Urban multivariable space-time prediction method based on large model agent

Abstract

The invention discloses a city multivariable space-time prediction method based on a large model agent, and belongs to the technical field of smart cities. In order to solve the problem of difficulty in urban operation condition prediction under the condition of incomplete data, the method comprises the following steps: defining urban multivariable space-time prediction as prediction of a discrete random process, and setting assumption conditions as a condition independence assumption and a Markov assumption for the probability of a system state at each moment; state decoupling is carried out, and the probability prediction decoupling of the system state is prediction of each channel state under the condition that the previous system state is given; constructing a space region, wherein the space region comprises a domain agent layer, a space-time agent layer and an individual agent layer; constructing an urban multivariable space-time prediction system based on a large model agent, including constructing an action space, a state space, a transfer function, a reward function and a memory flow; and performing task prediction on the constructed city multivariable space-time prediction system based on the large model agent. According to the invention, the accuracy and efficiency of city management can be improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of smart cities, and particularly relates to a method for predicting urban multi-variable spatio-temporal based on a large model proxy. Background Art

[0002] With the acceleration of the urbanization process and the rapid development of digital technologies, urban management faces unprecedented challenges and opportunities, such as unreasonable urban planning, inefficient resource utilization, traffic congestion and other problems. As the urban scale continues to expand and functions become increasingly complex, traditional urban planning and management often rely on experience, statistical data and expert opinions, and it is difficult to comprehensively and accurately grasp the complexity and diversity of urban development. In recent years, with the continuous maturity of technologies such as artificial intelligence, big data analysis, and multi-modal models, there is an urgent need for an interpretable reasoning technology driven by both data and knowledge to achieve computing for urban operation and control.

[0003] The invention patent with the application number 202411245264.3 and the invention name of "Method for Training a Large Language Model for Urban Domains and Urban Generative Intelligence Method and Device" relates to the field of artificial intelligence, and particularly relates to a method for training a large language model for urban domains and an urban generative intelligence method and device. The training method includes: obtaining a general large language model and an initial data set; obtaining a composite data set by setting agents in a virtual urban scenario to simulate various behaviors of humans in a real urban scenario; mixing the composite data set with the initial data set to form a pre-training data set, and using the pre-training data set to perform incremental pre-training on the general large language model to obtain a first urban large model; performing fine-tuning training on the first urban large model using a fine-tuning data set constructed for the urban domain to obtain a second urban large model; performing preference alignment training on the second urban large model using a human preference data set to obtain a target urban large model. Thus, a target urban large model with professional knowledge in the urban domain, general common sense in the world, and cognitive reasoning ability can be obtained. This model can achieve general question answering in the urban domain and retrieve urban data to perform tasks such as travel navigation, but essentially it is still an interface for data retrieval and does not have the ability to analyze complex knowledge and perform data calculations. Summary of the Invention

[0004] The problem to be solved by the present invention is the difficulty in predicting urban operation status under incomplete data, and a method for predicting urban multi-variable spatio-temporal based on a large model proxy is proposed.

[0005] To achieve the above object, the present invention is realized through the following technical solutions: A method for predicting urban multi-variable spatio-temporal based on a large model proxy, comprising the following steps: S1. Define the urban multivariate spatio-temporal prediction as the prediction of a discrete stochastic process. To predict the system state at each moment, then set the hypothesis conditions for the probability of the system state at each moment as the conditional independence hypothesis and the Markov hypothesis, and obtain the probability of the system state under the hypothesis conditions; S2. Decouple the state of the probability of the system state obtained in step S1. Decouple the probability prediction of the system state into the prediction of each channel state given the previous system state; S3. Based on the Markov hypothesis condition in step S1 and the state decoupling in step S2, construct a spatial region, which includes a domain agent layer, a spatio-temporal agent layer, and an individual agent layer; S4. Based on the domain agent layer, spatio-temporal agent layer, and individual agent layer constructed in step S3, construct an urban multivariate spatio-temporal prediction system based on large model agents, including constructing an action space, a state space, a transition function, a reward function, and a memory stream; S5. Perform task prediction on the urban multivariate spatio-temporal prediction system based on large model agents constructed in step S4. Record all decision events through the memory stream and record the feedback with reward values. Each agent will read relevant information from the event stream before the next action, and use the transition function to perform rolling prediction on the action strategy, and optimize and update the urban multivariate spatio-temporal prediction system based on large model agents.

[0006] Further, the specific implementation method of step S1 includes the following steps: S1.1. Set the urban multivariate spatio-temporal prediction definition as the prediction of a discrete stochastic process, determine the goal of the urban multivariate spatio-temporal prediction task as predicting the system state at each moment, and define as a stochastic process, , where, is 's sample space, is 's state space, is the set of time steps, , is the rd moment; is the system state random variable at the th moment, taking values from , , where, is 's number of channels, is 's dimension of each channel; Then predicting the system state at each moment is modeled as a based on the The joint distribution of the observed future values of the system state random variables at and before a certain moment is obtained, and the expression is: ; Among them, is the probability density function with as the parameter, is the N th channel in S1.2. Set the conditional independence assumption that given the historical state of the system, the state of a certain channel at the next moment is independent of other channels, that is: ; Among them, is the probability density function, is the n th channel in n is N any one in S1.3. Set the Markov assumption that the complete conditional probability is simplified to the transition probability between the last two states, and the expression is: .

[0007] Furthermore, the specific implementation method of step S2 is to decouple the probability prediction of the system state into the prediction of the state of each channel given the previous system state under the assumed conditions, and the expression is: .

[0008] Furthermore, the specific implementation method of step S3 includes the following steps: S3.1. Construct the domain agent layer: At the initial stage of the urban spatio-temporal state prediction task, set the action space of the i th domain agent to be ; Define the vector space composed of the directed unweighted adjacency matrix with all spatial regions as nodes to represent the association relationship between the spatial regions of the domain agent, is the j th spatial region, ; S3.2. Construct the individual agent layer: For each individual agent within the j th spatial region , by observing and 's state, set the action space of the k th individual agent to be , is the dimension number of individual decision-making, is the number of discretization intervals for individual decisions; S3.3. Construct the spatio-temporal agent layer: For the j th spatial region corresponding spatio-temporal agent, by observing and all the decision results of all individual agents within, set the action space of the j th spatio-temporal agent to be , is the number of predictive variables.

[0009] Furthermore, the specific implementation method of step S4 includes the following steps: S4.1. Construct the action space: The action space is all possible action combinations of all agents within the urban multi-variable spatio-temporal prediction system based on the large model agent within one time step, expressed as: ; where , , respectively represent the action spaces composed of all domain agents, spatio-temporal agents, and individual agents, , , respectively represent the total numbers of domain agents, spatio-temporal agents, and individual agents, where each spatial region corresponds to a spatio-temporal agent, that is ; S4.2. Construct the state space: The state space represents the integrated information about the urban environment, including individual positions, quantities, and all layouts of urban regions, given through images, vectors, and natural language descriptions; Set the expression for the spatio-temporal state at the th moment as: ; ; where is the facility layout map corresponding to the th moment and the j th spatial region, represented as a multi-channel image; is the vector composed of statistical variables corresponding to the th moment and the j th spatial region, including region size, region center position, resident population quantity, employed population quantity, energy consumption quantity, GDP quantity, enterprise quantity; is the th moment and thej The text description corresponding to a spatial region Indicates all sets of strings Indicates a character Thus, the system state at the th moment is represented as , all of the constitute the state space , all agents operate in the state space and perform different actions according to different environments at different times; S4.3. Construct the transition function: The transition function , according to the state space and the action space, converts an action into a state; for the individual agent action space and the spatio-temporal agent action space , Statistically analyze the individual change amounts in all regions and add them to the predicted value of the spatio-temporal agent to obtain the new system state ; S4.4. Construct the reward function: S4.4.1. Set the goal of the individual agent to be that the policy is more beneficial to the individual agent after the system state changes. Considering the living cost, commuting cost, and facility convenience, construct the reward function of the individual agent , and the expression is: ; Among them, is the living cost, is the commuting cost, is the facility convenience cost, , , respectively represent the reward coefficients of the living cost, commuting cost, and facility convenience cost; S4.4.2. Set the goal of the spatio-temporal agent to be that the predicted state at each time step is as close as possible to the actual state, so as to achieve the purpose of accurate spatio-temporal prediction. Considering the local state vector, construct the reward function of the spatio-temporal agent , and the expression is: ; Among them, and respectively represent the predicted value and the actual value of the local state vector at a time step, is the reward coefficient of the local fitting degree, is the 2-norm; Each dimension of the local state vector includes the number of resident population, the number of employed population, the amount of energy consumption, the amount of GDP, and the number of enterprises; S4.4.3. The goal of setting the domain agent is to limit the sparsity of the adjacency matrix while maintaining the global fitness, and obtain the reward function of the domain agent , and the expression is: ; Among them, is the 0-norm, is the total predicted value of the relevant variables under the domain agent, is the total actual value of the relevant variables under the domain agent, is the adjacency matrix, is the reward coefficient of sparsity, is the reward coefficient of global fitness; S4.5. Construct a memory stream, including an event stream and a perception stream. The event stream records the events that occur over time, including the actions of the agent, external intervention events, and environmental changes; the perception stream records the agent's tracking and understanding of the events in the event stream, and each node in the perception stream is linked to one or more nodes in the event stream.

[0010] Furthermore, the specific implementation method of step S5 includes the following steps: S5.1. Given the system state at the current moment , first, each domain agent gives its action at the current moment by observing this state. The action set of the domain agent layer is . Similarly, the action sets of the individual agent layer and the spatio-temporal agent layer are and respectively; S5.2. The state transition function converts the action sets of the domain agent layer, the individual agent layer, and the spatio-temporal agent layer in step S5.1 into a new system state at the next moment ; S5.3. All agents obtain their respective rewards from the memory stream according to time, adjust their action strategies at the next moment, and thus make new actions under the new system state , so as to complete the rolling prediction.

[0011] The beneficial effects of the present invention: A method for urban multivariate spatio-temporal prediction based on large model agents according to the present invention can improve the accuracy and efficiency of urban management. By integrating and analyzing multi-source data such as urban traffic, environment, population, and economy, the present invention can monitor the operation status of the city in real time, discover potential problems in a timely manner, and provide accurate decision-making support. For example, in traffic management, the model can predict traffic flow, optimize signal timing, reduce traffic congestion, and improve traffic efficiency.

[0012] A method for urban multivariate spatio-temporal prediction based on large model agents according to the present invention can optimize resource allocation and planning. By analyzing various regions of the city, the model identifies the imbalance in resource distribution and assists in formulating reasonable resource allocation plans. In urban planning, the present invention can simulate the impact of different planning schemes on urban development, help decision-makers select the optimal scheme, and enhance the sustainability of urban development.

[0013] A method for urban multivariate spatio-temporal prediction based on large model agents according to the present invention can strengthen emergency response and risk management. When emergencies or natural disasters occur, the model can quickly assess the scope of the event's impact, predict possible subsequent developments, assist in formulating emergency plans, and improve the speed and effectiveness of emergency response. For example, in flood warning, the present invention can predict the flood spread path, evacuate people in advance, and reduce losses.

[0014] A method for urban multivariate spatio-temporal prediction based on large model agents according to the present invention can promote the construction of smart cities. The present invention provides data support and decision-making basis for the construction of smart cities. By deeply mining various types of urban data, the present invention can provide technical support for fields such as intelligent transportation, intelligent healthcare, and intelligent security, and improve the intelligent level of urban services.

[0015] A method for urban multivariate spatio-temporal prediction based on large model agents according to the present invention can support policy formulation and evaluation. The present invention can simulate the impact of different policies on urban development, assist policy makers in evaluating the policy effects, and optimizing policy design. For example, when formulating environmental protection policies, the model can predict the change in air quality after the implementation of the policy and evaluate the effectiveness of the policy. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is a structural block diagram of a method for urban multivariate spatio-temporal prediction based on large model agents according to the present invention; Figure 2 is a flow chart of a method for urban multivariate spatio-temporal prediction based on large model agents according to the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0017] To make the objectives, technical solutions and advantages of the present invention more clear and understandable, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention, that is, the specific embodiments described are only a part of the embodiments of the present invention, rather than all of the specific embodiments. The components of the specific embodiments of the present invention usually described and shown in the drawings herein can be arranged and designed in various different configurations, and the present invention can also have other embodiments.

[0018] Therefore, the following detailed description of the specific embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents the selected specific embodiments of the present invention. All other specific embodiments obtained by those skilled in the art based on the specific embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0019] To further understand the content, features and effects of the present invention, the following specific embodiments are exemplified and accompanied by Figure 1 - attached Figure 2 The details are as follows:

[0020] Example 1: A method for urban multi-variable spatio-temporal prediction based on a large model proxy, comprising the following steps: S1. Define the urban multi-variable spatio-temporal prediction as the prediction of a discrete random process. To predict the system state at each moment, then set the assumption conditions for the probability of the system state at each moment as the conditional independence assumption and the Markov assumption, and obtain the probability of the system state under the assumption conditions; Further, the specific implementation method of step S1 includes the following steps: S1.1. Set the urban multi-variable spatio-temporal prediction as the prediction of a discrete random process, determine the goal of the urban multi-variable spatio-temporal prediction task as predicting the system state at each moment, and define as a random process, , where is the sample space of is the state space of is the set of time steps, , is the th moment; is the system state random variable at the th moment, taking values from , where , is The number of channels in is the dimension of each channel in Then, the system state prediction at each moment is modeled as a joint distribution of future values of a random variable based on the system state observations at the -th moment and previous moments, resulting in the expression: ; where is the probability density function with as the parameter, is the N -th channel in Furthermore, the problem of distribution drift becomes prominent as the dimension increases, making the parameter estimation of this high-dimensional distribution infeasible in practical calculations. To solve this problem, the urban multivariate spatio-temporal prediction system based on the large model proxy is established on the basis of the following two assumptions; S1.2. Set the conditional independence assumption that given the historical state of the system, the state of a certain channel at the next moment is independent of other channels, that is: ; where is the probability density function, is the n -th channel in n is N any one in S1.3. Set the Markov assumption that the complete conditional probability is simplified to the transition probability between the last two states, and the expression is: ; S2. Decouple the probability of the system state under the assumption conditions obtained in step S1, and decouple the probability prediction of the system state into the prediction of the state of each channel given the previous system state; Furthermore, the specific implementation method of step S2 is to decouple the probability prediction of the system state into the prediction of the state of each channel given the previous system state under the assumption conditions, and the expression is: ; Furthermore, decouple the determined target (i.e., the probability prediction of the system state at each moment) into the prediction of the state of each channel (i.e., each spatial region) given the previous system state; the urban multivariate spatio-temporal prediction system based on the large model agent in this embodiment is applicable to solving the probability prediction process proposed in this inference. That is to say, decouple the prediction of the system state probability distribution (specifically, the probability density function) into the prediction of the probability distribution of the state of each spatial region at the next moment given the previous system state. Furthermore, transform the decoupled system state prediction task into a reinforcement learning task: model the system state probability prediction task as a Markov Decision Process (MDP), that is, an interaction process between a group of agents and the system environment. These agents dynamically adjust their strategies to strengthen their prediction ability by evaluating the difference between the current state and the expected goal. The optimization task of this process is usually called a reinforcement learning task, expressed as a tuple , which includes the state space , the action space , the transition function , and the reward function . The group of agents observe the "state" (the current state of each spatial region), take "actions" (their respective predictions) at each step, and receive "rewards" set according to the expected goal issued by the environment, while the environment performs state "transitions" (the state changes of each spatial region) according to the actions of the agents.

[0021] S3. Based on the Markov hypothesis condition in step S1 and the state decoupling in step S2, construct spatial regions, which include the domain agent layer, the spatio-temporal agent layer, and the individual agent layer; Furthermore, the urban multivariate spatio-temporal prediction system designed in this embodiment consists of three groups of collaborative agent models: the domain agent layer, the spatio-temporal agent layer, and the individual agent layer, and each layer is located at a different spatial scale. The agent model in the domain agent layer plays the role of a "domain expert", and the agent models in the individual agent layer simulate the various behaviors of heterogeneous individuals (people, vehicles, enterprise entities, etc.) in different environments, such as travel, consumption, social interaction, production, etc., while the agent models in the spatio-temporal agent layer predict the state of each variable (energy consumption, population change, etc.) in the region at the next moment through the observed state.

[0022] Furthermore, the specific implementation method of step S3 includes the following steps: S3.1. Construct the domain agent layer: At the initial stage of the urban spatio-temporal state prediction task, set the action space of the i th domain agent to be ; Define the vector space composed of the directed unweighted adjacency matrix with all spatial regions as nodes, which is used to represent the association relationship between the spatial regions of the domain agents. Let j be the th spatial region; Furthermore, the domain agent layer predicts the spatial association. For example, for individual migration or travel behavior prediction, the decision result of the "traffic domain agent" is a definite directed unweighted adjacency matrix . For any region , other regions associated with in can be obtained. These regions are regarded as regions with strong OD (origin-destination) association with . On this premise, the individual agent and the spatio-temporal agent only observe a subset , thus greatly reducing the complexity of the state space. The number of domain agents is set to be variable, and the present invention does not discuss its value.

[0023] S3.2. Construct the individual agent layer: For each individual agent in the j th spatial region , by observing the states of and , set the action space of the k th individual agent to be , is the number of dimensions of the individual decision, is the number of discrete intervals of the individual decision; Furthermore, the individual agent layer predicts the individual strategy. The action space of the k th individual agent is a vector space with a dimension of , where the decision of each dimension is discretized between 1 and , is the unified number of discrete intervals. For example, if the first dimension is migration to other regions, its decision is one of the alternative regions; if the fourth dimension is consumption, its consumption value is classified between 1 and , and so on. The setting of the discrete interval is variable, and the present invention does not discuss its value. The setting of the number of dimensions of the individual decision is extensible. The decision dimensions implemented in the present invention include but are not limited to: residence choice, workplace choice, travel mode preference choice. More decision dimensions under this method are also protected by this patent.

[0024] S3.3. Construct the spatio-temporal agent layer: For thej a spatial region The corresponding spatio-temporal agent observes through and the decision results of all individual agents within, and sets the action space of the j th spatio-temporal agent to be , where is the number of predictive variables; Furthermore, the spatio-temporal agent layer predicts the spatio-temporal state of the system. The action space of the j th spatio-temporal agent is a vector space with a dimension of , which is the statistical value of a total of variables within this spatial region, including but not limited to the number of resident population, the number of employed population, the amount of energy consumption, the amount of GDP, and the number of enterprises. The number of predictive variables is extensible, and more predictive variables under this method are also protected by this patent.

[0025] S4. Based on the domain agent layer, spatio-temporal agent layer, and individual agent layer constructed in step S3, construct an urban multivariate spatio-temporal prediction system based on large model agents, including constructing an action space, a state space, a transition function, a reward function, and a memory stream; Furthermore, the specific implementation method of step S4 includes the following steps: S4.1. Construct an action space: The action space is all possible action combinations of all agents within an urban multivariate spatio-temporal prediction system based on large model agents within one time step, expressed as: ; where , , respectively represent the action spaces composed of all domain agents, spatio-temporal agents, and individual agents, , , respectively represent the total numbers of domain agents, spatio-temporal agents, and individual agents, where each spatial region corresponds to one spatio-temporal agent, that is, ; Furthermore, the overall action space is all possible action combinations of all agents within one time step; S4.2. Construct a state space: The state space represents the integrated information about the urban environment, including individual positions, quantities, and all layouts of urban regions, given through images, vectors, and natural language descriptions; Set the expression of the spatio-temporal state at the th moment to be: as: ; ; Among them, is the facility layout map corresponding to the th moment and the j th spatial region, represented by a multi-channel image; is the vector composed of statistical variables corresponding to the th moment and the j th spatial region, including region size, region center position, resident population quantity, employed population quantity, energy consumption quantity, GDP quantity, enterprise quantity; is the text description corresponding to the th moment and the j th spatial region, represents all string sets, represents a character; Thus, the system state at the th moment is represented as , and all constitute the state space . All agents run in the state space and make different actions according to different environments at different times; S4.3. Construct the transition function: The transition function , according to the state space and the action space, converts actions into states; for the individual agent action space and the spatio-temporal agent action space , statistically calculate the individual change amounts in all regions , and add them to the predicted value of the spatio-temporal agent to obtain the new system state ; Furthermore, in the new system state, each agent can have a new prediction for the next moment, so as to realize the rolling prediction of the current moment, the next moment, the moment after the next moment... and so on; S4.4. Construct the reward function: Furthermore, the reward function is the key to improving the autonomous learning and optimization of agents. In this design, three layers of measurement criteria are designed: individual needs, local fitness, and global fitness.

[0026] S4.4.1. Set the goal of the individual agent to be that the strategy is more beneficial to the individual agent after the system state changes. Considering the living cost, commuting cost, and facility convenience, construct the reward function of the individual agent, ; Among them, is the living cost, is the commuting cost, is the cost of convenient facilities, , , respectively represent the reward coefficients of living cost, commuting cost, and cost of convenient facilities; S4.4.2. The goal of setting the spatio-temporal agent is to make the predicted state as close as possible to the actual state at each time step, so as to achieve the purpose of accurate spatio-temporal prediction. Considering the local state vector, construct the reward function of the spatio-temporal agent , and the expression is: ; Among them, and respectively represent the predicted value and the actual value of the local state vector at a time step, is the reward coefficient of local fitting degree, is the 2-norm; Each dimension of the local state vector includes the number of resident population, the number of employed population, the amount of energy consumption, the amount of GDP, and the number of enterprises; S4.4.3. The goal of setting the domain agent is to limit the sparsity of the adjacency matrix while maintaining the global fitting degree, and obtain the reward function of the domain agent , and the expression is: ; Among them, is the 0-norm, is the total predicted value of the relevant variables under the domain agent, is the total actual value of the relevant variables under the domain agent, is the adjacency matrix, is the reward coefficient of sparsity, is the reward coefficient of global fitting degree; Furthermore, for example, the variables responsible for the "transportation domain agent" include the number of resident population, the number of employed population, and the amount of energy consumption.

[0027] S4.5. Construct the memory stream, including the event stream and the perception stream. The event stream records the events that occur over time, including the actions of the agent, external intervention events, and environmental changes; the perception stream records the agent's tracking and understanding of the events in the event stream, and each node in the perception stream is linked to one or more nodes in the event stream; Furthermore, different from traditional reinforcement learning, the urban multivariate spatio-temporal prediction system based on large model agents records all decision events through a "memory stream" and records the feedback with reward values. Before taking the next action, each agent reads relevant information from the event stream to update the action strategy, so as to complete the optimization and update of the entire agent system. The memory stream tracks events and perceives them over time steps. It consists of two types of memory streams: the event stream and the perception stream. The event stream records the events that occur over time, including the actions of agents, external intervention events, and environmental changes. These events are recorded in chronological order. The perception stream records the agent's tracking and understanding of the events in the event stream. Each node in the perception stream is linked to one or more nodes in the event stream, which reflects how the agent perceives or reacts to specific events. The behavior of the agent is driven by its current state, and the current state affects the decision-making process and the actions taken. By leveraging the memory stream, the agent adjusts its behavior over time in a way that reflects human cognition and expert decision-making, laying the foundation for evolution-based urban spatio-temporal prediction.

[0028] S5. Perform task prediction on the urban multivariate spatio-temporal prediction system based on large model agents constructed in step S4. Record all decision events through the memory stream and record the feedback with reward values. Before taking the next action, each agent reads relevant information from the event stream, uses the transfer function to perform rolling prediction on the action strategy, and optimizes and updates the urban multivariate spatio-temporal prediction system based on large model agents.

[0029] Furthermore, the specific implementation method of step S5 includes the following steps: S5.1. Given the system state at the current moment , first, each domain agent gives its action at the current moment by observing this state. The action set of the domain agent layer is . Similarly, the action sets of the individual agent layer and the spatio-temporal agent layer are and respectively; S5.2. The state transition function converts the action sets of the domain agent layer, the individual agent layer, and the spatio-temporal agent layer in step S5.1 into a new system state at the next moment ; S5.3. All agents obtain their respective rewards from the memory stream according to the moment, adjust their action strategies at the next moment, and thus make new actions in the new system state to complete the rolling prediction.

[0030] It should be noted that relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprising", "including" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the process, method, article or device comprising the said element.

[0031] Although the present application has been described above with reference to specific embodiments, various improvements can be made thereto and components thereof can be replaced with equivalents without departing from the scope of the present application. In particular, as long as there is no structural conflict, the various features in the specific embodiments disclosed in the present application can be combined with each other in any way, and the reason for not exhaustively describing the situations of these combinations in this specification is only to save space and resources. Therefore, the present application is not limited to the specific embodiments disclosed herein, but includes all technical solutions falling within the scope of the claims.

Claims

1. A method for urban multivariate spatiotemporal prediction based on a large model agent, characterized in that: The steps include: S1. Define urban multivariate spatiotemporal prediction as the prediction of discrete random processes, which is to predict the system state at each moment, and then set the assumptions for the probability of the system state at each moment as conditional independence assumption and Markov assumption, and obtain the probability of the system state under the assumptions; S2. Decoupling the probability of the system state under the assumptions obtained in step S1, decoupling the probability prediction of the system state into a prediction of each channel state given the previous system state; S3. Based on the Markov assumption of step S1 and the state decoupling of step S2, a spatial region is constructed, where the spatial region includes a domain agent layer, a spatiotemporal agent layer, and an individual agent layer; S4. Based on the domain agent layer, spatiotemporal agent layer, and individual agent layer constructed in step S3, a large model agent-based urban multivariate spatiotemporal prediction system is constructed, including constructing an action space, a state space, a transfer function, a reward function, and a memory flow; S5. Perform task prediction on the large-model agent-based urban multivariate spatiotemporal prediction system constructed in step S4, record all decision events through the memory stream, and record feedback with reward values. Each agent will read relevant information from the event stream before the next action, and use the transfer function to perform rolling prediction on the action strategy, so as to optimize and update the large-model agent-based urban multivariate spatiotemporal prediction system.

2. The urban multivariate spatiotemporal prediction method based on a large model agent according to claim 1 is characterized in that: The specific implementation method of step S1 includes the following steps: S1.

1. Set the urban multivariate spatiotemporal prediction as a discrete random process prediction, and define the urban multivariate spatiotemporal prediction task as predicting the system state at each moment. is a random process, ,in, yes The sample space of yes The state space of is the set of time steps, , It is a moment; For the The system state random variable at the moment is Take the value in ,in, for The number of channels in for The dimension of each channel in ; The system state at each moment is predicted as a The joint distribution of the future values ​​of the random variables of the system state at and before the moment is expressed as: ； in, For is the probability density function of the parameter, for The N Channels: S1.

2. Set the conditional independence assumption to be that given the historical state of the system, the next state of a channel is independent of other channels, that is: ； in, is the probability density function, for The n channels, n for N Any one of; S1.

3. Setting the Markov assumption, the complete conditional probability is simplified to the transition probability between the last two states, expressed as: 。 3. The urban multivariate spatiotemporal prediction method based on a large model agent according to claim 2 is characterized in that: The specific implementation method of step S2 is to decouple the probability prediction of the system state into the prediction of each channel state given the previous system state under the assumption that: 。 4. The urban multivariate spatiotemporal prediction method based on a large model agent according to claim 3 is characterized in that: The specific implementation method of step S3 includes the following steps: S3.

1. Constructing the domain proxy layer: In the initial stage of the urban spatiotemporal state prediction task, set up the first i The action space of the domain agent is ; Defined by all spatial regions It is a vector space composed of directed unweighted adjacency matrices of nodes, which is used to represent the relationship between the spatial regions of domain agents. For the j space area, ; S3.

2. Constructing the individual agent layer: For j Space area Each individual agent in and status, set the k The action space of each individual agent is , is the number of dimensions of individual decision making, The number of discretization intervals for individual decisions; S3.

3. Constructing the spatiotemporal proxy layer: For j Space area The corresponding space-time agent observes and The decision results of all individual agents in j The action space of a spatiotemporal agent is , is the number of predictor variables.

5. The urban multivariate spatiotemporal prediction method based on a large model agent according to claim 4 is characterized in that: The specific implementation method of step S4 includes the following steps: S4.

1. Constructing the Action Space: Action Space is all possible action combinations of all agents in a time step in the urban multivariate spatiotemporal prediction system based on a large model agent, expressed as: ； in, , , Respectively represent the action space composed of all domain agents, spatiotemporal agents and individual agents, , , They represent the total number of domain agents, spatiotemporal agents, and individual agents, respectively, where each spatial region corresponds to a spatiotemporal agent, i.e. ; S4.

2. Constructing the state space: State space Represents integrated information about the urban environment, including individual locations, quantities, and all layouts of urban areas, given by images, vectors, and natural language descriptions; Set the At this moment The space-time state The expression is: ； ； in, For the Moment j The facility layout map corresponding to each spatial area is represented by a multi-channel image; For the Moment j A vector of statistical variables corresponding to each spatial region, including region size, location of regional center, number of residents, number of employed people, number of energy consumption, number of GDP, and number of enterprises; For the Moment j The text description corresponding to the spatial region, Represents the set of all strings, Represents a character; Therefore, the The system state at a certain moment is expressed as , all Constructing the state space , all agents operate in the state space and take different actions at different times according to different environments; S4.

3. Constructing the transfer function: Transfer function , according to the state space and action space, the action is converted into state; for the individual agent action space and the spatiotemporal agent action space , Statistics for all regions The individual changes in Add together to get the new system state ; S4.

4. Constructing reward function: S4.4.

1. The goal of setting individual agents is to make the strategy more favorable to individual agents after the system state changes. Considering the cost of living, commuting costs, and the convenience of facilities, the reward function of individual agents is constructed. , the expression is: ； in, For living costs, For commuting costs, For the cost of facilities, , , Represent the reward coefficients of living cost, commuting cost, and facility convenience cost respectively; S4.4.

2. The goal of setting the spatiotemporal agent is to make the predicted state at each time step as close as possible to the actual state, so as to achieve accurate spatiotemporal prediction. Consider the local state vector to construct the reward function of the spatiotemporal agent , the expression is: ； in, and Respectively represent the predicted value and actual value of the local state vector at a time step, is the reward coefficient of local fitness, is the 2-norm; Each dimension of the local state vector includes the number of residents, the number of employed people, the amount of energy consumption, the amount of GDP, and the number of enterprises; S4.4.

3. The goal of setting the domain agent is to limit the sparsity of the adjacency matrix while maintaining the global fit, and obtain the reward function of the domain agent , the expression is: ； in, is the 0 norm, is the total predicted value of the relevant variables under the domain proxy, is the total actual value of the relevant variables under the domain agent, is the adjacency matrix, is the reward coefficient for sparsity, is the reward coefficient of global fitness; S4.

5. Construct memory streams, including event streams and perception streams. The event stream records events that occur over time, including the agent's actions, external intervention events, and environmental changes; the perception stream records the agent's tracking and understanding of events in the event stream. Each node in the perception stream is linked to one or more nodes in the event stream.

6. The urban multivariate spatiotemporal prediction method based on a large model agent according to claim 5 is characterized in that: The specific implementation method of step S5 includes the following steps: S5.

1. Given the current system state First, each domain agent gives its current action by observing the state. The action set of the domain agent layer is , similarly, the action sets of the individual agent layer and the spatiotemporal agent layer are and ; S5.

2. State transition function Convert the action set of the domain agent layer, the action set of the individual agent layer, and the action set of the spatiotemporal agent layer in step S5.1 into the new system state at the next moment ; S5.

3. All agents are fed from the memory stream according to Get their respective rewards at all times, adjust their action strategies for the next moment, and thus Make new actions to complete the rolling prediction.

Images