Urban system calculation simulation platform method driven by artificial intelligence and related device

Abstract

The invention discloses an artificial intelligence-driven city system calculation simulation platform method and a related device, and relates to the technical field of artificial intelligence and digital simulation. The method comprises the following steps: acquiring urban basic data to generate a plurality of initial strategies, and then sampling to obtain a first reference strategy; deducing the first reference strategy through the strategy calculation model to obtain a first reference deduction result, and adding the result to a strategy recommendation set; screening the at least two initial strategies through a target agent model to obtain a second reference strategy; and deducing each second reference strategy through the strategy calculation model to obtain a second benchmark deduction result, and updating the strategy recommendation set according to the second reference strategies and the second benchmark deduction result to perform strategy recommendation to the user. According to the method, city system strategy recommendation can be realized, and user regulation and control requirements are met.

Description

Technical Field

[0001] This application relates to the fields of artificial intelligence and digital simulation technologies, and particularly to a method and related device for an artificial intelligence-driven urban system computational simulation platform. Background Art

[0002] The operation and development status of an urban system is mainly dynamically composed of multiple subsystems such as land use evolution and transportation. All elements, flows, functions, facilities, environments, etc. are interconnected and interact with each other, supporting the life processes of the urban system such as growth, decline (shrinkage), illness, renewal, and self-adaptation.

[0003] Traditional urban system computational simulation technologies (e.g., UrbanSim model technology) take the main sectors (households, businesses, developers, etc.) in the urban development process as the research object, and conduct short-term to long-term dynamic simulations on an annual basis, mainly for predicting the long-term patterns of land use and transportation under various regulatory scenarios. However, for current urban system computational simulation technologies, their core functional modules mainly focus on simulation analysis rather than prediction functions. Most of these computational simulations adopt the BAU scenario (Business-as-Usual Scenario), that is, the baseline scenario, which is used to describe the natural state in which urban elements develop according to the existing trends without additional policy intervention, resulting in difficulties in effectively evaluating the policy responsiveness of the prediction results and predicting the specific effects and dynamic changes of the urban system after implementing the policies, thus limiting their application value in policy formulation and evaluation.

[0004] Therefore, how to recommend regulatory strategies for the urban system has become an urgent technical problem to be solved. Summary of the Invention

[0005] This application aims to at least solve one of the technical problems existing in the prior art. For this purpose, this application proposes a method, device, equipment, and medium for an artificial intelligence-driven urban system computational simulation platform, which can recommend strategies for the urban system and meet the regulatory requirements of users. In addition, it can also effectively evaluate and predict the specific effects and dynamic changes of the urban system after implementing the strategies.

[0006] To achieve the above object, a first aspect of the embodiments of this application proposes a method for an artificial intelligence-driven urban system computational simulation platform, and the method includes: Obtain urban basic data for urban system calculation, perform simulation calculation on the urban basic data, and obtain urban calculation element data; Generate at least two initial strategies according to the urban calculation element data; Sample the at least two initial strategies to obtain at least two first reference strategies; Deduce each of the first reference policies through a pre-trained policy calculation model to obtain a first benchmark deduction result, and add the first reference policy and the first benchmark deduction result to the policy recommendation set; Train an initial agent model according to the policy recommendation set to obtain a target agent model; the target agent model includes a plurality of base models and a meta-model for making predictions based on the outputs of the plurality of base models; Deduce each of the initial policies through the plurality of base models in the target agent model to obtain pseudo-calculation results; Screen the at least two initial policies according to the pseudo-calculation results to obtain second reference policies; Deduce each of the second reference policies through the policy calculation model to obtain a second benchmark deduction result, and update the policy recommendation set according to the second reference policy and the second benchmark deduction result; Obtain a regulation requirement target input by a user, determine a target policy from the policy recommendation set according to the regulation requirement target, and recommend the target policy to the user.

[0007] Optionally, the candidate policies in the policy recommendation set are used to execute multiple urban optimization tasks; The determining the target policy from the policy recommendation set according to the regulation requirement target includes: If the policy target is one of the multiple urban optimization tasks, set a first weight for the urban optimization task corresponding to the policy target, and set a second weight for the other urban optimization tasks, the first weight being greater than the second weight; Score the multiple candidate policies in the policy recommendation set according to the first weight and the second weight to obtain a policy recommendation score for each candidate policy; Sort the multiple candidate policies from high to low according to the policy recommendation scores to obtain a policy sorting result; Determine the candidate policy corresponding to the highest policy recommendation score as the target policy according to the policy sorting result.

[0008] Optionally, after determining the candidate policy corresponding to the highest policy recommendation score as the target policy according to the policy sorting result, the method further includes: Traverse the candidate policies in the policy sorting result in sequence, calculate the distance between the target policy and each candidate policy to obtain a policy distance; If the policy distance is greater than a predetermined threshold, determine the candidate policy as an auxiliary policy; Recommend the auxiliary policy to the user.

[0009] Optionally, screening the at least two initial strategies according to the pseudo-computation result to obtain a second reference strategy includes: Determining an expected improvement value generated by each of the initial strategies when executing each of a plurality of urban optimization tasks according to the pseudo-computation result; wherein, the plurality of urban optimization tasks include at least one of the following: implementation cost reduction task, environmental impact reduction task, implementation time shortening task; Obtaining an optimization weight set for each of the urban optimization tasks; Performing a weighted sum according to the optimization weight and the expected improvement value to obtain a target optimization score; Selecting the at least two initial strategies from high to low according to the target optimization score to obtain the second reference strategy.

[0010] Optionally, determining the expected improvement value generated by each of the initial strategies when executing each of a plurality of urban optimization tasks according to the pseudo-computation result includes: Calculating a pseudo-computation mean value and a pseudo-computation standard deviation according to the pseudo-computation results output by a plurality of the basic models; Performing a difference calculation according to the current best function value, the pseudo-computation mean value, and a preset balance parameter to obtain a pseudo-computation weight; Performing a ratio calculation according to the pseudo-computation weight and the pseudo-computation standard deviation to obtain a pseudo-computation ratio; Mapping the pseudo-computation ratio through a preset cumulative distribution function to obtain a pseudo-computation distribution function; Multiplying the pseudo-computation weight and the pseudo-computation distribution function to obtain a first function term; Mapping the pseudo-computation ratio through a preset probability density function to obtain a pseudo-computation probability density function; Multiplying the pseudo-computation standard deviation and the pseudo-computation probability density function to obtain a second function term; Summing the first function term and the second function term to obtain an expected improvement function; Calculating the expected improvement value according to the expected improvement function.

[0011] Optionally, training an initial surrogate model according to the strategy recommendation set to obtain a target surrogate model includes: Dividing the strategy recommendation set according to the total number of a plurality of basic models of the initial surrogate model to obtain a strategy recommendation subset for each of the basic models; Training the basic model according to the strategy recommendation subset for each of the basic models to update the basic model; Deduce the first reference policy in the policy recommendation subset through each of the base models to obtain a base prediction result; Train the meta-model of the initial proxy model based on the first reference policy and the base prediction result corresponding to the multiple base models to obtain the target proxy model.

[0012] Optionally, sampling the at least two initial policies to obtain at least two first reference policies includes: Stratify the at least two initial policies to obtain multiple policy layers; Randomly select the initial policies in each policy layer to obtain the first reference policy for each policy layer; Shuffle the first reference policies of the multiple policy layers to obtain the at least two first reference policies.

[0013] To achieve the above object, a second aspect of the embodiments of the present application proposes an artificial intelligence-driven urban system computing simulation platform device, the device includes: A data acquisition module, configured to acquire urban basic data for urban system computing, perform simulation calculations on the urban basic data to obtain urban computing element data; A policy generation module, configured to generate at least two initial policies according to the urban computing element data; A policy sampling module, configured to sample the at least two initial policies to obtain at least two first reference policies; A first deduction module, configured to deduce each of the first reference policies through a pre-trained policy calculation model to obtain a first benchmark deduction result, and add the first reference policy and the first benchmark deduction result to a policy recommendation set; A proxy model training module, configured to train an initial proxy model according to the policy recommendation set to obtain a target proxy model; the target proxy model includes multiple base models and a meta-model for predicting based on the outputs of the multiple base models; A second deduction module, configured to deduce each of the initial policies through the multiple base models in the target proxy model to obtain a pseudo-calculation result; A policy screening module, configured to screen the at least two initial policies according to the pseudo-calculation result to obtain a second reference policy; A third deduction module, configured to deduce each of the second reference policies through the policy calculation model to obtain a second benchmark deduction result, and update the policy recommendation set according to the second reference policy and the second benchmark deduction result; A policy recommendation module, configured to obtain a regulation requirement target input by a user, determine a target policy from the policy recommendation set according to the regulation requirement target, and recommend the target policy to the user.

[0014] To achieve the above object, a third aspect of the embodiments of the present application provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the method described in the first aspect above is implemented.

[0015] To achieve the above object, a fourth aspect of the embodiments of the present application provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the method described in the first aspect above is implemented.

[0016] The artificial intelligence-driven urban system computing simulation platform method, device, equipment, and medium proposed in the present application consider the uncertainty of urban system development, perform simulation operations on urban computing element data (such as multiple elements of the urban system scale, function, structure, spatial planning, etc.) to generate initial policies. Further, instead of directly evaluating and predicting the specific effects and dynamic changes of the urban system after implementing the policies for all initial policies, all initial policies are first sampled, and then a urban computing model is used for calculation to obtain an initial policy recommendation set. Further, a target agent model is trained based on the policy recommendation set, and multiple basic models in the target agent model are used to calculate the policies to obtain pseudo-calculation results; the most suitable policy is selected based on the pseudo-calculation results, and then a urban computing model is used for reasoning and added to the policy recommendation set. This process is executed asynchronously to continuously improve the policy recommendation set. Finally, according to the regulation requirement target input by the user and the policy recommendation set, the most suitable policy is screened and recommended to the user. Policy recommendation is used to help users find suitable policies in multiple target dimensions (implementation cost, environmental impact, implementation time) to meet the user's regulation requirements.

[0017] The additional aspects and advantages of the present application will be partially given in the following description, partially become obvious from the following description, or be understood through the practice of the present application. Description of the Drawings

[0018] Figure 1 It is an architecture diagram of the system to which the artificial intelligence-driven urban system computing simulation platform method provided by the embodiments of the present application is applied; Figure 2 It is a flowchart of the artificial intelligence-driven urban system computing simulation platform method provided by the embodiments of the present application; Figure 3It is another flowchart of the method for the artificial intelligence-driven urban system computing simulation platform provided by the embodiments of the present application; Figure 4 It is Figure 2 the flowchart of step 203 in Figure 5 It is Figure 2 the flowchart of step 207 in Figure 6 It is Figure 2 the flowchart of step 209 in Figure 7 It is the structural schematic diagram of the artificial intelligence-driven urban system computing simulation platform device provided by the embodiments of the present application. Detailed implementation manners

[0019] In order to make the objectives, technical solutions and advantages of the present application more clear and understandable, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0020] It should be noted that although functional module division is performed in the device schematic diagram and the logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different module division in the device or a different order in the flowchart. Terms such as "first" and "second" in the specification, claims and the above-mentioned drawings are used to distinguish similar objects and do not necessarily need to be used to describe a specific order or sequence.

[0021] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the technical field to which the present application belongs. The terms used herein are only for the purpose of describing the embodiments of the present application and are not intended to limit the present application.

[0022] The method for the artificial intelligence-driven urban system computing simulation platform provided by the embodiments of the present application can be applied to any one of the terminal and the server side, and can also be software running on the server side or the terminal. The server side can be configured as an independent physical server, or can be configured as a server cluster or distributed system composed of multiple physical servers, and can also be configured as a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; the software can be an application or computer program that implements the method for the artificial intelligence-driven urban system computing simulation platform, but is not limited to the above forms.

[0023] The method for an artificial intelligence-driven urban system computing simulation platform, the device for an artificial intelligence-driven urban system computing simulation platform, the computer device, and the computer-readable storage medium provided by the embodiments of the present application will be specifically described through the following embodiments. First, the method for an artificial intelligence-driven urban system computing simulation platform in the embodiments of the present application will be described.

[0024] The present application provides a method / technology for an artificial intelligence-driven urban system computing simulation platform. This platform is a digital space planning and urban planning decision-making assistance platform that integrates functions such as computing, simulation, evaluation, visualization, and data analysis. The platform is mainly applicable to key applications such as optimizing the spatial development pattern and improving the comprehensive transportation system.

[0025] Referring to Figure 1 , the present application realizes three major contents: a data management system, an urban computing engine, and a customer service platform based on artificial intelligence technologies such as SMBO, DeepSeek, and RAG.

[0026] I. Data management system, mainly used for data acquisition and management: The data management system is a process tool for data to enter the server (database) storage from local storage. It provides two main functions: data standardization verification and parameter automatic calculation, realizing data processing functions such as automatic review, import, cleaning calculation, reclassification, and uploading of initial complex data, and providing data to the urban computing engine. The data management platform includes two core functions. One is the "import" function, which mainly conducts automatic standardization review on various data in the "urban standard information library" to ensure that the data items are complete and the data attribute fields meet the requirements before importing into the urban computing engine. The other is the "calculation" function, which mainly realizes automatic calculation of regional parameters. This function mainly aims at different types and scales of urban systems, and obtains a set of regional parameters for the urban system through automatic calculation, so as to ensure the universality of the CitySPS platform when running in different urban systems.

[0027] (1) The "import" function specifically includes: 1. Standardization review: Review the folder name, file name, file type, file encoding, unique identifier, number of data rows, number of data columns, and format of each column of the data before warehousing. 2. Completeness review: Review whether the data quantity meets the calculation requirements. This function requires the parameter automatic function as a prerequisite. 3. Data cleaning and import into the database: In case of data that does not meet the calculation conditions, first return a review result list, and second, automatically clean and correct the data that can be automatically corrected, and return it to the data collection personnel for cleaning and modification if not. After meeting the conditions, import it into the production database.

[0028] (2) The "Calculation" function includes the automatic calculation of regional parameters, specifically calculating the parameters required for the subsequent formal "Urban Development Trend Deduction" calculation, and completing the process from data collection, parameter calculation to formal calculation. The specific calculation process includes: 1. Land module, calculating the number of buses in the grid, the growth of the number of buses, the number of subway lines, etc. 2. Population module, calculating the job-housing coefficient, preprocessing grid population data, etc. 3. Transportation module, generating bus data, calculating data such as car ownership rate, etc. The core calculation principle includes using methods such as crawlers and linear regression for parameter calculation, data acquisition, etc.

[0029] II. Urban Calculation Engine: The urban calculation engine is the simulation calculation support module of this platform, which is the code implementation of the core algorithm of this platform, that is, the principle algorithm of the urban full-system measurement model. With the support of the high-performance server software and hardware environment, it executes the core functions of urban calculation, supports the platform to achieve core calculations such as urban calculation and decision simulation, and provides the result data after analysis and calculation to users. The urban calculation engine contains four internal modules, namely the data import layer, the data persistence and management layer, the technical solution layer, and the function and access layer.

[0030] The function of the data import layer is to receive the standardized variables and regional parameters of the data management platform, quantify the data collected in the early stage, including data in the fields of population, land, transportation, industry, etc. of the city, and output the standard data format required for system calculation. At the same time, in view of the current situation that the data of each city has different regional characteristics, it is necessary to perform multi-model training on the historical data of the corresponding city to obtain the regional parameters unique to each city. Finally, the parameter and the standardized data are uniformly imported into the urban system calculation and simulation platform, and the system prediction calculation can be started.

[0031] The function of the data persistence and management layer is to perform systematic storage and management of various data based on database technology. The database of this platform involves different attribute data of multiple cities. Therefore, in terms of storage, a method of allocating one schema for each city is adopted to achieve data isolation and security guarantee, avoid data confusion and incorrect operations between different cities, and is more convenient for data management and maintenance. At the same time, in a single city schema, the platform sets up a user table to save the IDs corresponding to all users. At the same time, each user ID corresponds to multiple calculation projects created by it, and a project table is obtained, as shown in Tables 1 and 2.

[0032] Table 1

[0033] Table 2

[0034] The technical solution layer includes big data technology, quantitative simulation technology, urban element deduction technology, and scenario analysis technology that support platform computing, and it is also the core functional layer that supports the urban computing engine to complete calculations. First, the big data technology processes the underlying input data required for calculations, specifically including basic geographic information data (such as administrative division map data, road network vector data, land use data, etc.), multi-source spatio-temporal big data (such as mobile phone signaling data, POI data, etc.), statistical data (such as census data for multiple years, urban statistical yearbooks, etc.), and resident travel survey data; the quantitative simulation technology is the computing support for completing a "simulation of urban decision-making scenarios", specifically by conducting systematic quantitative analysis and simulation calculations on elements such as land, population, ecology, industrial chain, and transportation within the city; the urban element deduction technology is the computing support for completing a simulation and deduction of urban development trends, based on the core algorithm "technology for constructing and simulating an urban system model empowered by artificial intelligence"; the "scenario analysis technology" is the urban decision-making recommendation support technology for processing "simulations of urban decision-making scenarios", and can simulate and predict the urban population and employment distribution, traffic demand distribution, traffic mode share, and route allocation, etc.

[0035] The Function and Access Layer directly provides user application functions such as computing functions, data result browsing functions, and data access support to the customer service platform. It is reflected in the overall interface of the platform, such as modules for urban development trend deduction, urban decision-making scenario simulation, etc. The Function and Access Layer includes a "Function Module" and an "Access Module". First, the "Function Module" is related to the analysis and calculation module, data management module, and system management module, mainly providing functional services to users. The analysis and calculation module contains a city computing engine for a complete operation. Through the sequential operation of modules such as the urban land use simulation and evolution module, population and employment distribution module, traffic demand distribution module, traffic mode sharing and route allocation module, etc., calculations and analyses are carried out. Users obtain calculation results through the analysis and calculation module, and the calculation results are transmitted to the data management module in the form of output indicators; the data management module conducts standardized management on the received output indicator data and provides users with data storage and search functions based on calculation projects; finally, the system management module provides comprehensive account management functions, including background account management, software version management, help center, etc. The "Access Module" is a support module that ensures the normal operation of functional services and includes API gateway, reverse proxy, and load balancing functions. The API gateway function is an intermediate layer between the client and the backend. As an entry point, it processes multiple API calls from, including access authentication, user authorization, access control, etc. Users can access the platform through the API gateway function; the reverse proxy is a server. Users send requests to the reverse proxy server through the API gateway. The reverse proxy server selects the target server to obtain data and then returns this data to the client. Using the reverse proxy avoids direct interaction between the client and independent microservices, reduces the number of service calls, and controls traffic; the load balancer collects call information of backend services, such as the number of connections, response time, etc., horizontally scales the cluster according to the collected information, distributes requests to multiple processing nodes, shares the processing pressure of the system, and improves the operation efficiency and performance of the system.

[0036] This platform uses a fusion architecture for function development. The engineering architecture of the platform mainly operates based on the browser / server - Browser / Server mode (B / S architecture) and is built using the front-end and back-end separation mode. The front-end page is developed using the VUE framework. The back-end server provides basic services such as system management, service scheduling, and data access, responds to requests from the user side, and returns data. It interacts with the front-end through the Restful-API method and mainly transmits data in the lightweight data exchange format json.

[0037] The backend service hardware of this platform includes four types: interface server, computing server, database server, and scheduling server. The computing server is the core component of the backend module, mainly deploying and running the algorithm model code of the CitySPS platform model, performing analysis and calculations through high-performance computing servers, and pushing the results into the database server. The server hardware uses multi-core high-performance CPUs and has a relatively large amount of operating memory. The model algorithm code runs in multiple processes to improve the computing and processing efficiency.

[0038] III. Customer Service Platform: The customer service platform is an external service function module and a user operation platform. The customer service platform runs based on the browser side, realizing functions such as user demand interaction, displaying urban calculation results, and providing visual displays, and providing the core functions of urban intelligent decision-making to users: ① Deduction of urban development trends, ② Simulation of urban decision-making scenarios, ③ LLM artificial intelligence semantic assistance system.

[0039] It should be noted that the method of the artificial intelligence-driven urban system calculation and simulation platform provided in this application is mainly realized based on the urban decision-making scenario simulation function.

[0040] ① Deduction of urban development trends: The function of deducing urban development trends integrates spatial current situation information resources based on multi-source big data. By constructing an urban system model and using urban mechanism algorithms or urban machine learning algorithms, from the perspectives of total population and spatio-temporal distribution, land use scale and land function evolution, urban traffic routes and individual travel chains, and the supply-demand matching relationship of service facilities, etc., it simulates and evaluates the evolution trends of the core elements of the urban system within the planning period, providing a reference for urban system planning.

[0041] ② Simulation of urban decision-making scenarios: The function of simulating urban decision-making scenarios is oriented to the planning and development of space, and performs simulation operations on multiple elements such as the scale, function, structure, and spatial planning of the urban system. This function module conducts implementation simulations through more than twenty strategies, identifies and compares the typical characteristics before and after implementation, and provides a reference for urban construction and urban management. Build an intelligent decision-making support system, and use reinforcement learning algorithms to simulate and optimize various strategies of the urban system. For example, in urban planning, by simulating different land use strategies, evaluating their impacts on traffic and the environment, and selecting the optimal strategy. Establish a decision evaluation mechanism to evaluate the effects of the implemented strategies and provide feedback. Use deep learning algorithms to analyze the results after implementing the strategies, evaluate their impacts on the urban system, and provide a reference for subsequent decision-making.

[0042] ③ Semantic assistance system: The LLM (Large Language Model) language help system based on DeepSeek and RAG first classifies the user's question to determine whether it is relevant to the document. If the question is relevant to the document, the relevant context is first retrieved through RAG, and then the initial answer is generated, and the answer is further optimized and finally returned to the user. If the question is not related to the document, sentiment analysis is first performed, and then a polite response is generated and returned to the user.

[0043] In order to optimize the user experience, the DeepSeek model is used to classify user questions into two categories: document-related and document-irrelevant. This classification is based on the preset prompt and determines the strategy for subsequent answers: document-related questions will generate answers through knowledge base retrieval, while document-irrelevant questions will provide general answers.

[0044] For document-related questions, RAG (retrieval-augmented generation) technology is used for processing. First, the document is loaded through TextLoader, and the text is split into segments suitable for retrieval using TextSplitterFactory. Subsequently, FAISS is used to vectorize the text segments and store the results for fast retrieval. In the retrieval stage, documents related to the question are extracted from the document library by combining keyword matching and vector retrieval. Specifically, the keywords in the question are first extracted through Jieba word segmentation, and then two types of retrieval are performed in parallel: one is a retrieval based on keyword matching, which scores by calculating the frequency of keywords in the document; the other is a retrieval based on vector similarity, which obtains relevant documents by calculating the vector similarity between the question and the document. The two retrieval results are merged and sorted by weight (for example, keyword weight 0.6, vector weight 0.4), and the top 5 document contents are selected after deduplication to form the context. The entire retrieval process improves efficiency through asynchronous execution, and optimizes the quality of results through hybrid sorting, and finally generates the most relevant context content to the question. After the retrieval is completed, the DeepSeek model is used to generate the initial answer based on the retrieved context and the user question to ensure that the answer is concise and fully utilizes the context information. If no relevant context is found, the user is informed that the answer cannot be given. Finally, the DeepSeek model is used to optimize the answer, remove unnecessary or repeated content, retain the core information, and obtain the final answer.

[0045] For document-irrelevant questions, we first use the DeepSeek model to analyze the user’s emotions, including emotional inclination, urgency, and whether they need to be comforted. Based on the results of the sentiment analysis, we call the DeepSeek model again to generate friendly and professional responses to meet the user’s emotional needs.

[0046] The LLM semantic assistance system developed by this platform accurately realizes precise semantic understanding and generation functions based on existing artificial intelligence technologies, provides efficient and intelligent semantic assistance services for users, meets the diverse platform semantic processing requirements, and realizes the application of artificial intelligence in the urban complex system computing and simulation platform.

[0047] Please refer to Figure 2 , Figure 2 which discloses an alternative flowchart of a method for an artificial intelligence-driven urban system computing simulation platform. Figure 2 The method in

[0048] may include but is not limited to steps 201 to 209. Step 201, obtain urban basic data for urban system computing, perform simulation calculations on the urban basic data, and obtain urban computing element data; Step 202, generate at least two initial strategies according to the urban computing element data; Step 203, sample the at least two initial strategies to obtain at least two first reference strategies; Step 204, deduce each first reference strategy through a pre-trained policy calculation model to obtain a first benchmark deduction result, and add the first reference strategy and the first benchmark deduction result to the policy recommendation set; Step 205, train the initial agent model according to the policy recommendation set to obtain a target agent model; the target agent model includes multiple basic models and a meta-model for predicting based on the outputs of the multiple basic models; Step 206, deduce each initial strategy through the multiple basic models in the target agent model to obtain pseudo-computation results; Step 207, screen the at least two initial strategies according to the pseudo-computation results to obtain second reference strategies; Step 208, deduce each second reference strategy through the policy calculation model to obtain a second benchmark deduction result, and update the policy recommendation set according to the second reference strategy and the second benchmark deduction result;

[0049] Steps 201 to 209 shown in the embodiments of the present application perform simulation operations on urban computing element data (such as multiple elements like the scale, function, structure, and spatial planning of the urban system) to generate initial strategies. Further, instead of directly evaluating all initial strategies and predicting the specific effects after strategy implementation and the dynamic changes of the urban system, all initial strategies are first sampled, and then calculated using the urban computing model to obtain an initial set of strategy recommendations. Further, a target agent model is trained based on the set of strategy recommendations, and multiple basic models in the target agent model are used to calculate the strategies to obtain pseudo-calculation results; the most suitable strategy is selected based on the pseudo-calculation results, and then reasoning is performed using the urban computing model and added to the set of strategy recommendations, and this process is executed asynchronously to continuously improve the set of strategy recommendations. Finally, according to the regulation requirement target input by the user and the set of strategy recommendations, the most suitable strategy is selected and recommended to the user. The strategy recommendations are used to help the user find suitable strategies in multiple target dimensions (implementation cost, environmental impact, implementation time) to meet the user's needs.

[0050] In one example, referring to Figure 3 , the method of the artificial intelligence-driven urban system computing simulation platform may include: (1) Generate all strategy combinations: Generate at least two initial strategies; (2) Strategy sampling: Sample at least two initial strategies to obtain at least two first reference strategies; (3) Deduce strategies using the urban computing model: Deduce each first reference strategy through the urban computing model to obtain a first benchmark deduction result; (4) Update the set of strategy recommendations: Add the first reference strategy and the first benchmark deduction result to the set of strategy recommendations; (5) Agent model training: Train the initial agent model according to the set of strategy recommendations to obtain the target agent model; (6) Candidate strategy evaluation: Use the trained target agent model to predict each initial strategy to obtain approximate calculation results, and then filter the initial strategies according to the approximate calculation results; (7) Deduce the regulation results using the agent model: Deduce each initial strategy through multiple basic models in the target agent model to obtain pseudo-calculation results; (8) Select the optimal strategy: Screen at least two initial strategies according to the pseudo-calculation results to obtain a second reference strategy; (9) Deduce strategies using the urban computing model: Deduce each second reference strategy through the strategy calculation model to obtain a second benchmark deduction result, and then enter (4) to update the set of strategy recommendations according to the second reference strategy and the second benchmark deduction result; (10) User sets the regulation demand target: Obtain the regulation demand target input by the user; (11) Obtain the strategy recommendation result: Determine the target strategy from the strategy recommendation set according to the regulation demand target; (12) Conduct strategy recommendation: Recommend the target strategy to the user.

[0051] In step 201, the urban basic data includes basic geographic information data (such as administrative division map data, road network vector data, land use data, etc.), multi-source spatio-temporal big data (such as mobile phone signaling data, POI data, etc.), statistical data (such as population census data, economic census data, urban statistical yearbooks of multiple years, etc.) and resident travel survey data. The urban basic data can also include urban basic image data (population census data statistical charts, urban statistical annual inspection charts, etc.), urban basic text data (such as administrative division map data, road network vector data, land use data), etc.

[0052] The urban basic data can be simulated and deduced through quantitative simulation technology and urban element deduction technology to obtain urban calculation element data. The quantitative simulation technology is the computational support for completing a "simulation of urban decision-making scenarios". Specifically, it conducts systematic quantitative analysis and simulation calculations on elements such as land, population, ecology, industrial chain, and transportation within the city to obtain urban calculation element data. For a detailed introduction to the quantitative simulation technology, please refer to the above text and will not be elaborated here.

[0053] In step 202, at least two initial strategies are generated according to the urban calculation element data. The initial strategy includes at least one strategy, that is, the initial strategy is a strategy combination. For example, the initial strategy includes the strategy of "increasing the proportion of green buildings". Another example is that the initial strategy includes the strategy of "adjusting the land supply structure". Another example is that the initial strategy includes the strategy of "increasing the proportion of green buildings" and the strategy of "adjusting the land supply structure".

[0054] In an example, in urban planning, multiple strategies are generated according to the urban calculation element data, and at least two of the multiple strategies are permuted and combined to obtain various different land use strategies. It can be generated by a large language model or other methods, and the present application does not make specific limitations on this.

[0055] In step 203, at least two initial strategies are sampled to obtain at least two first reference strategies. The sampling method can be randomly sampled.

[0056] In one embodiment, referring to Figure 4 , step 203 may include: Step 401, stratify at least two initial strategies to obtain multiple strategy layers; Step 402: Randomly select the initial policies in each policy layer to obtain the first reference policies for each policy layer. Step 403: Shuffle the first reference policies of multiple policy layers to obtain at least two first reference policies.

[0057] In Step 401, the number of policies of different initial policies is different. The number of policies of at least two initial policies can be stratified to obtain multiple policy layers. For example, the number of policies in the first policy layer is 1, the number of policies in the second policy layer is 2, the number of policies in the third policy layer is 3, and so on. In this way, stratifying based on the number of policies can closely reflect the characteristic that the initial policy is a policy combination, greatly improving the stratification accuracy.

[0058] In Step 402, for at least one initial policy in each policy layer, at least one first reference policy can be randomly selected therefrom. For example, 2 first reference policies are randomly selected from the first policy layer, 2 first reference policies are randomly selected from the second policy layer, and 1 first reference policy is randomly selected from the third policy layer.

[0059] In one embodiment, Step 402 may include: determining a selection threshold according to the number of policies corresponding to each policy layer; randomly selecting the initial policies in each policy layer according to the selection area to obtain the first reference policies whose number of policies is less than or equal to the selection threshold. The relationship between the number of policies and the selection threshold is an inverse relationship, that is, the larger the number of policies, the smaller the selection threshold. For example, if the number of policies in the first policy layer is 1, the selection threshold is 10; if the number of policies in the second policy layer is 2, the selection threshold is 5; if the number of policies in the third policy layer is 3, the selection threshold is 3. In this way, determining the selection threshold for each policy layer based on the number of policies can effectively make the number of policies in the selected first reference policies smaller, increase the probability of the first reference policies passing subsequent evaluations, and improve the reliability and accuracy of policy selection.

[0060] In Step 403, shuffle the selected first reference policies to improve the randomness of subsequent policy evaluations, thereby improving the evaluation accuracy.

[0061] The advantage of the embodiments of the above Steps 401 to 403 is that it improves both the distribution rationality and randomness of the selected policies in the sampling process, and improves the sampling accuracy.

[0062] In step 204, each first reference policy is deduced through a pre-trained policy calculation model to obtain a first benchmark deduction result, and the first reference policy and the first benchmark deduction result are added to the policy recommendation set. The policy calculation model is a pre-trained machine learning model for policy deduction, which can deduce the execution effect / execution degree of executing multiple urban optimization tasks based on the input policy, but has more model parameters and a lower deduction speed.

[0063] Specifically, since the initial policy is a policy combination, different initial policies are actually combinations constructed by different policies. With a large number of policies and arbitrary combinations between them, the number of initial policies is large. Existing technologies often evaluate all initial policies separately to select the optimal policy, but the evaluation efficiency is extremely low and it is difficult to be widely used. In this application, sampling is first performed, and then each sampled first reference policy is deduced through the policy calculation model to reduce the number of policies to be deduced, thereby improving the deduction efficiency.

[0064] The multiple urban optimization tasks include at least one of the following: implementation cost reduction task, environmental impact reduction task, implementation time shortening task. For example, when the initial policy with the policy content of "increasing the proportion of green buildings" is input into the policy calculation model, the policy calculation model deduces it and then obtains a first benchmark deduction result, which usually includes various indicators in the predicted year, the implementation cost generated by this policy, and the implementation time, etc. Denote the first reference policy as x1, the first benchmark deduction result as f(x1), and the policy recommendation set as H. After adding x1 and f(x1) to H, we get H = {(x1, f(x1))}.

[0065] In step 205, the initial proxy model is trained according to the policy recommendation set to obtain a target proxy model. The target proxy model includes multiple base models and a meta-model for predicting based on the outputs of the multiple base models. The model parameters of the target proxy model are fewer than those of the urban calculation model, so the deduction speed is relatively higher.

[0066] In one embodiment, step 205 may include: predicting each first reference policy through each base model to obtain an initial pseudo-calculation result output by each base model; predicting the initial pseudo-calculation results output by the multiple base models through the meta-model to obtain an initial pseudo-deduction result; calculating the loss according to the initial pseudo-deduction result and the first benchmark deduction result to obtain a proxy loss function; adjusting the parameters of the multiple base models and the meta-model according to the proxy loss function to obtain the target proxy model.

[0067] The above loss calculation can use the mean square error loss function or other loss functions, and this application does not make specific limitations on this.

[0068] In one example, the initial proxy model is trained based on the policy recommendation set H to obtain the target proxy model S, which can efficiently learn the mapping and relationship between the input policy and the output result. To improve the performance of the target proxy model, an ensemble learning method is used to construct the proxy model, comprehensively leveraging the advantages of multiple machine learning algorithms to improve the overall prediction accuracy. The base models for ensemble learning include: Random Forest regression, Gradient Boosting regression, Multi-Layer Perceptron (MLP), XGBoost, and Gaussian Process regression. The meta-model uses Ridge regression and is trained on the prediction results of the base models.

[0069] In another embodiment, step 205 may include: dividing the policy recommendation set according to the total number of multiple base models of the initial proxy model to obtain a policy recommendation subset for each base model; training the base models according to the policy recommendation subsets of each base model to update the base models; inferring the first reference policy in the policy recommendation subset through each base model to obtain the base prediction results; and training the meta-model of the initial proxy model according to the first reference policies and the base prediction results corresponding to the multiple base models to obtain the target proxy model. The main difference between this embodiment and the previous one is that a policy recommendation subset is divided from the policy recommendation set for each base model, and the policy recommendation subsets between different base models do not affect each other, which can not only reduce the model training complexity to improve the model training efficiency, but also improve the inference accuracy.

[0070] In one example, all possible policies are combined to obtain the initial policy, and then the trained proxy model S is used to predict these combinations to obtain the target calculation result S(x). For K base models , , and the meta-model , when the input policy is x, each base model generates a pseudo-calculation result . The meta-model generates the target calculation result based on these pseudo-calculation results.

[0071] Before step 206, the initial policy can be preliminarily screened according to the target calculation result S(x) to further reduce the number of policies and improve the processing efficiency.

[0072] In step 206, through multiple base models in the target proxy model, each initial policy is inferred to obtain a pseudo-calculation result. For example, referring to the above, for K base models , , when the input policy is x, each base model generates a pseudo-calculation result .

[0073] In step 207, at least two initial strategies are screened according to the pseudo-computation result to obtain a second reference strategy.

[0074] In one embodiment, referring to Figure 5 , step 207 may include: Step 501, determining an expected improvement value generated by each of multiple urban optimization tasks executed by an initial strategy according to the pseudo-computation result; Step 502, obtaining an optimization weight set for each urban optimization task; Step 503, performing a weighted sum according to the optimization weight and the expected improvement value to obtain a target optimization score; Step 504, selecting at least two initial strategies from high to low according to the target optimization score to obtain a second reference strategy.

[0075] In one embodiment, step 501 may include: Calculating a pseudo-computation mean value and a pseudo-computation standard deviation according to the pseudo-computation results output by multiple basic models; Performing a difference calculation according to the current best function value, the pseudo-computation mean value, and a preset balance parameter to obtain a pseudo-computation weight; Performing a ratio calculation according to the pseudo-computation weight and the pseudo-computation standard deviation to obtain a pseudo-computation ratio; Mapping the pseudo-computation ratio through a preset cumulative distribution function to obtain a pseudo-computation distribution function; Multiplying the pseudo-computation weight and the pseudo-computation distribution function to obtain a first function term; Mapping the pseudo-computation ratio through a preset probability density function to obtain a pseudo-computation probability density function; Multiplying the pseudo-computation standard deviation and the pseudo-computation probability density function to obtain a second function term; Summing the first function term and the second function term to obtain an expected improvement function; Calculating an expected improvement value according to the expected improvement function.

[0076] Specifically, the most suitable strategy is screened from the calculation results based on multi-objective weighted expected improvement (EI), and the urban calculation model is used to infer this strategy to obtain an accurate calculation result . is added to the strategy recommendation set H. This process is continuously repeated to enrich the recommendation set. Expected improvement is used as an acquisition function for selecting the next evaluation point. For a minimization problem, the basic expected improvement formula is , where is the current best function value observed currently, and are the pseudo-computed mean and pseudo-computed standard deviation of the predicted values obtained by each base model for policy x. is the exploration-exploitation balance parameter, usually set to 0.01, , is the cumulative distribution function (CDF) of the standard normal distribution, is the probability density function (PDF) of the standard normal distribution. Since there are multiple optimization tasks in this application, multi-task optimization needs to be performed and calculated using a weighted method: . Where: is the target optimization score of the initial policy x, is the weight of task j, defined by the user, is the expected improvement value of target j. n is the number of tasks. In this application, tasks can include tasks for reducing implementation costs, reducing environmental impacts, and shortening implementation time.

[0077] The benefits of the embodiments of the above steps 501 to 504 are that they can perform multi-task evaluation on the initial policy, improve the screening accuracy, and have high applicability.

[0078] In step 208, each second reference policy is deduced through the policy calculation model to obtain a second benchmark deduction result, and the policy recommendation set is updated according to the second reference policy and the second benchmark deduction result. For example, the second reference policy is denoted as x2, the second benchmark deduction result is denoted as f(x2), and referring to the above, the policy recommendation set H = {(x1, f(x1))}. After adding x2 and f(x2) to H, H = {(x1, f(x1)), (x2, f(x2))}. The first reference policy and the second reference policy in the policy recommendation set are collectively referred to as candidate policies. The candidate policies in the policy recommendation set are used to perform multiple urban optimization tasks.

[0079] In step 209, the regulation requirement target input by the user is obtained, the target policy is determined from the policy recommendation set according to the regulation requirement target, and the target policy is recommended to the user.

[0080] The regulation requirement target refers to the urban optimization tasks that the user expects the policy to perform. For example, the regulation requirement target input by the user is "expect to control the implementation cost within 100 billion yuan". Another example is that the regulation requirement target input by the user is "expect the implementation duration to be less than 5 years". Another example is that the regulation requirement target input by the user is "expect to control the implementation cost within 100 billion yuan and expect the implementation duration to be less than 5 years".

[0081] In one embodiment, referring to Figure 6, determining the target policy from the policy recommendation set according to the regulation demand target in step 209 may include: Step 601, if the regulation demand target is one of multiple city optimization tasks, set a first weight for the city optimization task corresponding to the regulation demand target, and set a second weight for other city optimization tasks, where the first weight is greater than the second weight; Step 602, score multiple candidate policies in the policy recommendation set according to the first weight and the second weight to obtain the policy recommendation score of each candidate policy; Step 603, sort the multiple candidate policies from high to low according to the policy recommendation score to obtain the policy sorting result; Step 604, determine the candidate policy corresponding to the highest policy recommendation score as the target policy according to the policy sorting result.

[0082] For example, when the user inputs a specific regulation demand target, the present application filters in the policy recommendation set H according to the user's input, and finds the policies that meet the user's requirements and their deduction results. Finally, the target policy and its deduction result are recommended to the user to meet the user's needs.

[0083] The present application performs maximum-minimum normalization processing on each task, that is , after standardization, calculate the weighted score . Here is the preference weight when recommending to the user. For example, if the user hopes to view policies with lower implementation costs, the task of reducing implementation costs will have a higher weight, such as 0.8, and the weights of other tasks will be 0.1. The policy with lower implementation costs will get a higher score, and at the same time other costs will also affect the score.

[0084] In one embodiment, in order to ensure the diversity of the recommended policies, the present application adds a diversity guarantee mechanism. Then, after step 604, a method for an artificial intelligence-driven urban system computing simulation platform may further include: sequentially traversing the candidate policies in the policy sorting result, calculating the distance between the target policy and each candidate policy to obtain the policy distance; if the policy distance is greater than a predetermined threshold, determine the candidate policy as an auxiliary policy; recommend the auxiliary policy to the user.

[0085] For example, first sort the candidate policies according to the scores, take the policy with the highest score as the target policy and start adding it to the recommendation result. Then sequentially traverse each candidate policy, calculate the standardized Euclidean distance between the candidate policy and the selected policies. If the distance is greater than the threshold, then add it to the recommendation result. Repeat this step until the required number of policies is reached, so as to ensure that the recommended policies are different and facilitate the user to select a more suitable policy for themselves.

[0086] In one example, the regulatory requirement target input by the user is to optimize the land structure of the urban system, while expecting to control the implementation cost within 100 billion yuan and the implementation duration to be less than 5 years. Through the recommendation method of the artificial intelligence-driven urban system calculation and simulation platform provided by this application, three major categories of recommended strategies can be obtained, namely, the "lowest implementation cost strategy" for controlling costs, the "lowest environmental impact strategy" for controlling environmental impacts, and the "shortest implementation time strategy" for controlling the duration. Based on the regulatory requirement target, the three strategies under "the lowest implementation cost" are mainly compared. Through comparison, Strategy 1 will generate an implementation cost of 86.8 - 106.1 billion yuan by increasing the proportion of green buildings and is expected to be implemented for 36 months; Strategy 2 will generate an implementation cost of 2090.234 - 2554.7305 billion yuan by adjusting the land supply structure; Strategy 3 will generate an implementation cost of 2509.3576 - 3066.9926 billion yuan through comprehensive regulation (simultaneously increasing the proportion of green buildings and adjusting the land supply structure). The recommended strategy selected this time is Strategy 1 in the lowest implementation cost strategy, that is, increasing the proportion of green buildings.

[0087] The various technical features in the above embodiments can be combined arbitrarily as long as there is no conflict or contradiction between the features. However, due to space limitations, they are not described one by one. Therefore, any combination of the various technical features in the above embodiments also belongs to the scope disclosed in this specification.

[0088] The embodiment of this application also discloses an artificial intelligence-driven urban system calculation and simulation platform device. Refer to Figure 7, the artificial intelligence-driven urban system computational simulation platform device includes: a data acquisition module 701, configured to acquire urban basic text data for urban system computation, perform simulation computation on the urban basic text data, and obtain urban computation element data; a policy generation module 702, configured to generate at least two initial policies according to the urban computation element data; a policy sampling module 703, configured to sample the at least two initial policies to obtain at least two first reference policies; a first deduction module 704, configured to deduce each first reference policy through a pre-trained policy computation model to obtain a first benchmark deduction result, and add the first reference policy and the first benchmark deduction result to a policy recommendation set; an agent model training module 705, configured to train an initial agent model according to the policy recommendation set to obtain a target agent model; the target agent model includes a plurality of basic models and a meta-model for predicting based on the outputs of the plurality of basic models; a second deduction module 706, configured to deduce each initial policy through the plurality of basic models in the target agent model to obtain a pseudo-computation result; a policy screening module 707, configured to screen the at least two initial policies according to the pseudo-computation result to obtain second reference policies; a third deduction module 708, configured to deduce each second reference policy through the policy computation model to obtain a second benchmark deduction result, and update the policy recommendation set according to the second reference policies and the second benchmark deduction results; a policy recommendation module 709, configured to obtain a regulation requirement target input by a user, determine a target policy from the policy recommendation set according to the regulation requirement target, and recommend the target policy to the user.

[0089] In one embodiment, the artificial intelligence-driven urban system computational simulation platform device further includes an auxiliary recommendation module, configured to sequentially traverse candidate policies in a policy sorting result, calculate a policy distance between the target policy and each candidate policy, and obtain a policy distance; if the policy distance is greater than a predetermined threshold, determine the candidate policy as an auxiliary policy; and recommend the auxiliary policy to the user.

[0090] It should be noted that the specific implementation manner of the artificial intelligence-driven urban system computational simulation platform device is basically the same as the specific embodiments of the above-mentioned artificial intelligence-driven urban system computational simulation platform method, and will not be elaborated here.

[0091] An embodiment of this application also discloses a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the artificial intelligence-driven urban system computational simulation platform method as described above. The computer device is, for example, a mobile phone, a computer, or other devices.

[0092] The embodiments of the present application also provide a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the method of the artificial intelligence-driven urban system computing simulation platform as described above.

[0093] As a non-transitory computer-readable storage medium, a memory can be used to store non-transitory software programs and non-transitory computer-executable programs. In addition, the memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, a flash memory device, or other non-transitory solid-state storage devices. In some embodiments, the memory may optionally include memories that are remotely disposed relative to the processor, and these remote memories can be connected to the processor through a network. Examples of the above networks include, but are not limited to, the Internet, an enterprise intranet, a local area network, a mobile communication network, and combinations thereof.

[0094] The embodiments of the present application also provide a computer program product, which includes a computer program. The computer program is read and executed by a processor, so that when the processor executes the computer program, it implements the method of the artificial intelligence-driven urban system computing simulation platform as described above.

[0095] The embodiments described in the embodiments of the present application are for more clearly illustrating the technical solutions of the embodiments of the present application, and do not constitute a limitation on the technical solutions provided by the embodiments of the present application. Those skilled in the art will know that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of the present application are equally applicable to similar technical problems.

[0096] Those skilled in the art can understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present application, and may include more or fewer steps than those shown in the figures, or combine certain steps, or different steps.

[0097] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, that is, they may be located in one place, or may be distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0098] Those of ordinary skill in the art can understand that all or some of the steps in the methods disclosed above, and the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, and appropriate combinations thereof.

[0099] In the description of this application and the above-mentioned drawings, terms such as "first", "second", "third", "fourth", etc. (if any) are used to distinguish similar objects and do not necessarily describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances so that the embodiments of this application described here can be implemented in an order different from those illustrated or described here. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that comprises a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products, or devices.

[0100] It should be understood that in this application, "at least one (item)" means one or more, and "a plurality" means two or more. "And / or" is used to describe the association relationship of associated objects and indicates that three relationships can exist. For example, "A and / or B" can mean: only A exists, only B exists, and both A and B exist at the same time. Here, A and B can be singular or plural. The character " / " generally indicates that the associated objects before and after are in an "or" relationship. "At least one (one) of the following" or its similar expressions refer to any combination of these items, including any combination of single items (ones) or plural items (ones). For example, at least one (one) of a, b, or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0101] In several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the above-mentioned division of units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling, direct coupling, or communication connection to each other can be through some interfaces. The indirect coupling or communication connection of devices or units can be in electrical, mechanical, or other forms.

[0102] The units described above as separate components may or may not be physically separated. The components shown as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0103] In addition, in each embodiment of the present application, each functional unit may be integrated into one processing unit, may exist separately as individual physical units, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of a software functional unit.

[0104] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes multiple instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods in each embodiment of the present application. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs that can store programs.

[0105] The preferred embodiments of the embodiments of the present application have been described above with reference to the accompanying drawings, and thus do not limit the scope of the rights of the embodiments of the present application. Any modification, equivalent replacement, and improvement made by those skilled in the art without departing from the scope and essence of the embodiments of the present application shall be within the scope of the rights of the embodiments of the present application.

Claims

1. An artificial intelligence-driven method for a computational simulation platform of urban systems, characterized in that, The method includes: Obtaining urban basic data for urban system calculation, performing simulation calculation on the urban basic data to obtain urban calculation element data; Generating at least two initial strategies according to the urban calculation element data; Sampling the at least two initial strategies to obtain at least two first reference strategies; Deducing each of the first reference strategies through a pre-trained strategy calculation model to obtain a first benchmark deduction result, and adding the first reference strategy and the first benchmark deduction result to a strategy recommendation set; Training an initial agent model according to the strategy recommendation set to obtain a target agent model; the target agent model includes a plurality of basic models and a meta-model for predicting based on the outputs of the plurality of basic models; Deducing each of the initial strategies through the plurality of basic models in the target agent model to obtain pseudo-calculation results; Screening the at least two initial strategies according to the pseudo-calculation results to obtain second reference strategies; Deducing each of the second reference strategies through the strategy calculation model to obtain a second benchmark deduction result, and updating the strategy recommendation set according to the second reference strategy and the second benchmark deduction result; Obtaining a regulation demand target input by a user, determining a target strategy from the strategy recommendation set according to the regulation demand target, and recommending the target strategy to the user.

2. The method according to claim 1, characterized in that The candidate strategies in the strategy recommendation set are used to execute multiple urban optimization tasks; The determining the target strategy from the strategy recommendation set according to the regulation demand target includes: If the regulation demand target is one of the multiple urban optimization tasks, setting a first weight for the urban optimization task corresponding to the regulation demand target, and setting a second weight for the other urban optimization tasks, the first weight being greater than the second weight; Scoring the multiple candidate strategies in the strategy recommendation set according to the first weight and the second weight to obtain a strategy recommendation score for each candidate strategy; Sorting the multiple candidate strategies from high to low according to the strategy recommendation scores to obtain a strategy sorting result; Determining the candidate strategy corresponding to the highest strategy recommendation score as the target strategy according to the strategy sorting result.

3. The method according to claim 2, wherein After determining the candidate strategy corresponding to the highest strategy recommendation score as the target strategy according to the strategy sorting result, the method further includes: Sequentially traversing the candidate strategies in the strategy sorting result, calculating the distance between the target strategy and each candidate strategy to obtain a strategy distance; If the strategy distance is greater than a predetermined threshold, determining the candidate strategy as an auxiliary strategy; Recommending the auxiliary strategy to the user.

4. The method according to any one of claims 1 to 3, characterized in that The screening the at least two initial strategies according to the pseudo-calculation results to obtain second reference strategies includes: Determining the expected improvement value generated by each of the initial strategies for each of the multiple urban optimization tasks according to the pseudo-calculation results; wherein, the multiple urban optimization tasks include at least one of the following: implementation cost reduction task, environmental impact reduction task, implementation time shortening task; Obtain the optimization weights set for each of the city optimization tasks; Perform a weighted sum based on the optimization weights and the expected improvement value to obtain a target optimization score; Select the at least two initial strategies from high to low according to the target optimization score to obtain the second reference strategy.

5. The method according to claim 4, characterized in that, The determining the expected improvement value for each of the multiple city optimization tasks executed by the initial strategy according to the pseudo-computation result includes: Calculate a pseudo-computation mean and a pseudo-computation standard deviation based on the pseudo-computation results output by multiple of the base models; Perform a difference calculation based on the current best function value, the pseudo-computation mean, and a preset balance parameter to obtain a pseudo-computation weight; Perform a ratio calculation based on the pseudo-computation weight and the pseudo-computation standard deviation to obtain a pseudo-computation ratio; Map the pseudo-computation ratio through a preset cumulative distribution function to obtain a pseudo-computation distribution function; Multiply the pseudo-computation weight and the pseudo-computation distribution function to obtain a first function term; Map the pseudo-computation ratio through a preset probability density function to obtain a pseudo-computation probability density function; Multiply the pseudo-computation standard deviation and the pseudo-computation probability density function to obtain a second function term; Sum the first function term and the second function term to obtain an expected improvement function; Calculate the expected improvement value according to the expected improvement function.

6. The method according to any one of claims 1 to 3, characterized in that The training the initial surrogate model according to the policy recommendation set to obtain a target surrogate model includes: Divide the policy recommendation set according to the total number of multiple base models of the initial surrogate model to obtain a policy recommendation subset for each of the base models; Train the base model according to the policy recommendation subset for each of the base models to update the base model; Deduce the first reference strategy in the policy recommendation subset through each of the base models to obtain a base prediction result; Train the meta-model of the initial surrogate model according to the first reference strategy and the base prediction result corresponding to the multiple base models to obtain the target surrogate model.

7. The method according to any one of claims 1 to 3, characterized in that, The sampling the at least two initial strategies to obtain at least two first reference strategies includes: Stratify the at least two initial strategies to obtain multiple strategy layers; Randomly select the initial strategies in each of the strategy layers to obtain the first reference strategy for each of the strategy layers; Shuffle the first reference strategies of the multiple strategy layers to obtain the at least two first reference strategies.

8. An artificial intelligence-driven urban system computational simulation platform device, characterized in that, The device includes: A data acquisition module, configured to acquire city basic data for urban system calculation, perform simulation calculation on the city basic data to obtain city calculation element data; A strategy generation module, configured to generate at least two initial strategies according to the city calculation element data; A strategy sampling module, configured to sample the at least two initial strategies to obtain at least two first reference strategies; The first deduction module is used to deduce each of the first reference policies through a pre-trained policy calculation model to obtain a first benchmark deduction result, and add the first reference policy and the first benchmark deduction result to the policy recommendation set; The proxy model training module is used to train an initial proxy model according to the policy recommendation set to obtain a target proxy model; the target proxy model includes a plurality of base models and a meta-model for predicting based on the outputs of the plurality of base models; The second deduction module is used to deduce each of the initial policies through the plurality of base models in the target proxy model to obtain pseudo-calculation results; The policy screening module is used to screen the at least two initial policies according to the pseudo-calculation results to obtain second reference policies; The third deduction module is used to deduce each of the second reference policies through the policy calculation model to obtain a second benchmark deduction result, and update the policy recommendation set according to the second reference policy and the second benchmark deduction result; The policy recommendation module is used to obtain a regulation requirement target input by a user, determine a target policy from the policy recommendation set according to the regulation requirement target, and recommend the target policy to the user.

9. A computer device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the method described in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that, A computer program is stored on the computer-readable storage medium, and when the computer program is executed, the method described in any one of claims 1 to 7 is implemented.

Images