Method for constructing edge-computing-based house ticket management information system

Abstract

The application relates to the technical field of house ticket information management, and discloses a house ticket management information system construction method based on edge computing. An edge computing node is deployed to collect house ticket management original data in real time, which contains user qualification verification information, house property resource information and the like. A genetic algorithm is used to optimize multi-dimensional policy matching degrees of the original data, and a multi-dimensional feature matrix is generated. A house ticket transaction risk prediction model based on a capsule network is constructed, and a risk level and a potential illegal behavior identifier are output. A dynamic house ticket circulation graph is constructed according to the risk level and the identifier, and the topological relationship of relevant nodes and transaction paths is presented. A transaction strategy matching rule containing qualification check conditions and the like is generated based on the topological relationship. The rule is issued to the edge computing node to execute real-time transaction control. The method can realize the precision, dynamic and high efficiency of house ticket management, and improve the standardization and controllability of house ticket transactions.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of house ticket information management, in particular to a house ticket management information system construction method based on edge computing. BACKGROUND

[0002] In the current house ticket management field, the real-time performance of data collection is a common problem. Traditional centralized data collection mode often needs to collect various information to the center server for processing, which makes the user qualification verification information, real estate resource information, etc. prone to delay in the transmission process, and it is difficult to meet the requirement of information timeliness in house ticket transaction.

[0003] House ticket management involves many policy provisions and has multi-dimensional characteristics. The policy constraints of different regions and different time periods are different. The existing management mode relies more on manual operation or simple rule engine when matching policies, and cannot fully and accurately match multi-dimensional policy information, resulting in some house ticket transactions having problems due to insufficient policy adaptability.

[0004] Risk prediction in the process of house ticket transaction is also a big difficulty. Traditional risk assessment methods are mostly based on single feature or simple feature combination, which is difficult to capture the complex potential risk factors in house ticket transaction, and has weak identification ability for potential illegal behavior, which is easy to cause transaction disputes or illegal operation.

[0005] The house ticket circulation process lacks dynamic and intuitive graph display, and the topological relationship between the house ticket holding node, the real estate resource node and the transaction path is not clear, so the management personnel cannot fully grasp the circulation of the house ticket, which is not conducive to effective monitoring and management of the transaction process. In terms of transaction strategy, the existing rules are often fixed and cannot be adjusted in real time according to the dynamic changes of house ticket circulation, resulting in unreasonable resource allocation, non-optimal transaction path, etc., which affects the overall efficiency and fairness of house ticket management. SUMMARY

[0006] The purpose of the present application is to provide a house ticket management information system construction method based on edge computing to solve the problems raised in the background.

[0007] To achieve the above purpose, the present application provides a house ticket management information system construction method based on edge computing, which comprises:

[0008] Deploying edge computing nodes to collect house ticket management raw data in real time, the house ticket management raw data including user qualification verification information, real estate resource information, transaction dynamic information and regional policy constraint information;

[0009] The genetic algorithm is used to perform multi-dimensional policy matching degree optimization on the house ticket management original data, and a multi-dimensional feature matrix containing policy adaptability weights is generated.

[0010] A house ticket transaction risk prediction model based on a capsule network is constructed, the multi-dimensional feature matrix is input into the house ticket transaction risk prediction model, and a house ticket transaction risk level and a potential illegal behavior identifier are output.

[0011] According to the house ticket transaction risk level and the potential illegal behavior identifier, a dynamic house ticket circulation graph is constructed, which contains the topological relationship of house ticket holding nodes, real estate resource nodes and transaction path edges.

[0012] Based on the topological relationship of the dynamic house ticket circulation graph, a house ticket transaction strategy matching rule is generated, which contains qualification verification conditions, resource allocation priorities and transaction path constraint conditions.

[0013] The house ticket transaction strategy matching rule is issued to the edge computing node to perform real-time transaction control.

[0014] Preferably, the specific method of using the genetic algorithm to perform multi-dimensional policy matching degree optimization on the house ticket management original data comprises:

[0015] Extract the social security payment duration, household status and family population number in the user qualification verification information as individual gene coding;

[0016] Convert the regional policy constraint information into a fitness function, which contains policy clause coverage and constraint conflict penalty terms.

[0017] A new generation of policy adaptability weight set is generated through cross variation operation, and the individual solution with the highest fitness function value is decoded as the multi-dimensional feature matrix.

[0018] Preferably, the specific method of constructing a house ticket transaction risk prediction model based on a capsule network comprises:

[0019] The multi-dimensional feature matrix is split into a spatial correlation feature subset and a time sequence evolution feature subset;

[0020] In the primary capsule layer, the spatial correlation feature subset is subjected to a convolution compression operation, and the time sequence evolution feature subset is subjected to a gated recurrent encoding operation;

[0021] The coupling coefficient of the primary capsule output vector and the risk category capsule is calculated in the routing protocol layer, and the coupling coefficient is iteratively updated through a dynamic routing algorithm;

[0022] In the output layer, high-dimensional risk feature vectors are aggregated to generate the house ticket transaction risk level and the potential illegal behavior identifier.

[0023] Preferably, the specific method for constructing the dynamic house ticket flow map comprises:

[0024] Resolving the high-risk transaction path identification in the house ticket transaction risk level to generate a topology relationship blacklist;

[0025] Updating the credibility attribute of the house property resource node according to the potential illegal behavior identification;

[0026] Fusing the policy adaptability weight and the priority attribute of the house ticket holding node to reconfigure the connection weight of the transaction path edge;

[0027] Based on the updated node attribute and connection weight, an adjacency matrix of the dynamic house ticket flow map is established.

[0028] Preferably, the specific method for generating the house ticket transaction strategy matching rule comprises:

[0029] Traversing the adjacency matrix of the dynamic house ticket flow map to extract a house ticket holding node set satisfying the qualification verification condition;

[0030] Arranging the resource allocation priority in descending order according to the credibility attribute of the house property resource node;

[0031] Combining the connection weight of the transaction path edge to calculate the optimal transaction path constraint condition;

[0032] Compiling the qualification verification condition, resource allocation priority and transaction path constraint condition into executable strategy code.

[0033] Preferably, the specific method for executing real-time transaction control comprises:

[0034] Loading the executable strategy code on the edge computing node to real-time verify the user qualification information in the transaction request;

[0035] Comparing the target house property resource in the transaction request with the resource allocation priority list;

[0036] When detecting that the transaction path deviates from the optimal transaction path constraint condition, triggering a transaction termination instruction and recording the illegal behavior;

[0037] Encapsulating the transaction execution result and strategy deviation degree data into the transaction feedback data.

[0038] Preferably, the method further comprises:

[0039] Recording the transaction feedback data and strategy execution log;

[0040] The transaction feedback data and the strategy execution log are used to trigger a double-time-dimension model updating mechanism, and the house ticket transaction risk prediction model is iteratively optimized in parameters.

[0041] Preferably, the double-time-dimension model updating mechanism comprises the following specific methods:

[0042] The transaction success rate and the violation interception rate in the strategy execution log are collected in an hourly granularity;

[0043] The resource matching deviation value and the strategy deviation data in the transaction feedback data are aggregated in a daily granularity;

[0044] The hourly granularity data is input into the capsule routing protocol layer of the house ticket transaction risk prediction model for weight fine-tuning;

[0045] The daily granularity data is input into the primary capsule layer of the house ticket transaction risk prediction model for feature encoder reconstruction.

[0046] Preferably, the iteratively optimized specific method comprises the following steps:

[0047] The fine-tuned capsule routing protocol layer weight is extracted as the initial population of the genetic algorithm;

[0048] A new generation of spatial correlation feature subsets are generated according to the reconstructed feature encoder;

[0049] The strategy matching degree of the new generation of spatial correlation feature subsets is evaluated by an adaptability function;

[0050] When the strategy matching degree is improved by more than a preset threshold, the parameter set of the house ticket transaction risk prediction model is updated.

[0051] Preferably, the method further comprises a trusted transaction data certification mechanism, specifically, in the edge computing node executing the transaction control process, the transaction request, the strategy matching result and the transaction execution result are transmitted by an encrypted channel.

[0052] Compared with the prior art, the method has the following beneficial effects:

[0053] The edge computing-based house ticket management information system construction method can collect house ticket management original data in real time by deploying edge computing nodes, so that the data collection link is closer to the information source, the intermediate link of data transmission is reduced, various information can be quickly acquired and processed, the timeliness of user qualification verification, house property resource and other information is guaranteed, and subsequent management operations can be conveniently carried out in time.

[0054] The genetic algorithm is adopted to perform multi-dimensional policy matching degree optimization on the original data, a multi-dimensional feature matrix containing policy adaptability weights is generated, the correlation and difference between different policy clauses can be fully considered, the policy information is integrated and matched from multiple dimensions, the policy matching process is more comprehensive and accurate, and the adaptation between the housing ticket transaction and the policy constraint is more in line with the actual situation.

[0055] The housing ticket transaction risk prediction model based on the capsule network is constructed, the multi-dimensional feature matrix is input, and the risk level and potential illegal behavior identification are output, the capsule network can be used to extract and process complex features, potential risk factors in the housing ticket transaction can be deeply mined, possible illegal behaviors can be accurately identified, risk prediction is more in line with the actual risk situation of the transaction, and problems that can occur in the transaction can be avoided in advance.

[0056] The dynamic housing ticket circulation graph is constructed according to the risk level and the potential illegal behavior identification, the topological relationship of the housing ticket holding node, the real estate resource node and the transaction path edge is clearly presented, the management personnel can intuitively and comprehensively understand the circulation dynamics of the housing ticket, master the correlation between nodes, and facilitate effective monitoring and analysis of the housing ticket circulation process.

[0057] The housing ticket transaction strategy matching rule is generated based on the topological relationship of the dynamic housing ticket circulation graph, including qualification check conditions, resource allocation priority and transaction path constraint conditions, which can make the transaction strategy closely combined with the actual situation of the housing ticket circulation, make the qualification check more in line with the actual demand, make the resource allocation more reasonable, make the transaction path more optimal, and improve the overall standardization and efficiency of the housing ticket transaction.

[0058] The transaction strategy matching rule is issued to the edge computing node to perform real-time transaction control, which can make the transaction control instruction quickly act on the transaction link, timely respond to various changes in the housing ticket transaction, make the transaction process orderly under the constraint of the rule, and guarantee the fairness and stability of the housing ticket management. BRIEF DESCRIPTION OF DRAWINGS

[0059] Figure 1 A working principle diagram of the edge computing-based housing ticket management information system construction method is constructed;

[0060] Figure 2 A flowchart of multi-dimensional policy matching degree optimization using a genetic algorithm is constructed;

[0061] Figure 3 A flowchart of constructing a dynamic housing ticket circulation graph is constructed;

[0062] Figure 4 A flowchart of executing real-time transaction control is constructed. DETAILED DESCRIPTION

[0063] With reference to the drawings of the embodiments of the present application, the technical solutions in the embodiments of the present application will be described clearly and completely. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by a person of ordinary skill in the art without creative effort belong to the scope of protection of the present application.

[0064] With reference to Figure 1 The application provides an edge computing-based house ticket management information system construction method, which comprises the following steps:

[0065] In the system construction process, first, an edge computing node is deployed to collect house ticket management original data in real time. The edge computing node is distributed in each regional house ticket management terminal, and acquires user qualification verification information, house property resource information, transaction dynamic information and regional policy constraint information through a distributed data collection interface. The user qualification verification information covers social security payment records, household registration materials, family population registration information and the like; the house property resource information includes static attributes such as house source location, house type area and property right state, and dynamic attributes such as listing time and price fluctuation; the transaction dynamic information records time sequence data such as house ticket holding change, transaction initiation and confirmation state and fund flow record; and the regional policy constraint information includes text and structured data such as purchase qualification limitation clauses, house ticket flow transfer time limit provisions and resource allocation policy details.

[0066] After the collection is completed, a genetic algorithm is used to optimize the multi-dimensional policy matching degree of the house ticket management original data. In the optimization process, first, key parameters are extracted from the user qualification verification information as individual gene coding, and then the regional policy constraint information is converted into a fitness function, a policy adaptability weight set is generated through the cross variation operation of the genetic algorithm, and the optimal individual is decoded into a multi-dimensional feature matrix. Subsequently, a house ticket transaction risk prediction model based on a capsule network is constructed, the multi-dimensional feature matrix is split into a spatial correlation feature subset and a time sequence evolution feature subset, after processing by a primary capsule layer and iteration and updating of a routing protocol layer coupling coefficient, a risk level and a potential illegal behavior identifier are generated by an output layer aggregating features.

[0067] Based on the risk prediction result, a dynamic house ticket flow transfer graph is constructed. The graph clearly presents the topological relationship of house ticket holding nodes, house property resource nodes and transaction path edges by analyzing high-risk transaction paths, updating node credibility and reconstructing path edge weights to establish an adjacency matrix. According to the topological relationship, house ticket transaction strategy matching rules are generated, including qualification check conditions, resource allocation priority and transaction path constraint conditions, and are compiled into executable codes and issued to the edge computing node to realize real-time transaction control.

[0068] Embodiment 1: With reference to Figure 2In the multi-dimensional policy matching optimization of the house ticket management original data, the first step is to extract the key elements from the user qualification verification information. Among them, the length of social security payment is an important indicator to reflect the user's participation and stability in local economic activities. In actual operation, the system will accurately obtain the start date and current date of the user's social security payment, and calculate the payment length data in months. The household status is related to whether the user meets the local housing purchase preferential policy or restriction condition. The system records it in binary coding form, 1 represents local household, 0 represents non-local household. This simple and efficient coding method is convenient for subsequent algorithm processing. The number of family members also has a significant influence on housing qualification and housing resource allocation. The system accurately enters the number of registered population on the family household registration book and classifies and codes according to the preset population size interval, such as 1-3 people for small family, 4-6 people for medium family, etc. The combination of these three key information forms a specific length of gene sequence, which becomes the basis of individual gene coding for subsequent genetic algorithm operation.

[0069] Convert regional policy constraints information into fitness function. Regional policy constraints information contains a large amount of text and structured data, such as the minimum requirement of social security payment years in local housing policy, the number of housing units limited for different household registration, and the recommended area of house type for different family population size, etc. The system will analyze these policy provisions and convert them into quantifiable calculation rules. When calculating the policy clause coverage rate, each item of information in the individual gene coding will be compared with the policy clause one by one. Assuming that the policy requires 36 months of social security payment, if the individual gene coding shows that the social security payment time is equal to or greater than this value, it is considered to meet this policy clause. The proportion of the number of satisfied clauses to the total number of key policy clauses is the policy clause coverage rate. The constraint conflict penalty term is used to handle the possible conflict with the policy in the individual gene coding. For example, the policy stipulates that a local household with 24 months of social security payment can purchase two sets of houses. If the individual gene coding shows that the local household has only 12 months of social security payment but has a tendency to purchase two sets of houses, the system will deduct the fitness function value of this individual according to the preset weight, so as to reflect the conflict degree with the policy. The fitness function is finally calculated by the difference between the policy clause coverage rate and the constraint conflict penalty term. The higher the value, the better the matching degree of the individual and the regional policy.

[0070] After the fitness function is completed, the core operation of the genetic algorithm is entered, which is the crossover and mutation operation. When selecting parent individuals, the system sorts the individuals in the current population according to the fitness function value, and preferentially selects individuals with high fitness as parents. For example, the top 30% of individuals in terms of fitness are selected for subsequent operations. The crossover operation uses a single-point crossover method, and a non-key feature bit is randomly selected as the crossover point in the gene sequence of the parent individual. For example, if the randomly selected crossover point is between the social security payment length, the household registration status, and the family population size, then the gene fragments before and after the point of the two parent individuals will be exchanged, thereby generating two new offspring individual gene sequences. Mutation operation introduces new genetic diversity into the population, and the mutation probability is usually set within a reasonable range, such as 0.05. For mutation operation, the system will randomly select a certain proportion of offspring individuals for processing. During processing, the value of the social security payment length may be adjusted, such as increasing or decreasing a certain number of months based on the original value; the household registration status code may be flipped, changing 1 to 0 or 0 to 1; or the family population size code may be adjusted, such as moving up or down one interval based on the original interval. Through repeated crossover and mutation operations, a new generation of policy adaptability weight sets is continuously generated. After multiple iterations, usually set to 50-100 generations, the system will select the individual with the highest fitness function value from the last generation of the population, and decode its gene sequence into a multi-dimensional feature matrix. In the decoding process, the social security payment length, household registration status, and family population size codes in the gene sequence are converted back to their corresponding actual values and states, and are associated with the calculated policy adaptability weights to form a multi-dimensional feature matrix with user qualification features as rows and policy adaptability weights as columns.

[0071] In the stage of constructing a housing transaction risk prediction model based on a capsule network, the multi-dimensional feature matrix needs to be reasonably split. The spatial correlation feature subset mainly focuses on the static spatial relationship features between the user and the property. For example, the matching of the user's household registration location and the property's location area, if the user is a local household, the probability of choosing a local popular area property and the risk characteristics; the adaptation relationship between the family population size and the property size, such as large families choosing small property sizes may have a risk of overcrowding, which affects the stability of the transaction, these features are integrated into the spatial correlation feature subset. The time evolution feature subset focuses on dynamic features that change over time. The cumulative change of the social security payment length reflects the trend of the user's economic stability, if the social security payment length continues to increase, it indicates that the user's economic stability is enhanced, and the transaction risk is relatively reduced; the length of the housing ticket holding time also affects the transaction risk, and a short holding time may have a speculative transaction risk; the price fluctuation in a period of time is also an important time sequence feature, and large price fluctuations may indicate market instability or abnormal transactions.

[0072] The primary capsule layer adopts different processing methods for the two feature subsets. For the spatial correlation feature subset, a convolution compression operation is performed. The system uses 3x3 convolution kernels to perform convolution operations on spatial correlation feature data across multiple channels. Each convolution kernel is designed to capture spatial feature patterns, such as a convolution kernel that specifically identifies a specific combination pattern of user household registration matching the property area. The results of the convolution operation are subjected to a nonlinear transformation through an activation function, such as a ReLU function, to enhance the feature expression capability, and then form spatial feature capsule vectors that contain compressed and refined spatial correlation feature information. For the time evolution feature subset, a gated recurrent encoding operation is used to process time sequence data through a gated recurrent unit (GRU). GRU can effectively capture the variation of features over time, and when processing time series data such as social security payment duration, housing ticket holding time, and transaction price fluctuations, GRU will determine which information needs to be retained and which needs to be updated through a gating mechanism based on the current input and the hidden state at the last time step, thereby outputting capsule vectors that reflect the time evolution features.

[0073] The routing protocol layer is responsible for the calculation and update of the key coupling coefficients. In the initial stage, the relationship between the primary capsule output vector and the risk category capsule is normalized through a softmax function to obtain the initial coupling coefficients. Subsequently, based on the dynamic routing algorithm, the system continuously calculates the similarity between the primary capsule vector and the category capsule vector. If a primary capsule vector has a high similarity with a risk category capsule vector, the coupling coefficient between them is increased in subsequent iterations; otherwise, it is decreased. Usually after 3-5 iterations, the coupling coefficient matrix tends to be stable and can accurately reflect the correlation between different feature vectors and risk categories. In the output layer, the high-dimensional risk feature vector processed by the routing protocol layer is mapped through a fully connected layer. The neurons of the fully connected layer are connected with each element of the input vector, and through the linear transformation of the weight matrix and the nonlinear transformation of the activation function, the high-dimensional risk feature vector is mapped to the pre-set risk level space. For example, the pre-set risk levels are low, medium, and high, and the output layer determines the risk level of the housing ticket transaction through threshold judgment based on the calculated results. At the same time, a multi-label classifier is used to identify potential violations. The multi-label classifier learns the corresponding relationship between different risk features and violations through training, and outputs potential violation identifiers, such as false housing source risk identifier and transaction price anomaly identifier, after inputting the high-dimensional risk feature vector.

[0074] Example 2: see Figure 3The process of constructing a dynamic housing voucher circulation graph begins with an in-depth analysis of the risk levels of housing voucher transactions. The risk level of housing voucher transactions is output by a risk prediction model, which includes risk assessment results for various transaction paths. The system extracts identifiers for transaction paths marked as high-risk. These identifiers record detailed information such as the unique ID of the housing voucher-holding node participating in the transaction, the unique ID of the corresponding real estate resource node, the specific timestamp of the transaction initiation, and transaction status change records. The system organizes this information in chronological order of the transactions, removes duplicate records, and forms a set containing key information for all high-risk transaction paths—a topology blacklist.

[0075] After acquiring potential violation identifiers, the system updates the credibility attribute of property resource nodes accordingly. Potential violation identifiers cover various possible violation types, including but not limited to: fake listings (inconsistent information), unclear property rights (incomplete or disputed ownership documents), and duplicate listings (the same property listed on multiple platforms). For different types of violations, the system pre-sets corresponding credibility deduction rules. For example, for a property resource node marked as a fake listing, its initial credibility value will be reduced by a preset percentage; if there is an unclear property rights identifier, the percentage reduction is determined based on the severity of the ownership dispute; the credibility deduction percentage for duplicate listings is relatively low, but accumulates with the number of repetitions. The credibility attribute of property resource nodes uses a quantitative score of 0-100. The initial value is set based on the completeness and compliance of the information at the time of the property's initial registration. Each time a new potential violation identifier is detected, the system updates the corresponding node's credibility score in real time to ensure that the score accurately reflects the current compliance status of the property resource.

[0076] The system integrates policy adaptability weights and the priority attributes of housing voucher holders to reconstruct the connection weights of transaction path edges. The policy adaptability weights are derived from a multi-dimensional feature matrix generated after optimization using a genetic algorithm. This matrix quantifies the degree of matching between the eligibility information of housing voucher holders and regional policy provisions; a higher value indicates that the holder is more aligned with policy guidelines. The priority attributes of housing voucher holders incorporate multiple factors, including the duration of the housing voucher holding (longer holding time may increase priority), the level of housing need within the holder's family (e.g., families without housing have higher priority than families with housing), and the compliance status of past transaction records. These factors are weighted to form a quantified priority value. Connection weights are calculated using a weighted summation method, with policy adaptability weights accounting for 60% and housing voucher holder priority attributes accounting for 40%. The weighted result is normalized and mapped to a 0-1 range. Values ​​within this range directly reflect the compliance and rationality of the transaction path; higher values ​​indicate better overall performance in policy adaptability and resource matching.

[0077] Based on the updated node attributes and connection weights, the system proceeds to establish the adjacency matrix of the dynamic housing ticket flow graph. The number of rows of the adjacency matrix is consistent with the number of housing ticket holding nodes, and the number of columns corresponds to the number of real estate resource nodes. Each element value in the matrix represents the connection weight of the transaction path edge between the corresponding housing ticket holding node and real estate resource node. If a certain transaction path is included in the topology relationship blacklist, its corresponding matrix element value will be forcibly set to 0 to clearly mark the high-risk attribute of the path. At the same time, the adjacency matrix also has an additional attribute column for recording key information such as the priority attribute of each housing ticket holding node and the credibility attribute of each real estate resource node. These information and matrix element values together constitute the complete dynamic housing ticket flow graph data structure, which not only clearly shows the topological relationship between housing ticket holding nodes, real estate resource nodes and transaction path edges, but also dynamically reflects the compliance status and risk changes in the housing ticket flow process through real-time updating of node attributes and connection weights.

[0078] During the entire construction process, the system continuously monitors changes in node attributes and connection weights. When the qualification information of the housing ticket holding node changes (such as an increase in social security payment duration), the violation behavior identifier of the real estate resource node is removed or added, or the policy adaptability weight is adjusted due to the update of regional policy constraint information, the system will immediately trigger the dynamic updating mechanism to recalculate the attribute values of the relevant nodes and the connection weights of the path edges, and synchronously update the corresponding elements in the adjacency matrix. This real-time updating mechanism ensures that the dynamic housing ticket flow graph can accurately capture the latest changes in the housing ticket flow process, providing a dynamic and accurate topological relationship view for housing ticket management.

[0079] Embodiment 3: Refer to Figure 4 When generating the housing ticket transaction strategy matching rule, the system will comprehensively traverse the adjacency matrix of the dynamic housing ticket flow graph, and filter the qualified housing ticket holding nodes with the help of the pre-set qualification verification threshold. The qualification verification threshold includes multiple specific indicators. The minimum value of the policy adaptability weight is determined according to the strictness of the regional housing policy, for example, a higher value may be set in the core urban area to ensure that resources are preferentially allocated to groups with high policy adaptability; the minimum value of the housing ticket holding duration refers to the local housing ticket flow cycle regulation to prevent resource waste caused by short-term frequent transactions; and the minimum standard of the family's need level is determined based on the family's housing situation, member structure and other information, such as families without their own housing or families with elderly members, which correspond to more lenient standards. During the traversal process, the system extracts the associated attributes of the housing ticket holding nodes in the adjacency matrix row by row, and checks whether all threshold conditions are met one by one. Only the node IDs that meet all the conditions will be included in the qualified node set, and the nodes that do not meet the conditions will be marked as pending review status, and their subsequent transaction requests will need to go through an additional manual review process.

[0080] According to the credibility attribute of the real estate resource node, the system ranks the resource allocation priority from high to low according to the score. The credibility attribute comprehensively reflects the compliance status of the house source, covering the integrity of the property right certificate, the consistency of the house source information and the actual situation, the dispute records in the historical transactions and other contents. When sorting, all real estate resource nodes are first arranged in descending order according to the credibility score to form a preliminary sequence. If the credibility scores of multiple nodes are the same, the system will further combine the static attributes of the real estate for secondary sorting, such as preferentially selecting the house source with higher adaptability of the house type area and the family population, or the house source with more perfect surrounding supporting facilities, to form the final resource allocation priority list through multi-dimensional consideration. The house source at the top of the list will be preferentially recommended to the qualified house ticket holding node in the transaction matching stage.

[0081] When calculating the constraint condition of the optimal transaction path, the system will first count the connection weight data of all effective transaction path edges in the adjacency matrix, and determine the upper and lower threshold values of the path constraint through statistical analysis. The connection weight reflects the comprehensive compliance level of the transaction path, and its threshold calculation follows the formula:

[0082] T high =α+2β

[0083] T low =α-2β

[0084] Wherein, α represents the arithmetic mean of all effective transaction path connection weights in the current adjacency matrix, β represents the standard deviation of the connection weight, T high is the upper threshold value of the path weight, and T low is the lower threshold value of the path weight. The connection weight higher than the upper threshold value may exist abnormal situation of artificially raising the priority of the path, and the connection weight lower than the lower threshold value indicates that the path has deficiencies in policy adaptation or resource matching. At the same time, the system will set path taboo rules according to the topological relationship blacklist, and include the high-risk node pairs recorded in the blacklist into the prohibited transaction range, so as to clearly indicate that these nodes cannot directly establish transaction paths, and form a complete constraint boundary covering the weight threshold and taboo nodes.

[0085] After the constraint condition calculation is completed, the system converts the eligibility check condition, resource allocation priority, and transaction path constraint condition into executable policy code. The eligibility check condition is compiled into the conditional judgment logic of the rule engine, such as "when the policy adaptability weight of the house ticket holding node is ≥ 0.7 and the holding duration is ≥ 6 months, the eligibility check is passed"; the resource allocation priority is converted into sorting algorithm code, and the real-time update of the house source priority is realized by calling the dynamic sorting function; the transaction path constraint condition is compiled into path filtering logic, and the transaction path that meets the constraint is filtered out by comparing the path weight with the threshold value, checking whether the node is in the taboo list, and other operations. The generated code contains standardized input interfaces for receiving real-time transaction data uploaded by edge computing nodes; the logic processing module is responsible for executing various rule judgments and calculations; and the output interface returns transaction permission or interception instructions to ensure that the code can be directly run in the hardware environment of the edge computing node.

[0086] After the edge computing node loads the executable policy code, it enters the real-time transaction control process. When a transaction request is received, the system first extracts the user's eligibility information, including social security payment details, residence certificate number, family population registration information, etc., and compares it with the eligibility check condition in the policy rules one by one. During the comparison process, if it is found that the user's social security payment duration does not meet the threshold value, the residence status does not meet the policy requirements, etc., the system will immediately mark it as not eligible and suspend the transaction process. At the same time, the system will match the target house resource ID in the transaction request with the resource allocation priority list to query the specific ranking of the house source in the list, and if the ranking is low, it will push the information of the higher-ranked alternative house source to the user.

[0087] In the transaction path detection link, the system calculates the connection weight corresponding to the current transaction path and compares it with the upper and lower threshold values in the optimal transaction path constraint condition. If the connection weight is greater than the upper threshold value or less than the lower threshold value, or involves a node pair in the topology relationship blacklist, the system will immediately trigger the transaction termination instruction, suspend the transaction operation, and record the relevant information of the violation behavior in detail, including the violation type (such as path weight abnormality, involving high-risk nodes, etc.), the house ticket holding node ID, the house resource node ID, and the exact time of the violation. After the transaction is completed or terminated, the edge computing node will package the transaction result data, which includes the final state of the transaction (success or termination), the ranking of the matched house source in the priority list, the detailed information of the actual transaction path, etc.; the policy deviation degree data includes the difference between the actual path weight and the threshold value, the matching deviation value of the eligibility information and the check condition, etc. These data will be stored as historical records for subsequent system operation analysis. The entire process is realized through the localized processing of the edge computing node to ensure the timeliness and accuracy of transaction control.

[0088] In the running process of the edge computing-based house ticket management information system, the edge computing node continuously records transaction feedback data and policy execution logs. The transaction feedback data covers the final execution status of each transaction, including the identification of successfully completed transactions, the identification of terminated transactions due to risk interception, and the actual matching of property resource information, transaction parties' node IDs, etc. The policy execution log details the running track of the rule engine, such as the qualification check condition triggered by each transaction request, the resource priority comparison process, the path constraint check result, and the trigger time and execution duration of various instructions, etc. These data are classified and stored according to the timestamp of transaction occurrence, and each record contains complete field information, such as transaction ID, request time, user ID, house source ID, execution result code, rule matching details, etc., forming a structured historical data set to provide original data support for subsequent model updating.

[0089] The system realizes dynamic optimization through a double-time-dimension model updating mechanism. The transaction success rate and the violation interception rate are extracted from the policy execution log on an hourly basis. The transaction success rate is the ratio of the number of successfully executed transactions in a specific hour to the total number of transaction requests in that hour, and the violation interception rate is the ratio of the number of transactions that triggered termination instructions to the total number of transaction requests in the same hour. For example, during the 9:00-10:00 period of a working day, the system receives 120 transaction requests, of which 95 pass the qualification check and complete the transaction, and 25 are intercepted due to high risk level. The transaction success rate for this period is 95 / 120, and the violation interception rate is 25 / 120. After each hour ends, the system automatically calculates these indicators and generates hourly granularity data, which includes fields such as the start time of the period, the total number of transactions, the number of successful transactions, the number of intercepted transactions, the success rate, and the interception rate. These data are transmitted in real time to the model optimization module through a dedicated interface.

[0090] Table: Example of double-time-dimension model updating data

[0091]

[0092] When aggregating transaction feedback data at a daily granularity, the system calculates the daily average of resource matching deviation and strategy deviation data. Resource matching deviation reflects the difference between the actual allocated housing resources and the optimal recommended resources. It is calculated by summing the difference between the ranking of the actual selected housing resources in the priority list and the ranking of the optimal recommended resources in all transactions on that day, and then dividing by the total number of transactions on that day. For example, on a certain working day, there are 1200 transactions, and the total difference between the actual housing resource ranking and the optimal ranking is 1440, then the resource matching deviation is 1.2. Strategy deviation data considers both path weight deviation and qualification matching deviation. Path weight deviation is the absolute value of the difference between the connection weight of the actual transaction path and the optimal path weight. Qualification matching deviation is the average difference between the user's qualification information and the matching degree of the verification conditions. The two are weighted and summed according to a predetermined proportion to obtain the daily strategy deviation. For example, on a certain day, the path weight deviation is 0.4, the qualification matching deviation is 0.3, and the weighted strategy deviation is 0.35. After standardizing the daily granularity data, the dimensional differences between different indicators are eliminated, forming a unified scale of optimization data.

[0093] After inputting the hour granularity data into the capsule routing protocol layer of the housing ticket transaction risk prediction model, the system analyzes the correlation between transaction success rate, illegal interception rate and routing protocol layer coupling coefficient. If the illegal interception rate rises continuously for two hours, it indicates that the model's sensitivity to risk recognition is insufficient, and the system will increase the update step of the coupling coefficient, so that the routing protocol layer focuses on high-risk features faster in the iteration process. If the transaction success rate continues to decline, it may be that the model has over-intercepted normal transactions, so the update step will be reduced and the iteration times will be extended to make the coupling coefficient more accurately reflect the actual risk distribution. Through this dynamic adjustment, the model's recognition of recent transaction risks is consistent with the actual interception demand.

[0094] After inputting the daily granularity data into the primary capsule layer, the system will adjust the convolution kernel parameters of the spatial correlation feature according to the resource matching deviation value. If the resource matching deviation value is high, it indicates that the model's extraction of spatial adaptation features between users and housing resources is insufficient, and the system will optimize the size and number of convolution kernels to enhance the ability to capture key spatial features such as house area and regional location. For strategy deviation data, the system will adjust the parameters of the gated recurrent unit (GRU), such as the weights of the reset gate and the update gate, to more accurately encode dynamic features such as social security payment duration and housing ticket holding time in the time evolution feature subset. For example, when the strategy deviation shows that the qualification matching deviation is large, the update gate weight of the GRU will be tilted towards core qualification features such as social security payment records, improving the model's ability to capture long-term qualification evolution patterns. Through the coordinated optimization of the hour and day dimensions, the model can adapt to both short-term transaction fluctuations and long-term trend changes, continuously optimizing the accuracy of risk prediction and strategy matching.

[0095] In the parameter iteration optimization process, the system first extracts the weight parameters fine-tuned by the hour granularity data from the capsule routing protocol layer. These weight parameters are the key parameters formed by the model in the recent transaction risk identification process, reflecting the association strength between different primary capsules and risk category capsules. The system converts these weight parameters into an initial population that can be processed by the genetic algorithm, with each individual corresponding to a complete combination of routing weight parameters. The individual gene encoding length is consistent with the number of weight parameters, and the encoding value directly corresponds to the quantized value of the weight parameter. Through this conversion, the model parameters are effectively connected with the optimization algorithm, enabling the genetic algorithm to optimize based on the parameter distribution formed by the actual transaction data. The size of the initial population is determined according to the dimension of the weight parameters, ensuring sufficient diversity of the population.

[0096] After the feature encoder is reconstructed in the primary capsule layer, the system uses the updated encoder to extract features from the original house ticket management data, generating a new generation of spatial correlation feature subsets. The reconstruction of the feature encoder involves adjusting the convolution kernel parameters and the gated recurrent unit parameters, which are based on the daily granularity aggregated resource matching bias values and strategy deviation data. The updated encoder can more accurately capture the spatial correlation patterns between user qualifications and housing resources, such as more detailed encoding of matching features between user residence locations and housing resource areas, and adaptation features between family population size and house type area. The new generation of spatial correlation feature subsets contains more feature patterns reflecting actual trading rules compared to before the reconstruction, reducing the strategy matching errors caused by feature extraction bias.

[0097] The system evaluates the strategy matching degree of the new generation of spatial correlation feature subsets through the fitness function. The calculation of the fitness function takes the daily granularity aggregated resource matching bias values and strategy deviation data as the core input, where the resource matching bias values reflect the difference between the actual allocated housing resources and the optimal recommended housing resources, and the strategy deviation data reflects the deviation between the transaction path and the rule constraints. Both are weighted and summed according to the preset weight ratio, with the resource matching bias values having a higher weight to ensure that the optimization process prioritizes the accuracy of resource allocation. The lower the fitness function value, the better the strategy matching effect of the feature subset. The system evaluates the fitness of all individuals in the genetic algorithm population, retains individuals with lower fitness function values for the next generation iteration, and eliminates individuals with poor fitness, gradually improving the strategy matching ability of the feature subset through multiple iterations.

[0098] When the strategy matching degree improvement amplitude of three consecutive iterations exceeds the preset threshold, the system starts the model parameter update process. The setting of the threshold considers the typical improvement amplitude in historical optimization data and the actual business requirements for model stability, ensuring that only significant optimization results can trigger parameter updates, avoiding model fluctuations caused by frequent updates. The system selects the individual with the lowest fitness function value from the current population, decodes its gene code into specific model parameters, including the coupling coefficient matrix of the routing protocol layer, the convolution kernel parameters of the primary capsule layer, and the weight parameters of the gated recurrent unit. These parameters are updated to the corresponding modules of the house ticket transaction risk prediction model, enabling the model to apply the optimized parameter combination for subsequent risk prediction work and achieving continuous improvement of model performance.

[0099] During the entire process of transaction control at the edge computing node, the system uses a trusted evidence mechanism to ensure the security and integrity of transaction data. Transaction request data is symmetrically encrypted before transmission. The edge node and the central system generate a session key through a preset key negotiation mechanism, which is only valid for a single transaction session and automatically expires after the session ends. User eligibility information, property resource ID, transaction amount, and other sensitive data are encrypted using this session key. The encryption process is completed locally on the edge node, avoiding the risk of transmitting sensitive data in an unencrypted state.

[0100] After the strategy matching result is generated, the system signs the result data using a digital signature algorithm. The signature information includes the hash value of the result data and the unique identity of the edge node. The hash value is generated by one-way hash calculation of the result data and can uniquely identify the data content. Once the data is tampered with, the hash value will change. The digital signature is transmitted along with the strategy matching result. The receiver can confirm the authenticity of the data source and the integrity of the content by verifying the signature.

[0101] The transmission of transaction execution results uses the TLS protocol to establish an encrypted channel, ensuring the confidentiality of data during transmission. The TLS protocol negotiates encryption algorithms and session keys through a handshake process, encrypting and decrypting transmission data in real time to prevent data from being intercepted or tampered with in the transmission link. The transaction execution result data includes transaction status, involved node information, timestamp, and other content. After encapsulation, it is transmitted to the central system through the encrypted channel to form a complete transaction evidence record.

[0102] It is to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting; it is not intended to exclude myriad other embodiments of the present application that other inventors can develop based on the same general inventive concepts embodied by the described embodiments. That is, although the present application is described in terms of particular embodiments and illustrative figures, it should be apparent that the scope of the present application is not limited to these specific embodiments.

[0103] While the embodiments of the application have been shown and described herein, it is to be understood that the scope of the application, jointly pointed out in the appended claims, is not to be limited to the above-described embodiments but can be otherwise variously changed, modified, replaced, and altered within the principles and spirit of the present application.

Claims

1. A method for constructing a housing ticket management information system based on edge computing, characterized in that, include: Edge computing nodes are deployed to collect raw data on housing ticket management in real time. The raw data on housing ticket management includes user qualification verification information, real estate resource information, transaction dynamic information, and regional policy constraint information. A genetic algorithm is used to optimize the original data of housing ticket management in multiple dimensions to generate a multi-dimensional feature matrix containing policy adaptability weights. A housing voucher transaction risk prediction model based on capsule network is constructed. The multidimensional feature matrix is ​​input into the housing voucher transaction risk prediction model, and the housing voucher transaction risk level and potential violation behavior identifier are output. Based on the risk level and potential violation indicators of housing voucher transactions, a dynamic housing voucher circulation graph is constructed. This graph includes the topological relationships of housing voucher holding nodes, property resource nodes, and transaction path edges. The dynamic housing voucher circulation graph establishes an adjacency matrix by parsing high-risk transaction paths, updating node credibility, and reconstructing path edge weights, presenting the topological relationships of housing voucher holding nodes, property resource nodes, and transaction path edges, including: The high-risk transaction path identifiers in the risk level of the housing voucher transactions are analyzed to generate a topology blacklist; Update the credibility attribute of the property resource node based on the potential violation identifier; By integrating the policy adaptability weights and the priority attributes of housing voucher holder nodes, the connection weights of the transaction path edges are reconstructed. The adjacency matrix of the dynamic room ticket circulation graph is established based on the updated node attributes and connection weights; Based on the topological relationships of the dynamic housing voucher circulation graph, housing voucher transaction strategy matching rules are generated, including: Traverse the adjacency matrix of the dynamic housing ticket circulation graph and extract the set of housing ticket holding nodes that meet the eligibility verification conditions; Resource allocation priority is sorted in descending order based on the credibility attribute of real estate resource nodes; Calculate the optimal transaction path constraints by combining the connection weights of the transaction path edges; The qualification verification conditions, resource allocation priority, and transaction path constraints are compiled into executable strategy code. The matching rules for the housing voucher transaction strategy include eligibility verification conditions, resource allocation priority, and transaction path constraints. The matching rules for the housing voucher transaction strategy are distributed to edge computing nodes to execute real-time transaction control. The specific method for optimizing the multi-dimensional policy matching degree of the original data for housing ticket management using a genetic algorithm includes: Extract the social security payment duration, household registration status, and family size from the user qualification verification information as individual gene codes; The regional policy constraint information is converted into a fitness function, which includes policy clause coverage and constraint conflict penalty terms. The policy coverage rate is the proportion of the number of policy provisions whose information in an individual's genetic code satisfies the policy to the total number of key policy provisions. The constraint conflict penalty item is used to handle situations where individual genetic coding conflicts with policies; A new generation of policy adaptability weight set is generated through crossover and mutation operations, and the individual with the highest fitness function value is selected and decoded into the multidimensional feature matrix.

2. The method for constructing a housing ticket management information system based on edge computing according to claim 1, characterized in that, The specific method for constructing a housing ticket transaction risk prediction model based on capsule networks includes: The multidimensional feature matrix is ​​split into a spatial correlation feature subset and a temporal evolution feature subset; In the primary capsule layer, convolutional compression operations are performed on the spatial correlation feature subsets, and gated cyclic encoding operations are performed on the temporal evolution feature subsets. The coupling coefficient between the primary capsule output vector and the risk category capsule is calculated at the routing protocol layer, and the coupling coefficient is iteratively updated through a dynamic routing algorithm. The high-dimensional risk feature vector is aggregated at the output layer to generate the risk level of the housing voucher transaction and the identifier of potential violations.

3. The method for constructing a housing ticket management information system based on edge computing according to claim 2, characterized in that, The specific methods for implementing real-time transaction control include: The executable strategy code is loaded on the edge computing node to verify the user qualification information in the transaction request in real time; Compare the target property resources in the transaction request with the resource allocation priority list; When a transaction path is detected to deviate from the optimal transaction path constraints, a transaction termination instruction is triggered and the violation is recorded. The transaction execution results and strategy deviation data are encapsulated into transaction feedback data.

4. The method for constructing a housing ticket management information system based on edge computing according to claim 3, characterized in that, The method further includes: Record transaction feedback data and strategy execution logs; The model update mechanism, which uses the transaction feedback data and strategy execution logs to trigger a dual-time-dimensional mechanism, is used to iteratively optimize the parameters of the housing ticket transaction risk prediction model.

5. The method for constructing a housing ticket management information system based on edge computing according to claim 4, characterized in that, The specific methods of the dual-time-dimensional model update mechanism include: Collect the transaction success rate and violation interception rate from the strategy execution logs at an hourly granular level; Aggregate the resource matching deviation value and strategy deviation data from the transaction feedback data at a daily granularity. Hourly granular data is input into the capsule routing protocol layer of the housing ticket transaction risk prediction model for weight fine-tuning. The daily granular data is input into the primary capsule layer of the housing ticket transaction risk prediction model for feature encoder reconstruction.

6. The method for constructing a housing ticket management information system based on edge computing according to claim 5, characterized in that, The specific methods for parameter iterative optimization include: The fine-tuned capsule routing protocol layer weights are extracted as the initial population for the genetic algorithm; A new generation of spatially correlated feature subsets is generated based on the reconstructed feature encoder; The policy matching degree of the next-generation spatially associated feature subset is evaluated using a fitness function; When the strategy matching degree improves beyond a preset threshold, the parameter set of the housing ticket transaction risk prediction model is updated.

7. The method for constructing a housing ticket management information system based on edge computing according to claim 1, characterized in that, It also includes a trusted evidence storage mechanism for transaction data, specifically: during the transaction control process at the edge computing node, transaction requests, strategy matching results, and transaction execution results are transmitted via an encrypted channel.

Images