Escalator operation control method, system and computer device

Abstract

This invention provides an escalator operation control method, system, and computer equipment, relating to the field of escalator operation control technology. It aims to solve the problem that traditional escalator control relies on manual experience and struggles to cope with dynamic passenger flow changes, thereby improving the intelligence level of control and enhancing operational safety, efficiency, and energy efficiency. The method acquires operational source data of the controlled escalator, including historical operation data, historical passenger flow data, and operational environment data. Historical operation data includes historical operating speed control data of the controlled escalator. Based on the historical passenger flow data, it generates predicted passenger flow data for future periods. Based on the predicted passenger flow data and the historical operating speed control data of the controlled escalator, it generates predicted operational risks for the controlled escalator for future periods. Using the predicted operational risks, it generates an escalator operation control strategy for future periods. Finally, it controls the operation of the controlled escalator according to the escalator operation control strategy during the future periods.

Description

Technical Field

[0001] This invention belongs to the field of escalator operation control technology, specifically relating to urban rail transit and passenger flow management, and particularly to an escalator operation control method, system, and computer equipment. Background Technology

[0002] Existing subway stations generally face high-density passenger flow during morning and evening rush hours and holidays. Traditional escalator scheduling, which relies on manual experience or fixed strategies, is difficult to accurately cope with surges or sudden changes in passenger flow and is prone to congestion and safety hazards. On the other hand, simply increasing escalator speed will bring additional energy consumption and the risk of bottleneck accumulation. At the same time, there is a lack of quantitative measurement of passenger violations (such as escalator prohibition, changing direction, etc.), resulting in a lack of scientific comprehensive assessment methods for passenger flow time, congestion resistance, and violation costs. Therefore, it is necessary to combine machine learning technology for passenger flow prediction, and then use multi-index risk assessment and reinforcement learning algorithms to achieve dynamic adaptive control of escalator speed, thereby balancing safety and efficiency. Summary of the Invention

[0003] In order to at least partially solve the aforementioned problems, embodiments of the present invention propose an escalator operation control method, system, and computer equipment that integrates machine learning passenger flow prediction, risk assessment, and reinforcement learning decision-making.

[0004] In a first aspect, embodiments of the present invention provide an escalator operation control method, comprising:

[0005] Obtain the operation source data of the controlled escalator, which includes historical operation data, historical passenger flow data, and operation environment data. The historical operation data includes the historical operating speed control data of the controlled escalator.

[0006] Based on the historical passenger flow data, predictive passenger flow data for future periods is generated;

[0007] Based on the predicted passenger flow data and the historical operating speed control data of the controlled escalator, the predicted operating risk data of the controlled escalator for the future period is generated.

[0008] The escalator operation control strategy for the future time period is generated using the controlled escalator operation risk prediction data.

[0009] The controlled escalator is controlled to operate in the future time period according to the escalator operation control strategy.

[0010] Furthermore, generating the controlled escalator operation risk prediction data for the future period based on the predicted passenger flow data and the historical operating speed control data of the controlled escalator includes: acquiring key feature information for risk prediction, the key features including passenger flow data, risk event data, and potential hazard data; constructing risk prediction models based on different risk assessment tasks; and using the key feature information and the risk prediction models to predict risk indicators for the future period, the risk indicators including passenger flow time indicators, congestion indicators, and default risk indicators.

[0011] Furthermore, predicting risk indicators for the future period using the key feature information and the risk prediction model includes: calculating the objective weight of each risk indicator for the future period using the historical passenger flow data and the risk event data; calculating the fuzzy subjective weight of each risk indicator for the future period based on the fuzzy judgment matrix; and fusing the objective weight and the fuzzy subjective weight using a weighted average method to obtain the comprehensive weight of each risk indicator for the future period.

[0012] Furthermore, using the key feature information and the risk prediction model to predict the risk indicators for the future period also includes updating the weight allocation of each risk indicator through a game theory model.

[0013] Furthermore, using the key feature information and the risk prediction model to predict the risk indicators for the future period also includes: jointly training multiple risk assessment tasks; and fusing the output results of each risk assessment task through a weighted average fusion method to obtain a comprehensive risk value.

[0014] Furthermore, generating the escalator operation control strategy for the future period using the controlled escalator operation risk prediction data includes: using the predicted passenger flow data and the comprehensive risk value to form the state of the reinforcement learning agent; using the target operating speed or state set of the controlled escalator as the action space of the reinforcement learning agent; and generating and outputting the controlled escalator operation control strategy through the reinforcement learning strategy of the reinforcement learning agent.

[0015] Furthermore, generating the escalator operation control strategy for the future time period using the controlled escalator operation risk prediction data also includes: training the reinforcement learning strategy using a multi-step rolling strategy, wherein the multi-step rolling strategy includes: defining the current reinforcement learning state of the reinforcement learning agent based on the predicted passenger flow data and risk prediction data at the current moment; the reinforcement learning agent obtaining an immediate reward in the action space by specifying the escalator speed; and updating the current reinforcement learning state.

[0016] Furthermore, generating predicted passenger flow data for future periods based on the historical passenger flow data includes: converting the planar layout information of the space where the controlled escalator is located into two-dimensional grid or image data; inputting the two-dimensional grid or image data into a convolutional neural network to extract the spatial features of the space where the controlled escalator is located; constructing a graph structure from key areas within the space where the controlled escalator is located and related historical passenger flow data; and obtaining the passenger flow or passenger flow density of each key area in the future period using a knowledge-guided graph convolutional model based on the spatial features.

[0017] Secondly, embodiments of the present invention provide an escalator operation control system, comprising:

[0018] The data acquisition module is used to acquire the operation source data of the controlled escalator. The operation source data includes historical operation data, historical passenger flow data, and operation environment data. The historical operation data includes the historical operating speed control data of the controlled escalator.

[0019] The future passenger flow prediction module is used to generate predicted passenger flow data for future periods based on the historical passenger flow data.

[0020] The future risk prediction module is used to generate the controlled escalator operation risk prediction data for the future period based on the predicted passenger flow data and the historical operating speed control data of the controlled escalator.

[0021] The control strategy generation module is used to generate the escalator operation control strategy for the future time period using the controlled escalator operation risk prediction data;

[0022] The operation control module is used to control the operation of the controlled escalator in the future time period according to the escalator operation control strategy.

[0023] Thirdly, embodiments of the present invention provide a computer device, including: at least one processor and at least one memory; the memory stores executable instructions of the processor; the processor is configured to execute the aforementioned escalator operation control method.

[0024] This invention provides an escalator operation control method, system, and computer equipment. The method generates predicted passenger flow data for future periods based on the operation source data of the controlled escalator, and generates predicted escalator operation risk data for future periods based on the predicted passenger flow data and the historical operating speed control data of the controlled escalator. Then, it uses the predicted escalator operation risk data to generate escalator operation control strategies for future periods. Specifically, it uses a convolutional neural network (CNN) and a knowledge-guided graph convolutional model (KGCS) to perform multi-step rolling predictions of the subway station's floor plan layout and passenger flow at key nodes to accurately grasp the passenger flow trend for future periods. Based on the Projection Pursuit Model (PP) and the Starfish Algorithm (SFOA), multiple indicators such as passenger flow time, congestion, and default cost are comprehensively weighted and projected in a low dimension to output a quantitative value representing the risk of peak passenger flow. Reinforcement learning (RL) is used to take the risk value and passenger flow prediction results as state inputs and learn the optimal strategy in a series of escalator speed actions. Thus, the escalator speed is adaptively adjusted under a real-time rolling update mechanism to achieve a comprehensive balance between congestion, safety, and energy consumption. This solves the problem that traditional escalator control relies on human experience and is difficult to cope with dynamic changes in passenger flow, thereby improving the level of intelligence of control and the safety, efficiency, and energy efficiency of operation. Attached Figure Description

[0025] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0026] Figure 1 This is a flowchart illustrating an escalator operation control method provided in an embodiment of the present invention;

[0027] Figure 2 This is a functional structure diagram of an escalator operation control system according to an embodiment of the present invention;

[0028] Figure 3 This is a schematic diagram of the system architecture for implementing the escalator operation control method provided in the embodiments of the present invention;

[0029] Figure 4 This is a schematic diagram of the process for implementing the escalator operation control method provided in the embodiments of the present invention;

[0030] Figure 5 This is a schematic diagram of the structure of the knowledge-guided graph convolution model in an embodiment of the present invention;

[0031] Figure 6 This is a schematic diagram of the prediction model and risk assessment process in an embodiment of the present invention;

[0032] Figure 7 This is a schematic diagram of the rolling update process according to an embodiment of the present invention. Detailed Implementation

[0033] To enable those skilled in the art to better understand this prediction method, the escalator operation control method provided by this invention will be clearly described below in conjunction with specific implementation methods and accompanying drawings. It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used in this invention have the same meaning as commonly understood by those skilled in the art to which this invention pertains.

[0034] It should be understood that the term "and / or" as used in this specification and appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes such combinations. Furthermore, in the description of this specification and appended claims, the terms "first," "second," "third," etc., are used only for distinguishing descriptions and should not be construed as indicating or implying relative importance. It should be noted that, unless otherwise specified, the embodiments and features in the embodiments of this invention can be combined with each other. The escalator operation control method provided by the embodiments of this invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0035] With the rapid development of urban rail transit, subway stations have become an important part of daily transportation. Escalators are one of the main pieces of equipment for passenger flow within subway stations, especially during peak hours, where their operational efficiency directly impacts passenger experience and the overall operational efficiency of the subway station. However, traditional escalator operation control methods rely on fixed adjustment strategies or manual experience, often failing to cope with drastic fluctuations in passenger flow. For example, during peak hours or holidays, passenger flow increases dramatically, and traditional methods struggle to adapt to rapidly changing demand, potentially leading to congestion, delays, and even safety hazards. On the other hand, simply increasing escalator speed can improve transportation efficiency, but excessively high speeds may cause passenger discomfort and safety issues; conversely, excessively low speeds may result in long waiting times, affecting passenger comfort and station mobility.

[0036] Traditional escalator speed control methods typically do not consider energy consumption. Escalators can consume significant amounts of electricity during operation, especially under high loads, and existing control strategies often fail to optimize for energy consumption, leading to energy waste. Furthermore, the complex and highly unpredictable passenger flow in subway stations makes it difficult for traditional control methods to provide flexible and precise adjustment schemes, further exacerbating energy waste and system inefficiency.

[0037] With the continuous advancement of artificial intelligence technology, adaptive control methods based on machine learning, deep learning, and reinforcement learning are increasingly being applied to the optimization and adjustment of escalator operating speed. Utilizing advanced technologies such as convolutional neural networks and graph convolutional networks, passenger flow prediction models can more accurately predict passenger flow and distribution in different areas of subway stations, thus providing data support for escalator speed adjustment. Furthermore, reinforcement learning technology is widely used in dynamic decision-making. By training an agent to output the optimal escalator speed under different passenger flow and risk conditions, adaptive adjustment of escalator operating speed is achieved. This allows for rapid responses based on real-time conditions, improving system flexibility and operational efficiency.

[0038] Existing reinforcement learning methods still face several challenges. While reinforcement learning can achieve dynamic adjustment based on real-time data, it may suffer from low learning efficiency, insufficient prediction accuracy, and poor adaptability to new situations when dealing with complex environments and multi-task problems. Furthermore, risk assessment in related technologies mostly relies on single-task models, failing to fully consider the combined effects of multiple factors. Therefore, how to combine multi-task learning and ensemble methods for accurate passenger flow prediction and risk assessment, and how to dynamically optimize escalator speed through reinforcement learning, has become a key technical challenge in current intelligent escalator control systems.

[0039] To address the aforementioned issues, the objective of this invention is to provide a more precise, intelligent, and energy-efficient escalator operation control scheme by combining multi-task learning with an integration method, thereby resolving problems in related technologies and achieving efficient, safe, and energy-saving escalator speed regulation.

[0040] like Figure 1 As shown, the present invention discloses an escalator operation control method, including:

[0041] S101: Obtain the operating source data of the controlled escalator.

[0042] The operational source data includes historical operational data, historical passenger flow data, and operational environment data. Historical operational data includes historical operational speed control data of the controlled escalators.

[0043] It should be understood that the escalator operation control method provided in this embodiment of the invention can be applied to various scenarios, such as subway stations, bus stations, airport terminals, and other places with unevenly distributed pedestrian traffic. This embodiment of the invention specifically uses the operation of escalators in subway stations as an example. The application of the escalator operation control method provided by this invention in other scenarios is similar to that in subway stations, and will not be described in detail here.

[0044] Acquire multi-source basic data on the operation of subway station escalators, including historical operation data, historical environmental data, and historical data of the controlled escalators. Historical operation data includes historical mill flow data of the subway station; environmental data includes geographical data and equipment status data; and historical data includes time-series data, operation records, and user behavior data.

[0045] After obtaining multi-source basic data, the multi-source basic data is preprocessed in a unified manner to remove outliers and fill in missing data, so as to ensure the continuity and availability of the data. The specific methods and processes of preprocessing are not described in detail in this embodiment of the invention. In engineering practice, those skilled in the art can choose the appropriate implementation method as needed.

[0046] S102: Generate predicted passenger flow data for future periods based on historical passenger flow data.

[0047] In some alternative embodiments, multi-step rolling prediction is performed by combining convolutional neural networks and knowledge-guided graph convolutional networks to obtain passenger flow or density at key nodes in the subway station in future time periods.

[0048] First, historical passenger flow data for the subway station is collected, including passenger volume or density at key nodes during different time periods. Simultaneously, the subway station's floor plan layout information is collected and converted into two-dimensional grid or image data. This two-dimensional grid or image data will serve as input to a convolutional neural network, with key nodes including escalator entrances, turnstiles, and bottleneck areas.

[0049] It should be understood that spatial features of subway stations can be extracted using convolutional neural networks. In this process, convolution operations are performed on two-dimensional grids or image data. The convolution kernel and bias term work together to filter the input data, resulting in convolutional output features. This process transforms the subway station's planar layout information into spatial features that can be used for subsequent predictions.

[0050] To further improve the accuracy of passenger flow forecasting, spatial features extracted by convolutional neural networks are combined with historical passenger flow data to construct a graph structure. Based on this graph structure, a knowledge-guided graph convolutional network is used for processing. Specifically, KGCS can be used to extract the relationship information between nodes in the graph structure and combine it with historical passenger flow data for more accurate predictions.

[0051] Knowledge-guided graph convolutional networks can effectively capture the spatiotemporal dependencies between different areas within a subway station through graph convolution operations, thereby more accurately predicting passenger flow in future time periods.

[0052] Finally, through joint prediction using convolutional neural networks and knowledge-guided graph convolutional networks, a multi-step rolling prediction method is adopted to gradually obtain the predicted passenger flow or passenger density of key nodes in the subway station within future time periods. During the rolling prediction process, the prediction results at the current moment are used as the input data for the next moment, thereby ensuring the continuity and dynamism of the prediction.

[0053] S103: Generate prediction data for the operation risk of controlled escalators in future periods based on predicted passenger flow data and historical operating speed control data of the controlled escalators.

[0054] In order to predict the operational risks of controlled escalators, some optional embodiments choose to construct a risk element system. The risk element system includes key feature information closely related to the risk status of the controlled escalator, and maps the key feature information into a structured set of risk elements to provide an input basis for subsequent risk assessment. Key features include passenger flow data, risk event data, and hazard data.

[0055] This paper proposes a joint modeling approach for multiple risk assessment tasks, establishing a risk assessment model based on multi-task learning. This model evaluates passenger flow data and risk events through multiple tasks, allowing for parallel processing of these tasks to improve overall modeling efficiency and enhance the model's generalization ability. Furthermore, it proposes joint learning and optimization of shared feature representations across multiple tasks to improve the stability and discriminative power of risk feature expression.

[0056] S104: Generate escalator operation control strategies for future periods using controlled escalator operation risk prediction data.

[0057] For different risk assessment tasks, various risk indicators are predicted, and then the different risk indicators are integrated and processed to obtain a unified comprehensive risk assessment result, which serves as the high-level risk state input for reinforcement learning strategy decision-making and provides input for the generation of subsequent escalator operation control policies.

[0058] S105: Control the operation of the controlled escalator in the future time period according to the escalator operation control strategy.

[0059] In some alternative embodiments, the comprehensive risk assessment result is used as the reinforcement learning state, and the corresponding speed adjustment action is generated through the policy function and mapped to the escalator target running speed.

[0060] It should be understood that the embodiments of the present invention periodically collect real-time passenger flow data of subway stations and update the prediction results of the passenger flow prediction step and the evaluation results of the risk assessment step, and then dynamically adjust the escalator running speed according to the reinforcement learning strategy to achieve continuous control of the high-density passenger flow environment.

[0061] In some optional embodiments, the spatial features extracted by the convolutional neural network are further processed through a knowledge-guided graph convolutional model, and the features of each key node in the subway station are updated. Specifically, in the knowledge-guided graph convolutional model, the feature vector of each key node is iteratively updated through graph convolution operations. The knowledge-guided graph convolutional model uses the neural network structure shown in the figure below to update node features:

[0062] ,in, As a key node At any moment The updated feature vector, Indicates key nodes The set of adjacent key nodes, As a key node and key nodes Similarity or attention weight between them The parameters of the graph convolutional layer are... Key Nodes At any moment The input feature vector, It is a non-linear activation function.

[0063] In some alternative embodiments, S103 may be implemented, but is not limited to, by the following process (not shown in the figure):

[0064] S1031: Obtain key feature information for risk prediction, including passenger flow data, risk event data, and potential hazard data.

[0065] S1032: Construct risk prediction models based on different risk assessment tasks.

[0066] S1033: Utilize key feature information and risk prediction models to predict risk indicators for future periods. Risk indicators include passenger flow time indicators, crowding indicators, and default risk indicators.

[0067] In some alternative embodiments, S1033 may be implemented, but is not limited to, by the following process (not shown in the figure):

[0068] S10331: Calculate the objective weight of each risk indicator in the future period using historical passenger flow data and risk event data.

[0069] S10332: Calculate the fuzzy subjective weight of each risk indicator in the future time period based on the fuzzy judgment matrix.

[0070] S10333: The weighted average method is used to integrate objective weights and fuzzy subjective weights to obtain the comprehensive weight of each risk indicator in the future period.

[0071] S10334: Update the weight allocation of each risk indicator using a game theory model.

[0072] S10335: A multi-task learning framework is used to jointly train multiple risk assessment tasks.

[0073] It should be understood that the multi-task learning framework is not only about parallel processing of multiple tasks, but also includes the interrelationship and shared learning between tasks. In this framework, although multiple tasks can be processed in parallel, they influence and collaborate with each other by sharing underlying network layers (such as feature extraction layers). Tasks are interconnected through shared representations and weight adjustments, learning collaboratively to improve overall learning performance and generalization ability. Therefore, tasks are not only independent and parallel, but also interact and influence each other during training. It should be noted that this multi-task learning framework is not limited to the implementation provided in the embodiments of this invention, but can also be other implementations designed by those skilled in the art according to engineering needs.

[0074] S10336: The output results of each risk assessment task are merged by weighted average fusion to obtain a comprehensive risk value.

[0075] In some alternative embodiments, S104 may be implemented, but is not limited to, by the following process (not shown in the figure):

[0076] S1041: The state of the reinforcement learning agent is composed of predicted passenger flow data and comprehensive risk value.

[0077] S1042: The target running speed or state set of the controlled escalator is used as the action space of the reinforcement learning agent.

[0078] S1043: Generate and output the control strategy for the controlled escalator through the reinforcement learning strategy of the reinforcement learning agent.

[0079] S1044 employs a multi-step rolling strategy to train reinforcement learning strategies.

[0080] The multi-step scrolling strategy can be implemented, but is not limited to, through the following process (not shown in the figure):

[0081] S1: Define the current reinforcement learning state of the reinforcement learning agent based on the predicted passenger flow data and risk prediction data at the current moment.

[0082] S2: The reinforcement learning agent receives an immediate reward for moving up a specified escalator speed in the action space.

[0083] S3: Update the current reinforcement learning state.

[0084] In some alternative embodiments, S102 may be implemented, but is not limited to, by the following process (not shown in the figure):

[0085] S1021: Convert the planar layout information of the space where the controlled escalator is located into two-dimensional grid or image data.

[0086] S1022: Input two-dimensional grid or image data into a convolutional neural network to extract the spatial features of the space where the controlled escalator is located.

[0087] S1023: Construct a graphical structure of key areas and relevant historical passenger flow data within the space where the controlled escalator is located.

[0088] S1024: Based on spatial features, use a knowledge-guided graph convolutional model to obtain the passenger flow or passenger flow density of each key area in the future time period.

[0089] The escalator operation control method provided in this embodiment utilizes machine learning, risk assessment, and reinforcement learning techniques to offer escalator operation control strategies for subway stations and adjust escalator speed. It predicts passenger flow using convolutional neural networks and knowledge-guided graph convolutional models, obtaining passenger flow density and distribution for future time periods. Risk assessments are then performed on passenger flow time, congestion levels, and default risks to obtain the subway station's risk value. Subsequently, reinforcement learning is used to adjust the escalator speed based on the risk value and passenger flow predictions, optimizing efficiency and reducing energy consumption. This ensures that the escalator speed can be flexibly adjusted under different conditions and provides real-time feedback to optimize subway station operational efficiency and passenger experience. This effectively solves the problem of traditional escalator control relying on manual experience and struggling to cope with dynamic passenger flow changes, thereby improving the intelligence level of control and enhancing operational safety, efficiency, and energy efficiency.

[0090] To better describe the implementation of the embodiments of the present invention, some contents of the embodiments of the present invention will be described in detail below.

[0091] In S103, risk assessment uses an improved analytic hierarchy process (AHP) and a centralized approach to objectively calculate the weights of objectives. It then combines historical data and risk indicator data through a transformational approach, standardizing both historical passenger flow data and risk event data. Historical passenger flow data... and risk event data Corresponding to historical periods The data includes passenger flow data and risk event data. These two types of data are then standardized to obtain standardized data. and The standardized calculation formula is:

[0092] ,in, and These are the mean and standard deviation of historical passenger flow data, respectively. and The mean and standard deviation of the risk event data. For the first Historical passenger flow data for the time period In order to be with the first Risk event data for the corresponding time period based on historical passenger flow data.

[0093] In S104, the combined weights of risk data and risk indicator data are calculated separately. and The calculation formula is as follows:

[0094] ,in, This represents the total number of data points. This is a time period number.

[0095] Then, set the fusion factor according to the actual needs of the task. The final objective weight is calculated using the following formula. :

[0096] ,in, It is a fusion factor used to adjust the weight of historical passenger flow data and risk event data in the overall weighting.

[0097] Finally, based on the actual task requirements and the task integration factors, the final comprehensive risk weight is obtained. This is to facilitate subsequent decision-making and risk control.

[0098] In S102, to further optimize the model's learning performance, an adaptive learning strategy is used to update the features of historical passenger flow data. Spatial features extracted using a convolutional neural network include: information on the floor plan layout within the subway station. The data is converted into two-dimensional grids or image data, and the converted result is input into a convolutional neural network to extract spatial features. In the convolution operation, the convolution kernel... With bias term It is used to convolve the input data to obtain the output features after convolution. The formula is expressed as:

[0099] Where * denotes a two-dimensional convolution operation, The activation function is the output feature after convolution. Represents the time step Below, the convolutional layer... The activation values ​​of each feature map.

[0100] During the training of convolutional neural networks and graph convolutional networks, the weights of each convolutional layer and graph convolutional layer are updated by adjusting the weights of historical passenger flow data and the parameters of graph convolution, so as to better fit the data and improve model performance.

[0101] Specifically, the joint prediction part of the convolutional neural network and the knowledge-guided graph convolutional network uses an adaptive learning strategy to train on historical passenger flow data with weights, and adjusts the weights of the convolutional layers and graph convolutional layers in real time. During training, the knowledge-guided graph convolutional model automatically adjusts the weights of each convolutional kernel by weighting historical and real-time passenger flow data. And the parameters of the graph convolutional layer To adapt to passenger flow characteristics at different times, this adaptive learning strategy adjusts the weights of the convolutional neural network with convolutional layers and the knowledge-guided graph convolutional network. It will be updated based on the current training state and historical learning state. This is achieved by comparing the weights with previous ones. Combined with the current training results and the learning rate We adjust the weights of the convolutional and graph convolutional networks to better learn the features of historical passenger flow data in each training round.

[0102] .

[0103] Meanwhile, the parameters in the knowledge-guided graph convolutional network The weights will also be adjusted accordingly based on the weights of the convolutional layers and the graph convolutional layers. (Graph convolutional layer weights) This will be adjusted through a similar update method, enabling the graph convolutional network to better learn the passenger flow characteristics of different time periods with each update. This process ensures that the model can dynamically adjust its weights to adapt to changes in different time periods, and is achieved through the following formula:

[0104] ,in, and These are the old weights of the current convolutional layer and the graph convolutional layer, respectively. and These are the gradients of the parameters of the convolutional layer and the graph convolutional layer, respectively. For learning rate, This is the loss function.

[0105] Through the above adjustment process, the weights of the convolutional layer and the graph convolutional layer are dynamically updated in order to better capture the passenger flow patterns and regularities in different areas of the subway station.

[0106] The core idea of ​​the global attention mechanism is to dynamically adjust the contribution of different nodes to feature updates by weighted aggregation of information between nodes, thereby improving the model's ability to capture spatiotemporal dependencies.

[0107] First, for each key node Calculate its updated feature representation. Specifically, for node First, through the set of adjacent nodes The node features in the dataset are weighted and aggregated. Here, the weights are... Reflects the nodes and nodes The similarity or importance between nodes. Graph convolutional networks extract features from each node. And combined with similarity weight These features are weighted and summed, and finally passed through a nonlinear activation function. Update node Feature representation:

[0108] ,in, Represents a node At any moment Updated features It is a node The set of adjacent nodes, It is a node For nodes Attention weights These are the weights of the graph convolutional layer. It is a node At any moment Its characteristics.

[0109] Next, we calculate the attention weights. In this case, the softmax function is used to normalize the similarity between nodes. This normalization process ensures that the contributions of all neighboring nodes can be uniformly represented in probabilistic form, thus making the information aggregation process more reasonable.

[0110] ,in, Represents a node and nodes Similarity or attention weight between them.

[0111] Ultimately, through this global attention mechanism, nodes Features The result will be updated to a weighted sum of all adjacent nodes, enabling information to be transmitted more effectively from surrounding nodes to the target node. This process ensures that the model can dynamically adjust the weights of information transmission based on the relationships between different nodes and effectively capture the spatiotemporal dependencies of key areas within the subway station.

[0112] During the global attention mechanism phase, information between key nodes is weighted and aggregated using the following formula:

[0113] ,in, The weights are calculated using the softmax function and are used to measure key nodes. For key nodes At any moment Information contribution.

[0114] In S103, the subjective and objective weights of each risk indicator are calculated, and the comprehensive weight of each risk indicator is calculated through game theory combination weighting.

[0115] Constructing a fuzzy judgment matrix Each element This indicates that experts have a view on risk indicators. and The relative importance of is determined by using fuzzy numbers. To represent uncertainty, where The lower realm The most likely value, The upper boundary.

[0116] Next, the judgment matrix is ​​processed using a fuzzy synthesis method. Fuzzy operations are used to synthesize the fuzzy judgments of each indicator to obtain fuzzy weights. Finally, fuzzy normalization is applied to calculate the fuzzy subjective weight of each risk indicator. The formula is:

[0117] ,in, The fuzzy weight assigned to each risk indicator represents the relative importance of that factor.

[0118] Objective and subjective weights are combined using a weighted fusion method to obtain a comprehensive weight. Specifically, a weighted average method is used to fuse fuzzy subjective and objective weights, and the calculation formula is as follows:

[0119] ,in, As a fusion factor, its value is usually determined by the needs of the actual task and the characteristics of different risk indicators. For objective weighting, Subjective weighting, For the final overall weight, The value is usually determined by the needs of the actual task and the characteristics of different risk indicators.

[0120] Furthermore, considering the interaction of multiple factors in risk assessment, this invention introduces a game-theoretic weighting method to optimize the weight allocation of each risk indicator through a game-theoretic model. A utility function for each risk indicator is defined. This represents the contribution of each risk indicator to the overall assessment result, and the final weight is determined by solving for the equilibrium point of the game:

[0121] .

[0122] A multi-task learning framework is employed to jointly train multiple risk assessment tasks. Specifically, the framework simultaneously trains several related tasks, including passenger flow prediction, congestion assessment, and default risk assessment. The output of each task is... ,in Indicates the index of the task. Indicates the time.

[0123] All tasks share some network layers, and the outputs of each task are merged through a weighted average fusion method to obtain the final comprehensive output. The calculation formula is as follows:

[0124] ,in, For the first The weights of each task, where N is the total number of tasks. The weight of each task... It is automatically learned through backpropagation during the training process. Ultimately, the resulting comprehensive risk assessment is... ,in, For the first The risk value calculated for each task This is the final comprehensive risk value. This result serves as the basis for decisions regarding escalator operating speed control.

[0125] In S104, reinforcement learning is introduced to optimize escalator operation control decisions. Reinforcement learning allows the agent to learn optimal strategies through interaction with the environment, selecting actions that maximize cumulative rewards at each time step to improve system efficiency and reduce energy consumption. Through reinforcement learning, the system can adaptively adjust the escalator's operating speed in complex dynamic environments.

[0126] Specifically, the reward function in the reinforcement learning decision-making process integrates risk value, energy consumption factors, and comfort factors. The reward function is defined as follows:

[0127] ,in, For the comprehensive risk value, This represents the energy loss at the current moment, reflecting the energy consumption of the escalator. Passenger comfort ratings reflect the comfort level of the escalator. and These are the balancing factors for energy loss and comfort score, respectively.

[0128] Furthermore, the reinforcement learning strategy is trained using the Proximal Policy Optimization (PPO) algorithm to update the escalator speed control strategy. The PPO algorithm updates the escalator speed control strategy by updating the objective function, which includes risk value, energy loss, and passenger comfort. The PPO algorithm updates the strategy using the following objective function:

[0129] ,in, It is the ratio of the current strategy to the old strategy. It is the advantage function (whose calculation depends on the reward function). This includes risk value. Energy loss Comfort rating (information) It is a hyperparameter. The clip function is used to limit the policy ratio within a certain range to prevent the policy from changing too much, thereby stabilizing the learning process.

[0130] It should be understood that a rolling update method can be used to dynamically adjust the risk assessment model and passenger flow forecasting model to ensure that the model can adapt to changes in the environment in real time. This process is mainly based on real-time updates of historical passenger flow data and updates to the parameters of the risk assessment model based on new passenger flow forecast results, ensuring the accuracy and timeliness of the escalator operation control strategy.

[0131] Specifically, firstly, real-time passenger flow data from various areas of the subway station is collected, including passenger flow information obtained from sensing devices in each area. By collecting and analyzing this real-time data, passenger density and flow information for each area are obtained, serving as input data for further prediction and risk assessment. Based on this latest data, the parameters of the passenger flow prediction model and risk assessment model are updated, generating new prediction results. The models are updated at each time step. Sensing devices include sensors and surveillance cameras.

[0132] The passenger flow forecasting model is updated based on the latest passenger flow data for each area within the subway station. Risk assessment indicators for each area are calculated and relevant parameters are updated. Subsequently, based on the latest forecast results and the adjusted risk assessment data, the risk value is recalculated. The system then updates its risk assessment model based on this value. Ultimately, this rolling update process ensures that the system can accurately predict changes in passenger flow with each real-time data update and adjust risk assessments and strategy implementation in a timely manner.

[0133] Through rolling updates, reinforcement learning models can progressively improve strategies to adapt to new passenger flow patterns and risk changes. After each update, the agent adjusts the escalator speed control strategy based on new passenger flow forecast data and risk assessment results, ensuring that the updated strategy maximizes the system's efficiency at the current moment.

[0134] Rolling updates involve periodically collecting real-time passenger flow data from subway stations and updating passenger flow prediction and risk assessment models based on this data to dynamically adjust escalator operating speeds. The rolling update process includes the following steps:

[0135] Real-time passenger flow data collection: Regularly acquire real-time passenger flow data from sensors, cameras, and other devices in the subway station. The real-time passenger flow data includes the number of people entering and exiting the station, passenger density in different areas of the station, and dwell time. The real-time passenger flow data is then converted into a time-series dataset. This serves as input for subsequent forecasting and risk assessment;

[0136] Passenger flow forecast update: Update the input data of the convolutional neural network and the knowledge-guided graph convolutional network based on real-time passenger flow data, retrain or fine-tune the passenger flow forecast model, and generate the latest passenger flow or density forecast results for the current time period. In addition, by combining historical passenger flow data and subway station layout information, multi-step rolling forecasts are conducted to accurately predict passenger flow changes in future periods.

[0137] Risk assessment model update: Based on the updated passenger flow forecast results, the risk indicators for each area within the subway station are recalculated. These risk indicators include passenger flow time, congestion level, and default risk. An improved multi-task learning and ensemble method is used to weight and fuse multiple risk indicators to obtain an updated comprehensive risk value. The escalator speed control strategy is updated based on the comprehensive risk value.

[0138] Strategy Adjustment and Feedback: After updating the passenger flow forecasting and risk assessment models, the reinforcement learning agent updates its strategy based on the new forecast results and risk assessment outputs using the PPO algorithm. This adjusts the escalator speed to cope with the current passenger flow and risk status. The adjustment of the escalator speed control strategy is fed back to the escalator speed control system in real time. Periodic Update: The rolling update step is executed periodically. Each time, the passenger flow forecast, risk assessment, and reinforcement learning strategy are recalculated based on real-time passenger flow data.

[0139] The ultimate goal of decision-making and control is to develop the optimal escalator operation control strategy based on all the data processing and optimization in the previous steps. This step involves integrating the results of previous risk assessments, passenger flow forecasts, and reinforcement learning to generate the final decision output, which controls the escalator's operating speed and ensures optimal performance in a changing environment.

[0140] Constraints and safety corrections are applied to the target operating speed to ensure that the speed output by the reinforcement learning strategy meets the safety operation specifications of the escalator equipment.

[0141] Combining real-time updated passenger flow forecast data and risk assessment results, a comprehensive model is used to derive the optimal escalator speed control decision for each time step. Specifically, the system determines the optimal escalator speed control based on the current state. Target risk value Based on the current passenger flow forecast, the strategy is updated using the reinforcement learning PPO algorithm to select an optimal action. That is, to determine the operating speed of the escalator.

[0142] In the final decision-making process, the system adjusts the escalator's operating speed in real time according to the optimized PPO control strategy to cope with different passenger flow densities and risk levels. This process optimizes the escalator's operating speed control by maximizing the comprehensive reward function. Ultimately, through the PPO algorithm, the reinforcement learning agent can gradually adjust its strategy during training, optimize the escalator's operating speed, maximize the expected reward, and ensure that each update does not lead to excessive policy changes, thus maintaining training stability.

[0143] To implement the above-mentioned escalator operation control method, embodiments of the present invention provide an escalator operation control system, such as... Figure 2 As shown, the system includes:

[0144] The data acquisition module 21 is used to acquire the operation source data of the controlled escalator. The operation source data includes historical operation data, historical passenger flow data and operation environment data. The historical operation data includes the historical operation speed control data of the controlled escalator.

[0145] The future passenger flow prediction module 22 is used to generate predicted passenger flow data for future periods based on historical passenger flow data.

[0146] The future risk prediction module 23 is used to generate prediction data of the operation risk of the controlled escalator for future periods based on the predicted passenger flow data and the historical operating speed control data of the controlled escalator.

[0147] The control strategy generation module 24 is used to generate escalator operation control strategies for future periods using the controlled escalator operation risk prediction data.

[0148] The operation control module 25 is used to control the operation of the controlled escalator in the future time period according to the escalator operation control strategy.

[0149] Figure 3 This is a schematic diagram of the system architecture for implementing the escalator operation control method provided in the embodiments of the present invention. See [link / reference]. Figure 3The system comprises a data layer, a model layer, a decision layer, and an execution layer. The data layer is used for passenger flow data collection; the model layer includes a prediction model for passenger flow forecasting and risk assessment; the decision layer uses reinforcement learning; and the execution layer controls elevator speed and feeds relevant data back to the data layer.

[0150] Figure 4 The implementation process of the escalator operation control method based on the aforementioned system architecture is shown, such as... Figure 4 As shown, the process includes: a data acquisition module collects passenger flow data, a prediction module predicts passenger flow, and then, combined with the spatial features of a convolutional neural network, a knowledge-guided graph convolutional model is used to perform multi-step rolling prediction to obtain the passenger flow or density at each node in the future time period. The risk assessment module then performs risk assessment, performs three-dimensional risk quantification, outputs risk values, and reinforces learning decision-making. The execution module then executes elevator speed control and adjusts the elevator speed in real time, and sends the control data to the data acquisition module through a feedback mechanism.

[0151] Figure 5 The diagram illustrates the structure of the knowledge-guided graph convolutional model provided in an embodiment of the present invention, as shown below. Figure 6 As shown, the structure includes: input features, adjacency matrix, graph convolution propagation, graph convolution operation, first convolutional layer, first activation function, second convolutional layer, second activation function, fully connected layer, global information aggregation, and output prediction.

[0152] Figure 6 The flowchart of the prediction model and risk assessment in an embodiment of the present invention is shown.

[0153] Figure 7 A diagram illustrating the rolling update mechanism in an embodiment of the present invention is shown. It includes how various indicators are calculated, weighted, and summed to arrive at the final output of the risk value.

[0154] This invention discloses an escalator operation control method. This method combines convolutional neural networks and knowledge-guided graph convolutional networks to dynamically adjust escalator speed through accurate passenger flow prediction and risk assessment. First, a passenger flow prediction module collects real-time passenger flow data and, combined with the planar layout information of key nodes within the subway station, uses a convolutional neural network to extract spatial features, converting the station information into two-dimensional grids or image data. Then, a knowledge-guided graph convolutional network is used to perform graph-structured processing on this data, performing multi-step rolling predictions to obtain passenger flow and density in various areas during future time periods. Based on the prediction results, a risk assessment module calculates multiple risk indicators, including passenger flow time, congestion level, and default risk, and uses multi-task learning and ensemble methods for joint training to obtain the final comprehensive risk assessment value.

[0155] The reinforcement learning module, based on the obtained comprehensive risk assessment value and passenger flow prediction results, uses the PPO algorithm for training and outputs the optimal escalator speed control strategy. At each time step, the reinforcement learning agent adjusts the escalator's operating speed according to the current risk value and passenger flow status, ensuring that the escalator can operate efficiently and safely in high-density passenger flow and high-risk environments.

[0156] Furthermore, the rolling update module periodically collects real-time passenger flow data and updates the passenger flow prediction and risk assessment model to ensure that the system can adapt to changes in passenger flow in real time and dynamically adjust the escalator speed. Through this rolling update mechanism, the present invention can intelligently adjust the escalator speed, achieving efficient, safe, and energy-saving escalator operation.

[0157] This disclosure provides a computer device, including: at least one processor and at least one memory; the memory stores executable instructions of the processor; the processor is configured to perform the aforementioned escalator operation control method.

[0158] This invention provides an escalator operation control method, system, and computer equipment. It uses a convolutional neural network and a knowledge-guided graph convolutional model to perform multi-step rolling predictions of the subway station's layout and passenger flow at key nodes, accurately grasping future passenger flow trends. Based on the prediction results of the passenger flow prediction steps, it calculates passenger flow time indicators, passenger congestion indicators, and default risk indicators within the subway station. Based on an improved analytic hierarchy process and ensemble method, it calculates the subjective and objective weights of each indicator and obtains the comprehensive weight of each risk indicator through a game theory-based weighting method.

[0159] By using reinforcement learning to take risk values ​​and passenger flow predictions as state inputs, and learning the optimal strategy through a series of escalator speed actions, the escalator speed is adaptively adjusted under a real-time rolling update mechanism to achieve a comprehensive balance between congestion, safety and energy consumption.

[0160] Those skilled in the art will understand that all or some of the steps, functional modules / units in the system disclosed above can be implemented as software, firmware, hardware, or appropriate combinations thereof.

[0161] In hardware implementations, the division between functional modules / units mentioned in the above description does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be executed by several physical components working together.

[0162] Some or all of the physical components may be implemented as software executed by a processor, such as a central processing unit (CPU), digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit (ASIC). Such software may be distributed on a computer-readable medium, which may include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technique for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media include, but are not limited to, random access memory (RAM, more specifically SDRAM, DDR, etc.), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory (FLASH) or other disk storage; read-only optical disc (CD-ROM), digital versatile disc (DVD) or other optical disc storage; magnetic cartridges, magnetic tapes, disk storage or other magnetic storage; and any other media that can be used to store desired information and can be accessed by a computer. Furthermore, as is known to those skilled in the art, communication media typically contain computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0163] This disclosure has disclosed exemplary embodiments, and although specific terminology has been used, it is for general illustrative purposes only and should not be construed as limiting. In some instances, it will be apparent to those skilled in the art that features, characteristics, and / or elements described in conjunction with particular embodiments may be used alone, or in combination with features, characteristics, and / or elements described in conjunction with other embodiments, unless otherwise expressly indicated. Therefore, those skilled in the art will understand that various changes in form and detail may be made without departing from the scope of this disclosure as set forth by the appended claims.

Claims

1. An escalator operation control method, characterized in that, include: Obtain the operation source data of the controlled escalator, which includes historical operation data, historical passenger flow data, and operation environment data. The historical operation data includes the historical operating speed control data of the controlled escalator. Based on the historical passenger flow data, predictive passenger flow data for future periods is generated; Based on the predicted passenger flow data and the historical operating speed control data of the controlled escalator, the predicted operating risk data of the controlled escalator for the future period is generated. The escalator operation control strategy for the future time period is generated using the controlled escalator operation risk prediction data. The controlled escalator is controlled to operate in the future time period according to the escalator operation control strategy described above. The predicted passenger flow data and the historical operating speed control data of the controlled escalator are used to generate the predicted operating risk data of the controlled escalator for the future period, including: Obtain key feature information for risk prediction, including passenger flow data, risk event data, and potential hazard data; Build risk prediction models based on different risk assessment tasks; Using the key feature information, the risk indicators for the future period are predicted based on the risk prediction model. The risk indicators include passenger flow time indicators, crowding indicators, and default risk indicators. Using the aforementioned key feature information, the risk indicators for the future period predicted based on the aforementioned risk prediction model include: The objective weight of each risk indicator in the future time period is calculated using the historical passenger flow data and the risk event data. The fuzzy subjective weight of each risk indicator in the future time period is calculated based on the fuzzy judgment matrix; The objective weights and the fuzzy subjective weights are fused using a weighted average method to obtain the comprehensive weight of each risk indicator in the future time period.

2. The method according to claim 1, characterized in that, Using the aforementioned key feature information, the risk indicator prediction for the future period based on the risk prediction model also includes: The weight allocation of each risk indicator is updated using a game theory model.

3. The method according to claim 2, characterized in that, Using the aforementioned key feature information, the risk indicator prediction for the future period based on the risk prediction model also includes: Joint training is performed on multiple risk assessment tasks described above; The outputs of each risk assessment task are combined using a weighted average fusion method to obtain a comprehensive risk value.

4. The method according to claim 3, characterized in that, Generating the escalator operation control strategy for the future time period using the controlled escalator operation risk prediction data includes: The predicted passenger flow data and the comprehensive risk value constitute the state of the reinforcement learning agent; The target running speed or state set of the controlled escalator is used as the action space of the reinforcement learning agent; The controlled escalator operation control strategy is generated and output through the reinforcement learning strategy of the reinforcement learning agent.

5. The method according to claim 4, characterized in that, The generation of the escalator operation control strategy for the future period using the controlled escalator operation risk prediction data also includes: The reinforcement learning policy is trained using a multi-step rolling strategy, which includes: Based on the current predicted passenger flow data and risk prediction data, define the current reinforcement learning state of the reinforcement learning agent; The reinforcement learning agent receives an immediate reward in the action space by moving at a specified escalator speed. Update the current reinforcement learning state.

6. The method according to any one of claims 1-3, characterized in that, Generating predicted passenger flow data for future periods based on the historical passenger flow data includes: The planar layout information of the space where the controlled escalator is located is converted into two-dimensional grid or image data; The two-dimensional grid or the image data is input into a convolutional neural network to extract the spatial features of the space where the controlled escalator is located. Construct a graph structure from the key areas and relevant historical passenger flow data within the space where the controlled escalator is located; Based on the spatial characteristics, the passenger flow or passenger flow density of each key area in the future time period is obtained using a knowledge-guided graph convolution model.

7. An escalator operation control system, characterized in that, include: The data acquisition module is used to acquire the operation source data of the controlled escalator. The operation source data includes historical operation data, historical passenger flow data, and operation environment data. The historical operation data includes the historical operating speed control data of the controlled escalator. The future passenger flow prediction module is used to generate predicted passenger flow data for future periods based on the historical passenger flow data. The future risk prediction module is used to generate the controlled escalator operation risk prediction data for the future period based on the predicted passenger flow data and the historical operating speed control data of the controlled escalator. The control strategy generation module is used to generate the escalator operation control strategy for the future time period using the controlled escalator operation risk prediction data; The operation control module is used to control the operation of the controlled escalator in the future time period according to the escalator operation control strategy.

8. A computer device, characterized in that, include: At least one processor and at least one memory; The memory stores executable instructions of the processor; the processor is configured to execute the escalator operation control method according to any one of claims 1-7.

Images