping relationship between the change of the support force of each support point and the deformation of the nodes other than the support point, the joint influence of the change of the support force of each support point on the deformation of the nodes other than the support point is solved:

[0038] in, The change of the supporting force of each supporting point The combined influence of the deformation of each node; For the The influence reduction coefficient of the supporting point with changing supporting force; For the The support point of the support force change to the The single-step influence coefficient of each node; For the The supporting point of the first supporting force change is The shortest path step length on the node connection line; For the When the support force of the first support point decreases, The deformation of each node; For the Support point to The single-step influence coefficient of each node; For the When the support force of the first support point decreases, Support point to The node number on the shortest path of nodes; For the Support point to The error correction coefficient of each node; For the When the support force of the first support point decreases, The deformation of each node; According to the joint influence of the change of supporting force of each supporting point on the deformation of nodes other than the supporting point, the special-shaped steel structure is used as the intelligent agent in the deep reinforcement learning network model. The state space, candidate action space and reward of the special-shaped steel structure are constructed respectively. Based on the state space, candidate action space and reward of the special-shaped steel structure, the support unloading strategy is solved.

[0039] In this embodiment, the deformation propagated from the support point to the node affects the fitting formula Able to obtain , because during the support unloading process, after the temporary support is unloaded, the support point will become an ordinary node. Based on the deformation propagation from the support point to the node, the deformation at the support point that is transformed into a node can be given. is 0, which means that The unloading of a support point cannot affect the current node. Then, after the support point becomes a node, it is very direct to get which support points will affect the deformation at the support point that becomes the node: the support points that affect the adjacent nodes of the support point that becomes the node.

[0040] The expression of the state space is:

[0041] in, for The state space at the moment; is a set of support points to be unloaded; is the total deformation of the steel structure; for Action at all times The weight of influence; for The final execution action at the moment; for The change of the support force of each support point at the moment The combined influence of the deformation of each node; for The number of the support point unloaded in the final execution action at the moment; For the The influence reduction coefficient of the supporting point with changing supporting force; For the The support point of the support force change to the The single-step influence coefficient of each node; For the The supporting point of the first supporting force change is The shortest path step length on the node connection line.

[0042] In this embodiment, there are two requirements for the support unloading strategy: ① unloading all temporary supports; ② ensuring the unloading effect.

[0043] The expression of the candidate action space is:

[0044] in, for The candidate action space at time , denoted and All possible combinations; is the number of support points for unloading; To be unloaded from the support point collection Selected Support points; Unload the upper limit for the support point; Unload the maximum value for the batch support point; is the number of support points to be unloaded; For the The deformation of each node; It is the node deformation warning value.

[0045] In this embodiment, since there is a joint influence between the supporting points of the steel structure, the best effect will occur when several supporting points are unloaded at the same time. Therefore, the candidate action space is set as a combination of supporting points.

[0046] The solution method for the maximum value of batch support point unloading is: B1. Use dichotomy to select Support points for unloading; The upper limit of the number of support points, the initial value is the total number of support points in the current unloading area; B2. Model the special-shaped steel structure and use simulation technology to select different The support points are unloaded, and the total deformation of each node under each working condition is calculated to obtain the error between the maximum total deformation and the minimum total deformation; under each working condition, the deformation of each node is less than the node deformation warning value; B3. Determine whether the error is within the threshold. If so, As the batch support point unload the maximum value, otherwise, Updated to , and returns B1.

[0047] In this embodiment, the error between the maximum total deformation and the minimum total deformation is used as the judgment condition for the maximum value of batch support point unloading in order to ensure the balance when selecting the unloading strategy and avoid errors caused by extreme or special circumstances.

[0048] In this embodiment, the maximum value of batch support point unloading is highly dependent on the threshold. Under the current threshold, if a suitable value cannot be found, the threshold needs to be adjusted.

[0049] The reward expression is:

[0050]

[0051] in, For reward; is the normalization function; is the upper limit of deformation deviation; for After the final execution action is executed, The deformation of each node; for The mean deformation of each node after the final execution action is executed at the moment; for The total number of nodes at the moment; is the node deformation warning value; To support the objective function of the offloading task; is the first objective, which means minimizing the deformation fluctuation of each node; For the The deformation of each node; is the mean deformation of each node; is the second objective, which means minimizing the total deformation of the steel structure; is the deformation fluctuation weight; is the deformation warning weight; is the deformation weight.

[0052] In this embodiment, a reward is constructed based on the objective function of the unloading task. The three items in the reward are used to balance the uniformity of the distribution of the deformation of each node after the support point is unloaded, the convergence of the deformation of each node to 0 after the support point is unloaded, and the control of the total deformation after the support point is unloaded. The first goal ensures that there are no particularly unstable nodes, and the deformation of all nodes is uniform. When all nodes change uniformly, the overall stability of the steel structure can be guaranteed to a certain extent; the second goal ensures that the impact of support unloading on the steel structure is minimized. The balance between the first and second goals can make the final unloading strategy optimal.

[0053] The support unloading strategy is solved according to the state space, candidate action space and reward of the special-shaped steel structure, specifically: A1. Determine the current unloading area and initialize the Q network, target network, experience pool and state space of the deep reinforcement learning network model; A2. Obtain candidate action space based on the current state space; A3. Use the Q network to calculate the action values ​​of all candidate actions in the candidate action space; A4. According to the action values ​​of all candidate actions, use the greedy strategy to select the final execution action; A5. Execute the final execution action and feedback the reward to obtain the updated state space; A6. According to the current state space , final execution action ,award and the next state space , obtain the state transition sequence, and store the state transition sequence in the experience pool; A7. Select several groups of state transition sequences from the experience pool as training samples; A8. Update the Q network and target network using training samples; A9, return to step A2, until all the support points in the current unloading area are unloaded, and obtain the deep reinforcement learning network model corresponding to the current unloading area, that is, the support point unloading strategy obtaining model of the current unloading area; A10, return to step A1, until all the support points of the unloading area are unloaded, and obtain the unloading strategy model of the support points of each unloading area; A11. Use the support point unloading strategy of each unloading area to obtain the model, and obtain the support unloading strategy of each unloading area respectively.

[0054] In this embodiment, the whole process of obtaining the unloading strategy is based on the unloading area as a unit.

Claims

1. A large-scale special-shaped steel structure support unloading analysis method based on data processing, characterized in that: include: A connection topology diagram of the special-shaped steel structure is constructed with the component connection points as nodes, and based on the division structure when the special-shaped steel structure is installed, the connection topology diagram is divided into several unloading areas to obtain a connection topology diagram marked with unloading area information; Mark the position of the temporary support in the connection topology diagram marked with the unloading area information, mark it as a support point, and obtain a support unloading topology diagram; According to the support unloading topology diagram, the mapping relationship between the change of the support force of each support point and the deformation of the nodes other than the support point is constructed respectively with the unloading area as the unit; According to the mapping relationship between the change of supporting force of each supporting point and the deformation of nodes other than the supporting point, the support unloading strategy is solved by using the deep reinforcement learning network model.

2. According to claim 1, the large-scale special-shaped steel structure support unloading analysis method based on data processing is characterized in that: According to the support unloading topology diagram, the mapping relationship between the change of the support force of each support point and the deformation of the node other than the support point is constructed in units of the unloading area, specifically: According to the support unloading topology diagram, obtain the connection information between each support point and each node in each unloading area: in, For the The first support point and the Connection information of each node; For the The first support point and the The shortest path step length on the node connection line; For the The first support point and the The set of nodes that the shortest path on the node connection line passes through; Model the special-shaped steel structure, use simulation technology to simulate the working conditions of the supporting force changes of each supporting point, and obtain the response of each node under different working conditions; According to the connection information between each support point and each node and the response of each node under different working conditions, the mapping relationship between the change of the support force of each support point and the deformation of the nodes except the support point is obtained.

3. According to claim 2, the large-scale special-shaped steel structure support unloading analysis method based on data processing is characterized in that: According to the connection information between each support point and each node and the response of each node under different working conditions, the mapping relationship between the change of the support force of each support point and the deformation of the node other than the support point is obtained, which is specifically: According to the response of each node under different working conditions, the deformation of each node is obtained when the support force of a single support point decreases; According to the connection information between each support point and each node and the deformation of each node when the support force of a single support point decreases, a mapping relationship between the change of the support force of each support point and the deformation of the nodes other than the support point is obtained.

4. According to claim 3, the large-scale special-shaped steel structure support unloading analysis method based on data processing is characterized in that: The expression for the mapping relationship between the change of the supporting force of each supporting point and the deformation of the nodes other than the supporting point is: in, For the The first support point and the The mapping relationship of the deformation of each node; For the When the support force of the first support point decreases, Support point to The shortest path through the nodes nodes; For the When the support force of the first support point decreases, The deformation of a node.

5. According to claim 1, the large-scale special-shaped steel structure support unloading analysis method based on data processing is characterized in that: According to the mapping relationship between the change of the support force of each support point and the deformation of the nodes other than the support point, the support unloading strategy is solved by using the deep reinforcement learning network model, which is specifically: According to the mapping relationship between the change of the support force of each support point and the deformation of the nodes other than the support point, the joint influence of the change of the support force of each support point on the deformation of the nodes other than the support point is solved: in, The change of the supporting force of each supporting point The combined influence of the deformation of each node; For the The influence reduction coefficient of the supporting point with changing supporting force; For the The support point of the support force change to the The single-step influence coefficient of each node; For the The supporting point of the first supporting force change is The shortest path step length on the node connection line; For the When the support force of the first support point decreases, The deformation of each node; For the Support point to The single-step influence coefficient of each node; For the When the support force of the first support point decreases, Support point to The node number on the shortest path of nodes; For the Support point to The error correction coefficient of each node; For the When the support force of the first support point decreases, The deformation of each node; According to the joint influence of the change of supporting force of each supporting point on the deformation of nodes other than the supporting point, the special-shaped steel structure is used as the intelligent agent in the deep reinforcement learning network model. The state space, candidate action space and reward of the special-shaped steel structure are constructed respectively. Based on the state space, candidate action space and reward of the special-shaped steel structure, the support unloading strategy is solved.

6. The large-scale special-shaped steel structure support unloading analysis method based on data processing according to claim 5 is characterized in that: The expression of the state space is: in, for The state space at the moment; is a set of support points to be unloaded; is the total deformation of the steel structure; for Action at all times The weight of influence; for The final execution action at the moment; for The change of the support force of each support point at the moment The combined influence of the deformation of each node; for The number of the support point unloaded in the final execution action at the moment; For the The influence reduction coefficient of the supporting point with changing supporting force; For the The support point of the support force change to the The single-step influence coefficient of each node; For the The supporting point of the first supporting force change is The shortest path step length on the node connection line.

7. The large-scale special-shaped steel structure support unloading analysis method based on data processing according to claim 5 is characterized in that: The expression of the candidate action space is: in, for The candidate action space at time , denoted and All possible combinations; is the number of support points for unloading; To be unloaded from the support point collection Selected Support points; Unload the upper limit for the support point; Unload the maximum value for the batch support point; is the number of support points to be unloaded; For the The deformation of each node; It is the node deformation warning value.

8. The large-scale special-shaped steel structure support unloading analysis method based on data processing according to claim 7 is characterized in that: The solution method for the maximum value of batch support point unloading is: B1. Use dichotomy to select Support points for unloading; The upper limit of the number of support points, the initial value is the total number of support points in the current unloading area; B2. Model the special-shaped steel structure and use simulation technology to select different The support points are unloaded, and the total deformation of each node under each working condition is calculated to obtain the error between the maximum total deformation and the minimum total deformation; under each working condition, the deformation of each node is less than the node deformation warning value; B3. Determine whether the error is within the threshold. If so, As the batch support point unload the maximum value, otherwise, Updated to , and returns B1.

9. The large-scale special-shaped steel structure support unloading analysis method based on data processing according to claim 5 is characterized in that: The reward expression is: in, For reward; is the normalization function; is the upper limit of deformation deviation; for After the final execution action is executed, The deformation of each node; for The mean deformation of each node after the final execution action is executed at the moment; for The total number of nodes at the moment; is the node deformation warning value; To support the objective function of the offloading task; is the first objective, which means minimizing the deformation fluctuation of each node; For the The deformation of each node; is the mean deformation of each node; is the second objective, which means minimizing the total deformation of the steel structure; is the deformation fluctuation weight; is the deformation warning weight; is the deformation weight.

10. The large-scale special-shaped steel structure support unloading analysis method based on data processing according to claim 5 is characterized in that: The support unloading strategy is solved according to the state space, candidate action space and reward of the special-shaped steel structure, specifically: A1. Determine the current unloading area and initialize the Q network, target network, experience pool and state space of the deep reinforcement learning network model; A2. Obtain candidate action space based on the current state space; A3. Use the Q network to calculate the action values ​​of all candidate actions in the candidate action space; A4. According to the action values ​​of all candidate actions, use the greedy strategy to select the final execution action; A5. Execute the final execution action and feedback the reward to obtain the updated state space; A6. According to the current state space , final execution action ,award and the next state space , obtain the state transition sequence, and store the state transition sequence in the experience pool; A7. Select several groups of state transition sequences from the experience pool as training samples; A8. Update the Q network and target network using training samples; A9, return to step A2, until all the support points in the current unloading area are unloaded, and obtain the deep reinforcement learning network model corresponding to the current unloading area, that is, the support point unloading strategy obtaining model of the current unloading area; A10, return to step A1, until all the support points of the unloading area are unloaded, and obtain the unloading strategy model of the support points of each unloading area; A11. Use the support point unloading strategy of each unloading area to obtain the model, and obtain the support unloading strategy of each unloading area respectively.

Images