Construction site tower crane layout method and device and terminal equipment

Abstract

The invention discloses a construction site tower crane layout method and device and terminal equipment, and belongs to the field of crane construction. The method comprises the steps that construction state information representing a construction site is obtained; inputting the construction state information into a pre-trained tower crane layout strategy model to iteratively decide a layout scheme of the crane, and extracting a first state space feature representing a construction site space association relationship from the construction state information through a shared backbone network of the tower crane layout strategy model during each iteration decision, based on a decision network of the tower crane layout strategy model, mapping according to the first state space feature to obtain a first optimal deployment action of the crane in the current state until the current layout scheme meets a preset termination condition, and generating a final layout scheme of the crane; and the position of the construction site tower crane is arranged according to the final arrangement scheme. According to the invention, the adaptability of the generated layout scheme to a complex and dynamic construction scene can be improved.

Description

Technical Field

[0001] This application relates to the field of tower crane construction, and in particular to a method, apparatus and terminal equipment for the layout of tower cranes on construction sites. Background Technology

[0002] In the construction of large and complex buildings, tower cranes serve as core vertical transportation equipment. Their quantity, model, and spatial location directly affect the efficiency of material coverage, operational safety, and overall construction progress. A reasonable tower crane layout not only ensures that all material demand points on the construction site are effectively covered, reducing blind spots and repetitive operations, but also avoids interference and collisions between tower cranes, and between tower cranes and buildings or construction obstacles, thereby reducing safety risks and controlling equipment operating costs.

[0003] In the current process of tower crane layout on construction sites, the placement of tower cranes is usually determined based on manual experience, standard parameters, or planning models based on static working conditions. This approach often assumes that the construction site conditions are relatively stable and only completes the layout once at the beginning of construction. It is difficult to effectively cope with complex situations such as changes in the work area, adjustments to the stacking position of components, and dynamic changes in the collaborative operation relationship of multiple tower cranes during the construction phase. At the same time, some existing automated layout methods rely heavily on manually preset evaluation indicators and constraints, which have limited dimensions for expressing the construction site conditions. When faced with highly complex and frequently changing construction conditions, it is difficult to fully reflect the comprehensive impact of tower crane layout on construction efficiency, safety, and space utilization, which can easily lead to insufficient adaptability of the final layout scheme. Summary of the Invention

[0004] This application provides a method, apparatus, and terminal equipment for the layout of tower cranes on construction sites, which can solve the problem that the layout schemes generated by the prior art are not adaptable enough to complex and dynamic construction scenarios.

[0005] This application provides a method for arranging a tower crane on a construction site in some embodiments, including: Obtain construction status information that characterizes the construction site, wherein the construction status information includes an obstacle distribution map, a crane distribution map, and a construction area distribution map; The construction status information is input into a pre-trained tower crane layout strategy model to iteratively decide on the crane layout scheme. During each iteration, the shared backbone network of the tower crane layout strategy model extracts first state space features representing the spatial relationships of the construction site from the construction status information. Based on the decision network of the tower crane layout strategy model, the first optimal deployment action of the crane in the current state is obtained by mapping the first state space features. This process continues until the current layout scheme meets a preset termination condition, generating the final layout scheme of the crane. The tower crane layout strategy model is trained and obtained based on the historical layout state value obtained by processing historical state space features. The locations of the tower cranes on the construction site are arranged according to the final layout plan.

[0006] Compared with existing technologies, the above embodiments have the following beneficial effects: By continuously representing the construction site status as a state feature map during the layout generation stage, and performing multiple rounds of deployment actions based on a pre-trained decision network during the inference process, the generation of the layout scheme no longer depends on static working condition assumptions, but can be gradually adjusted with changes in the construction status, thus providing a foundation for the continuous evolution of the layout scheme under dynamic construction conditions; During the model training stage, by evaluating the value of historical state space features, the training process is guided to optimize the deployment strategy from the perspective of the overall construction stage and long-term effects, thereby suppressing the local optimal deployment strategy that only obtains short-term benefits based on the current construction status, and prompting the decision network to gradually learn more stable and generalizable deployment rules in the evolution of different construction stages, so that the deployment actions based solely on the output of the decision network during the inference stage can also better adapt to complex and dynamically changing construction scenarios, significantly reducing the risk of layout mismatch caused by frequent changes in construction status, and improving the overall adaptability of the final layout scheme.

[0007] Furthermore, the tower crane layout strategy model is trained and obtained based on the historical layout state value obtained after processing the historical state space features, including: The value assessment network based on the tower crane layout strategy model evaluates the characteristics of the historical state space and outputs the historical layout state value, which represents the expected total revenue of the current layout state. The decision network is used to output the second optimal deployment action of the crane in the current layout state based on the historical state space characteristics, and the real-time layout return generated after executing the second optimal deployment action is evaluated based on the task coverage increment and installation cost change. The layout gain evaluation value is calculated based on the historical layout state value and the real-time layout return, and the model parameters of the tower crane layout strategy model are updated using the layout gain evaluation value based on the approximate strategy optimization algorithm.

[0008] Compared to existing technologies, the above embodiments offer the following advantages: By employing state value assessment based on historical state space characteristics and combining it with real-time layout rewards to train the strategy model for approximate strategy optimization, the tower crane layout strategy simultaneously considers the coverage gain, cost changes, and long-term overall layout benefit trends brought about by the current deployment action during the learning process. This training mechanism avoids the strategy making decisions solely based on single-step coverage effects, effectively suppressing locally optimal but overall unreasonable layout behaviors, and enabling the strategy network to gradually develop a global cognitive ability for multi-round deployment processes. Therefore, even relying solely on the deployment actions output by the strategy network during the inference phase, it can maintain good overall coverage efficiency, cost control capabilities, and layout stability in complex construction scenarios.

[0009] Furthermore, the evaluation of the immediate deployment returns based on the task coverage increment and installation cost changes after executing the second optimal deployment action includes: Based on the crane working radius determined by the second optimal deployment action, calculate the overlap area between the crane working radius and the existing crane coverage area at the task requirement point, and generate a negative redundant coverage penalty based on the overlap area. Based on the coverage area corresponding to the second optimal deployment action, identify the number of task requirement points that the second optimal deployment action adds coverage to within the construction site, and generate an effective coverage reward value based on the number of task requirement points. Calculate the physical distance between the deployment position corresponding to the second optimal deployment action and the preset obstacle boundary or the deployed crane base at the construction site. When the physical distance does not meet the preset construction safety distance, generate a collision penalty value. Obtain the model of the crane selected in the second optimal deployment action, and generate a deployment cost penalty value based on the model; The instant layout reward is calculated based on the negative redundancy coverage penalty, the coverage reward value, the collision penalty value, and the deployment cost penalty value.

[0010] Compared to existing technologies, the above embodiments offer the following advantages: by incorporating factors such as coverage overlap, effective new coverage, safety distance, and crane model cost into the calculation of real-time layout rewards, the evaluation results of each deployment action can truly reflect the comprehensive trade-off between efficiency, safety, and economy on the construction site. Under this reward mechanism, the strategy actively avoids redundant coverage and potential collision risks during training, while prioritizing deployment schemes that provide effective task coverage and are cost-effective. This reduces coverage waste and safety hazards commonly found in multi-tower layouts, improving the feasibility and reliability of the final generated layout scheme under actual engineering conditions.

[0011] Furthermore, the value assessment network includes a first fully connected layer, a first activation function, and a second fully connected layer; the value assessment network based on the tower crane layout strategy model evaluates the historical state space characteristics and outputs the historical layout state value representing the expected total revenue of the current layout state, including: The historical state spatial features are input into the first fully connected layer for linear weighted operation to obtain the intermediate layer feature vector representing the spatial location fusion information of the task point, obstacles and deployed cranes. The intermediate layer feature vector is nonlinearly transformed by the first activation function to extract deep evaluation features related to the expected total revenue of the current layout state. The deep evaluation features are input into the second fully connected layer for dimensionality compression to obtain the historical layout state value, which represents the expected total revenue of the current layout state.

[0012] Compared to existing technologies, the above embodiments offer the following advantages: By utilizing a multi-layer fully connected structure to nonlinearly map and compress historical state space features, the value assessment network can extract key features highly correlated with overall returns from complex spatial layout information and output stable state value estimation results. This structure helps reduce evaluation noise introduced by the complexity of construction scenarios and the high state dimension, enabling state values ​​to more accurately reflect the potential return level of the current layout in subsequent deployment processes. This provides a low-variance, reliable value reference for policy updates, improving the stability and convergence efficiency of the reinforcement learning training process.

[0013] Further, the shared backbone network includes: a first convolution operator, a second convolution operator, and a flattening operator; the shared backbone network of the tower crane layout strategy model extracts first state space features representing the spatial relationships of the construction site from the construction state information, including: The obstacle distribution map, the crane distribution map, and the construction area distribution map are overlaid to generate a multi-channel distribution map; The first convolution operator is used to perform sliding sampling on the multi-channel distribution map to extract the local geometric correlation information between the task points and obstacles in the construction site, and the first nonlinear mapping operation is used to remove negative redundant features to obtain the primary layout feature map. The primary layout feature map is aggregated by the second convolution operator to capture the global topological constraint relationship of the crane coverage range between different construction areas, and the saliency of key layout features is enhanced by the second nonlinear mapping operation to obtain a high-dimensional semantic feature vector representing the complete situation of the construction site. The high-dimensional semantic feature vector is arranged by flattening operator to reduce its dimensionality, and the feature matrix with spatial topological structure is transformed into a first state space feature of a preset dimension.

[0014] Compared to existing technologies, the above embodiments have the following advantages: By overlaying obstacle distribution, crane distribution, and construction area distribution across multiple channels, and combining convolutional feature extraction and feature aggregation operations, the model can simultaneously capture local spatial relationships and global topological structures on the construction site. This feature extraction method can effectively characterize the geometric constraints between task points and obstacles, as well as the overall coordination relationships between the coverage areas of multiple cranes, providing a spatially consistent state representation for subsequent deployment decisions, thereby reducing layout misjudgments caused by fragmented or insufficient spatial information.

[0015] Further, the decision network based on the tower crane layout strategy model obtains the first optimal deployment action of the crane in the current state according to the first state space feature mapping, including: The first state space features are subjected to a combination of linear and nonlinear transformations to extract deployment decision feature vectors that are related to the geographical constraints and coverage requirements of the construction site. The layout decision feature vector is mapped to a preset layout action space to generate original action scores corresponding to different candidate deployment actions; The scores of each original action are normalized and mapped to construct a decision probability distribution representing each candidate deployment action in the current construction state; Based on the decision probability distribution, the candidate crane deployment action with the most significant probability response is determined as the first optimal deployment action.

[0016] Compared to existing technologies, the above embodiments have the following advantages: by mapping state space features to layout action space and constructing a probability distribution of candidate deployment actions, the deployment decision-making process can comprehensively evaluate the relative merits of multiple feasible solutions under the current construction state. This decision-making method avoids the problem of rigid selection based on fixed rules or a single scoring function, making the strategy more flexible and robust when facing complex constraints. It is beneficial to select the action with better overall benefits from multiple approximately feasible deployment solutions, thereby improving the adaptability of the layout results to changes in complex construction scenarios.

[0017] Furthermore, the decision network includes a third fully connected layer and a second activation function; the step of performing a linear and nonlinear combination transformation on the first state space features to extract a deployment decision feature vector that relates to the geographical constraints and coverage requirements of the construction site includes: The first state space features are input into the third fully connected layer for linear weighting. By aggregating the spatial distribution weights of each task point within the construction area, a decision intermediate vector containing local crane location clues is obtained. The decision intermediate vector is nonlinearly transformed using the second activation function to simulate the nonlinear interference relationship between different crane coverage areas and the obstruction constraints of geographical barriers, thereby extracting the deployment decision feature vector.

[0018] Compared to existing technologies, the above embodiments have the following advantages: By aggregating the spatial distribution weights of each task point within the construction area through a fully connected layer, and combining this with a nonlinear activation function to model coverage interference relationships and geographical barrier effects, the deployment decision characteristics can reflect the nonlinear interactions between tower crane coverage areas. This approach helps the strategy accurately identify potential interference, occlusion, or collaborative relationships between different deployment locations, thereby avoiding configurations in actual layouts that meet single-point coverage requirements but have poor overall collaboration, and improving space utilization efficiency under multi-tower collaborative operation conditions.

[0019] Furthermore, the acquisition of construction status information characterizing the construction site includes: Obtain a remote sensing map of the construction site, and map the construction site into a grid map of a preset size based on the remote sensing map; Collect building information within each grid cell of the grid map, and generate the obstacle distribution map based on the building information within each grid cell; The deployment of cranes within each grid cell is statistically analyzed, and a crane distribution map is generated based on the crane deployment data. The construction points within each grid cell are counted, and a construction area distribution map is generated based on these construction points.

[0020] Compared to existing technologies, the above embodiments have the following advantages: Based on remote sensing maps, the construction site is uniformly mapped into a gridded spatial model, and obstacle distribution maps, crane distribution maps, and construction area distribution maps are generated accordingly, giving construction status information a unified scale and structured expression. This data modeling method facilitates the transformation of heterogeneous information in complex construction sites into standard inputs suitable for deep neural network processing, thereby reducing human modeling errors, improving the integrity and consistency of status information, and providing a reliable data foundation for subsequent layout decisions and strategy reasoning.

[0021] Another embodiment of this application provides a site tower crane layout device, including: an acquisition module, an iteration module, and a layout module; The acquisition module acquires construction status information representing the construction site, including an obstacle distribution map, a crane distribution map, and a construction area distribution map. The iterative module is used to input the construction status information into a pre-trained tower crane layout strategy model to iteratively decide on the crane layout scheme. During each iteration, the shared backbone network of the tower crane layout strategy model extracts first state space features representing the spatial relationships of the construction site from the construction status information. Based on the decision network of the tower crane layout strategy model, the first optimal deployment action of the crane in the current state is obtained by mapping the first state space features. This process continues until the current layout scheme meets a preset termination condition, generating the final layout scheme of the crane. The tower crane layout strategy model is trained and obtained based on the historical layout state value obtained by processing historical state space features. The layout module is used to arrange the positions of the tower cranes on the construction site according to the final layout scheme.

[0022] Another embodiment of this application also provides a terminal device, including: a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, it implements the steps of the construction site tower crane layout method of this application. Attached Figure Description

[0023] To more clearly illustrate the technical solution of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0024] Figure 1 This is a flowchart illustrating a method for arranging a tower crane on a construction site, provided in some embodiments of this application. Figure 2 This is a schematic diagram of a two-dimensional workspace at a construction site provided in some embodiments of this application; Figure 3 This is an obstacle distribution map provided in some embodiments of this application; Figure 4 This is a crane distribution diagram provided in some embodiments of this application; Figure 5 This is a construction area distribution map provided in some embodiments of this application; Figure 6 This is a structural schematic diagram of a tower crane layout strategy model provided in some embodiments of this application; Figure 7This is a diagram showing the task coverage results under different numbers of buildings after the layout of the construction site tower crane layout method provided in some embodiments of this application; Figure 8 This is a schematic diagram showing the layout results of the construction site tower crane layout method provided in some embodiments of this application after layout; Figure 9 This is a structural schematic diagram of a construction site tower crane layout device provided in some embodiments of this application. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0026] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the application; the terms “comprising” and “having”, and any variations thereof, in the specification, claims, and foregoing description of the drawings are intended to cover non-exclusive inclusion.

[0027] In the description of the embodiments of this application, technical terms such as "first" and "second" are used only to distinguish different objects and should not be construed as indicating or implying relative importance or implicitly specifying the number, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly defined.

[0028] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0029] In the description of the embodiments in this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0030] In the description of the embodiments of this application, the term "multiple" refers to two or more (including two), similarly, "multiple sets" refers to two or more (including two sets), and "multiple pieces" refers to two or more (including two pieces).

[0031] In the description of the embodiments of this application, unless otherwise expressly specified and limited, technical terms such as "installation," "connection," "joining," and "fixing" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. For those skilled in the art, the specific meaning of the above terms in the embodiments of this application can be understood according to the specific circumstances.

[0032] In the current process of tower crane layout on construction sites, the placement of tower cranes is usually determined based on manual experience, standard parameters, or planning models based on static working conditions. This approach often assumes that the construction site conditions are relatively stable and only completes the layout once at the beginning of construction. It is difficult to effectively cope with complex situations such as changes in the work area, adjustments to the stacking position of components, and dynamic changes in the collaborative operation relationship of multiple tower cranes during the construction phase. At the same time, some existing automated layout methods rely heavily on manually preset evaluation indicators and constraints, which have limited dimensions for expressing the construction site conditions. When faced with highly complex and frequently changing construction conditions, it is difficult to fully reflect the comprehensive impact of tower crane layout on construction efficiency, safety, and space utilization, which can easily lead to insufficient adaptability of the final layout scheme.

[0033] Please refer to Figure 1 To address the problem that existing technologies lack adaptability in generating layout schemes for complex and dynamic construction scenarios, this application provides a method for laying out tower cranes on construction sites.

[0034] To better understand the construction environment at the construction site and the tower crane layout problem solved by the embodiments of this application, the following will be combined with... Figure 2 The model shown illustrates the layout problem of tower cranes.

[0035] The tower crane layout problem addressed in this application involves determining the optimal type, quantity, and location of cranes to efficiently transport materials within a construction site while satisfying geometric, operational, and safety constraints. This application focuses on the final tower crane layout scheme, assuming all buildings are completed and no dynamic construction process is involved. Furthermore, this application only considers the placement of tower cranes and not their operating trajectories; therefore, there are no obstacles other than buildings in the analysis. To achieve learning-based optimization, this paper constructs a mathematical model to characterize the core elements of the layout problem, including construction task points, crane operating areas, material supply points, and building obstacles.

[0036] Because of the tower crane layout problem solved by the embodiments of this application, the tower crane must be anchored to the building facade and maintain a fixed distance from the main structure. Therefore, refer to Figure 2 The construction site is modeled as a two-dimensional workspace. The set of candidate deployment locations for the tower crane is denoted as... Each of them All satisfy site boundary and obstacle constraints. Binary decision variables. Indicates whether it is in the candidate position Install cranes. Each tower crane has an operating radius. Maximum lifting capacity And installation costs. At the same time, based on practical design experience, each tower crane maintains a fixed distance from the building. .

[0037] To meet the basic requirements for tower crane layout, the purpose of this application is to maximize the coverage of tower cranes over all buildings while minimizing redundant coverage and overall installation costs. Therefore, the tower crane layout problem solved by this application can be modeled as shown in the following formula.

[0038] ; This issue involves The tower crane task requirements that need to be covered and One candidate installation location. Binary decision variable x j ∈{0,1} indicates whether it is in position Tower crane installation. Parameters Indicate whether the tower crane at position j can cover the task point. The installation cost can be pre-calculated based on spatial constraints. Each candidate location j corresponds to an installation cost. Two weighting coefficients are introduced. and Each control is set to manage the penalties associated with redundant coverage and installation costs.

[0039] Furthermore, the tower crane layout problem solved by the embodiments of this application includes at least constraints to avoid collisions between crane installation positions, as shown in the following formula.

[0040] ; Each candidate tower crane location is associated with a coordinate. , In the two-dimensional space of construction, a minimum safe distance must be maintained between any two cranes to prevent collisions. If two cranes are installed at the same location... and (Right now Then the Euclidean distance between their positions must be no less than 1 / 3. .

[0041] Thus, the tower crane layout problem to be solved in this application embodiment has been explained through mathematical modeling. To achieve intelligent and autonomous tower crane layout, this application embodiment constructs a reinforcement learning environment based on a low-level mathematical model. In this environment, the agent learns the optimal placement strategy through trial-and-error interaction with a simulated construction site. This environment is carefully designed to reflect the characteristics of tower crane planning, including the definitions of state space, action space, reward function, and event termination conditions. Accordingly, the tower crane layout problem is modeled as a Markov decision process.

[0042] Furthermore, the present application provides a method for arranging a tower crane on a construction site, including the following steps S101 to S103: S101: Obtain construction status information representing the construction site, wherein the construction status information includes an obstacle distribution map, a crane distribution map, and a construction area distribution map.

[0043] Furthermore, in some embodiments of this application, obtaining construction status information characterizing the construction site includes: Obtain a remote sensing map of the construction site, and map the construction site into a grid map of a preset size based on the remote sensing map; Collect building information within each grid cell of the grid map, and generate the obstacle distribution map based on the building information within each grid cell; The deployment of cranes within each grid cell is statistically analyzed, and a crane distribution map is generated based on the crane deployment data. The construction points within each grid cell are counted, and a construction area distribution map is generated based on these construction points.

[0044] Preferably, refer to Figure 3The obstacle distribution map shown uses buildings as obstacles to form a binary grid map, where grid cells covered by buildings have a value of 1 and grid cells not covered by buildings have a value of 0, resulting in the final obstacle distribution map.

[0045] Preferably, refer to Figure 4 The crane distribution map shown is obtained by setting the value of the grid cells with installed crane towers to 1 and the value of the grid cells in other areas to 0.

[0046] Preferably, refer to Figure 5 The construction area distribution map shown has grid cells in the construction area set to 1, while grid cells in other areas are set to 0, thus obtaining the construction area distribution map.

[0047] Based on remote sensing maps, the construction site is uniformly mapped into a gridded spatial model, and obstacle distribution maps, crane distribution maps, and construction area distribution maps are generated accordingly, giving construction status information a unified scale and structured expression. This data modeling method is beneficial for transforming heterogeneous information in complex construction sites into standard inputs suitable for deep neural network processing, thereby reducing human modeling errors, improving the completeness and consistency of status information, and providing a reliable data foundation for subsequent layout decisions and strategy reasoning.

[0048] S102: The construction status information is input into a pre-trained tower crane layout strategy model to iteratively decide the crane layout scheme. During each iteration, the shared backbone network of the tower crane layout strategy model extracts the first state space features representing the spatial relationship of the construction site from the construction status information. Based on the decision network of the tower crane layout strategy model, the first optimal deployment action of the crane in the current state is obtained by mapping the first state space features. This continues until the current layout scheme meets the preset termination condition, generating the final layout scheme of the crane. The tower crane layout strategy model is trained and obtained based on the historical layout state value obtained after processing the historical state space features.

[0049] Furthermore, in some embodiments of this application, the tower crane layout strategy model is trained and obtained based on the historical layout state value obtained after processing the historical state space features, including: The value assessment network based on the tower crane layout strategy model evaluates the characteristics of the historical state space and outputs the historical layout state value, which represents the expected total revenue of the current layout state. The decision network is used to output the second optimal deployment action of the crane in the current layout state based on the historical state space characteristics, and the real-time layout return generated after executing the second optimal deployment action is evaluated based on the task coverage increment and installation cost change. The layout gain evaluation value is calculated based on the historical layout state value and the real-time layout return, and the model parameters of the tower crane layout strategy model are updated using the layout gain evaluation value based on the approximate strategy optimization algorithm.

[0050] Preferably, in some embodiments of the application, the step of calculating the layout gain evaluation value based on the historical layout state value and the immediate layout return, and updating the model parameters of the tower crane layout strategy model using the layout gain evaluation value based on the approximate strategy optimization algorithm includes: using the historical layout state value output by the value assessment network to approximate the expected return, providing a low variance benchmark for the advantage calculation of the truncation target of the approximate strategy optimization algorithm.

[0051] This application employs state value assessment based on historical state space characteristics and combines it with real-time layout rewards to train the strategy model for approximate strategy optimization. This allows the tower crane layout strategy to simultaneously consider the coverage gain, cost changes, and long-term benefit trends of the overall layout during the learning process. This training mechanism avoids the strategy making decisions solely based on single-step coverage effects, effectively suppressing locally optimal but overall unreasonable layout behaviors, and enabling the strategy network to gradually develop a global cognitive ability for multi-round deployment processes. Therefore, even relying solely on the deployment actions output by the strategy network during the inference phase, it can maintain good overall coverage efficiency, cost control capabilities, and layout stability in complex construction scenarios.

[0052] Furthermore, in some embodiments of this application, the evaluation of the immediate deployment return based on the task coverage increment and installation cost change after performing the second optimal deployment action includes: Based on the crane working radius determined by the second optimal deployment action, calculate the overlap area between the crane working radius and the existing crane coverage area at the task requirement point, and generate a negative redundant coverage penalty based on the overlap area. Based on the coverage area corresponding to the second optimal deployment action, identify the number of task requirement points that the second optimal deployment action adds coverage to within the construction site, and generate an effective coverage reward value based on the number of task requirement points. Calculate the physical distance between the deployment position corresponding to the second optimal deployment action and the preset obstacle boundary or the deployed crane base at the construction site. When the physical distance does not meet the preset construction safety distance, generate a collision penalty value. Obtain the model of the crane selected in the second optimal deployment action, and generate a deployment cost penalty value based on the model; The instant layout reward is calculated based on the negative redundancy coverage penalty, the coverage reward value, the collision penalty value, and the deployment cost penalty value.

[0053] Preferably, in some embodiments of this application, the instantaneous layout reward is the reward value in reinforcement learning. Therefore, a reasonable reward function can promote an ideal layout strategy and guide the policy network to achieve efficient learning. Instantaneous layout rewards can encourage full coverage of all building areas while penalizing redundant coverage, excessively high installation costs, and unsafe crane configurations. The formula for calculating the instantaneous layout reward is shown below.

[0054] ; in, for Real-time strategic returns under time-step; for The coverage improvement value is calculated based on the number of newly covered task requirement points at each time step; for Coverage at each time step; This represents the increment of the total installation cost; The penalty for violating the minimum safe distance requirement for cranes reflects the actual construction safety requirements and ensures that the learned strategies avoid dangerous configurations. The overlapping area between the crane's working radius and the existing crane coverage area at the task requirement point; exist At each time step, the value is 1 if the deployment action selected by the decision network is invalid, and 0 otherwise. This helps the strategy learn quickly, thereby eliminating inefficient or infeasible actions. , which is a weighting coefficient used to control the contribution of each item in the real-time layout return calculation process.

[0055] Factors such as coverage overlap, effective new coverage, safety distance, and crane model cost are all incorporated into the calculation of real-time layout returns. This ensures that the evaluation results of each deployment action truly reflect the comprehensive trade-off between efficiency, safety, and economy on the construction site. Under this return mechanism, the strategy will proactively avoid redundant coverage and potential collision risks during training, while prioritizing deployment schemes that provide effective task coverage and are cost-effective. This reduces coverage waste and safety hazards commonly found in multi-tower layouts, improving the feasibility and reliability of the final generated layout scheme under actual engineering conditions.

[0056] Furthermore, in some embodiments of this application, the shared backbone network includes: a first convolution operator, a second convolution operator, and a flattening operator; the shared backbone network of the tower crane layout strategy model extracts first state space features representing the spatial relationships of the construction site from the construction state information, including: The obstacle distribution map, the crane distribution map, and the construction area distribution map are overlaid to generate a multi-channel distribution map; The first convolution operator is used to perform sliding sampling on the multi-channel distribution map to extract the local geometric correlation information between the task points and obstacles in the construction site, and the first nonlinear mapping operation is used to remove negative redundant features to obtain the primary layout feature map. The primary layout feature map is aggregated by the second convolution operator to capture the global topological constraint relationship of the crane coverage range between different construction areas, and the saliency of key layout features is enhanced by the second nonlinear mapping operation to obtain a high-dimensional semantic feature vector representing the complete situation of the construction site. The high-dimensional semantic feature vector is arranged by flattening operator to reduce its dimensionality, and the feature matrix with spatial topological structure is transformed into a first state space feature of a preset dimension.

[0057] Preferably, refer to Figure 6 The first convolution operator includes a first convolutional layer and a third activation function, wherein the first convolutional layer is Conv2d (two-dimensional convolution) and the third activation function is ReLU (activation function); the second convolution operator includes a second convolutional layer and a fourth activation function, wherein the second convolutional layer is Conv2d (two-dimensional convolution) and the fourth activation function is ReLU (activation function).

[0058] This application employs multi-channel overlay of obstacle distribution, crane distribution, and construction area distribution, combined with convolutional feature extraction and feature aggregation operations, enabling the model to simultaneously capture local spatial relationships and global topological structure within the construction site. This feature extraction method effectively characterizes the geometric constraints between task points and obstacles, as well as the overall coordination relationships between the coverage areas of multiple cranes, providing a spatially consistent state representation for subsequent deployment decisions. This reduces layout misjudgments caused by fragmented or insufficient spatial information.

[0059] Furthermore, in some embodiments of this application, the value evaluation network includes a first fully connected layer, a first activation function, and a second fully connected layer; the value evaluation network based on the tower crane layout strategy model evaluates the historical state space characteristics and outputs a historical layout state value representing the expected total revenue of the current layout state, including: The historical state spatial features are input into the first fully connected layer for linear weighted operation to obtain the intermediate layer feature vector representing the spatial location fusion information of the task point, obstacles and deployed cranes. The intermediate layer feature vector is nonlinearly transformed by the first activation function to extract deep evaluation features related to the expected total revenue of the current layout state. The deep evaluation features are input into the second fully connected layer for dimensionality compression to obtain the historical layout state value, which represents the expected total revenue of the current layout state.

[0060] Preferably, refer to Figure 6 During training, both the value evaluation network and the policy network simultaneously receive historical state space features extracted from historical construction status information by the shared backbone network. The value evaluation network then sequentially passes these historical state space features through a first fully connected layer, a first activation function, and a second fully connected layer to output the historical layout state value. The first activation function employs the ReLU activation function. This feature-sharing mechanism reduces the number of parameters, provides implicit regularization, and improves data efficiency and training stability under diverse layouts.

[0061] By employing a multi-layer fully connected structure to nonlinearly map and compress historical state space features, the value assessment network can extract key features highly correlated with overall benefits from complex spatial layout information and output stable state value estimates. This structure helps reduce assessment noise introduced by the complexity of construction scenarios and the high state dimensionality, enabling state values ​​to more accurately reflect the potential benefit level of the current layout in subsequent deployments. This provides a low-variance, reliable value reference for policy updates, improving the stability and convergence efficiency of the reinforcement learning training process.

[0062] Furthermore, in some embodiments of this application, the decision network based on the tower crane layout strategy model obtains the first optimal deployment action of the crane in the current state according to the first state space feature mapping, including: The first state space features are subjected to a combination of linear and nonlinear transformations to extract deployment decision feature vectors that are related to the geographical constraints and coverage requirements of the construction site. The layout decision feature vector is mapped to a preset layout action space to generate original action scores corresponding to different candidate deployment actions; The scores of each original action are normalized and mapped to construct a decision probability distribution representing each candidate deployment action in the current construction state; Based on the decision probability distribution, the candidate crane deployment action with the most significant probability response is determined as the first optimal deployment action.

[0063] Understandably, in a reinforcement learning environment, each action of the agent is taken from a predefined set. The candidate positions for crane deployment are selected sequentially, so the layout action space of the agent is as shown in the following formula.

[0064] ; in, To create space for movement; This means that the candidate deployment action will deploy to the candidate location. As the final deployment location for the heavy machinery; This means skipping the current building and not setting up a crane for it. This represents the total number of candidate deployment locations. The decision network of the tower crane layout strategy model is used to generate the decision probability distribution of each candidate deployment action in the action space under the current construction state.

[0065] By mapping state-space features to the layout action space and constructing a probability distribution of candidate deployment actions, the deployment decision-making process can comprehensively evaluate the relative merits of multiple feasible options under the current construction state. This decision-making approach avoids the problem of rigid selection based on fixed rules or a single scoring function, making the strategy more flexible and robust when facing complex constraints. It is beneficial to select the action with better overall benefits from multiple approximately feasible deployment options, thereby improving the adaptability of the layout results to changes in complex construction scenarios.

[0066] Furthermore, in some embodiments of this application, the decision network includes a third fully connected layer and a second activation function; the step of performing a linear and nonlinear combination transformation on the first state space features to extract a deployment decision feature vector that relates to the geographical constraints and coverage requirements of the construction site includes: The first state space features are input into the third fully connected layer for linear weighting. By aggregating the spatial distribution weights of each task point within the construction area, a decision intermediate vector containing local crane location clues is obtained. The decision intermediate vector is nonlinearly transformed using the second activation function to simulate the nonlinear interference relationship between different crane coverage areas and the obstruction constraints of geographical barriers, thereby extracting the deployment decision feature vector.

[0067] Preferably, in some embodiments of this application, reference is made to Figure 6 After the tower crane layout strategy model is trained, during the inference process, the parameters of the value assessment network are frozen, and inference is performed only through the shared backbone network and the decision network. The shared backbone network receives the current construction status information and extracts the first state space features based on the construction status information. Then, the decision network sequentially passes the first state space features through the third fully connected layer, the second activation function, the fourth fully connected layer, and the normalization layer to output the first optimal deployment action in the current state. Among them, the fourth fully connected layer maps the layout decision feature vector to a preset layout action space to generate the original action scores corresponding to different candidate deployment actions; the second activation function adopts the ReLU activation function; and the normalization layer adopts the Softmax function.

[0068] This application aggregates the spatial distribution weights of each task point within the construction area through a fully connected layer and models coverage interference relationships and geographical barrier effects using a nonlinear activation function. This enables deployment decision characteristics to reflect the nonlinear interactions between tower crane coverage areas. This approach helps the strategy accurately identify potential interference, occlusion, or collaborative relationships between different deployment locations, thereby avoiding configurations that meet single-point coverage requirements but exhibit poor overall collaboration in actual layouts, and improving space utilization efficiency under multi-tower collaborative operation conditions.

[0069] S103: Arrange the positions of the tower cranes on the construction site according to the final layout scheme.

[0070] To further illustrate the effectiveness of the construction site tower crane layout method provided in this application embodiment, a specific example will be used for further explanation below. In this example, the training parameters during the model training process are set as follows: learning rate of 0.0003, discount factor of 10.99, pruning ratio of approximate strategy optimization algorithm of 0.2, update rounds of 1000, and mini-batch size of 32.

[0071] refer to Figure 7 The diagram illustrates the relationship between the number of buildings and the percentage of tower crane coverage, achieved by the proposed algorithm. The horizontal axis represents the number of buildings (ranging from 5 to 10), while the vertical axis displays the corresponding average coverage percentage. The results show that the construction site tower crane layout method provided in this application maintains high performance under different site densities. The coverage is highest (99.55%) when the number of buildings is 6, and lowest (96.53%) when the number of buildings is 10. The coverage decreases slightly after the number of buildings exceeds 8, which may be due to increased spatial complexity and overlapping constraints.

[0072] refer to Figure 8 The tower crane layout scheme for nine buildings demonstrates efficient spatial distribution. The selected crane types and their corresponding locations together achieve approximately 96% coverage of the total building area. The overall layout is balanced and rational, allowing each building to be independently serviced. Furthermore, the overlapping coverage areas in critical work zones effectively reduce the risk of single points of failure and significantly improve the system reliability of important areas.

[0073] In summary, the construction site tower crane layout method provided in this application has the following advantages compared with the prior art: By continuously representing the construction site state as a state feature map during the layout generation stage, and performing multiple rounds of deployment actions based on a pre-trained decision network during the inference process, the generation of the layout scheme no longer depends on static working condition assumptions, but can be gradually adjusted with changes in the construction state, thus providing a basis for the continuous evolution of the layout scheme under dynamic construction conditions; During the model training stage, by evaluating the value of historical state space features, the training process is guided to optimize the deployment strategy from the perspective of the overall construction stage and long-term effects, thereby suppressing the local optimal deployment strategy that only obtains short-term benefits based on the current construction state, and prompting the decision network to gradually learn more stable and generalizable deployment rules in the evolution of different construction stages, so that the deployment actions based solely on the output of the decision network during the inference stage can also better adapt to complex and dynamically changing construction scenarios, significantly reducing the risk of layout mismatch caused by frequent changes in construction state, and improving the overall adaptability of the final layout scheme.

[0074] like Figure 9 As shown, based on the above method embodiments, an embodiment of this application provides a construction site tower crane layout device, including: an acquisition module 201, an iteration module 202, and a layout module 203; The acquisition module 201 acquires construction status information representing the construction site, wherein the construction status information includes an obstacle distribution map, a crane distribution map, and a construction area distribution map. The iterative module 202 is used to input the construction status information into a pre-trained tower crane layout strategy model to iteratively decide on the crane layout scheme. During each iteration, the shared backbone network of the tower crane layout strategy model extracts first state space features representing the spatial relationships of the construction site from the construction status information. Based on the decision network of the tower crane layout strategy model, the first optimal deployment action of the crane in the current state is obtained by mapping the first state space features. This process continues until the current layout scheme meets a preset termination condition, generating the final layout scheme of the crane. The tower crane layout strategy model is trained and obtained based on the historical layout state value obtained after processing historical state space features. The layout module 203 is used to arrange the positions of the construction site tower cranes according to the final layout scheme.

[0075] Furthermore, in some embodiments of this application, the tower crane layout strategy model is trained and obtained based on the historical layout state value obtained after processing the historical state space features, including: The value assessment network based on the tower crane layout strategy model evaluates the characteristics of the historical state space and outputs the historical layout state value, which represents the expected total revenue of the current layout state. The decision network is used to output the second optimal deployment action of the crane in the current layout state based on the historical state space characteristics, and the real-time layout return generated after executing the second optimal deployment action is evaluated based on the task coverage increment and installation cost change. The layout gain evaluation value is calculated based on the historical layout state value and the real-time layout return, and the model parameters of the tower crane layout strategy model are updated using the layout gain evaluation value based on the approximate strategy optimization algorithm.

[0076] Furthermore, in some embodiments of this application, the evaluation of the immediate deployment return based on the task coverage increment and installation cost change after performing the second optimal deployment action includes: Based on the crane working radius determined by the second optimal deployment action, calculate the overlap area between the crane working radius and the existing crane coverage area at the task requirement point, and generate a negative redundant coverage penalty based on the overlap area. Based on the coverage area corresponding to the second optimal deployment action, identify the number of task requirement points that the second optimal deployment action adds coverage to within the construction site, and generate an effective coverage reward value based on the number of task requirement points. Calculate the physical distance between the deployment position corresponding to the second optimal deployment action and the preset obstacle boundary or the deployed crane base at the construction site. When the physical distance does not meet the preset construction safety distance, generate a collision penalty value. Obtain the model of the crane selected in the second optimal deployment action, and generate a deployment cost penalty value based on the model; The instant layout reward is calculated based on the negative redundancy coverage penalty, the coverage reward value, the collision penalty value, and the deployment cost penalty value.

[0077] Furthermore, in some embodiments of this application, the value evaluation network includes a first fully connected layer, a first activation function, and a second fully connected layer; the value evaluation network based on the tower crane layout strategy model evaluates the historical state space characteristics and outputs a historical layout state value representing the expected total revenue of the current layout state, including: The historical state spatial features are input into the first fully connected layer for linear weighted operation to obtain the intermediate layer feature vector representing the spatial location fusion information of the task point, obstacles and deployed cranes. The intermediate layer feature vector is nonlinearly transformed by the first activation function to extract deep evaluation features related to the expected total revenue of the current layout state. The deep evaluation features are input into the second fully connected layer for dimensionality compression to obtain the historical layout state value, which represents the expected total revenue of the current layout state.

[0078] Further, in some embodiments of this application, the shared backbone network includes: a first convolution operator, a second convolution operator, and a flattening operator; the iteration module 202 includes: a channel overlay unit, a first convolution unit, a second convolution unit, and a flattening unit; the iteration module 202 is used to extract first state space features representing the spatial correlation of the construction site from the construction state information through the shared backbone network of the tower crane layout strategy model, including: The channel overlay unit is used to overlay the obstacle distribution map, the crane distribution map, and the construction area distribution map to generate a multi-channel distribution map. The first convolutional unit is used to perform sliding sampling on the multi-channel distribution map through the first convolutional operator to extract local geometric correlation information between task points and obstacle distribution in the construction site, and to remove negative redundant features through the first nonlinear mapping operation to obtain a primary layout feature map. The second convolutional unit is used to perform feature aggregation on the primary layout feature map through the second convolutional operator to capture the global topological constraint relationship of the crane coverage range between different construction areas, and to enhance the saliency of key layout features through the second nonlinear mapping operation to obtain a high-dimensional semantic feature vector representing the complete situation of the construction site. The flattening unit is used to perform dimensionality reduction and arrangement of the high-dimensional semantic feature vector through a flattening operator, transforming the feature matrix with spatial topological structure into a first state space feature of a preset dimension.

[0079] Further, in some embodiments of this application, the iteration module 202 includes: a feature transformation unit, a feature mapping unit, a normalization unit, and a decision selection unit; the iteration module 202 is used to obtain the first optimal deployment action of the crane in the current state based on the decision network of the tower crane layout strategy model according to the first state space feature mapping, including: The feature transformation unit is used to perform linear and nonlinear combination transformations on the first state space features to extract deployment decision feature vectors that are associated with the geographical constraints and coverage requirements of the construction site. The feature mapping unit is used to map the layout decision feature vector to a preset layout action space to generate original action scores corresponding to different candidate deployment actions. The normalization unit is used to normalize and map the scores of each of the original actions to construct a decision probability distribution representing each of the candidate deployment actions in the current construction state. The decision selection unit is used to determine the candidate crane deployment action with the most significant probability response based on the decision probability distribution, as the first optimal deployment action.

[0080] Furthermore, in some embodiments of this application, the decision network includes a third fully connected layer and a second activation function; the feature transformation unit is used to perform a linear and nonlinear combination transformation on the first state space features to extract a deployment decision feature vector that relates to the geographical constraints and coverage requirements of the construction site, including: The first state space features are input into the third fully connected layer for linear weighting. By aggregating the spatial distribution weights of each task point within the construction area, a decision intermediate vector containing local crane location clues is obtained. The decision intermediate vector is nonlinearly transformed using the second activation function to simulate the nonlinear interference relationship between different crane coverage areas and the obstruction constraints of geographical barriers, thereby extracting the deployment decision feature vector.

[0081] Further, in some embodiments of this application, the acquisition module 201 includes: a grid division unit, a first map generation unit, a second map generation unit, and a third map generation unit; the acquisition module 201 is used to acquire construction status information representing the construction site, including: The grid division unit is used to acquire a remote sensing map of the construction site and map the construction site into a grid map of a preset size based on the remote sensing map; The first map generation unit is used to collect building information in each grid cell of each grid map and generate the obstacle distribution map based on the building information in each grid cell. The second map generation unit is used to statistically analyze the crane deployment in each grid cell and generate the crane distribution map based on the crane deployment. The third map generation unit is used to count the construction points within each grid unit and generate the construction area distribution map based on the construction points.

[0082] In summary, the construction site tower crane layout device provided in this application has the following advantages compared with the prior art: By continuously representing the construction site state as a state feature map during the layout generation stage, and performing multiple rounds of deployment actions based on a pre-trained decision network during the inference process, the generation of the layout scheme no longer depends on static working condition assumptions, but can be gradually adjusted with changes in the construction state, thus providing a basis for the continuous evolution of the layout scheme under dynamic construction conditions; During the model training stage, by evaluating the value of historical state space features, the training process is guided to optimize the deployment strategy from the perspective of the overall construction stage and long-term effects, thereby suppressing the local optimal deployment strategy that only obtains short-term benefits based on the current construction state, and prompting the decision network to gradually learn more stable and generalizable deployment rules in the evolution of different construction stages, so that the deployment actions based solely on the output of the decision network during the inference stage can also better adapt to complex and dynamically changing construction scenarios, significantly reducing the risk of layout mismatch caused by frequent changes in construction state, and improving the overall adaptability of the final layout scheme.

[0083] It is understood that the above-described device embodiments correspond to the method embodiments of this application, and can implement the construction site tower crane layout method provided by any of the above-described method embodiments of this application.

[0084] It should be noted that the device embodiments described above are merely illustrative, and some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided in this application, the connection relationships between modules indicate that they have communication connections, which can specifically be implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0085] Based on the above embodiments of the construction site tower crane layout method, another embodiment of this application provides a terminal device, which includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the construction site tower crane layout method of any embodiment of this application.

[0086] For example, in this embodiment, the computer program can be divided into one or more modules, which are stored in the memory and executed by the processor to complete this application. The one or more module units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the terminal device.

[0087] The terminal device may be a desktop computer, laptop, handheld computer, or cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0088] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the terminal device, connecting all parts of the terminal device via various interfaces and lines.

[0089] Based on the above-described method embodiments, another embodiment of this application provides a computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to execute the construction site tower crane layout method described in any of the above-described method embodiments of this application.

[0090] The modules / units integrated in the device / terminal equipment, if implemented as software functional units and sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

Claims

1. A method for arranging tower cranes on construction sites, characterized in that, include: Obtain construction status information that characterizes the construction site, wherein the construction status information includes an obstacle distribution map, a crane distribution map, and a construction area distribution map; The construction status information is input into a pre-trained tower crane layout strategy model to iteratively decide on the crane layout scheme. During each iteration, the shared backbone network of the tower crane layout strategy model extracts first state space features representing the spatial relationships of the construction site from the construction status information. Based on the decision network of the tower crane layout strategy model, the first optimal deployment action of the crane in the current state is obtained by mapping the first state space features. This process continues until the current layout scheme meets a preset termination condition, generating the final layout scheme of the crane. The tower crane layout strategy model is trained and obtained based on the historical layout state value obtained by processing historical state space features. The locations of the tower cranes on the construction site are arranged according to the final layout plan.

2. The method for arranging a tower crane on a construction site as described in claim 1, characterized in that, The tower crane layout strategy model is trained and obtained based on the historical layout state value obtained after processing the historical state space features, including: The value assessment network based on the tower crane layout strategy model evaluates the characteristics of the historical state space and outputs the historical layout state value, which represents the expected total revenue of the current layout state. The decision network is used to output the second optimal deployment action of the crane in the current layout state based on the historical state space characteristics, and the real-time layout return generated after executing the second optimal deployment action is evaluated based on the task coverage increment and installation cost change. The layout gain evaluation value is calculated based on the historical layout state value and the real-time layout return, and the model parameters of the tower crane layout strategy model are updated using the layout gain evaluation value based on the approximate strategy optimization algorithm.

3. The method for arranging a tower crane on a construction site as described in claim 2, characterized in that, The evaluation of the immediate deployment returns based on the task coverage increment and installation cost changes after executing the second optimal deployment action includes: Based on the crane working radius determined by the second optimal deployment action, calculate the overlap area between the crane working radius and the existing crane coverage area at the task requirement point, and generate a negative redundant coverage penalty based on the overlap area. Based on the coverage area corresponding to the second optimal deployment action, identify the number of task requirement points that the second optimal deployment action adds coverage to within the construction site, and generate an effective coverage reward value based on the number of task requirement points. Calculate the physical distance between the deployment position corresponding to the second optimal deployment action and the preset obstacle boundary or the deployed crane base at the construction site. When the physical distance does not meet the preset construction safety distance, generate a collision penalty value. Obtain the model of the crane selected in the second optimal deployment action, and generate a deployment cost penalty value based on the model; The instant layout reward is calculated based on the negative redundancy coverage penalty, the coverage reward value, the collision penalty value, and the deployment cost penalty value.

4. The method for arranging a tower crane on a construction site as described in claim 2, characterized in that, The value assessment network includes a first fully connected layer, a first activation function, and a second fully connected layer. The value assessment network based on the tower crane layout strategy model evaluates the historical state space characteristics and outputs a historical layout state value representing the expected total revenue of the current layout state, including: The historical state spatial features are input into the first fully connected layer for linear weighted operation to obtain the intermediate layer feature vector representing the spatial location fusion information of the task point, obstacles and deployed cranes. The intermediate layer feature vector is nonlinearly transformed by the first activation function to extract deep evaluation features related to the expected total revenue of the current layout state. The deep evaluation features are input into the second fully connected layer for dimensionality compression to obtain the historical layout state value, which represents the expected total revenue of the current layout state.

5. The method for arranging a tower crane on a construction site as described in claim 1, characterized in that, The shared backbone network includes: a first convolution operator, a second convolution operator, and a flattening operator; the shared backbone network of the tower crane layout strategy model extracts first state space features representing the spatial relationships of the construction site from the construction state information, including: The obstacle distribution map, the crane distribution map, and the construction area distribution map are overlaid to generate a multi-channel distribution map; The first convolution operator is used to perform sliding sampling on the multi-channel distribution map to extract the local geometric correlation information between the task points and obstacles in the construction site, and the first nonlinear mapping operation is used to remove negative redundant features to obtain the primary layout feature map. The primary layout feature map is aggregated by the second convolution operator to capture the global topological constraint relationship of the crane coverage range between different construction areas, and the saliency of key layout features is enhanced by the second nonlinear mapping operation to obtain a high-dimensional semantic feature vector representing the complete situation of the construction site. The high-dimensional semantic feature vector is arranged by flattening operator to reduce its dimensionality, and the feature matrix with spatial topological structure is transformed into a first state space feature of a preset dimension.

6. The method for arranging a tower crane on a construction site as described in claim 1, characterized in that, The decision network based on the tower crane layout strategy model obtains the first optimal deployment action of the crane in the current state according to the first state space feature mapping, including: The first state space features are subjected to a combination of linear and nonlinear transformations to extract deployment decision feature vectors that are related to the geographical constraints and coverage requirements of the construction site. The layout decision feature vector is mapped to a preset layout action space to generate original action scores corresponding to different candidate deployment actions; The scores of each original action are normalized and mapped to construct a decision probability distribution representing each candidate deployment action in the current construction state; Based on the decision probability distribution, the candidate crane deployment action with the most significant probability response is determined as the first optimal deployment action.

7. A method for arranging tower cranes on construction sites as described in claim 6, characterized in that, The decision network includes a third fully connected layer and a second activation function; the step of performing a linear and nonlinear combination transformation on the first state space features to extract a deployment decision feature vector that relates to the geographical constraints and coverage requirements of the construction site includes: The first state space features are input into the third fully connected layer for linear weighting. By aggregating the spatial distribution weights of each task point within the construction area, a decision intermediate vector containing local crane location clues is obtained. The decision intermediate vector is nonlinearly transformed using the second activation function to simulate the nonlinear interference relationship between different crane coverage areas and the obstruction constraints of geographical barriers, thereby extracting the deployment decision feature vector.

8. A method for arranging tower cranes on construction sites as described in any one of claims 1 to 7, characterized in that, The acquisition of construction status information characterizing the construction site includes: Obtain a remote sensing map of the construction site, and map the construction site into a grid map of a preset size based on the remote sensing map; Collect building information within each grid cell of the grid map, and generate the obstacle distribution map based on the building information within each grid cell; The deployment of cranes within each grid cell is statistically analyzed, and a crane distribution map is generated based on the crane deployment data. The construction points within each grid cell are counted, and a construction area distribution map is generated based on these construction points.

9. A layout device for a construction site tower crane, characterized in that, include: The module includes an acquisition module, an iteration module, and a layout module. The acquisition module acquires construction status information representing the construction site, including an obstacle distribution map, a crane distribution map, and a construction area distribution map. The iterative module is used to input the construction status information into a pre-trained tower crane layout strategy model to iteratively decide on the crane layout scheme. During each iteration, the shared backbone network of the tower crane layout strategy model extracts first state space features representing the spatial relationships of the construction site from the construction status information. Based on the decision network of the tower crane layout strategy model, the first optimal deployment action of the crane in the current state is obtained by mapping the first state space features. This process continues until the current layout scheme meets a preset termination condition, generating the final layout scheme of the crane. The tower crane layout strategy model is trained and obtained based on the historical layout state value obtained by processing historical state space features. The layout module is used to arrange the positions of the tower cranes on the construction site according to the final layout scheme.

10. A terminal device, characterized in that, It includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement a site tower crane layout method as described in any one of claims 1 to 8.

Images