An artificial intelligence-based industrial internet service optimization method and system

Abstract

This invention discloses an artificial intelligence-based method and system for optimizing industrial internet services, comprising the following steps: generating an industrial internet service dataset; generating a spatiotemporal calibration state feature set; constructing a dynamic causal influence graph and performing redundant causal chain pruning to generate a pruned causal state representation result; performing trend prediction to generate a service operation trend prediction result; using an improved ChronoSpike network for joint optimization calculation to generate an industrial internet service optimization strategy; executing differentiated flexible control to generate industrial internet service optimization execution results; collecting actual service response latency and actual resource consumption data, updating the parameters of the improved ChronoSpike network, thereby achieving accurate prediction of industrial internet service status and collaborative resource optimization, improving service response efficiency, resource utilization, and the operational stability of the industrial internet system.

Description

Technical Field

[0001] This invention relates to the field of industrial internet and intelligent service optimization, and in particular to an industrial internet service optimization method and system based on artificial intelligence. Background Technology

[0002] With the rapid development of Industrial Internet technology, a large number of industrial devices, edge nodes, and business service systems are connected through networks, resulting in a continuous increase in equipment operation data, network communication data, service request data, and resource status data during industrial production. Industrial Internet platforms need to allocate service tasks, schedule computing resources, and optimize service response processes based on real-time collected data to ensure the continuous operation and service quality of industrial businesses. Currently, Industrial Internet service optimization mainly relies on data collection and analysis, rule-based scheduling algorithms, and traditional artificial intelligence prediction models. By analyzing equipment status, network load, and service request conditions, service resource configuration schemes are determined.

[0003] Existing industrial internet service optimization methods typically begin by collecting operational data from industrial equipment and network nodes, then use data analysis models to extract equipment status characteristics, predict changes in service demand based on historical operational patterns, and generate service scheduling schemes using task scheduling algorithms or resource optimization algorithms. However, when dealing with multi-node collaborative service scenarios in the industrial internet, existing technologies usually directly perform time-series analysis on data from different sources, without fully considering the data time offset issues caused by different data collection times, network transmission delays, and differences in edge node processing speeds between different service nodes. This results in inconsistent status representations of service status data acquired within the same time frame, leading to errors in subsequent service status analysis and optimization decisions. Summary of the Invention

[0004] One objective of this invention is to propose an industrial internet service optimization method and system based on artificial intelligence. This invention fully utilizes artificial intelligence, dynamic causal analysis, spatiotemporal state calibration, and improved ChronoSpike network technology, possessing advantages such as high service optimization accuracy, high resource utilization efficiency, and strong system operational stability.

[0005] An industrial internet service optimization method based on artificial intelligence according to an embodiment of the present invention includes the following steps:

[0006] Collect and preprocess multi-source operational data from service nodes and network edge nodes in the industrial internet system to generate an industrial internet service dataset.

[0007] Spatiotemporal misalignment detection is performed on the industrial internet service dataset to identify state representation deviations between service nodes caused by data transmission delays and processing time differences, and a spatiotemporal misalignment compensation factor is constructed to generate a spatiotemporal calibration state feature set.

[0008] Causal correlation analysis is performed on the spatiotemporal calibration state feature set to construct a dynamic causal influence graph between service states. Redundant causal chains are pruned from the dynamic causal influence graph to generate pruned causal state characterization results.

[0009] Predict the changing trends of key causal paths in the causal state characterization results after pruning, and generate service operation trend prediction results.

[0010] The service operation trend prediction results are input into the improved ChronoSpike network for joint optimization calculation to generate industrial Internet service optimization strategies.

[0011] Based on the industrial internet service optimization strategy, differentiated and flexible control is performed on the service task allocation process, resource scheduling process, and service response process in the industrial internet system to generate industrial internet service optimization execution results.

[0012] Collect data on actual service response latency and actual resource consumption after the service optimization execution results are obtained, and update the parameters of the improved ChronoSpike network.

[0013] Optionally, the multi-source operational data includes device operational data, network communication data, service request data, and resource status data, and the preprocessing includes data cleaning, data denoising, time synchronization, and format conversion.

[0014] Optionally, the generation of the spatiotemporal calibration state feature set specifically includes:

[0015] Perform node association processing on the industrial internet service dataset to obtain service node status characteristics and service node time sequence information;

[0016] Perform differential analysis on the timing information of service nodes to determine the differences in data transmission latency and processing time between different service nodes, and generate spatiotemporal misalignment quantities.

[0017] Based on the spatiotemporal misalignment, identify the state representation deviations between service nodes caused by data transmission delays and processing time differences, and construct a spatiotemporal misalignment compensation factor.

[0018] The service node state characteristics are compensated based on the spatiotemporal misalignment compensation factor to generate spatiotemporal calibration state characteristics.

[0019] The spatiotemporal calibration status features corresponding to each service node are aggregated to generate a spatiotemporal calibration status feature set.

[0020] Optionally, the generation of the causal state characterization result after pruning specifically includes:

[0021] The spatiotemporal calibration state feature set is processed into nodes to generate a service state node set;

[0022] Based on the time sequence information of each service status node in the service status node set, the sequential relationship of state changes between each service status node is identified, and a candidate causal chain set is generated.

[0023] Based on the set of candidate causal chains, calculate the causal association strength value corresponding to each candidate causal chain, and construct a dynamic causal influence graph;

[0024] Redundant causal chains are pruned from the dynamic causal influence graph to generate a causal state representation result after pruning.

[0025] Optionally, the generation of the service operation trend prediction results specifically includes:

[0026] The key causal paths in the causal state characterization results after pruning are sorted to generate a key causal path sequence;

[0027] Based on the service status nodes corresponding to each key causal path in the key causal path sequence, generate path status change quantities.

[0028] Path trend driving values ​​are generated based on path state changes and causal correlation strength values, and then accumulated to form service operation trend values.

[0029] Based on the service operation trend value, determine the direction and magnitude of change of the key causal path at the next time sequence position, and generate service operation trend prediction results.

[0030] Optionally, the generation of the industrial internet service optimization strategy specifically includes:

[0031] The service operation trend prediction results are input into the improved ChronoSpike network. The improved ChronoSpike network includes a trend phase locking module, a causal potential field projection module, a resource contention culling module, and a service strategy reconstruction module. The improvement of the improved ChronoSpike network is as follows: The traditional ChronoSpike network uses time pulse sequences to trigger pulses and accumulate states for the temporal changes of the input data, and outputs prediction results based on the pulse response intensity. The improved ChronoSpike network introduces the key causal paths, change directions, and change magnitudes in the service operation trend prediction results into the pulse state calculation process. It identifies the trend reversal points of key causal paths through phase gap locations, marks the continuous enhancement segments of key causal paths through phase locking objects, and introduces a service load lattice mapping mechanism and lattice folding processing. It transforms the load occupancy relationship of key causal paths in service nodes, temporal sequence positions, and change directions into resource contention culling results, generating industrial internet service optimization strategies.

[0032] In the trend phase locking module, phase breakpoints are identified in the key causal paths, change directions and change magnitudes in the service operation trend prediction results. Key causal paths where the change direction reverses are identified as phase gap locations, and key causal paths where the change magnitude continuously increases are identified as phase locking objects. Trend phase locking results are generated based on the phase gap locations and phase locking objects.

[0033] In the causal potential field projection module, a service load lattice mapping mechanism is introduced to map the key causal paths in the trend phase locking results to the service load lattice units according to the service node, time sequence position and change direction. The lattice occupancy intensity of the service load lattice unit is determined according to the change amplitude. Lattice folding processing is performed on the key causal paths that are in the same service load lattice unit and have opposite change directions to generate causal potential field projection results.

[0034] In the resource contention elimination module, the resource contention relationship of the critical causal path in the service request processing process, resource allocation process and service response process is determined based on the causal potential field projection result. Based on the resource contention relationship, the critical causal path is eliminated to generate the resource contention elimination result.

[0035] In the service strategy reconstruction module, the key causal paths to be retained are determined based on the resource contention elimination results. Then, the service request processing order, resource allocation ratio, and service response priority are determined based on the service nodes, time sequence positions, and lattice occupancy strength corresponding to the retained key causal paths, thereby generating an industrial internet service optimization strategy.

[0036] Optionally, the generation of the industrial internet service optimization execution result specifically includes:

[0037] Read the service request processing order, resource allocation ratio, and service response priority from the industrial internet service optimization strategy;

[0038] Based on the service request processing order, the service tasks corresponding to each service node in the industrial internet system are arranged to generate a service task allocation queue.

[0039] Based on the resource allocation ratio, the computing resources, storage resources, and network bandwidth resources in the service nodes and network edge nodes are configured to generate resource scheduling configurations;

[0040] Based on the service response priority, determine the response time limit, retry count and transmission priority of each service node, and generate service response control parameters;

[0041] The service task allocation process is executed according to the service task allocation queue, the resource scheduling process is executed according to the resource scheduling configuration, the service response process is executed according to the service response control parameters, and the industrial internet service optimization execution results are generated.

[0042] Optionally, the parameter update of the improved ChronoSpike network specifically includes:

[0043] Based on the results of industrial internet service optimization, data on actual service response latency and actual resource consumption of each service node are collected to generate service execution feedback data.

[0044] Based on the service execution feedback data, calculate the response latency deviation corresponding to the actual service response latency and the resource consumption deviation corresponding to the actual resource consumption data, and generate the service execution deviation result.

[0045] Based on the service execution deviation results, the industrial internet service optimization strategy is fed back and labeled, and the strategy feedback labeling results are generated.

[0046] The service execution feedback data, service execution deviation results, and policy feedback annotation results are input into the improved ChronoSpike network to update the parameters of the improved ChronoSpike network.

[0047] An industrial internet service optimization system based on artificial intelligence according to an embodiment of the present invention includes:

[0048] The data acquisition and preprocessing module is used to collect and preprocess multi-source operational data from service nodes and network edge nodes in the industrial internet system to generate industrial internet service datasets.

[0049] The spatiotemporal calibration module is used to detect spatiotemporal misalignment in industrial internet service datasets, construct spatiotemporal misalignment compensation factors, and generate a set of spatiotemporal calibration state features.

[0050] The causal analysis module is used to perform causal correlation analysis on the spatiotemporal calibration state feature set, construct a dynamic causal influence diagram, and perform redundant causal chain pruning to generate causal state characterization results after pruning.

[0051] The trend prediction module is used to predict the changing trends of key causal paths in the causal state characterization results after pruning, and generate service operation trend prediction results.

[0052] The optimization strategy generation module is used to input the service operation trend prediction results into the improved ChronoSpike network for joint optimization calculation and generate industrial Internet service optimization strategies.

[0053] The execution control module is used to perform differentiated and flexible control on the service task allocation process, resource scheduling process, and service response process according to the industrial internet service optimization strategy, and generate industrial internet service optimization execution results.

[0054] The feedback update module is used to collect data on the actual service response latency and actual resource consumption after the execution of industrial internet service optimization, and to update the parameters of the improved ChronoSpike network.

[0055] The beneficial effects of this invention are:

[0056] This invention reduces state representation bias caused by data acquisition time, network transmission delay, and differences in node processing by performing spatiotemporal misalignment detection and state calibration on multi-source operational data of service nodes and network edge nodes in an Industrial Internet system. This enables the operational status of different service nodes to be analyzed based on a unified time benchmark, improving the accuracy of Industrial Internet service status perception. Simultaneously, by performing causal correlation analysis on the spatiotemporally calibrated state feature set, a dynamic causal influence graph is constructed, and redundant causal chains are pruned to reduce interference from invalid state associations on service trend prediction, improving the ability to identify key causal paths. This allows the generated service operation trend prediction results to more accurately reflect the changing patterns of Industrial Internet service status.

[0057] This invention improves the ChronoSpike network to jointly optimize service operation trend prediction results. It introduces a service load lattice mapping mechanism and a resource contention elimination mechanism to dynamically analyze the competitive relationships in service request processing, resource allocation, and service response, generating industrial internet service optimization strategies. This achieves coordinated optimization of service task allocation, resource scheduling, and service response processes. By updating the parameters of the improved ChronoSpike network using actual service response latency and resource consumption data, the optimization strategy can be continuously adjusted according to the actual operating status of the industrial internet, improving the dynamic adaptability of the service optimization process, reducing service response latency, increasing resource utilization efficiency, and enhancing the operational stability of the industrial internet system. Attached Figure Description

[0058] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0059] Figure 1 This is an overall flowchart of an industrial internet service optimization method and system based on artificial intelligence proposed in this invention;

[0060] Figure 2 This is a schematic diagram of the structure of an improved ChronoSpike network for an industrial internet service optimization method and system based on artificial intelligence proposed in this invention. Detailed Implementation

[0061] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0062] refer to Figures 1-2 An artificial intelligence-based method for optimizing industrial internet services includes the following steps:

[0063] Collect and preprocess multi-source operational data from service nodes and network edge nodes in the industrial internet system to generate an industrial internet service dataset.

[0064] Spatiotemporal misalignment detection is performed on the industrial internet service dataset to identify state representation deviations between service nodes caused by data transmission delays and processing time differences, and a spatiotemporal misalignment compensation factor is constructed to generate a spatiotemporal calibration state feature set.

[0065] Causal correlation analysis is performed on the spatiotemporal calibration state feature set to construct a dynamic causal influence graph between service states. Redundant causal chains are pruned from the dynamic causal influence graph to generate pruned causal state characterization results.

[0066] Predict the changing trends of key causal paths in the causal state characterization results after pruning, and generate service operation trend prediction results.

[0067] The service operation trend prediction results are input into the improved ChronoSpike network for joint optimization calculation to generate industrial Internet service optimization strategies.

[0068] Based on the industrial internet service optimization strategy, differentiated and flexible control is performed on the service task allocation process, resource scheduling process, and service response process in the industrial internet system to generate industrial internet service optimization execution results.

[0069] Collect data on actual service response latency and actual resource consumption after the service optimization execution results are obtained, and update the parameters of the improved ChronoSpike network.

[0070] In this embodiment, the multi-source operational data includes device operation data, network communication data, service request data, and resource status data. Preprocessing includes data cleaning, data denoising, time synchronization, and format conversion.

[0071] In this embodiment, the generation of the spatiotemporal calibration state feature set specifically includes:

[0072] Perform node association processing on the industrial internet service dataset to obtain service node status characteristics and service node time sequence information;

[0073] The process of acquiring service node status characteristics and service node time sequence information is as follows: Industrial Internet multi-source operational data in the Industrial Internet service dataset is classified according to service node identifiers; data belonging to the same service node is associated and aggregated to obtain operational status data corresponding to each service node; feature extraction processing is performed on service load status, resource occupancy status, service request status, and service response status in the operational status data to generate service node status characteristics; based on the data acquisition time, data transmission time, and data processing time carried in the Industrial Internet multi-source operational data, the data generation order and time change relationship of each service node are analyzed to establish the time sequence relationship corresponding to the service node and generate service node time sequence information.

[0074] Perform differential analysis on the timing information of service nodes to determine the differences in data transmission latency and processing time between different service nodes, and generate spatiotemporal misalignment quantities.

[0075] The process of generating the spatiotemporal misalignment is as follows: Based on the time sequence information of service nodes, the time sequence analysis is performed on the data acquisition time, data transmission time, and data processing completion time of each service node. The data transmission delay and data processing delay of each service node are calculated respectively. The data transmission delay is obtained by calculating the time difference between the data reception time and the data acquisition time of the service node, and the data processing delay is obtained by calculating the time difference between the data processing completion time and the data reception time of the service node. The difference in data transmission delay between different service nodes is calculated to obtain the difference in data transmission delay between service nodes, and the difference in data processing delay between different service nodes is calculated to obtain the difference in processing time between service nodes. The weighted sum of the difference in data transmission delay and the difference in processing time is then calculated to obtain the spatiotemporal misalignment.

[0076] Based on the spatiotemporal misalignment, identify the state representation deviations between service nodes caused by data transmission delays and processing time differences, and construct a spatiotemporal misalignment compensation factor.

[0077] The construction process of the spatiotemporal misalignment compensation factor is as follows: Based on the spatiotemporal misalignment amount, the state representation deviation between each service node is classified and identified. The spatiotemporal misalignment amount is compared with a preset misalignment benchmark value to determine the degree of misalignment impact between different service nodes. When the spatiotemporal misalignment amount is less than 10 milliseconds, the state representation deviation between the corresponding service nodes is judged as a slight deviation; when the spatiotemporal misalignment amount is greater than 10 milliseconds but less than 50 milliseconds, the state representation deviation between the corresponding service nodes is judged as a moderate deviation; when the spatiotemporal misalignment amount is greater than 50 milliseconds, the state representation deviation between the corresponding service nodes is judged as a severe deviation. A spatiotemporal misalignment compensation factor is generated based on the degree of influence of the state representation deviation on the state characteristics of the service nodes. The spatiotemporal misalignment compensation factor decreases as the spatiotemporal misalignment amount increases. The spatiotemporal misalignment compensation factor for slight deviation ranges from 0.85 to 1.00, for moderate deviation from 0.60 to 0.85, and for severe deviation from 0.30 to 0.60.

[0078] The service node state characteristics are compensated based on the spatiotemporal misalignment compensation factor to generate spatiotemporal calibration state characteristics.

[0079] The process of generating spatiotemporal calibration state features is as follows: Service node state features are compensated based on a spatiotemporal misalignment compensation factor. Each service node state feature is matched with its corresponding spatiotemporal misalignment compensation factor according to the service node identifier and time sequence to determine the reliability of the service node state feature at the current service time. When the spatiotemporal misalignment compensation factor is close to 1, the main state values ​​in the service node state features are retained. When the spatiotemporal misalignment compensation factor decreases, the service request load, resource occupancy ratio, task queuing length, and service response latency in the service node state features are weighted according to the spatiotemporal misalignment compensation factor to reduce the impact of offset state values ​​caused by data transmission delays and processing time differences on the current service state. Based on the service node time sequence information, the service node state features with time offsets are adjusted to the corresponding unified service time, ensuring that the state features of different service nodes are under the same time reference. The service node state features after weight correction and time reference adjustment are then determined as the spatiotemporal calibration state features.

[0080] The spatiotemporal calibration status features corresponding to each service node are aggregated to generate a spatiotemporal calibration status feature set.

[0081] The process of generating the spatiotemporal calibration state feature set is as follows: The spatiotemporal calibration state features corresponding to each service node are classified according to the service node identifier; the spatiotemporal calibration state features belonging to the same service node are sorted according to time order, and the service node identifier and time order information corresponding to each spatiotemporal calibration state feature are recorded; based on the data association relationship between service nodes, the correspondence between different service nodes is determined; the spatiotemporal calibration state features corresponding to each service node with the same time range are arranged according to a preset order to form a service node state feature sequence; the service node state feature sequence is indexed and stored according to the service node identifier to generate a spatiotemporal calibration state feature set containing the state feature sequences of each service node.

[0082] In this embodiment, the generation of the causal state characterization result after pruning specifically includes:

[0083] The spatiotemporal calibration state feature set is processed into nodes to generate a service state node set;

[0084] The service status node set generation process is as follows: Spatiotemporal calibration status features in the spatiotemporal calibration status feature set are classified according to service node identifiers; spatiotemporal calibration status features belonging to the same service node are sorted according to time sequence information; and the spatiotemporal calibration status features corresponding to each service node at a time sequence position are determined as a service status node. Each service status node includes a corresponding service node identifier, time sequence information, and spatiotemporal calibration status features. All service nodes at each time sequence position are numbered, and a correspondence between service status nodes is established according to the service node identifier and time sequence information to generate the service status node set.

[0085] Based on the time sequence information of each service status node in the service status node set, the sequential relationship of state changes between each service status node is identified, and a candidate causal chain set is generated.

[0086] The specific process of generating the candidate causal chain set is as follows: Based on the service node identifier, time sequence information, and spatiotemporal calibration state characteristics corresponding to each service state node in the service state node set, the service state nodes are traversed in chronological order. The service state node with the earlier time sequence is determined as the preceding service state node, and the service state node with the later time sequence is determined as the following service state node. When the following service state node changes in service load state, resource occupancy state, task queuing state, or service response state relative to the preceding service state node within 1 to 3 sampling periods, a candidate causal chain pointing from the preceding service state node to the following service state node is established.

[0087] Based on the set of candidate causal chains, calculate the causal association strength value corresponding to each candidate causal chain, and construct a dynamic causal influence graph;

[0088] The specific process of constructing the dynamic causal influence graph is as follows: For each candidate causal chain in the candidate causal chain set, read the preceding and following service state nodes corresponding to the candidate causal chain. Calculate the difference between the spatiotemporal calibration state feature corresponding to the preceding service state node and the spatiotemporal calibration state feature of the previous time sequence position to obtain the preceding state change. Calculate the difference between the spatiotemporal calibration state feature corresponding to the following service state node and the spatiotemporal calibration state feature of the previous time sequence position to obtain the following state change. Based on the preceding state change, the following state change, and the time interval between the preceding and following service state nodes, calculate the causal association strength value corresponding to the candidate causal chain. Use the service state node set as the graph nodes of the dynamic causal influence graph, the candidate causal chain set as the directed connection edge of the dynamic causal influence graph, and the causal association strength value as the edge weight of the corresponding directed connection edge to construct the dynamic causal influence graph.

[0089] Redundant causal chain pruning is performed on the dynamic causal influence graph to generate the causal state representation results after pruning.

[0090] The specific process for generating the causal state characterization results after pruning is as follows: Candidate causal chains in the dynamic causal influence graph are sorted according to their causal association strength values. Candidate causal chains with causal association strength values ​​less than a preset pruning threshold (set to 0.35) are deleted from the dynamic causal influence graph. For candidate causal chains where the same preceding service state node points to multiple subsequent service state nodes, the top 5 candidate causal chains are retained according to their causal association strength values ​​from highest to lowest. For candidate causal chains pointing to the same subsequent service state node with a time interval exceeding 3 sampling periods, candidate causal chains with long time intervals and low causal association strength values ​​are deleted. Candidate causal chains in the dynamic causal influence graph whose causal association strength values ​​reach the preset pruning threshold and whose time sequence satisfies the chronological relationship of state changes are retained. The retained candidate causal chains are identified as key causal paths. Based on the key causal paths, the service state nodes corresponding to the key causal paths, and the causal association strength values ​​corresponding to the key causal paths, the causal state characterization results after pruning are formed.

[0091] In this embodiment, the generation of service operation trend prediction results specifically includes:

[0092] The key causal paths in the causal state characterization results after pruning are sorted to generate a key causal path sequence;

[0093] The key causal path sequence generation process is as follows: Key causal paths are obtained from the pruned causal state representation results, and the preceding service state nodes, succeeding service state nodes, temporal sequence information, and causal association strength value corresponding to each key causal path are read. The key causal paths are sorted from earliest to latest according to the temporal sequence information of the preceding service state nodes. If the temporal sequence information of the preceding service state nodes is the same, they are sorted from earliest to latest according to the temporal sequence information of the succeeding service state nodes. If both the temporal sequence information of the preceding and succeeding service state nodes are the same, they are sorted from largest to smallest according to the causal association strength value. The sorted key causal paths are then arranged sequentially to generate the key causal path sequence.

[0094] Based on the service status nodes corresponding to each key causal path in the key causal path sequence, generate path status change quantities.

[0095] The specific process for generating path state change is as follows: Read the preceding and subsequent service state nodes corresponding to each key causal path sequence, and read the corresponding spatiotemporal calibration state features from both nodes. Calculate the difference between the spatiotemporal calibration state features of the subsequent and preceding service state nodes according to the same feature dimensions. These same feature dimensions include service load status, resource occupancy status, task queuing status, and service response status. Take the absolute value of the difference for each of the same feature dimensions, add all the absolute values, and divide by the number of the same feature dimensions to obtain the path state change corresponding to the key causal path.

[0096] Path trend driving values ​​are generated based on path state changes and causal correlation strength values, and then accumulated to form service operation trend values.

[0097] The process of forming the service operation trend value is as follows: The path state change and causal association strength values ​​corresponding to each key causal path are read sequentially according to the key causal path sequence. The path state change and causal association strength values ​​corresponding to the same key causal path are multiplied to generate the path trend driving value corresponding to the key causal path. The path state change represents the magnitude of state change between preceding and subsequent service state nodes. At the same time sequence position, the path trend driving values ​​corresponding to all key causal paths are accumulated, and the accumulated result is determined as the service operation trend value corresponding to that time sequence position. The service operation trend value characterizes the overall change intensity of the key causal path at the time sequence position.

[0098] Based on the service operation trend value, determine the direction and magnitude of change of the key causal path at the next time sequence position, and generate service operation trend prediction results;

[0099] The specific process for generating service operation trend prediction results is as follows: Service operation trend values ​​corresponding to multiple consecutive time sequence positions are read in chronological order, and the difference between service operation trend values ​​between adjacent time sequence positions is calculated. When the difference in service operation trend values ​​is positive, the direction of change of the critical causal path at the next time sequence position is determined as the strengthening direction; when the difference in service operation trend values ​​is negative, the direction of change of the critical causal path at the next time sequence position is determined as the weakening direction; when the difference in service operation trend values ​​is zero, the direction of change of the critical causal path at the next time sequence position is determined as the maintaining direction. The absolute value of the difference in service operation trend values ​​is determined as the magnitude of change of the critical causal path at the next time sequence position, and the service operation trend prediction result is generated based on the direction and magnitude of change.

[0100] In this embodiment, the generation of industrial internet service optimization strategies specifically includes:

[0101] The service operation trend prediction results are input into the improved ChronoSpike network. The improved ChronoSpike network includes a trend phase locking module, a causal potential field projection module, a resource contention culling module, and a service strategy reconstruction module. The improvement of the improved ChronoSpike network is as follows: The traditional ChronoSpike network uses time pulse sequences to trigger pulses and accumulate states for the temporal changes of the input data, and outputs prediction results based on the pulse response intensity. The improved ChronoSpike network introduces the key causal paths, change directions, and change magnitudes in the service operation trend prediction results into the pulse state calculation process. It identifies the trend reversal points of key causal paths through phase gap locations, marks the continuous enhancement segments of key causal paths through phase locking objects, and introduces a service load lattice mapping mechanism and lattice folding processing. It transforms the load occupancy relationship of key causal paths in service nodes, temporal sequence positions, and change directions into resource contention culling results, generating industrial internet service optimization strategies.

[0102] The improved ChronoSpike network adopts a sequential connection structure, connecting the trend phase locking module, causal potential field projection module, resource contention culling module, and service strategy reconstruction module in sequence. Service operation trend prediction results are input in a three-dimensional format: "number of critical causal paths, number of temporal positions, and feature dimensions," where the feature dimensions include critical causal paths, direction of change, and magnitude of change. The trend phase locking module outputs trend phase locking results, which, based on the original three-dimensional format, add phase gap locations and phase locking object markers. The causal potential field projection module receives the trend phase locking results and establishes service load lattice units according to service nodes, temporal positions, and direction of change, outputting causal potential field projection results containing service nodes, temporal positions, direction of change, and lattice occupancy strength. The resource contention culling module receives the causal potential field projection results and outputs resource contention culling results containing retained critical causal paths, lattice occupancy strength after culling, service nodes, and temporal positions. The service strategy reconstruction module receives the resource contention culling results and outputs an industrial internet service optimization strategy containing service request processing order, resource allocation ratio, and service response priority.

[0103] The training data for the improved ChronoSpike network comes from the industrial internet service dataset, service operation trend prediction results, industrial internet service optimization strategies, and industrial internet service optimization execution results generated by the industrial internet system during its historical operation cycle. The annotation method is as follows: the actual service response latency and actual resource consumption data corresponding to the industrial internet service optimization execution results are used as supervision annotation data. Execution results with a service response latency of less than 120 milliseconds and actual resource consumption less than 80% of the set resource capacity are labeled as valid optimization samples, while execution results with a service response latency greater than 120 milliseconds or actual resource consumption greater than 80% of the set resource capacity are labeled as samples to be optimized. A joint loss function is used during training, consisting of policy matching loss, response latency loss, resource consumption loss, and lattice occupancy sparsity loss. The policy matching loss is used to constrain the industrial internet service optimization strategy and the number of supervision annotations. Based on consistency, response latency loss is used to constrain the difference between predicted service response latency and actual service response latency; resource consumption loss is used to constrain the difference between predicted resource consumption and actual resource consumption data; and lattice occupancy sparsity loss is used to suppress excessive simultaneous occupancy of critical causal paths in the service load lattice units. The weights of policy matching loss, response latency loss, resource consumption loss, and lattice occupancy sparsity loss are set to 0.35, 0.30, 0.25, and 0.10, respectively. The training parameters are set as follows: batch size of 32, initial learning rate of 0.002, weight decay coefficient of 0.001, maximum number of training epochs of 200 epochs, learning rate decayed to 60% of the current learning rate every 20 epochs, and upper limit of gradient norm of 5. When the joint loss of the validation set does not decrease for 15 consecutive epochs, the improved ChronoSpike network is considered to have reached convergence, and the model parameters corresponding to the epoch with the minimum joint loss of the validation set are saved.

[0104] In the trend phase locking module, phase breakpoints are identified in the key causal paths, change directions and change magnitudes in the service operation trend prediction results. Key causal paths where the change direction reverses are identified as phase gap locations, and key causal paths where the change magnitude continuously increases are identified as phase locking objects. Trend phase locking results are generated based on the phase gap locations and phase locking objects.

[0105] The trend phase locking result generation process is as follows: Key causal paths in the service operation trend prediction results are read in chronological order, and the direction and magnitude of change corresponding to each key causal path are read; the direction of change of the same key causal path at two adjacent time-series positions is compared. When the direction of change changes from an enhancing direction to a weakening direction, or from a weakening direction to an enhancing direction, the key causal path at the corresponding time-series position is determined as a phase gap position; the magnitude of change of the same key causal path at three consecutive time-series positions is compared. When the magnitude of change increases sequentially, the key causal path is determined as a phase locking object; the direction of change of the key causal path is marked as a breakpoint according to the phase gap position, and the magnitude of change of the key causal path is marked as a lock according to the phase locking object, thus generating the trend phase locking result.

[0106] In the causal potential field projection module, a service load lattice mapping mechanism is introduced to map the key causal paths in the trend phase locking results to the service load lattice units according to the service node, time sequence position and change direction. The lattice occupancy intensity of the service load lattice unit is determined according to the change amplitude. Lattice folding processing is performed on the key causal paths that are in the same service load lattice unit and have opposite change directions to generate causal potential field projection results.

[0107] The process of generating the causal potential field projection result is as follows: The service load lattice mapping mechanism uses service nodes, temporal positions, and change directions as lattice positioning dimensions to locate each key causal path in the trend phase locking result into the corresponding service load lattice unit. During the mapping process, the service nodes corresponding to the key causal paths are used as the node coordinates of the service load lattice unit, the temporal positions corresponding to the key causal paths are used as the temporal coordinates of the service load lattice unit, the change directions corresponding to the key causal paths are used as the directional coordinates of the service load lattice unit, and the change magnitude corresponding to the key causal paths is used as the service load lattice unit's coordinates. The lattice occupancy intensity of the unit is calculated. When multiple critical causal paths fall into the same service load lattice unit and have the same direction of change, the lattice occupancy intensity is accumulated. When multiple critical causal paths fall into the same service load lattice unit and have opposite directions of change, lattice folding is performed. The lattice folding process subtracts the lattice occupancy intensities corresponding to opposite directions of change and retains the direction of change corresponding to the side with the larger lattice occupancy intensity. After completing the service load lattice mapping and lattice folding processes, the causal potential field projection results are generated based on the service node, time sequence position, direction of change, and lattice occupancy intensity corresponding to each service load lattice unit.

[0108] In the resource contention elimination module, the resource contention relationship of the critical causal path in the service request processing process, resource allocation process and service response process is determined based on the causal potential field projection result. Based on the resource contention relationship, the critical causal path is eliminated to generate the resource contention elimination result.

[0109] The process of generating resource contention occlusion results is as follows: Based on the service nodes, temporal order positions, change directions, and lattice occupancy strengths corresponding to each key causal path in the causal potential field projection results, it is determined whether different key causal paths simultaneously point to the service request processing process, resource allocation process, or service response process at the same service node and temporal order position. When two key causal paths correspond to the same service request processing process, the same resource allocation process, or the same service response process at the same service node and temporal order position, a resource contention relationship is determined between the two key causal paths. Based on the resource contention relationship, the key causal paths are occupied. Key causal paths with the same change direction are sorted according to their lattice occupancy strength from largest to smallest. The key causal path with the largest lattice occupancy strength is retained, and the lattice occupancy strengths of the other key causal paths are weakened. Key causal paths with opposite change directions are canceled according to their lattice occupancy strengths, and the key causal path corresponding to the change direction with the larger lattice occupancy strength after cancellation is retained. Based on the key causal paths after occupancy occupancy occupancy, the corresponding service nodes, and temporal order positions, the resource contention occupancy occupancy results are generated.

[0110] In the service strategy reconstruction module, the key causal paths to be retained are determined based on the resource contention elimination results. Based on the service nodes, time sequence positions and lattice occupancy strength corresponding to the retained key causal paths, the service request processing order, resource allocation ratio and service response priority are determined to generate industrial internet service optimization strategies.

[0111] The process of generating an industrial internet service optimization strategy is as follows: First, the retained critical causal paths are read from the resource competition elimination results, and the corresponding service nodes, temporal positions, and lattice occupancy strengths are obtained. Second, the retained critical causal paths are sorted from highest to lowest lattice occupancy strength, and the service nodes corresponding to the highest-ranked retained critical causal paths are identified as priority processing objects, thus determining the service request processing order. Third, the resource allocation ratio for each service node is determined based on the proportion of the lattice occupancy strength of each retained critical causal path to the total lattice occupancy strength of all retained critical causal paths. Fourth, the service response priority is determined based on the temporal position of the retained critical causal paths; retained critical causal paths with earlier temporal positions and higher lattice occupancy strengths have higher service response priority. Finally, an industrial internet service optimization strategy is generated based on the service request processing order, resource allocation ratio, and service response priority.

[0112] In this embodiment, the generation of industrial internet service optimization execution results specifically includes:

[0113] Read the service request processing order, resource allocation ratio, and service response priority from the industrial internet service optimization strategy;

[0114] Based on the service request processing order, the service tasks corresponding to each service node in the industrial internet system are arranged to generate a service task allocation queue.

[0115] The service task allocation queue generation process is as follows: Based on the service request processing order in the industrial internet service optimization strategy, the service tasks corresponding to each service node in the industrial internet system are read and sorted according to the service request processing order; when multiple service tasks correspond to the same service request processing order, they are sorted again according to the service response priority of the service node; when multiple service tasks correspond to the same service request processing order and the same service response priority, they are sorted according to the order in which the service tasks entered the industrial internet system; after sorting, each service task is associated with its corresponding service node, and written into the queue position according to the sorting result to generate the service task allocation queue.

[0116] Based on the resource allocation ratio, the computing resources, storage resources, and network bandwidth resources in the service nodes and network edge nodes are configured to generate resource scheduling configurations;

[0117] The resource scheduling configuration generation process is as follows: Based on the resource allocation ratio in the industrial internet service optimization strategy, the currently available computing resources, storage resources, and network bandwidth resources of service nodes and network edge nodes are read, and the corresponding computing resource allocation, storage resource allocation, and network bandwidth resource allocation for each service node are determined respectively. Specifically, the computing resource allocation is obtained by dividing the available computing resources of service nodes and network edge nodes according to the resource allocation ratio; the storage resource allocation is obtained by dividing the available storage resources of service nodes and network edge nodes according to the resource allocation ratio; and the network bandwidth resource allocation is obtained by dividing the available network bandwidth resources of service nodes and network edge nodes according to the resource allocation ratio. The computing resource allocation, storage resource allocation, and network bandwidth resource allocation for each service node are written into the corresponding resource scheduling record, and a resource scheduling record is established according to the service node identifier and network edge node identifier to generate the resource scheduling configuration.

[0118] Based on the service response priority, determine the response time limit, retry count and transmission priority of each service node, and generate service response control parameters;

[0119] The service response control parameter generation process is as follows: Based on the service response priority in the industrial internet service optimization strategy, each service node is classified into response levels. Service nodes with the highest service response priority are designated as Level 1 response nodes, those with a priority between 30% and 70% are designated as Level 2 response nodes, and those with a priority below 70% are designated as Level 3 response nodes. For Level 1 response nodes, a response time limit of 50 milliseconds, 3 retries, and a transmission priority of first priority are set. For Level 2 response nodes, a response time limit of 100 milliseconds, 2 retries, and a transmission priority of second priority are set. For Level 3 response nodes, a response time limit of 200 milliseconds, 1 retry, and a transmission priority of third priority are set. The response time limit, retries, and transmission priority for each service node are then written into the control parameter table according to the service node identifier to generate the service response control parameters.

[0120] The service task allocation process is executed according to the service task allocation queue, the resource scheduling process is executed according to the resource scheduling configuration, the service response process is executed according to the service response control parameters, and the industrial internet service optimization execution results are generated.

[0121] The specific process for generating industrial internet service optimization execution results is as follows: Service tasks are distributed to their corresponding service nodes according to the queue order in the service task allocation queue, executing the service task allocation process; resource allocation is performed on service nodes and network edge nodes according to the computing resource allocation, storage resource allocation, and network bandwidth resource allocation recorded in the resource scheduling configuration, executing the resource scheduling process; service responses of each service node are controlled according to the response time limit, retry count, and transmission priority recorded in the service response control parameters, executing the service response process; after the service task allocation process, resource scheduling process, and service response process are completed, the task distribution results, resource configuration results, and service response results corresponding to each service node are recorded, generating the industrial internet service optimization execution results.

[0122] In this embodiment, the improved parameter update of the ChronoSpike network specifically includes:

[0123] Based on the results of industrial internet service optimization, data on actual service response latency and actual resource consumption of each service node are collected to generate service execution feedback data.

[0124] The service execution feedback data generation process is as follows: After the industrial internet service optimization execution results are generated, the service start time, service completion time, and resource usage records of each service node are read according to the service node identifier; the actual service response latency is determined based on the time difference between the service completion time and the service start time, and the computing resources, storage resources, and network bandwidth resources consumed by each service node during service execution are statistically analyzed based on the resource usage records to generate actual resource consumption data; the actual service response latency and actual resource consumption data are recorded according to the service node identifier and time sequence position to generate service execution feedback data.

[0125] Based on the service execution feedback data, calculate the response latency deviation corresponding to the actual service response latency and the resource consumption deviation corresponding to the actual resource consumption data, and generate the service execution deviation result.

[0126] The service execution deviation result generation process is as follows: Based on the service execution feedback data, the actual service response latency and actual resource consumption data corresponding to each service node are read, and the response latency deviation and resource consumption deviation are calculated one by one according to the service node identifier. The response latency deviation is the difference between the actual service response latency and the corresponding preset response latency threshold. The preset response latency threshold for a Level 1 response node is 50 milliseconds, for a Level 2 response node it is 100 milliseconds, and for a Level 3 response node it is 200 milliseconds. The resource consumption deviation is the difference between the resource consumption percentage corresponding to the actual resource consumption data and the preset resource consumption threshold, which is set to 80% of the set resource capacity. When the response latency deviation is less than or equal to 0 and the resource consumption deviation is less than or equal to 0, the execution status of the corresponding service node is determined to be in a satisfactory state. When the response latency deviation is greater than 0 or the resource consumption deviation is greater than 0, the execution status of the corresponding service node is determined to be in a deviation state. Based on the response latency deviation, resource consumption deviation, and execution status of each service node, the service execution deviation result is generated.

[0127] Based on the service execution deviation results, the industrial internet service optimization strategy is fed back and labeled, and the strategy feedback labeling results are generated.

[0128] The strategy feedback annotation result generation process is as follows: Based on the service execution deviation results, read the response latency deviation, resource consumption deviation, and execution status corresponding to each service node, and correlate the execution status of each service node with the service request processing order, resource allocation ratio, and service response priority in the industrial internet service optimization strategy; when the execution status of a service node is in a qualified state, mark the industrial internet service optimization strategy corresponding to the service node as an effective optimization strategy; when the execution status of a service node is in a deviation state, mark the industrial internet service optimization strategy corresponding to the service node as an optimization strategy to be corrected, and retain the corresponding response latency deviation and resource consumption deviation; record the effective optimization strategy, the optimization strategy to be corrected, the response latency deviation, and the resource consumption deviation according to the service node identifier to generate the strategy feedback annotation result;

[0129] The service execution feedback data, service execution deviation results, and policy feedback annotation results are input into the improved ChronoSpike network to update the parameters of the improved ChronoSpike network.

[0130] The process of improving the ChronoSpike network parameter update is as follows: After inputting service execution feedback data, service execution deviation results, and policy feedback annotation results into the improved ChronoSpike network, the actual service response latency and actual resource consumption data in the service execution feedback data are read, the response latency deviation and resource consumption deviation in the service execution deviation results are read, and the effective optimization policies and optimization policies to be corrected in the policy feedback annotation results are read. When the policy feedback annotation result is an effective optimization policy, the trend phase locking module, causal potential field projection module, resource contention concealment module, and service policy in the improved ChronoSpike network are maintained. The model parameters corresponding to the reconstructing module are adjusted as follows: When the policy feedback annotation result is a policy to be corrected and optimized, the model parameters corresponding to the service response priority in the trend phase locking module are adjusted according to the response latency deviation, the model parameters corresponding to the lattice occupancy intensity in the causal potential field projection module are adjusted according to the resource consumption deviation, and the model parameters corresponding to the retained key causal paths in the resource contention concealment module are adjusted. The service policy reconstructing module re-determines the service request processing order, resource allocation ratio and service response priority according to the adjusted model parameters, and updates the model parameters of the improved ChronoSpike network with the goal of reducing response latency deviation and resource consumption deviation.

[0131] An artificial intelligence-based industrial internet service optimization system includes:

[0132] The data acquisition and preprocessing module is used to collect and preprocess multi-source operational data from service nodes and network edge nodes in the industrial internet system to generate industrial internet service datasets.

[0133] The spatiotemporal calibration module is used to detect spatiotemporal misalignment in industrial internet service datasets, construct spatiotemporal misalignment compensation factors, and generate a set of spatiotemporal calibration state features.

[0134] The causal analysis module is used to perform causal correlation analysis on the spatiotemporal calibration state feature set, construct a dynamic causal influence diagram, and perform redundant causal chain pruning to generate causal state characterization results after pruning.

[0135] The trend prediction module is used to predict the changing trends of key causal paths in the causal state characterization results after pruning, and generate service operation trend prediction results.

[0136] The optimization strategy generation module is used to input the service operation trend prediction results into the improved ChronoSpike network for joint optimization calculation and generate industrial Internet service optimization strategies.

[0137] The execution control module is used to perform differentiated and flexible control on the service task allocation process, resource scheduling process, and service response process according to the industrial internet service optimization strategy, and generate industrial internet service optimization execution results.

[0138] The feedback update module is used to collect data on the actual service response latency and actual resource consumption after the execution of industrial internet service optimization, and to update the parameters of the improved ChronoSpike network.

[0139] Example 1: In an industrial internet service optimization scenario at a smart manufacturing plant, the plant deploys production equipment nodes, edge computing nodes, and industrial internet service nodes to support production task scheduling, equipment status monitoring, and production data processing. During actual operation, the industrial internet system generates a large amount of multi-source operational data daily, including equipment operation data, network communication data, service request data, and resource status data. Due to differences in data collection cycles, network transmission paths, and edge node computing capabilities among different service nodes, the data uploaded by different service nodes exhibits time offsets. This makes it impossible to accurately match service status data within the same time frame, further impacting subsequent service trend analysis and resource scheduling.

[0140] In this application scenario, multi-source operational data generated by various service nodes and network edge nodes in the industrial internet system are first collected. The collected data is then cleaned, denoised, time-synchronized, and format-converted to generate an industrial internet service dataset. In this embodiment, operational data generated by 128 service nodes and 32 network edge nodes within the production area are collected. The data collection period is set to 100 milliseconds, and continuous collection for 24 hours generates approximately 860,000 sets of industrial internet service data.

[0141] To address the differences in data transmission latency and processing time among different service nodes, spatiotemporal misalignment detection is performed on the industrial internet service dataset. By analyzing the data acquisition time, data transmission time, and data processing completion time of each service node, the time offset between different service nodes is calculated, and a spatiotemporal misalignment compensation factor is constructed based on the time offset. For example, when the data transmission latency between two service nodes is 8 milliseconds, the corresponding service status data is maintained with high reliability; when the data transmission latency reaches 65 milliseconds, the corresponding service status characteristics are significantly corrected to ensure that the status data of different service nodes are under a unified time reference, generating a spatiotemporally calibrated status feature set.

[0142] After completing service status calibration, causal correlation analysis is performed on the spatiotemporal calibration status feature set. The status features corresponding to each service node are converted into service status nodes, and a dynamic causal influence graph is constructed based on the sequential relationship of status changes between service nodes. During the construction of the dynamic causal influence graph, the relationship between changes in service load status, resource occupancy status, task queuing status, and service response status is analyzed, and the causal correlation strength value between different service status nodes is calculated. Correlations with a causal correlation strength value lower than 0.35 are deleted, while key causal paths with high causal correlation strength are retained. This reduces data associations with no actual impact and improves the accuracy of identifying service status change relationships.

[0143] Based on the causal state characterization results after pruning, trend change analysis is performed on key causal paths. By reading the direction and magnitude of change of key causal paths at different time sequence positions, the operational trend of industrial internet services in the next time period is predicted. For example, when the number of task requests for a certain service node increases from 420 times per second to 560 times per second for three consecutive time periods, and the corresponding resource utilization ratio increases from 65% to 82%, this key causal path is identified as an enhanced trend path, and a service operation trend prediction result is generated.

[0144] The service operation trend prediction results are input into the improved ChronoSpike network for joint optimization calculations. The improved ChronoSpike network first determines the trend change characteristics based on the direction and magnitude of change in key causal paths. Then, using a service load lattice mapping mechanism, it maps different service nodes, their temporal positions, and change directions to corresponding service load lattice cells. Based on lattice occupancy strength, it identifies the resource competition relationships between different service nodes. When multiple service tasks simultaneously occupy the resources of the same network edge node, it analyzes the competition relationship based on lattice occupancy strength and reduces the interference of low-impact paths on the generation of optimization strategies, ultimately generating an industrial internet service optimization strategy.

[0145] Based on the generated industrial internet service optimization strategy, adjustments are made to the service task allocation process, resource scheduling process, and service response process in the industrial internet system. For example, during peak production periods, high-priority service tasks are allocated to network edge nodes with sufficient computing resources, while the network bandwidth resource ratio of critical service nodes is increased. Ordinary service tasks are then reassigned to service nodes with lower loads, achieving dynamic allocation of service resources.

[0146] After the optimization strategy is implemented, actual service response latency and resource consumption data continue to be collected, and the actual operating results are fed back to the improved ChronoSpike network to update the model parameters. This embodiment runs continuously for 7 days to test the operating status of industrial internet services before and after optimization. The test results show that this method can reduce the state deviation between service nodes, improve the accuracy of service trend prediction, and enhance the resource utilization efficiency of industrial internet services.

[0147] Table 1. Performance Comparison Data Before and After Application of Industrial Internet Service Optimization Methods

[0148]

[0149] As shown in Table 1, this invention reduces the average time deviation of industrial internet service status data from 42.6 milliseconds to 8.7 milliseconds through spatiotemporal misalignment detection and compensation, making the data status of different service nodes more consistent and improving the reliability of subsequent causal analysis and trend prediction. Simultaneously, through dynamic causal influence graph construction and redundant causal chain pruning, the accuracy of key causal path identification is increased from 81.3% to 95.6%, effectively reducing the impact of invalid associations on service status analysis.

[0150] By improving the ChronoSpike network to jointly optimize service operation trend prediction results and combining it with a service load lattice mapping mechanism to identify resource contention relationships, the response latency of industrial internet services was reduced from 186 milliseconds to 96 milliseconds, and resource utilization was increased from 68.5% to 86.2%. Simultaneously, by continuously updating model parameters through feedback data, the service task scheduling success rate reached 97.9%, reducing the frequency of resource contention anomalies and improving the service optimization capabilities and stability of the industrial internet system in dynamic operating environments.

[0151] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for optimizing industrial internet services based on artificial intelligence, characterized in that, Includes the following steps: Collect and preprocess multi-source operational data from service nodes and network edge nodes in the industrial internet system to generate an industrial internet service dataset. Spatiotemporal misalignment detection is performed on the industrial internet service dataset to identify state representation deviations between service nodes caused by data transmission delays and processing time differences, and a spatiotemporal misalignment compensation factor is constructed to generate a spatiotemporal calibration state feature set. Causal correlation analysis is performed on the spatiotemporal calibration state feature set to construct a dynamic causal influence graph between service states. Redundant causal chains are pruned from the dynamic causal influence graph to generate pruned causal state characterization results. Predict the changing trends of key causal paths in the causal state characterization results after pruning, and generate service operation trend prediction results. The service operation trend prediction results are input into the improved ChronoSpike network for joint optimization calculation to generate industrial Internet service optimization strategies. Based on the industrial internet service optimization strategy, differentiated and flexible control is performed on the service task allocation process, resource scheduling process, and service response process in the industrial internet system to generate industrial internet service optimization execution results. Collect data on actual service response latency and actual resource consumption after the service optimization execution results are obtained, and update the parameters of the improved ChronoSpike network.

2. The method for optimizing industrial internet services based on artificial intelligence according to claim 1, characterized in that, The multi-source operational data includes equipment operation data, network communication data, service request data, and resource status data. The preprocessing includes data cleaning, data denoising, time synchronization, and format conversion.

3. The method for optimizing industrial internet services based on artificial intelligence according to claim 1, characterized in that, The generation of the spatiotemporal calibration state feature set specifically includes: Perform node association processing on the industrial internet service dataset to obtain service node status characteristics and service node time sequence information; Perform differential analysis on the timing information of service nodes to determine the differences in data transmission latency and processing time between different service nodes, and generate spatiotemporal misalignment quantities. Based on the spatiotemporal misalignment, identify the state representation deviations between service nodes caused by data transmission delays and processing time differences, and construct a spatiotemporal misalignment compensation factor. The service node state characteristics are compensated based on the spatiotemporal misalignment compensation factor to generate spatiotemporal calibration state characteristics. The spatiotemporal calibration status features corresponding to each service node are aggregated to generate a spatiotemporal calibration status feature set.

4. The method for optimizing industrial internet services based on artificial intelligence according to claim 1, characterized in that, The generation of the causal state characterization results after pruning specifically includes: The spatiotemporal calibration state feature set is processed into nodes to generate a service state node set; Based on the time sequence information of each service status node in the service status node set, the sequential relationship of state changes between each service status node is identified, and a candidate causal chain set is generated. Based on the set of candidate causal chains, calculate the causal association strength value corresponding to each candidate causal chain, and construct a dynamic causal influence graph; Redundant causal chains are pruned from the dynamic causal influence graph to generate a causal state representation result after pruning.

5. The method for optimizing industrial internet services based on artificial intelligence according to claim 1, characterized in that, The generation of the service operation trend prediction results specifically includes: The key causal paths in the causal state characterization results after pruning are sorted to generate a key causal path sequence; Based on the service status nodes corresponding to each key causal path in the key causal path sequence, generate path status change quantities. Path trend driving values ​​are generated based on path state changes and causal correlation strength values, and then accumulated to form service operation trend values. Based on the service operation trend value, determine the direction and magnitude of change of the key causal path at the next time sequence position, and generate service operation trend prediction results.

6. The method for optimizing industrial internet services based on artificial intelligence according to claim 1, characterized in that, The generation of the industrial internet service optimization strategy specifically includes: The service operation trend prediction results are input into the improved ChronoSpike network. The improved ChronoSpike network includes a trend phase locking module, a causal potential field projection module, a resource contention culling module, and a service strategy reconstruction module. The improvement of the improved ChronoSpike network is as follows: The traditional ChronoSpike network uses time pulse sequences to trigger pulses and accumulate states for the temporal changes of the input data, and outputs prediction results based on the pulse response intensity. The improved ChronoSpike network introduces the key causal paths, change directions, and change magnitudes in the service operation trend prediction results into the pulse state calculation process. It identifies the trend reversal points of key causal paths through phase gap locations, marks the continuous enhancement segments of key causal paths through phase locking objects, and introduces a service load lattice mapping mechanism and lattice folding processing. It transforms the load occupancy relationship of key causal paths in service nodes, temporal sequence positions, and change directions into resource contention culling results, generating industrial internet service optimization strategies. In the trend phase locking module, phase breakpoints are identified in the key causal paths, change directions and change magnitudes in the service operation trend prediction results. Key causal paths where the change direction reverses are identified as phase gap locations, and key causal paths where the change magnitude continuously increases are identified as phase locking objects. Trend phase locking results are generated based on the phase gap locations and phase locking objects. In the causal potential field projection module, a service load lattice mapping mechanism is introduced to map the key causal paths in the trend phase locking results to the service load lattice units according to the service node, time sequence position and change direction. The lattice occupancy intensity of the service load lattice unit is determined according to the change amplitude. Lattice folding processing is performed on the key causal paths that are in the same service load lattice unit and have opposite change directions to generate causal potential field projection results. In the resource contention elimination module, the resource contention relationship of the critical causal path in the service request processing process, resource allocation process and service response process is determined based on the causal potential field projection result. Based on the resource contention relationship, the critical causal path is eliminated to generate the resource contention elimination result. In the service strategy reconstruction module, the key causal paths to be retained are determined based on the resource contention elimination results. Then, the service request processing order, resource allocation ratio, and service response priority are determined based on the service nodes, time sequence positions, and lattice occupancy strength corresponding to the retained key causal paths, thereby generating an industrial internet service optimization strategy.

7. The method for optimizing industrial internet services based on artificial intelligence according to claim 1, characterized in that, The generation of the optimized execution results of the industrial internet services specifically includes: Read the service request processing order, resource allocation ratio, and service response priority from the industrial internet service optimization strategy; Based on the service request processing order, the service tasks corresponding to each service node in the industrial internet system are arranged to generate a service task allocation queue. Based on the resource allocation ratio, the computing resources, storage resources, and network bandwidth resources in the service nodes and network edge nodes are configured to generate resource scheduling configurations; Based on the service response priority, determine the response time limit, retry count and transmission priority of each service node, and generate service response control parameters; The service task allocation process is executed according to the service task allocation queue, the resource scheduling process is executed according to the resource scheduling configuration, the service response process is executed according to the service response control parameters, and the industrial internet service optimization execution results are generated.

8. The method for optimizing industrial internet services based on artificial intelligence according to claim 1, characterized in that, The parameter updates for the improved ChronoSpike network specifically include: Based on the results of industrial internet service optimization, data on actual service response latency and actual resource consumption of each service node are collected to generate service execution feedback data. Based on the service execution feedback data, calculate the response latency deviation corresponding to the actual service response latency and the resource consumption deviation corresponding to the actual resource consumption data, and generate the service execution deviation result. Based on the service execution deviation results, the industrial internet service optimization strategy is fed back and labeled, and the strategy feedback labeling results are generated. The service execution feedback data, service execution deviation results, and policy feedback annotation results are input into the improved ChronoSpike network to update the parameters of the improved ChronoSpike network.

9. An artificial intelligence-based industrial internet service optimization system, executing the artificial intelligence-based industrial internet service optimization method according to any one of claims 1 to 8, characterized in that, include: The data acquisition and preprocessing module is used to collect and preprocess multi-source operational data from service nodes and network edge nodes in the industrial internet system to generate industrial internet service datasets. The spatiotemporal calibration module is used to detect spatiotemporal misalignment in industrial internet service datasets, construct spatiotemporal misalignment compensation factors, and generate a set of spatiotemporal calibration state features. The causal analysis module is used to perform causal correlation analysis on the spatiotemporal calibration state feature set, construct a dynamic causal influence diagram, and perform redundant causal chain pruning to generate causal state characterization results after pruning. The trend prediction module is used to predict the changing trends of key causal paths in the causal state characterization results after pruning, and generate service operation trend prediction results. The optimization strategy generation module is used to input the service operation trend prediction results into the improved ChronoSpike network for joint optimization calculation and generate industrial Internet service optimization strategies. The execution control module is used to perform differentiated and flexible control on the service task allocation process, resource scheduling process, and service response process according to the industrial internet service optimization strategy, and generate industrial internet service optimization execution results. The feedback update module is used to collect data on the actual service response latency and actual resource consumption after the execution of industrial internet service optimization, and to update the parameters of the improved ChronoSpike network.

Images