Distributed Real-time Processing Method and System for Spatiotemporal Data

Abstract

The present application relates to the technical field of data processing, and discloses a spatio-temporal data distributed real-time processing method and system. The method includes: extracting features of data source nodes in a smart city spatio-temporal big data AI platform to obtain a data source node feature vector set; performing channel allocation on edge nodes in the data source node feature vector set to obtain an edge node power allocation matrix and a channel allocation indication matrix; performing utility calculation and task iterative solution according to the edge node power allocation matrix and the channel allocation indication matrix to obtain a task allocation strategy matrix; calculating a target score for each task in the task allocation strategy matrix, and selecting an optimal data processing link based on the target score, thereby effectively reducing the task processing failure rate and reducing data transmission conflicts of edge nodes.

Description

Technical Field

[0001] This application relates to the technical field of data processing, and particularly to a method and system for distributed real-time processing of spatio-temporal data. Background Art

[0002] With the rapid development of smart cities, the demand for real-time processing of spatio-temporal big data is increasing day by day. The massive spatio-temporal data generated by various urban sensing devices and monitoring systems has characteristics such as strong timeliness, large data volume, and wide distribution, which pose great challenges to the real-time processing of data. Traditional centralized processing methods are difficult to meet the real-time requirements, and existing distributed processing methods still have many deficiencies in processing node selection and task allocation.

[0003] Especially in the case where the distance between the data source node and the processing node is relatively far, problems such as large processing delay and transmission interruption are likely to occur, affecting the stability and reliability of the entire system. In addition, existing methods often only consider traditional factors such as computing resources and network bandwidth when performing task allocation, ignoring the affinity relationship between nodes and the reliability of links, resulting in unsatisfactory task allocation effects. Summary of the Invention

[0004] This application provides a method and system for distributed real-time processing of spatio-temporal data, thereby effectively reducing the task processing failure rate and reducing data transmission conflicts at edge nodes.

[0005] In the first aspect of this application, a method for distributed real-time processing of spatio-temporal data is provided. The method for distributed real-time processing of spatio-temporal data includes:

[0006] Extract features from data source nodes in the spatio-temporal big data AI platform of a smart city to obtain a set of data source node feature vectors;

[0007] Perform channel allocation on edge nodes in the set of data source node feature vectors to obtain an edge node power allocation matrix and a channel allocation indication matrix;

[0008] Perform utility calculation and task iterative solution according to the edge node power allocation matrix and the channel allocation indication matrix to obtain a task allocation strategy matrix;

[0009] Calculate the target score of each task in the task allocation strategy matrix, and select the optimal data processing link based on the target score.

[0010] In the second aspect of this application, a system for distributed real-time processing of spatio-temporal data is provided. The system for distributed real-time processing of spatio-temporal data includes:

[0011] A feature extraction module, which is used to extract features from data source nodes in the spatio-temporal big data AI platform of a smart city to obtain a set of feature vectors of the data source nodes;

[0012] An allocation module, which is used to allocate channels to edge nodes in the set of feature vectors of the data source nodes to obtain an edge node power allocation matrix and a channel allocation indication matrix;

[0013] A solution module, which is used to perform utility calculation and task iterative solution according to the edge node power allocation matrix and the channel allocation indication matrix to obtain a task allocation strategy matrix;

[0014] A selection module, which is used to calculate the target score of each task in the task allocation strategy matrix and select the optimal data processing link based on the target score.

[0015] Compared with the prior art, the present application has the following beneficial effects: By introducing a node affinity evaluation mechanism and combining with a link reliability probability matrix, it can accurately reflect the cooperation effect between nodes, effectively reducing the task processing failure rate; Channel allocation is performed on edge nodes, and through the dynamic adjustment of the power allocation matrix, the channel utilization rate of the system is improved, and data transmission conflicts between edge nodes are reduced; Based on the task allocation strategy of the hybrid game model, the balanced allocation of resources among the three layers of edge nodes, regional nodes, and cloud computing centers is realized, avoiding resource competition and waste; Through the improved particle swarm optimization algorithm with a phased adaptive inertia weight and a random search strategy, the global search ability of the task allocation scheme is improved, effectively avoiding the problem of falling into local optimum. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0017] The structures, ratios, sizes, etc. shown in the drawings of this specification are only used to cooperate with the content disclosed in the specification for those who are familiar with this technology to understand and read, and are not used to limit the limited conditions under which the present invention can be implemented. Therefore, they do not have a substantial technical meaning. Any modification of the structure, change of the ratio relationship, or adjustment of the size, without affecting the effects that the present invention can produce and the purposes that can be achieved, should still fall within the scope that the technical content disclosed by the present invention can cover.

[0018] Figure 1 It is a schematic flowchart of a spatio-temporal data distributed real-time processing method provided by an embodiment of the present invention;

[0019] Figure 2 It is a schematic block diagram of the structure of the spatio-temporal data distributed real-time processing system provided by the embodiments of the present invention. Detailed implementation manners

[0020] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0021] The flowcharts shown in the accompanying drawings are only illustrative examples, and do not necessarily include all contents and operations / steps, nor do they necessarily need to be executed in the described order. For example, some operations / steps can also be decomposed, combined or partially merged, so the actual execution order may be changed according to the actual situation.

[0022] It should also be understood that the terms used in the specification of this application are only for the purpose of describing specific embodiments and are not intended to limit this application. As used in the specification of this application and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include the plural forms.

[0023] It should be further understood that the term "and / or" used in the specification of this application and the appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations. Please refer to Figure 1 , an embodiment of the spatio-temporal data distributed real-time processing method in the embodiments of this application includes:

[0024] Step 100: Extract features from the data source nodes in the smart city spatio-temporal big data AI platform to obtain a set of data source node feature vectors;

[0025] It can be understood that the execution subject of this application can be a spatio-temporal data distributed real-time processing system, or a terminal or a server, and specific limitations are not made here. In the embodiments of this application, the server is taken as an example of the execution subject for illustration.

[0026] Specifically, collect the geographical location information of each data source node distributed in the spatio-temporal big data AI platform of the smart city to obtain the longitude and latitude coordinate information of each data source node. Through a positioning system (such as GPS) or a geographic information service (such as a GIS platform), provide the longitude and latitude coordinate information of each data source node in real time with high precision. Convert the longitude and latitude coordinates into plane rectangular coordinates. Through a geographic coordinate conversion algorithm, for example, using the Gauss-Krüger projection or the UTM (Universal Transverse Mercator) projection method, map the longitude and latitude coordinates on the sphere to the plane rectangular coordinate system to obtain the position representation of each data source node on the two-dimensional plane. Collect the data traffic characteristics of the data source nodes. Obtain the data traffic rate by monitoring the network activities of the data source nodes in real time. This rate is expressed as the number of bytes transmitted per unit time of data transmission, such as Mbps or Gbps. Through the collection of the rate, reflect the load status of the data source nodes in the time dimension. Collect and classify the types of data traffic. The data traffic type describes the purpose or content attribute of the data stream, such as whether it is a video stream, an audio stream, a text data stream, or an image data stream. Parse the protocol layer of the data packet or through a predefined classification algorithm, for example, effectively identify different data stream types through deep packet inspection technology. At the same time, evaluate the data traffic level of each type. The data traffic level represents the priority or importance of this traffic. For example, the data traffic level is obtained through comprehensive evaluation of factors such as the timeliness of the task, the requirement for real-time performance, and the contribution to the overall utility of the system. Construct a data source node feature vector set with the data traffic rate, data traffic type, and data traffic level. The feature vector of each data source node is jointly composed of its geographical location feature and plane rectangular coordinates, and data traffic features (including data traffic rate, data traffic type, and data traffic level).

[0027] Step 200: Perform channel allocation on the edge nodes in the data source node feature vector set to obtain an edge node power allocation matrix and a channel allocation indication matrix;

[0028] Specifically, measure the channel state corresponding to the edge nodes in the data source node feature vector set. Use channel sounding technology to characterize the quality characteristics of the channel by obtaining parameters such as the signal-to-noise ratio and channel gain of the signal. Through statistical analysis of the channel quality data, quantify the channel conditions corresponding to each edge node. Sort the channel quality data in descending order to obtain the node priority sequence. Perform channel detection on the edge nodes in the node priority sequence to obtain the channel noise and the bandwidth distribution of the system. The channel noise is obtained by measuring the background interference level in the signal, while the system bandwidth is the data capacity that the channel can carry, which is jointly determined by the spectral width and the communication protocol. According to the channel detection results, calculate the target data rate of the nodes. The target data rate refers to the transmission rate that the edge node expects to achieve under the current channel conditions and is estimated by the Shannon formula. The Shannon formula is , where is the data rate,[[]] is the system bandwidth, and SNR is the signal-to-noise ratio. Through this formula, determine the theoretical maximum transmission rate for each edge node. At the same time, considering the power constraints in the actual system, calculate the power upper limit for each node based on the target data rate. The power upper limit ensures that the node transmission power does not exceed the physical limits of the hardware and the system by setting the maximum allowable power output range. After determining the power upper limit, combine the target data rate, system bandwidth, channel noise, and channel quality data to calculate the power of the edge nodes. Optimize the power allocation strategy to ensure the maximization of resource utilization and the minimization of system interference. For example, construct an optimization problem based on a utility function, where the utility function is defined as the sum of the channel transmission rates minus the cost of interference and power loss, and then solve this problem through the Lagrange multiplier method or convex optimization techniques to obtain the specific power value of each node. After power calculation, allocate the power values of each node to form an initial power matrix. Each element of the initial power matrix represents the power allocation of a certain edge node on a certain channel. On this basis, to reduce the interference inside the system, suppress the interference of the initial power matrix. Through an adaptive interference control algorithm, for example, by adjusting the transmit power of the node or switching channels to avoid co-channel interference, generate the power allocation matrix of the edge nodes. At the same time, mark the channel resources of the system to form an initial channel allocation matrix. The initial channel allocation matrix is a binary matrix, where each element represents whether a certain channel has been allocated. According to the power allocation matrix of the edge nodes, update the initial channel allocation matrix to ensure the coordination of channel resource allocation and power allocation. Through the channel occupancy update algorithm, check the current occupancy status of each channel one by one, and allocate the edge node with a higher priority to each channel according to the power allocation matrix to generate the final channel allocation indication matrix.

[0029] Step 300: Perform utility calculation and task iterative solution according to the edge node power allocation matrix and the channel allocation indication matrix to obtain the task allocation policy matrix;

[0030] It should be noted that the execution time of edge node tasks is calculated to obtain the time benefit. The time benefit is evaluated based on the timeliness of task completion and the corresponding time window. For example, for tasks with high real-time requirements, if they can be completed within the specified time, the benefit is high, while delayed completion will lead to a decrease in the benefit. At the same time, the energy consumption is calculated. The energy consumption is directly related to the power allocation of the edge node and is estimated by the product of power and task processing time. The lower the energy consumption, the higher the operating efficiency of the system. The usage amount of computing resources is counted, and the resource consumption is quantified by the processing capacity allocated by the edge node to the task (such as the usage ratio of CPU or GPU). The higher the resource usage amount, the greater the consumption. Based on these three factors of time benefit, energy consumption, and resource consumption, an edge layer utility function is constructed. The utility function is expressed in the form of benefit minus consumption, reflecting the comprehensive effectiveness of the overall task processing in the edge layer. Analyze and construct the task processing utility of the regional layer. Calculate the task processing duration of the regional node to obtain the processing benefit, which is measured by the efficiency of the node in completing tasks. For example, it is determined by the number of tasks completed per unit time. At the same time, calculate the service metrics. The service benefit is related to the user experience or service quality. For example, the reduction of latency and the timely response of tasks will increase the service benefit. Calculate the cost of resource allocation for the regional node. The allocation consumption mainly measures the cost incurred by the regional node when allocating communication resources, computing resources, and other related system resources. By comprehensively processing the benefit, service benefit, and allocation consumption, a regional layer utility function is constructed, which reflects the global processing ability of the regional node and its resource usage efficiency. Model and calculate the utility of the cloud layer. The key to the cloud layer is to process a large amount of task data that requires further calculation and provide the global scheduling ability of resources. In the cloud layer, calculate the processing latency of tasks. The latency benefit is evaluated by the relationship between the response time of task completion and the latency tolerance limit of the system. The lower the latency, the higher the benefit. At the same time, calculate the utilization status of cloud resources. The utilization benefit reflects the efficient use of cloud resources, such as the utilization rate of resources such as CPU and storage. A higher resource utilization rate means better benefit. Calculate the cloud operation cost, including energy cost, equipment depreciation cost, and other operation and maintenance costs. The lower the operation consumption, the higher the overall utility of the system. Based on the latency benefit, utilization benefit, and operation consumption, construct a cloud layer utility function. Input the edge layer utility function, regional layer utility function, and cloud utility function into the game solver to obtain the equilibrium solution of the system. The game solver is based on the theory of Nash equilibrium and, through the method of iterative optimization, searches for the optimal task allocation strategy between different layers to maximize the global utility. After obtaining the equilibrium solution, calculate the task ratio of the solution, and determine the allocation ratio of each task in the edge, regional, and cloud layers according to the utility values in the equilibrium solution. For example, allocate the specific processing amount of task data through the weight values of the solution to generate a task allocation strategy matrix.

[0031] Resource scanning is performed on various data processing nodes in the smart city spatio-temporal big data AI platform to obtain key information such as the computing capacity, available computing resources, available storage space, and processing type of the nodes. The computing capacity of the nodes is quantified by its hardware configuration, such as parameters like the number of CPU cores, main frequency, and GPU computing power; the available computing resources is a dynamic metric, representing the remaining allocable resources of the current node after running tasks, and its value changes over time; the available storage space mainly reflects the storage capacity of the node, including the unoccupied disk or memory capacity; the processing type refers to the types of computing tasks supported by the node, for example, whether it supports image processing, video stream analysis, or large-scale data aggregation. By standardizing the above information, a vector set containing the characteristics of all processing nodes is constructed. Based on the vector set of processing node characteristics, the planar rectangular coordinates of the data source node are used to calculate the distance, in order to evaluate the physical position relationship between the nodes. Using the planar rectangular coordinates of the data source node and the planar rectangular coordinates of the processing nodes, the Euclidean distance between the two is calculated through geometric formulas. The physical distances between all node pairs are constructed into a node distance matrix, and each element of the matrix represents the physical distance between a pair of nodes. The communication quality between the nodes is statistically analyzed to evaluate the stability and reliability of the link. The communication quality is quantified by statistically analyzing the number of successful communications and the total number of communications between each pair of nodes, where the number of successful communications represents the number of transmissions successfully completed between the node pair, and the total number of communications refers to all attempted transmissions, including successful and failed communications. By calculating the ratio of the number of successful communications to the total number of communications, a quality metric reflecting the link reliability is obtained. This metric is used to construct a link quality matrix, where each element represents the reliability level of the link between the node pair, and the value range is usually between 0 and 1, and the higher the value, the more stable and reliable the link. At the same time, the cooperation history between the nodes is statistically analyzed to evaluate the cooperation relationship and cooperation frequency between the nodes. By recording the historical cooperation times between each pair of nodes, the cooperation times reflect the past task allocation and processing interaction between the node pair. By accumulating the historical cooperation data, a node cooperation matrix is constructed, where each element represents the cumulative cooperation times between the node pair. The cooperation relationship matrix can reveal the long-term cooperation patterns between the nodes, which helps to preferentially utilize the node pairs with existing cooperation bases in subsequent task allocation, thereby improving the cooperation efficiency and reducing potential communication overhead.

[0032] Step 400: Calculate the target score of each task in the task allocation strategy matrix, and select the optimal data processing link based on the target score.

[0033] Specifically, for each task in the task assignment strategy matrix, extract its corresponding execution time to obtain the task latency value. Latency is the time it takes for a task to be assigned and finally completed, and it is an important indicator for evaluating task execution efficiency. Standardize the latency value. Use the linear normalization method to map the latency value to a fixed interval (such as 0 to 1) to obtain the standard latency index. Extract the energy consumption of the task to obtain the energy consumption value. The energy consumption value reflects the energy resources consumed during task execution. Normalize the energy consumption value through standardization to generate the standard energy consumption index. At the same time, extract the execution quality of each task to obtain the quality value. The execution quality is determined by factors such as the completion accuracy of the task, the reliability of the result, or the accuracy of the response. Standardize the quality value to generate the standard quality index. Calculate the standard latency index, the standard energy consumption index, and the standard quality index comprehensively to generate the target score. Use the weighted sum method to reflect the relative importance of each index by assigning weight coefficients to each index. Through this comprehensive score, quantitatively evaluate the performance of each task in resource utilization and execution effect. Based on the target score, update the node cooperation matrix to reflect the dynamic cooperation relationship during the task assignment process. The updated node cooperation matrix can more accurately describe the real-time cooperation state between nodes. At the same time, combine the link quality matrix and the node distance matrix to comprehensively evaluate the data transmission conditions and physical location relationships between nodes. The link quality matrix represents the reliability of communication between node pairs, while the node distance matrix reflects the physical proximity between node pairs. Assign weights to the node relationship matrix, the link quality matrix, and the node distance matrix respectively to form a group of weight coefficients. The weight assignment is adjusted based on the importance of different indicators. For example, in a latency-sensitive scenario, assign a higher weight to the node distance matrix, while in a high-reliability requirement scenario, give priority to the weight of the link quality matrix. Based on the group of weight coefficients, perform a weighted combination of the three matrices to generate the total link weight. By comparing all possible values of the total link weight, select the link with the highest weight value as the optimal data processing link. This link can achieve the best balance among latency, energy consumption, and quality, and combine the cooperation relationship, communication quality, and physical location between nodes to ensure the overall optimal efficiency of task assignment and execution.

[0034] Encode the optimal data processing link, and represent the selection status of the link in the form of a binary matrix. Each matrix element is used to indicate whether the link is selected, where 1 means selected and 0 means not selected. This encoding method can abstract the complex link selection problem and provide clear initial conditions for the particle swarm algorithm, including the initial position and initial velocity of each particle. The initial position directly corresponds to the encoded allocation scheme, while the initial velocity is set by introducing random perturbations within a small range to ensure the initial diversity of the particle swarm. Divide the optimization process of the particle swarm algorithm into different stages to adapt to the diverse requirements of the task allocation problem. The optimization is divided into a global stage and a local stage, which are optimized for exploring the potential solution space and fine-tuning the current solution respectively. The basis for stage division is the threshold of the number of iterations. By setting different optimization strategies, the global stage is more inclined to expand the search range to find more possible solutions, while the local stage is more focused on adjusting details to improve the quality of the solution. Dynamic weights play a key role in this process, and their values are dynamically adjusted according to the current optimization stage, so as to balance the relationship between exploration and exploitation. Larger weights in the initial stage help particles conduct extensive searches, while smaller weights in the later stage help particles focus on the vicinity of the optimal solution. In each round of iteration of the optimization, record the historical states of the particles to determine the individual optimal solution and the global optimal solution. The individual optimal solution refers to the state in which each particle performs optimally among its historical positions, while the global optimal solution is the best state among all particles. These optimal solutions are dynamically updated, and in each iteration, it is judged whether to update by comparing the fitness of the current state with the historical state. Fitness is used to measure the quality of the task allocation scheme, and a higher fitness indicates that the allocation scheme is more in line with the optimization goal. Based on the dynamic weights, individual optimal solutions, and global optimal solutions, update the velocity and position of the particles. Introduce a random perturbation mechanism in the updated position to enhance the global search ability of the algorithm. Random perturbation is achieved by adding small random changes to the positions of the particles. This change can break the local optimal state that the algorithm may fall into, while ensuring the diversity of solutions and the robustness of the algorithm. Through this mechanism, particles explore new solution spaces, thereby increasing the possibility of finding the global optimal solution. Decode the updated position into a specific task allocation scheme and evaluate it through a fitness function. The evaluation of fitness comprehensively considers multiple indicators such as delay, energy consumption, and execution quality, and weights these indicators through preset weights to obtain the comprehensive quality value of the allocation scheme. The evaluated fitness is used to adjust the state of the particles and guide the next optimization direction. During the whole process, judge the convergence of the fitness of the allocation scheme. If the fitness value tends to be stable in several iterations or reaches the set optimization goal, it is considered that the task allocation scheme has been optimized. At this time, the algorithm terminates and outputs the current optimal scheme.

[0035] In the embodiments of the present application, by introducing a node affinity evaluation mechanism and combining with the link reliability probability matrix, the cooperation effect between nodes can be accurately reflected, effectively reducing the task processing failure rate; channel allocation is performed on edge nodes, and through the dynamic adjustment of the power allocation matrix, the channel utilization rate of the system is improved, and data transmission conflicts among edge nodes are reduced; based on the task allocation strategy of the hybrid game model, the balanced allocation of resources among the three layers of edge nodes, regional nodes, and cloud computing centers is realized, avoiding resource competition and waste; through the improved particle swarm optimization algorithm with a phased adaptive inertia weight and a random search strategy, the global search ability of the task allocation scheme is improved, effectively avoiding the problem of falling into local optima.

[0036] In a specific embodiment, the process of executing step 100 may specifically include the following steps:

[0037] Collect the geographical locations of data source nodes in the smart city spatio-temporal big data AI platform to obtain the longitude and latitude coordinates of the data source nodes, and convert the longitude and latitude coordinates into the plane rectangular coordinates of the data source nodes;

[0038] Collect the data traffic of the data source nodes to obtain the data traffic rate, data traffic type, and data traffic level, and construct the data traffic rate, data traffic type, and data traffic level into a data source node feature vector set.

[0039] Specifically, collect the geographical location information of the data source nodes. The geographical location of each data source node is represented in the form of longitude and latitude, where the longitude (Longitude, λ) and latitude (Latitude, ) respectively describe the position of the node in the east-west direction and the north-south direction. The longitude and latitude data are obtained in real time through a Global Positioning System (GPS) device or a Geographic Information Service (GIS). Convert the longitude and latitude coordinates into plane rectangular coordinates. The conversion from longitude and latitude to plane rectangular coordinates uses the Gauss-Kruger projection, based on the formula:

[0040] ;

[0041] ;

[0042] Among them, is the plane rectangular coordinate; is the radius of the prime vertical; is the semi-major axis of the earth; is the first eccentricity of the earth, , where is the semi-minor axis of the earth; is the reference meridian and the longitude offset. At the same time, the data traffic characteristics of each data source node are collected. The characteristics of data traffic include rate, type, and level. The data traffic rate (Rate, ) describes the amount of traffic passing through a node per unit time, and the calculation formula is:

[0043] ;

[0044] where is the total amount of data transmitted; is the transmission time. The data traffic type (Type, ) is used to describe the content attribute of the data transmitted by a node. For example, represents a video stream, represents an image stream, represents text data, etc. These types are identified through packet analysis techniques (such as deep packet inspection, DPI). For example, if the traffic of a node is mainly a video stream, it is marked as . The data traffic level (Level, ) is used to represent the priority of data traffic, which is determined by system settings or task requirements. The level takes discrete numerical values. For example, represents high priority, represents low priority. The determination of the level usually considers factors such as real-time performance and importance. For example, the traffic priority of an emergency monitoring video is higher than that of ordinary network log data. By collecting the rate, type, and level of data traffic, a feature vector set of data source nodes is constructed. Suppose the feature vector of a certain node is:

[0045] ;

[0046] where is the plane rectangular coordinate, is the traffic rate, is the encoded value of the traffic type, is the traffic level.

[0047] In a specific embodiment, the process of executing step 200 may specifically include the following steps:

[0048] Measure the channel state of the edge nodes in the data source node feature vector set to obtain channel quality data, and perform a descending order sorting on the channel quality data to obtain a node priority sequence;

[0049] Perform channel detection on the edge nodes in the node priority sequence to obtain channel noise and system bandwidth;

[0050] Calculate the data rate for the edge nodes in the node priority sequence to obtain the target data rate, and calculate the power upper limit based on the target data rate;

[0051] Perform power calculation on the edge nodes according to the power upper limit, target data rate, system bandwidth, channel noise, and channel quality data to obtain the node power value;

[0052] Allocate the node power value to obtain the initial power matrix, and perform interference suppression on the initial power matrix to obtain the edge node power allocation matrix;

[0053] Mark the system channel resources to obtain the initial channel allocation matrix, and update the channel occupancy according to the edge node power allocation matrix to obtain the channel allocation indication matrix.

[0054] Specifically, measure the channel state of the edge nodes. The channel state is evaluated by the signal strength and interference information received by the nodes. The channel quality data is represented by the signal-to-noise ratio (SNR), and its calculation formula is:

[0055] ;

[0056] where is the signal-to-noise ratio of node ; is the signal power received by node ; is the noise power received by node . By measuring and calculating the of all edge nodes, a set of channel quality data is obtained. In order to preferentially allocate resources to nodes with better channel quality, sort the in descending order to generate a node priority sequence. The nodes ranked higher in this sequence have better channel quality and will be preferentially considered in resource allocation. Perform channel detection on the edge nodes in the priority sequence to obtain key parameters such as channel noise and system bandwidth. Channel noise is determined by measuring the background interference level in the channel. System bandwidth reflects the spectral range supported by the channel. Through these parameters, estimate the target data rate of each node. The target data rate is the maximum transmission rate that the edge node can achieve under the current channel conditions, and its calculation formula is:

[0057] ;

[0058] where is the target data rate of node ; is the system bandwidth; is the node 's signal-to-noise ratio. Calculate the power upper limit according to the target data rate. The power upper limit is the node 's maximum allowable transmission power under the current channel conditions, which is achieved by constraining the total system power. Assume that the total power limit of the system is , then the power upper limit is allocated by the following formula:

[0059] ;

[0060] where is the power upper limit of the node ; is the signal-to-noise ratio of the node ; is the set of all edge nodes; is the total power limit of the system. Based on the power upper limit, target data rate, system bandwidth, channel noise, and channel quality data, calculate the power of the edge nodes. Solve the optimal power value of the node through an optimization problem. The optimization goal is to maximize the system utility (such as the total transmission rate) and minimize the interference. The power allocation formula is expressed as:

[0061] ;

[0062] where is the actual allocated power of the node ; is the target data rate of the node ; is the channel noise power; is the system bandwidth. Organize the power values of all nodes into an initial power matrix , where each row represents the power allocation of a node on different channels. To reduce the internal interference of the system, perform interference suppression on the initial power matrix, such as by adjusting the power values or reallocating channels, to generate the power allocation matrix of the edge nodes. At the same time, mark the channel resources of the system to generate the initial channel allocation matrix . The element of this matrix indicates whether the channel is allocated to the node . 1 means allocated, 0 means not allocated. Based on the power allocation matrix, update the initial channel allocation matrix by checking the occupancy of each channel to generate the channel allocation indication matrix . For example, if the power value of the node on the channel is non-zero, then mark

[0063] In a specific embodiment, the process of executing step 300 may specifically include the following steps:

[0064] Calculate the task execution time to obtain the time benefit, calculate the energy consumption to obtain the energy cost, and at the same time, count the amount of computing resources used to obtain the resource cost, and construct an edge layer utility function based on the time benefit, energy cost, and resource cost;

[0065] Calculate the processing duration of the regional node tasks to obtain the processing benefit, calculate the service metrics to obtain the service benefit, and at the same time, calculate the cost of resource allocation for the regional nodes to obtain the allocation cost, and construct a regional layer utility function based on the processing benefit, service benefit, and allocation cost;

[0066] Calculate the cloud processing delay to obtain the delay benefit, calculate the resource utilization status to obtain the utilization benefit, and at the same time, calculate the cloud operation cost to obtain the operation cost, and construct a cloud utility function based on the delay benefit, utilization benefit, and operation cost;

[0067] Input the edge layer utility function, regional layer utility function, and cloud utility function into the game solver to obtain an equilibrium solution, and calculate the task proportion of the equilibrium solution to obtain a task allocation strategy matrix.

[0068] Specifically, perform quantitative calculations on the task execution time, energy consumption, and amount of computing resources used, and construct an edge layer utility function based on this. On the edge node, the execution time of each task is determined by the ratio between the amount of computing required for the task and the computing power of the node. The formula is:

[0069] ;

[0070] where represents the computing requirement of the task, with the unit of CPU cycles; represents the computing power of the edge node. Based on the execution time, define the time benefit , whose value increases as the execution time decreases. The typical form is:

[0071] ;

[0072] where is the weight factor of the time benefit, reflecting the requirement of the task for timeliness. At the same time, calculate the energy consumption of the task on the edge node. Its calculation formula is:

[0073] ;

[0074] where is the energy consumption coefficient; is the execution time of the task. Statistically calculate the usage of computing resources , for example, the resource usage is directly equal to the computing requirements of the task . Based on the time benefit, energy overhead, and resource usage, the edge layer utility function is defined as:

[0075] ;

[0076] Among them, and are the weight factors of energy and resource overhead respectively. On the regional node, calculate the task processing duration, service benefit, and resource allocation overhead. The task processing duration is the time required for the task to be processed from arriving at the regional node to completion, and the formula is:

[0077] ;

[0078] Among them represents the queuing time of the task at the regional node. The service benefit reflects the contribution of task completion to the overall service quality, and its value is set according to the type and priority of the task. For example, urgent tasks have a higher service benefit:

[0079] ;

[0080] Among them is the weight factor of the service benefit. The resource allocation overhead measures the cost of the regional node allocating computing, storage, and network resources for the task. For example, set the cost per unit of resource as a constant , then:

[0081] ;

[0082] The regional layer utility function combines these metrics and is expressed as:

[0083] ;

[0084] Among them and are the weight factors. At the cloud layer, calculate the processing delay, resource utilization benefit, and operating cost. The processing delay is the completion time of the task in the cloud, including the upload time , the processing time and the download time in total:

[0085] ;

[0086] Delayed revenue Increases as the delay decreases, expressed as:

[0087] ;

[0088] Where is the weight of the delayed revenue. Resource utilization revenue Measures the efficient use of cloud computing and storage resources, defined as a linear function of the utilization rate:

[0089] ;

[0090] Where is the proportionality coefficient of the utilization revenue. Operating cost Is the overhead of running tasks in the cloud, including energy consumption and resource usage costs, calculated as:

[0091] ;

[0092] Where is the operating cost per unit time. The cloud utility function is defined as:

[0093] ;

[0094] Input the utility functions of the above-mentioned edge layer, regional layer, and cloud layer into the game solver. According to the interest conflicts and coordination requirements of different layers, solve the equilibrium solution of task allocation. The solver calculates the equilibrium point through iterative optimization and outputs the proportion of task allocation for each layer. Based on the equilibrium solution, the task proportion is adjusted according to the weight allocation of the utility functions of each layer, thereby generating the final task allocation strategy matrix.

[0095] In a specific embodiment, the method for executing spatio-temporal data distributed real-time processing further includes the following steps:

[0096] Perform resource scanning on the data processing nodes in the smart city spatio-temporal big data AI platform to obtain the node computing capacity, available computing resources, available storage space, and processing type, and construct the node computing capacity, available computing resources, available storage space, and processing type into a processing node feature vector set;

[0097] Based on the processing node feature vector set, calculate the physical distance between the plane rectangular coordinates of the data source nodes, and obtain the physical distance between node pairs, and construct the physical distance between node pairs into a node distance matrix;

[0098] Statistically analyze the communication quality between nodes to obtain the number of successful communications and the total number of communications, and construct the ratio of the number of successful communications to the total number of communications into a link quality matrix;

[0099] Statistically analyze the collaboration history between nodes to obtain the number of historical collaborations, and construct the number of historical collaborations into a node collaboration matrix.

[0100] Specifically, perform a resource scan on the data processing nodes to obtain core performance parameters, including the computing capacity of the nodes , the available computing resource volume , the available storage space and the processing types supported by the nodes . The computing capacity of a node represents the maximum computing task volume that it can theoretically handle. Its calculation method is:

[0101] ;

[0102] where the number of cores is the physical core number of the node's CPU, and the main frequency is the processing speed of each core. The available computing resource volume represents the computing resources that are currently not occupied by the node, and the calculation formula is:

[0103] ;

[0104] where is the computing resource volume that has been allocated to the node currently. The available storage space represents the remaining storage capacity of the node, and its calculation formula is:

[0105] ;

[0106] where and are the total storage capacity and the used storage volume of the node respectively. The processing type is a discrete variable used to identify the task categories supported by the node. For example represents image processing, represents video analysis, etc. Through the above parameters, construct a feature vector for each node:

[0107] ;

[0108] Calculate the physical distance between nodes. The position of each node is represented by a plane rectangular coordinate , and the physical distance between node pairs is calculated through the Euclidean distance formula:

[0109] ;

[0110] where and are the plane rectangular coordinates of the two nodes respectively. By calculating the distances of all node pairs, construct a node distance matrix , where the matrix element represents the node and the node is the physical distance between them. The communication quality between nodes is statistically analyzed. The communication quality is calculated by the ratio of the number of successful communications and the total number of communications as follows:

[0111] ;

[0112] where is the number of communications successfully completed between node pairs, is the total number of communication attempts. By statistically analyzing the communication quality of all node pairs, a link quality matrix is constructed, where the matrix element represents the reliability of the communication link between the node and the node . The cooperation history between nodes is statistically analyzed. The cooperation history is the total number of past cooperations between node pairs, obtained by accumulating task assignment records. The cooperation history constructs a node cooperation matrix in the following way, where the matrix element represents the historical cooperation times between the node and the node . Through the above steps, the following several key matrices are formed: a set of node feature vectors: reflecting the computing power, storage resources, and task adaptability of each node; a node distance matrix describing the physical proximity between nodes; a link quality matrix : measuring the reliability of the communication link between nodes. A node cooperation matrix : recording the historical cooperation relationship of node pairs.

[0113] In a specific embodiment, the process of executing step 400 may specifically include the following steps:

[0114] Extract the execution time of each task in the task assignment policy matrix to obtain a delay value, and standardize the delay value to obtain a standard delay index;

[0115] Extract the energy consumption of each task in the task assignment policy matrix to obtain an energy consumption value, and standardize the energy consumption value to obtain a standard energy consumption index;

[0116] Extract the execution quality of each task in the task assignment policy matrix to obtain a quality value, and standardize the quality value to obtain a standard quality index;

[0117] Perform comprehensive calculations on the standard delay index, standard energy consumption index, and standard quality index to obtain the target score, and update the node cooperation matrix according to the target score to obtain the node relationship matrix;

[0118] Perform weight allocation on the node relationship matrix, link quality matrix, and node distance matrix respectively to obtain a weight coefficient group;

[0119] Weight the node relationship matrix, link quality matrix, and node distance matrix according to the weight coefficient group to obtain the total link weight, and compare and select the total link weight to obtain the optimal data processing link.

[0120] Specifically, extract the execution time of each task from the task allocation strategy matrix and calculate the corresponding delay value. The execution time of the task represents the total time consumed from the start of task allocation to completion, including two parts: communication time and processing time, and is calculated using the following formula:

[0121] ;

[0122] Among them, is the data transmission time of the task, which depends on the data size and the link transmission rate :

[0123] ;

[0124] is the processing time of the task, which depends on the task calculation requirements and the computing power of the processing node :

[0125] ;

[0126] After extraction, perform standardization processing on the delay values of all tasks to convert the delays of different tasks into a unified dimension. The standardization formula is:

[0127] ;

[0128] Among them, is the standardized delay value; and are respectively the minimum and maximum values of the delays among all tasks. Extract the energy consumption of each task from the task allocation strategy matrix. The energy consumption value of the task is the total energy consumed during task execution and is calculated using the following formula:

[0129] ;

[0130] Among them is the power consumption of the processing node during task execution. Similar to the delay, the energy consumption value is normalized, and the normalization formula is:

[0131] ;

[0132] where is the normalized energy consumption value; and are the minimum and maximum energy consumptions among all tasks respectively. The execution quality of each task is extracted from the task assignment strategy matrix. The execution quality is defined by the accuracy, reliability, or task requirement satisfaction of task completion and is represented in the form of a score. The quality value is also normalized, and the formula is:

[0133] ;

[0134] where is the normalized quality value; and are the minimum and maximum qualities among all tasks respectively. After normalizing the delay, energy consumption, and quality, the target score is calculated through weighted synthesis, and the formula is:

[0135] ;

[0136] where is the target score; are the weights of the delay, energy consumption, and quality respectively, satisfying . Based on the target score of each task, the node cooperation matrix is updated. Each element of the node cooperation matrix represents the cooperation degree between node and node . After the update, the cooperation matrix reflects the impact of task completion quality on node relationships. Assuming that tasks with higher scores have a greater positive impact on node cooperation relationships, the cooperation value is updated using the following formula:

[0137] ;

[0138] where is the increment of the task score on the node relationship. Combining the node relationship matrix , the link quality matrix and the node distance matrix for weighted calculation to obtain the total link weight. Weights are assigned to each matrix, and the weight coefficient group satisfies . The calculation formula for the total link weight is:

[0139] ;

[0140] wherein, is the total weight of the link and ; is the node cooperation value; is the link quality; is the reciprocal of the node distance, which is used to encourage the selection of links with a relatively short physical distance. By calculating the total weight of all links, the advantages and disadvantages of different links are compared. The link with the highest weight is selected as the optimal data processing link.

[0141] In a specific embodiment, the spatio-temporal data distributed real-time processing method further includes the following steps:

[0142] Encode the optimal data processing link to obtain a binary matrix, and generate the initial position and initial velocity of the particles of the particle swarm algorithm according to the binary matrix;

[0143] Divide the optimization process of the particle swarm algorithm into stages to obtain the iteration thresholds of the global stage and the local stage, and calculate the dynamic weight according to the iteration thresholds;

[0144] Record the historical state of the particles of the particle swarm algorithm to obtain the individual optimal solution, and update the population state to obtain the global optimal solution;

[0145] Calculate the velocity according to the dynamic weight, the individual optimal solution and the global optimal solution to obtain the new particle velocity, and calculate the position according to the new particle velocity to obtain the new particle position;

[0146] Generate random perturbations to obtain random values, and perform position perturbations according to the random values to obtain the perturbed position vector;

[0147] Decode the perturbed position vector to obtain the allocation scheme, evaluate the allocation scheme to obtain the scheme fitness, and perform convergence judgment on the scheme fitness. When the termination condition is met, output the task allocation optimization scheme.

[0148] Specifically, encode the optimal data processing link and convert it into a binary matrix. The binary matrix is used to represent the selection state of the link, where the rows of the matrix represent different data source nodes and the columns represent available edge or cloud nodes. Each element of the matrix is represented by a binary value 0 or 1, where 1 means the link is selected and 0 means not selected. By encoding the optimal link, the initial position of the particles required for the particle swarm algorithm is generated , and the position of each particle is represented by a vector corresponding to the binary matrix. To introduce the dynamic characteristics of particle movement, generate the initial velocity of the particles , its initial value is set to a random number vector within a small range to simulate the initial perturbation in the search space. During the particle swarm optimization process, the iterations are divided into stages to balance global search and local search. The entire optimization process is divided into a global stage and a local stage, and the two stages are distinguished by setting an iteration threshold . In the global stage, the focus of optimization is to expand the search range and explore potential optimal solutions, so a larger weight coefficient is required for the random search of particles. In the local stage, the focus of optimization turns to the convergence and fine-tuning of the solution. At this time, a smaller weight is needed to reduce randomness and thus focus on the refinement of the current solution. The dynamic weight is a key parameter in particle swarm optimization, and its value is dynamically adjusted with the number of iterations , and the formula is:

[0149] ;

[0150] where, is the dynamic weight of the th iteration; and are the initial weight and the minimum weight respectively; is the maximum number of iterations. During the optimization process, the particle swarm algorithm records the historical states of each particle to update the individual best solution and the global best solution. The individual best solution refers to the position with the highest fitness found by the particle in its historical movement trajectory, and the global best solution is the position with the highest fitness among all particles. At each iteration, the current fitness of each particle is calculated and compared with its historical best fitness. If the current solution is better, the individual best solution is updated; if the fitness of this solution is better than the global best solution, the global best solution is updated. Based on the dynamic weight, the individual best solution and the global best solution, the velocity update formula of the particle is calculated:

[0151] ;

[0152] where, is the velocity of the particle in the th iteration; is the velocity of the particle in the th iteration; is the dynamic weight; and are learning factors, which control the influence of the individual and global solutions respectively; and are random numbers with a value range in [0, 1]; is the individual best solution of the particle; is the global optimal solution; is the particle at the th iteration. Calculate the next position of the particle according to the updated velocity:

[0153] ;

[0154] where is the new position of the particle . To enhance the global search ability of the algorithm, a random perturbation mechanism is introduced. By generating a random value and applying it to the position of the particle, the perturbed position vector is obtained:

[0155] ;

[0156] where is the random perturbation vector, which usually follows a uniform distribution or a normal distribution. Decode the perturbed position vector into a specific allocation scheme, that is, generate a task allocation strategy according to the decoding rules of the binary matrix. Evaluate the allocation scheme and calculate its fitness through the fitness function . The fitness function comprehensively considers the delay, energy consumption and resource utilization efficiency of the allocation scheme. The typical form is:

[0157] ;

[0158] where are weight factors, which respectively control the contributions of delay, energy consumption and execution quality to the fitness. Judge the convergence of the fitness of the scheme. When the change in fitness is lower than the set threshold in several iterations, or the number of iterations reaches the upper limit, the algorithm stops and outputs the allocation scheme with the highest current fitness as the task allocation optimization result. This scheme ensures a dynamic balance between global optimality and local adjustment, and realizes the optimal optimization of resource allocation and task processing.

[0159] In this embodiment, the method further includes performing adaptive event-triggered control on the task allocation optimization scheme, specifically including: performing real-time sampling on the computing load, storage state, and communication delay of the data processing nodes to obtain the node state vector xi(t), calculating the state error ei(t) between adjacent nodes according to the node state vector xi(t), and constructing the node trigger matrix E(t) according to the state error ei(t); performing error analysis on the synchronization states of the nodes in the node trigger matrix E(t) to obtain the synchronization deviation vector δi(t), constructing the Lyapunov function V(t) based on the synchronization deviation vector δi(t), and deriving the lag synchronization criterion S(t) according to the convergence of the Lyapunov function V(t); performing dynamic threshold calculation on the lag synchronization criterion S(t) to obtain the reference trigger threshold σi, adaptively adjusting the reference trigger threshold σi according to the node communication quality qi(t) to obtain the adaptive threshold function σi(t), and updating the event trigger condition of the node according to the adaptive threshold function σi(t); extracting the time interval τk between adjacent trigger events to obtain the trigger time series {tk}, calculating the minimum trigger interval τmin according to the trigger time series {tk}, and constructing the Zeno behavior determination condition Z(t); performing theoretical analysis on the Zeno behavior determination condition Z(t) to obtain the lower bound of the trigger interval , and according to the lower bound of the trigger interval prove that τmin > 0, thereby verifying the non-existence of Zeno behavior; analyzing the topological structure of the task processing network, extracting the edge connection relationship between nodes to obtain the edge connection matrix G(t), calculating the edge weight wij(t) according to the edge connection matrix G(t), and constructing the edge trigger controller C(t); calculating the optimal transmission timing T(t) according to the edge trigger controller C(t), and jointly optimizing the node trigger mechanism in combination with the adaptive threshold function σi(t) to obtain the hybrid trigger strategy H(t), generating the node trigger moment set Γ(t) based on the hybrid trigger strategy H(t); performing Lyapunov analysis on the stability of the system under the hybrid trigger strategy H(t), calculating the change rate of the system energy function E(t), and when the energy function E(t) is monotonically decreasing and bounded, outputting the final task allocation execution plan that satisfies real-time performance and synchronization.

[0160] The above describes the spatio-temporal data distributed real-time processing method in the embodiment of the present application. Next, the spatio-temporal data distributed real-time processing system 10 in the embodiment of the present application will be described. Please refer to Figure 2 , an embodiment of the spatio-temporal data distributed real-time processing system 10 in the embodiment of the present application includes:[[]]

[0161] A feature extraction module 11, configured to extract features from the data source nodes in the smart city spatio-temporal big data AI platform to obtain a data source node feature vector set;

[0162] The allocation module 12 is used to perform channel allocation on the edge nodes in the data source node feature vector set to obtain an edge node power allocation matrix and a channel allocation indication matrix;

[0163] The solution module 13 is used to perform utility calculation and task iterative solution according to the edge node power allocation matrix and the channel allocation indication matrix to obtain a task allocation strategy matrix;

[0164] The selection module 14 is used to calculate the target score of each task in the task allocation strategy matrix and select the optimal data processing link based on the target score.

[0165] Through the collaborative cooperation of the above-mentioned various components, by introducing a node affinity evaluation mechanism and combining with the link reliability probability matrix, it can accurately reflect the cooperation effect between nodes, effectively reducing the task processing failure rate; performing channel allocation on edge nodes and improving the channel utilization rate of the system and reducing data transmission conflicts among edge nodes through dynamic adjustment of the power allocation matrix; based on the task allocation strategy of the hybrid game model, achieving balanced allocation of resources among the three layers of edge nodes, regional nodes, and cloud computing centers, avoiding resource competition and waste; through the improved particle swarm algorithm with phased adaptive inertia weight and random search strategy, improving the global search ability of the task allocation scheme and effectively avoiding the problem of falling into local optimum.

[0166] Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working processes of the above-described system, system, and unit can refer to the corresponding processes in the foregoing method embodiments and will not be described herein again.

[0167] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing an electronic device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs that can store program codes.

[0168] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than limiting it; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A distributed real-time processing method for spatio-temporal data, characterized in that, The method includes: Performing feature extraction on data source nodes in the spatio-temporal big data AI platform of the smart city to obtain a set of data source node feature vectors; Performing channel allocation on edge nodes in the set of data source node feature vectors to obtain an edge node power allocation matrix and a channel allocation indication matrix; specifically including: measuring the channel state of edge nodes in the set of data source node feature vectors to obtain channel quality data, and performing a descending order sorting on the channel quality data to obtain a node priority sequence; performing channel detection on edge nodes in the node priority sequence to obtain channel noise and system bandwidth; calculating the target data rate for edge nodes in the node priority sequence to obtain the target data rate, and calculating the power upper limit according to the target data rate; calculating the node power value for edge nodes according to the power upper limit, the target data rate, the system bandwidth, the channel noise, and the channel quality data; allocating the node power value to obtain an initial power matrix, and performing interference suppression on the initial power matrix to obtain an edge node power allocation matrix; marking the system channel resources to obtain an initial channel allocation matrix, and performing channel occupancy update according to the edge node power allocation matrix to obtain a channel allocation indication matrix; Performing utility calculation and task iterative solution according to the edge node power allocation matrix and the channel allocation indication matrix to obtain a task allocation strategy matrix; Calculating the target score of each task in the task allocation strategy matrix, and selecting the optimal data processing link based on the target score.

2. The spatio-temporal data distributed real-time processing method according to claim 1, wherein The performing feature extraction on data source nodes in the spatio-temporal big data AI platform of the smart city to obtain a set of data source node feature vectors includes: Collecting the geographical location of data source nodes in the spatio-temporal big data AI platform of the smart city to obtain the longitude and latitude coordinates of the data source nodes, and converting the longitude and latitude coordinates into the plane rectangular coordinates of the data source nodes; Collecting the data traffic of data source nodes to obtain the data traffic rate, data traffic type, and data traffic level, and constructing a set of data source node feature vectors from the data traffic rate, data traffic type, and data traffic level.

3. The spatio-temporal data distributed real-time processing method according to claim 1, wherein The performing utility calculation and task iterative solution according to the edge node power allocation matrix and the channel allocation indication matrix to obtain a task allocation strategy matrix includes: Calculating the time benefit by calculating the task execution time, calculating the energy consumption by calculating the energy overhead, and at the same time, counting the amount of computing resources used to obtain the resource overhead, and constructing an edge layer utility function according to the time benefit, the energy consumption, and the resource overhead; Calculating the processing benefit by calculating the task processing duration of regional nodes, calculating the service benefit by calculating the service metrics, and at the same time, calculating the cost of resource allocation for regional nodes to obtain the allocation overhead, and constructing a regional layer utility function according to the processing benefit, the service benefit, and the allocation overhead; Calculate the cloud processing delay to obtain the delay benefit, calculate the resource utilization status to obtain the utilization benefit, and at the same time, calculate the cloud operation cost to obtain the operation overhead, and construct a cloud utility function according to the delay benefit, the utilization benefit and the operation overhead; Input the edge layer utility function, the regional layer utility function and the cloud utility function into a game solver to obtain an equilibrium solution, and calculate the task proportion of the equilibrium solution to obtain a task allocation strategy matrix.

4. The spatio-temporal data distributed real-time processing method according to claim 2, characterized in that The spatio-temporal data distributed real-time processing method further includes: Scan the resources of the data processing nodes in the smart city spatio-temporal big data AI platform to obtain the node computing capacity, available computing resources, available storage space and processing type, and construct the node computing capacity, the available computing resources, the available storage space and the processing type into a processing node feature vector set; Based on the processing node feature vector set, calculate the physical distance between the plane rectangular coordinates of the data source nodes to obtain the physical distance between node pairs, and construct the physical distance between node pairs into a node distance matrix; Statistically analyze the communication quality between nodes to obtain the number of successful communications and the total number of communications, and construct the ratio of the number of successful communications to the total number of communications into a link quality matrix; Statistically analyze the historical cooperation between nodes to obtain the number of historical cooperation times, and construct the number of historical cooperation times into a node cooperation matrix.

5. The spatio-temporal data distributed real-time processing method according to claim 4, characterized in that Calculating the target score of each task in the task allocation strategy matrix and selecting the optimal data processing link based on the target score includes: Extract the execution time of each task in the task allocation strategy matrix to obtain a delay value, and standardize the delay value to obtain a standard delay index; Extract the energy consumption of each task in the task allocation strategy matrix to obtain an energy consumption value, and standardize the energy consumption value to obtain a standard energy consumption index; Extract the execution quality of each task in the task allocation strategy matrix to obtain a quality value, and standardize the quality value to obtain a standard quality index; Comprehensively calculate the standard delay index, the standard energy consumption index and the standard quality index to obtain a target score, and update the node cooperation matrix according to the target score to obtain a node relationship matrix; Assign weights to the node relationship matrix, the link quality matrix and the node distance matrix respectively to obtain a weight coefficient group; Weight the node relationship matrix, the link quality matrix and the node distance matrix according to the weight coefficient group to obtain the total link weight, and compare and select the total link weight to obtain the optimal data processing link.

6. The spatio-temporal data distributed real-time processing method according to claim 5, characterized in that The spatio-temporal data distributed real-time processing method further includes: Encode the optimal data processing link to obtain a binary matrix, and generate the initial position and initial velocity of the particles of the particle swarm algorithm according to the binary matrix; Divide the optimization process of the particle swarm algorithm into stages to obtain the iteration thresholds of the global stage and the local stage, and calculate the dynamic weight according to the iteration thresholds; Record the particle historical states of the particle swarm algorithm to obtain the individual optimal solution, and update the swarm state to obtain the global optimal solution; Calculate the particle's new velocity according to the dynamic weight, the individual optimal solution and the global optimal solution, and calculate the particle's new position according to the particle's new velocity; Generate random perturbations to obtain random values, and perform position perturbations according to the random values to obtain a perturbed position vector; Decode the perturbed position vector to obtain an allocation scheme, evaluate the allocation scheme to obtain the fitness of the scheme, and judge the convergence of the fitness of the scheme. When the termination condition is met, output the optimized task allocation scheme.

7. A distributed real-time processing system for spatio-temporal data, characterized in that, For implementing the spatio-temporal data distributed real-time processing method as claimed in claim 1, the spatio-temporal data distributed real-time processing system comprises: A feature extraction module, configured to extract features from data source nodes in the spatio-temporal big data AI platform of the smart city to obtain a set of data source node feature vectors; An allocation module, configured to perform channel allocation on edge nodes in the set of data source node feature vectors to obtain an edge node power allocation matrix and a channel allocation indication matrix; A solution module, configured to perform utility calculation and task iterative solution according to the edge node power allocation matrix and the channel allocation indication matrix to obtain a task allocation strategy matrix; A selection module, configured to calculate the target score of each task in the task allocation strategy matrix, and select the optimal data processing link based on the target score.

Images