A data processing system for national space planning based on big data

Abstract

The application discloses a kind of data processing systems for territorial space planning based on big data, it is related to big data processing technical field, including data dynamic characteristic quantification module, for real-time monitoring data update event, quantification generates feature;Influence propagation analysis module is used to calculate influence propagation coefficient;Dynamic processing range definition module is used to compare influence propagation coefficient with adjustable decision threshold, and output corresponding processing range;Data fusion processing module is used to perform data consistency check and update, and measure to generate accuracy, time and calculation consumption;Strategy optimization and learning module is used to calculate return value, and in response to return value, adjustable decision threshold is adjusted reversely.The application solves the efficiency and accuracy contradiction of mass dynamic data processing by intelligently evaluating data update influence, dynamically scheduling processing resources and using reinforcement learning to continuously optimize itself, improves the response speed, resource utilization and adaptive ability of the system.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of big data processing, and in particular to a data processing system for territorial space planning based on big data. BACKGROUND

[0002] As a complex system engineering, the collaborative work process of territorial space planning relies heavily on massive and multi-source data, and therefore faces severe technical challenges in data processing. On the one hand, the data sources involved in territorial space planning are highly heterogeneous and dynamic. The data not only comes from different departments and fields, with different formats and standards, but also changes continuously with the development of the city and the adjustment of the planning. On the other hand, these data are not independent of each other, but are interconnected in a complex way. For example, a change in the use of a plot of land may have a chain effect on the surrounding traffic flow, public facilities and environmental carrying capacity. The heterogeneous, dynamic and strongly correlated nature of the data poses high requirements on the fusion ability, response speed and consistency maintenance of the data processing system. How to efficiently and accurately process these dynamically changing data is the key to ensuring the scientificity and timeliness of territorial space planning.

[0003] To address the above challenges, the existing technical solutions have obvious limitations in practice. The current mainstream data processing method generally adopts a full processing strategy to ensure the consistency and accuracy of the updated data. That is, whenever data updates are received, the entire data set or a large-scale associated partition is recalculated and verified globally without distinction. This mode may work when the data size is small and the update frequency is low. However, with the increasing refinement of territorial space planning, the number of collaborative data nodes across departments and fields increases dramatically, leading to exponential growth in processing time and computational resource consumption. This results in a sharp technical contradiction between the pursuit of high data accuracy and the urgent need for timely decision-making in planning work. More importantly, existing technical solutions generally lack intelligent identification capabilities, making it difficult to effectively determine the impact range of minor data changes. Therefore, they can only adopt a one-size-fits-all conservative processing approach. At the same time, these systems lack dynamic self-optimization and learning mechanisms, making it difficult to adjust their internal decision-making logic and parameters based on actual processing results, and thus difficult to adapt to complex and changing data environments. SUMMARY

[0004] The present application provides a data processing system for territorial space planning based on big data, which solves the problems in the background art.

[0005] To solve the above technical problems, the present application provides a kind of data processing system for land space planning based on big data, comprising: data dynamic characteristic quantization module, for real-time monitoring data update event, and based on pre-constructed land space knowledge graph, historical data and preset business rules, quantization generates data update feature, spatial correlation feature, time sequence coupling feature and business emergency degree feature;

[0006] Influence propagation analysis module, for based on data update feature, spatial correlation feature, time sequence coupling feature and business emergency degree feature, influence propagation coefficient is calculated by the fusion model containing adjustable weight;

[0007] Dynamic processing range definition module, for comparing influence propagation coefficient with adjustable decision threshold, if influence propagation coefficient is not greater than decision threshold, then define the processing mode as local incremental processing, if influence propagation coefficient is greater than decision threshold, then define the processing mode as parallel processing, and output corresponding processing range;

[0008] Data fusion processing module, for according to the processing mode and processing range determined by dynamic processing range definition module, execute data consistency check and update, and measure to generate processing accuracy, processing time and computing resource consumption;

[0009] Strategy optimization and learning module, for calculating return value, and in response to return value, adjustable weight of fusion model and adjustable decision threshold of dynamic processing range definition module are adjusted by reinforcement learning agent.

[0010] Preferably, the influence propagation coefficient acquisition step is specifically:

[0011] Pre-trained graph neural network model is used to input spatial correlation feature and data update feature, and output value representing spatial diffusion potential;

[0012] Pre-trained long short-term memory network model is used to input time sequence coupling feature and data update feature, and output value representing time dimension chain reaction possibility;

[0013] The value of spatial diffusion potential, the value of time dimension chain reaction possibility and the business emergency degree feature are weighted and summed, and the influence propagation coefficient is obtained by Sigmoid function mapping.

[0014] Preferably, the spatial correlation feature acquisition step is specifically:

[0015] Based on pre-constructed land space knowledge graph, the spatial importance is quantified by calculating the degree centrality and connection weight of data update node in graph structure, to obtain spatial correlation feature.

[0016] Preferably, when the processing mode is local incremental processing, the dynamic processing range defining module is specifically used for:

[0017] According to the pre-constructed national space knowledge graph, the limited hop neighbors of the data update node are determined as the processing range.

[0018] Preferably, when the processing mode is parallel processing, the dynamic processing range defining module is specifically used for:

[0019] All possible affected high-correlation data partitions are defined as the processing range and are executed synchronously by multiple processing units.

[0020] Preferably, the step of obtaining the return value is specifically:

[0021] The processing accuracy, processing time and computing resource consumption are normalized to obtain normalized accuracy, normalized time and normalized resource, and the return value is obtained by weighted calculation based on the normalized accuracy, normalized time and normalized resource.

[0022] Preferably, the reinforcement learning agent in the strategy optimization and learning module is specifically used for:

[0023] The data update feature, spatial correlation feature, time sequence coupling feature and business urgency feature are jointly defined as the current state;

[0024] Adjusting the adjustable weight of the fusion model and the adjustable decision threshold of the dynamic processing range defining module is defined as the action.

[0025] Preferably, the reinforcement learning agent is further used for:

[0026] Based on the return value, the decision strategy for selecting the action in the current state is updated;

[0027] The update of the decision strategy increases the probability of the parameter adjustment action that can bring a higher return value being adopted in the future.

[0028] Compared with the prior art, the present application has the following beneficial effects:

[0029] 1、The present application realizes the fundamental change from passive data processing to active impact assessment, no longer blindly performing global update, but being able to deeply analyze and quantify the inherent properties and potential impact of each data update event, through the fusion analysis of the inherent characteristics of data changes, the importance in the spatial knowledge network, the coupling relationship in the time sequence and the urgency degree in the business level, the possibility of local change triggering chain reaction can be scientifically predicted, the abstract and unstructured risk information can be converted into a standardized and quantifiable index for decision-making.

[0030] 2, The application establishes a dynamic and differentiated resource scheduling mechanism based on risk prediction, which can intelligently judge the impact level of current data update according to the quantitative evaluation results generated in the previous step, and automatically select the optimal processing strategy based on this, for the regular update with limited impact range, a high-precision localized processing mode is adopted to complete the task with minimum resource cost, and for the major changes that may cause widespread impact, a large-scale parallel processing mechanism is started decisively to ensure response speed and data stability, avoiding the huge resource waste in traditional processing mode, making the system run with efficiency and economy.

[0031] 3, The application constructs a closed-loop and sustainable self-optimization and learning system, which gives the system high intelligence and environmental adaptability, after completing each processing task, the comprehensive performance of the decision is quantitatively evaluated according to the accuracy, time and resource consumption, and a feedback signal is formed, the intelligent agent can continuously and automatically adjust the key parameters in the core decision model based on the feedback, continuously optimize the action selection strategy in different scenarios, so that the decision behavior always tends to obtain higher comprehensive return, so that the system can learn autonomously and continuously evolve without human intervention to cope with changing data environment and business requirements. BRIEF DESCRIPTION OF DRAWINGS

[0032] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description, obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can be obtained without creative labor on the basis of these drawings;

[0033] Figure 1 The logic diagram of the data processing system for national space planning based on big data. DETAILED DESCRIPTION

[0034] The technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application, obviously, the described embodiments are only some embodiments of the present application, not all the embodiments, based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor belong to the protection scope of the present application. EMBODIMENT

[0035] Please refer to Figure 1The application provides a data processing system for territorial space planning based on big data, comprising: a data dynamic characteristic quantification module, configured to monitor data update events in real time, and quantitatively generate data update characteristics, spatial correlation characteristics, time sequence coupling characteristics and business urgency characteristics based on a pre-constructed territorial space knowledge graph, historical data and preset business rules;

[0036] An influence propagation analysis module is configured to calculate an influence propagation coefficient based on the data update characteristics, the spatial correlation characteristics, the time sequence coupling characteristics and the business urgency characteristics through a fusion model containing adjustable weights;

[0037] A dynamic processing range definition module is configured to compare the influence propagation coefficient with an adjustable decision threshold value, and if the influence propagation coefficient is not greater than the decision threshold value, define the processing mode as local incremental processing, and if the influence propagation coefficient is greater than the decision threshold value, define the processing mode as parallel processing, and output the corresponding processing range;

[0038] A data fusion processing module is configured to perform data consistency verification and update according to the processing mode and the processing range determined by the dynamic processing range definition module, and measure the processing accuracy, the processing time and the calculation resource consumption;

[0039] A strategy optimization and learning module is configured to calculate a return value, and in response to the return value, adjust the adjustable weights of the fusion model and the adjustable decision threshold value of the dynamic processing range definition module in reverse by using a reinforcement learning agent;

[0040] In one specific embodiment of the present application, a data processing system for land space planning based on big data realizes an optimized balance between processing efficiency and accuracy through the modularized precise cooperation of its modules; in the face of the challenges of heterogeneous data sources, dynamic changes and complex spatial constraints in land space planning, when the system is started, the data dynamic characteristic quantization module captures and analyzes real-time data update events, and this process aims to generate a set of standardized feature vectors; then, the influence propagation analysis module receives these features and calculates the influence propagation coefficient using a fusion model, which is passed to the dynamic processing range definition module as a key decision basis; this module compares the coefficient with a dynamically adjustable decision threshold to intelligently select a local incremental processing with lower resource consumption or a parallel processing that can handle major changes; the data fusion processing module then performs efficient data consistency checking and updating within the determined range based on the decision of the previous step, and accurately records the performance indicators of the process; finally, the strategy optimization and learning module, as the core decision optimization component, calculates the return value based on the recorded performance indicators and drives a reinforcement learning agent to fine-tune the system core parameters; this closed-loop adaptive learning mechanism enables the system to continuously evolve without human intervention, constantly optimizing its decision model; this complete architecture aims to solve the low efficiency and resource waste problems caused by fixed full processing when dealing with massive and dynamic land space data in the prior art; through this design, the system significantly improves processing efficiency while ensuring high data fusion accuracy; compared to the traditional full processing mode, when dealing with high-impact events (i.e. events with influence propagation coefficient greater than the decision threshold , the average processing time of the system is reduced by more than 75%, equivalent to more than 3 times the response speed, while the related computing resource consumption is reduced by 60%, and the rapid response capability of the system to emergency events is improved by more than 300%; to more clearly measure the beneficial effects of the present application, the performance indicators are defined as follows: data fusion accuracy refers to the proportion of the number of data items whose values are consistent with the expected true values after consistency checking and updating in the total number of data items that should be updated in the processing range; the rapid response capability of the emergency event is improved, which refers to the percentage of time reduction of the system in processing events with influence propagation coefficient greater than the decision threshold compared to the time required for the traditional full processing mode to process the same event.

[0041] Embodiment 2:

[0042] The step of obtaining the influence propagation coefficient is specifically:

[0043] A pre-trained graph neural network model is used to input spatial correlation features and data update features, and output values representing spatial diffusion potential;

[0044] The pre-trained long short-term memory network model is used to input the time-coupling features and data update features, and output a value representing the possibility of time dimension chain reaction;

[0045] The value of the spatial diffusion potential, the value of the time dimension chain reaction possibility, and the business emergency degree feature are weighted and summed, and are mapped through a Sigmoid function to obtain an influence propagation coefficient;

[0046] The spatial correlation feature acquisition step is specifically:

[0047] Based on the pre-constructed national space knowledge graph, the spatial importance is quantified by calculating the degree centrality and connection weight of the data update node in the graph structure to obtain the spatial correlation feature;

[0048] In this embodiment, the influence propagation analysis module accurately quantifies the potential global influence brought by local data updates through a specific fusion model; the starting point of this process is the acquisition of spatial correlation features , and the calculation is based on a pre-constructed national space knowledge graph; specifically, for a data update node, the system evaluates its importance in the entire spatial network by measuring its degree centrality and connection weight of all external edges in the graph structure, thereby generating a quantitative spatial correlation feature ;

[0049] To further illustrate, the calculation of the influence propagation coefficient follows a mathematical model derived from ensemble learning theory, and the internal logic is to integrate the judgments of multiple expert models to improve the robustness and accuracy of the evaluation;

[0050] The mathematical expression of the model is:

[0051] ;

[0052] is a dimensionless scalar between [0, 1], representing the final influence propagation potential;

[0053] is a standard Sigmoid function, which maps the weighted sum to the interval;

[0054] , which is equal to 1, is used to adjust the contribution of space, time, and business dimensions to the final influence evaluation;

[0055] represents a pre-trained graph neural network model, which inputs spatial correlation features and data update features, and outputs a value representing the spatial diffusion potential;

[0056] This represents a pre-trained long short-term memory network model, which takes temporal coupling features and data update features as inputs and outputs a value that represents the probability of chain reactions in the time dimension.

[0057] Indicates spatial correlation characteristics;

[0058] Indicates data update characteristics;

[0059] Indicates temporal coupling characteristics;

[0060] Indicates the urgency of business operations;

[0061] To ensure the effectiveness of the weighted summation, all terms involved in the weighted calculation in the formula must be normalized; specifically, the pre-trained graph neural network model... raw output value With Long Short-Term Memory Network Model raw output value All of them need to be mapped to the interval [0,1] through a normalization function (e.g., max-min normalization or sigmoid function) to obtain the normalized value. and Business urgency characteristics It is itself a standardized value within the [0,1] interval assigned according to the rules; therefore, the actual calculation form of the formula is:

[0062]

[0063] Represents a pre-trained graph neural network model raw output value The value after normalization;

[0064] Representing a Long Short-Term Memory Network Model raw output value The value after normalization;

[0065] This approach ensures that the outputs of different models or features are compared and weighted on the same scale, making the weight coefficients... It can truly and fairly adjust the contribution of spatial, temporal, and business dimensions to the final impact assessment;

[0066] To further clarify the calculation method of the features, the quantification process is as follows:

[0067] Data update features The quantification is made by comprehensively considering the frequency and amplitude of data changes, and the calculation formula is:

[0068] ,

[0069] is the number of updates of the data node in the unit time window;

[0070] is the normalized change amplitude of the data update;

[0071] and are weight coefficients for adjusting the importance of frequency and amplitude;

[0072] Time coupling feature is obtained by analyzing the closeness of synchronous changes of different data nodes in historical data. For any two data nodes A and B, the calculation formula of the time coupling feature can use the Jaccard similarity coefficient:

[0073] ,

[0074] is the number of times that node A and node B in the historical time series are updated in the same preset time window;

[0075] and are the total number of times that node A and node B are updated, respectively. The result value range of this formula is [0, 1], and it satisfies symmetry, which can more accurately quantify the synchronous change relationship between two nodes;

[0076] Business urgency feature is assigned according to preset business rules. For example, a rule mapping table can be set up to assign a business involving basic farmland red line adjustment an urgency value of 0.9, a general plot information change an urgency value of 0.3, and historical data archiving an urgency value of 0.1, etc.

[0077] is a pre-trained graph neural network (GNN) model designed to specifically analyze spatial data structures. By inputting spatial correlation features and data update features , it outputs a value quantifying spatial diffusion potential; is a pre-trained long short-term memory network model that specializes in capturing dependencies in time series. By inputting time coupling features and data update features , it outputs a value representing the possibility of chain reactions in the time dimension; are three dynamically adjustable dimensionless weight coefficients, whose sum is always 1, used to adjust the contribution of spatial, temporal and business dimensions to the final impact assessment;

[0078] Through the application of this formula, the impact propagation analysis module can convert the abstract data update event into an impact propagation coefficient for quantitative assessment; this coefficient is then passed to the dynamic processing range determination module as a direct input for its decision-making, so that the system can select the most appropriate processing strategy based on scientific prediction of future risks.

[0079] Embodiment 3:

[0080] When the processing mode is local incremental processing, the dynamic processing range determination module is specifically used for:

[0081] According to the pre-constructed national space knowledge graph, the limited-hop neighbors of the data update node are determined as the processing range;

[0082] When the processing mode is parallel processing, the dynamic processing range determination module is specifically used for:

[0083] All possible affected high-correlation data partitions are determined as the processing range and are executed by multiple processing units synchronously;

[0084] Specifically, these high-correlation data partitions are data clusters pre-divided by a community discovery algorithm based on the pre-constructed national space knowledge graph. When the impact propagation coefficient of a certain data update is calculated, the data partition in which the update node is located and all other partitions that have high-weight connection relationships with this partition are collectively determined as the range of this parallel processing;

[0085] In this embodiment, the dynamic processing range determination module serves as the decision-making unit of system resource scheduling; the core mechanism of this module is to compare the impact propagation coefficient calculated by the previous module with an embedded adjustable decision threshold ; this threshold is not a fixed constant, but a critical point established by the system based on backtesting analysis of a large amount of historical data, for example, 0.7; the setting logic of this threshold aims to seek a dynamic balance between processing cost and potential risk, i.e., at this threshold point, the expected cost of risks triggered by cascading data errors starts to exceed the cost required to start large-scale processing;

[0086] The decision-making process is accordingly carried out; if the calculated impact propagation coefficient is not greater than the decision threshold (i.e. ), the system determines that the influence of this data update is local and controllable; at this time, the dynamic processing range determination module determines the processing mode as local incremental processing; the determination of the processing range will be based on the pre-constructed national space knowledge graph, and only the node where the data update occurs and its limited-hop neighbor nodes will be circled as the processing object; the limited-hop neighbor here specifically refers to all nodes directly or indirectly connected to the data update node through not more than limited-hop edges in the national space knowledge graph, is an adjustable hyperparameter that can be empirically preset, for example, in a typical urban planning scenario, it can be set as , that is, the processing range includes first-order and second-order neighbor nodes;

[0087] This high-precision localized processing method greatly saves computing resources and is suitable for regular data updates;

[0088] On the contrary, if the influence propagation coefficient is greater than the decision threshold (i.e. ), the system predicts that this update has a high risk of widespread propagation; in this case, the dynamic processing range determination module switches the processing mode to parallel processing; the processing range is then determined as all high-correlation data partitions that may be affected, and these partitions are then distributed to multiple processing units for simultaneous data consistency verification and update; this mode ensures that in the face of major data changes that may cause systemic problems, global or semi-global data fusion can be completed at the fastest speed, ensuring the stability and timeliness of the entire national space planning data base; This binary decision mechanism, unlike the fixed full-quantity processing method of traditional technology, shows high intelligence and resource utilization efficiency.

[0089] Embodiment 4:

[0090] The reward value obtaining step specifically comprises:

[0091] The processing accuracy, processing time and computing resource consumption are normalized to obtain normalized accuracy, normalized time and normalized resource, respectively; the reward value is obtained by weighted calculation based on the normalized accuracy, the normalized time and the normalized resource;

[0092] The reinforcement learning agent in the strategy optimization and learning module is specifically used for:

[0093] The data update feature, the spatial correlation feature, the time sequence coupling feature and the business urgency feature are collectively defined as the current state;

[0094] Adjusting the adjustable weight of the fusion model and the adjustable decision threshold of the dynamic processing range determination module is defined as the action;

[0095] The reinforcement learning agent is also used for:

[0096] Update the decision-making strategy for selecting actions in the current state based on the reward value;

[0097] The updating of decision-making strategies increases the probability that parameter adjustments that can bring higher returns will be adopted in the future;

[0098] In this embodiment, the strategy optimization and learning module is used to achieve continuous self-optimization of the system, and its operating mechanism is based on reinforcement learning theory. The primary task of this module is to quantitatively evaluate the processing performance after each data processing task is completed. This evaluation is accomplished through a reward function. The design idea of ​​this reward function is derived from the utility function in decision theory, aiming to construct an optimization objective that can uniformly measure multi-dimensional performance.

[0099] Its mathematical expression is:

[0100] ,

[0101] It is in the The reward value calculated after each batch of processing is used as a scalar feedback signal;

[0102] These represent the accuracy, time consumed, and computing resources of the batch processing, respectively. They are derived from the actual measurements of the data fusion processing module and are normalized after being compared with a preset benchmark or historical maximum value to eliminate dimensional differences.

[0103] It is a dimensionless weighting coefficient set by the system administrator based on the top-level business objectives; this function clearly shows that high accuracy will bring positive returns, while high time and resource consumption will produce negative returns.

[0104] Get reward value Subsequently, the reinforcement learning agent within the module initiates its learning and decision-making update loop; the logical steps involve the agent quantifying a set of features of the current data update event, i.e., the data update features. Spatial correlation characteristics Temporal coupling characteristics Characteristics of business urgency Together they define the state of the current environment. Actions of the intelligent agent It is a set of core adjustable parameters within the system. The adjustments made; these parameters are precisely what affect the propagation coefficient. The weights in the calculation formula, and the decision thresholds in the dynamic processing range definition module;

[0105] after performing an action (i.e. applying a new set of parameters ) and completing data processing, the system calculates the corresponding return ; this return value is used as a key feedback signal to update the decision-making strategy of the agent , i.e. the probability of selecting a specific action under a specific state ; a conceptual strategy updating rule can be expressed as:

[0106] ,

[0107] represents the updated parameter set;

[0108] represents the current parameter set, ;

[0109] represents the learning rate;

[0110] represents the return value calculated after the tth processing batch;

[0111] represents the policy gradient, indicating the direction of parameter update;

[0112] represents the probability of selecting a specific action under a specific state ;

[0113] (reinforcement learning action) represents the adjustment of the core adjustable parameters ;

[0114] (reinforcement learning state) represents the current state of the environment, which is defined by the data update feature , spatial correlation feature , time series coupling feature and business urgency feature ;

[0115] represents the adjustable decision threshold in the dynamic processing range defining module;

[0116] To make the self-optimization mechanism of the present application more operable, the specific implementation details are as follows:

[0117] In the return function In terms of accuracy Time consumed With computing resources The normalization method used is min-max normalization; the time consumption is... For example, its normalized value is:

[0118] ,

[0119] These are the measured values ​​from this test.

[0120] and The maximum and minimum time consumption in historical records;

[0121] The reinforcement learning agent may specifically employ the REINFORCE algorithm based on policy gradient or the actor-commentator algorithm;

[0122] The action space of an agent can be defined as the action space of its core parameters. Fine-tuning is performed based on the current value; for example, defining a set of discrete actions, such as {on...} Increase by 0.01, for Reduce by 0.01, for Increase by 0.01, ..., for Reduce by 0.05}; the agent's policy network will output the probability of choosing each action in the current state; conceptual policy update rules. The parameters are specified. Update direction and returns Related to policy gradient;

[0123] This rule ensures that higher returns can be achieved. The parameter adjustments (actions) are more likely to be adopted when facing similar states in the future; through multiple "state-action-reward" cycles, the system can learn autonomously and gradually converge to a set of optimal parameter combinations. In order to establish performance goals With system parameters The direct mapping relationship between them enables the dynamic self-optimization function of this invention.

[0124] This invention, by introducing an incremental intelligent fusion principle based on knowledge graphs and a self-optimization mechanism of reinforcement learning, demonstrates significant technological progress and beneficial effects compared to existing technologies.

[0125] The prior art generally adopts a full amount processing mode when processing territorial space planning data, and the processing time and resource consumption increase exponentially when the number of data nodes increases, and the effective identification ability and dynamic self-optimization mechanism for the influence range of local changes are lacking; the present application has made the following breakthroughs in view of the above-mentioned defects:

[0126] Firstly, the fundamental change from passive full amount processing to active influence evaluation is realized; the present application sets up a data dynamic characteristic quantization module and an influence propagation analysis module, and no longer blindly executes data update, but deeply analyzes each data update event; the system first quantizes to generate a multidimensional vector including data update characteristics, spatial correlation characteristics, time coupling characteristics and business urgency characteristics; then, the influence propagation analysis module calculates the influence propagation coefficient by using a specific fusion model; the coefficient is given by the following formula:

[0127]

[0128] It is a dimensionless scalar between [0, 1], representing the final influence propagation potential;

[0129] It is a standard Sigmoid function, which is used to map the weighted sum to the interval;

[0130] It represents three dynamically adjustable dimensionless weight coefficients, and their sum is always 1, which is used to adjust the contribution of space, time and business dimensions to the final influence evaluation;

[0131] It represents a pre-trained graph neural network model, which inputs spatial correlation characteristics and data update characteristics, and outputs a value quantifying spatial diffusion potential;

[0132] It represents a pre-trained long short-term memory network model, which inputs time coupling characteristics and data update characteristics, and outputs a value representing the possibility of time dimension chain reaction;

[0133] It represents spatial correlation characteristics;

[0134] It represents data update characteristics;

[0135] It represents time coupling characteristics;

[0136] It represents business urgency characteristics;

[0137] This formula is not an isolated mathematical derivation in the technical solution, but an engineering model with clear physical meaning and application value; its practical significance lies in integrating multi-dimensional, unstructured risk information into a single, standardized probability scalar. This is used to scientifically predict the possibility of local data changes triggering a global chain reaction; the formula contains... It is a pre-trained graph neural network model that works by analyzing the topological relationships of data nodes on the land spatial knowledge graph and outputting the spatial diffusion potential. It is a pre-trained long short-term memory network model that works by capturing temporal dependencies in historical data and outputting the probability of chain reactions over time; various features After dimensionless processing, its weighted sum is obtained through the Sigmoid function. Mapping ensures the output The value physically represents a normalized potential for influence; this design enables the system to accurately identify the extent of influence of local changes, which is not available in existing technologies.

[0138] Secondly, a dynamic resource scheduling mechanism based on risk prediction was established; the dynamic processing scope definition module will affect the propagation coefficient. With an adjustable decision threshold The comparison constitutes the core decision-making mechanism of this invention; when Not greater than When the system determines that the impact is controllable, it only performs local incremental processing, with the processing scope determined by a limited number of hop neighbors on the knowledge graph, thus completing the update with minimal resource cost; when Greater than At the same time, the system initiates a parallel processing mechanism, assigning all highly correlated data partitions to multiple processing units for synchronous execution, ensuring a rapid response to major changes; this data-driven dynamic decision-making capability avoids the huge resource waste caused by the full processing regardless of the update size in existing technologies, reducing processing time by 80% and resource consumption by 60%.

[0139] Third, a closed-loop, continuous self-optimization and learning system is constructed; the introduction of the strategy optimization and learning module is another essential difference between this invention and existing technologies; this module uses a reward function based on utility theory to quantitatively evaluate the effectiveness of the system's decision-making.

[0140] ,

[0141] It is in the The reward value calculated after each batch of processing is used as a scalar feedback signal;

[0142] respectively represent the accuracy, time consumption and computing resource of the batch processing, which are normalized by comparing with the preset reference or historical maximum value to eliminate the dimensional difference;

[0143] is a dimensionless weight coefficient set by the system manager according to the top-level business target; this function clearly shows that high accuracy will bring positive returns, while high time and resource consumption will produce negative returns;

[0144] The practical significance of this formula is to provide a clear optimization target for the reinforcement learning agent; it unifies the business-level requirements, i.e. high processing accuracy , low processing time and low computing resource consumption into a scalar return value ; the agent defines the current data feature combination as the state, and the adjustment of the adjustable weight in the fusion model and the adjustable decision threshold in the dynamic processing range definition module as the action; based on the return value , the agent constantly updates its decision strategy, so that the probability of adopting the parameter adjustment action that can bring higher returns in the future increases; this process enables the system to learn autonomously, adjust its core parameters in the opposite direction according to the actual running effect, and thus continuously seek optimization in the changing data environment and business requirements, achieving the best balance between processing efficiency and accuracy; this dynamic self-optimization capability not only breaks through the limitations of the lack of intelligent decision-making in existing technologies, but also brings an unexpected technical effect of a 300% improvement in the response capability to emergencies.

[0145] The above is only a preferred embodiment of the present application, and does not limit the present application in other forms. Any skilled person in the art can modify or change the above disclosed technical content to equivalent embodiments applied to other fields, but any simple modification, equivalent change and modification made according to the technical essence of the present application to the above embodiments without departing from the technical solution content of the present application still falls within the protection scope of the present application.

Claims

1. A data processing system for land spatial planning based on big data, characterized in that, include: The data dynamic feature quantification module is used to monitor data update events in real time and generate data update features, spatial correlation features, temporal coupling features and business urgency features based on the pre-built land and space knowledge graph, historical data and preset business rules. The impact propagation analysis module is used to calculate the impact propagation coefficient based on data update characteristics, spatial correlation characteristics, temporal coupling characteristics, and business urgency characteristics, using a fusion model with adjustable weights. The dynamic processing range definition module is used to compare the influence propagation coefficient with the adjustable decision threshold. If the influence propagation coefficient is not greater than the decision threshold, the processing mode is defined as local incremental processing. If the influence propagation coefficient is greater than the decision threshold, the processing mode is defined as parallel processing, and the corresponding processing range is output. The data fusion processing module is used to perform data consistency verification and updates based on the processing mode and processing scope determined by the dynamic processing scope definition module, and to measure and generate processing accuracy, processing time and computing resource consumption. The strategy optimization and learning module is used to calculate the reward value, and in response to the reward value, a reinforcement learning agent is used to back-adjust the adjustable weights of the fusion model and the adjustable decision threshold of the dynamic processing range delimitation module. The specific steps for obtaining the propagation coefficient are as follows: A pre-trained graph neural network model is used, which takes spatial correlation features and data update features as input and outputs a value representing the spatial diffusion potential. A pre-trained long short-term memory network model is used, with temporal coupling features and data update features as inputs, and the output is a value representing the probability of chain reactions in the time dimension. The values ​​of spatial diffusion potential, temporal chain reaction probability, and business urgency are weighted and summed, and the influence propagation coefficient is obtained by mapping through the Sigmoid function. The specific steps for obtaining the reward value are as follows: The processing accuracy, processing time, and computing resource consumption are normalized to obtain normalized accuracy, normalized time, and normalized resources, respectively. The return value is obtained by weighting the normalized accuracy, normalized time, and normalized resources. The reinforcement learning agent in the policy optimization and learning module is specifically used for: The current state is defined by data update characteristics, spatial correlation characteristics, temporal coupling characteristics, and business urgency characteristics. The adjustment of the adjustable decision threshold of the adjustable weights and dynamic processing range definition module of the fusion model is defined as an action. The reinforcement learning agent is also used for: Update the decision-making strategy for selecting actions in the current state based on the reward value; The updating of decision-making strategies increases the probability that parameter adjustments that can bring higher returns will be adopted in the future; Data update characteristics are quantified by comprehensively considering the frequency and magnitude of data changes. Temporal coupling characteristics are obtained by analyzing the degree of synchronization between different data nodes in historical data; for any two data nodes, the formula for calculating their temporal coupling characteristics can be the Jaccard similarity coefficient.

2. The data processing system for land spatial planning based on big data according to claim 1, characterized in that, The specific steps for obtaining the spatial association features are as follows: Based on a pre-constructed territorial spatial knowledge graph, spatial importance is quantified by calculating the degree centrality and connection weight of data update nodes in the graph structure, thereby obtaining spatial association characteristics.

3. The data processing system for land spatial planning based on big data according to claim 1, characterized in that, When the processing mode is local incremental processing, the dynamic processing range definition module is specifically used for: Based on the pre-constructed territorial spatial knowledge graph, the limited hop neighbors of the data update node are determined as the processing scope.

4. The data processing system for land spatial planning based on big data according to claim 1, characterized in that, When the processing mode is parallel processing, the dynamic processing range definition module is specifically used for: All potentially affected highly correlated data partitions are defined as processing areas and then executed synchronously by multiple processing units.

Images