Cross-system cooperative control method and system based on digital thread

Abstract

The invention provides a cross-system cooperative control method and system based on a digital thread, and the method comprises the steps: building a cross-system interaction architecture, and building real-time data transmission and instruction interaction among a plurality of business systems through a distributed node communication protocol; collecting dynamic operation characteristic data of each service system based on the cross-system interaction architecture, and generating a cross-system cooperative control logic; converting the cross-system cooperative control logic into a standardized control instruction, distributing the standardized control instruction to a corresponding service system, executing real-time cooperative control, and collecting state feedback characteristics in a control process; performing dynamic deviation analysis on the state feedback characteristics to generate cooperative control strategy adjustment parameters; and synchronizing the cooperative control strategy adjustment parameter to each service system through the cross-system interaction architecture. According to the invention, the efficiency of cross-system cooperative control is improved.

Description

Technical Field

[0001] The present invention relates to the field of digital control, and in particular to a cross-system collaborative control method and system based on digital threads. Background Art

[0002] With the development of industrial digitalization and system integration, cross-system collaborative control technology has been proposed. This technology is used to enable multiple independent business systems to collaboratively execute tasks, allocate resources, and optimize operations under a unified goal. It is widely used in complex industrial scenarios such as intelligent manufacturing, intelligent transportation, and the energy internet. Its core goal is to address data interaction, task scheduling, and dynamic adaptation between different business systems. Currently, common cross-system collaborative control technologies typically rely on fixed communication links to enable inter-system data transmission. Control instructions are generated based on preset rules or the operating status of a single system. These instructions are then distributed to each business system through a centralized node for execution. During execution, the system relies on regular manual inspections or system logs to obtain operating status, allowing for offline adjustments to the control strategy. However, this collaborative approach, which relies on fixed communication links and preset rules, struggles to adapt to the dynamically changing operating characteristics of different business systems, resulting in a poor match between control instructions and actual system requirements. Furthermore, the centralized node's instruction distribution model is susceptible to single points of failure, and the lag in offline status acquisition and policy adjustments prevents real-time resolution of resource contention and task conflicts during cross-system collaboration, impacting the stability and efficiency of the overall collaborative control. Based on this, how to improve the dynamic adaptability and real-time response capability of cross-system collaborative control has become a current research hotspot. Summary of the Invention

[0003] The present invention provides a cross-system collaborative control method and system based on digital thread.

[0004] In a first aspect, an embodiment of the present invention provides a cross-system collaborative control method based on a digital thread, comprising: Build a cross-system interaction architecture and establish real-time data transmission and instruction interaction between multiple business systems through distributed node communication protocols; Based on the cross-system interaction architecture, dynamic operation characteristic data of each business system is collected to generate cross-system collaborative control logic; Convert the cross-system collaborative control logic into standardized control instructions, distribute them to the corresponding business systems, and execute real-time collaborative control, collecting state feedback features during the control process; Performing dynamic deviation analysis on the state feedback characteristics to generate collaborative control strategy adjustment parameters; The collaborative control strategy adjustment parameters are synchronized to each business system through the cross-system interaction architecture.

[0005] In a second aspect, an embodiment of the present invention provides a computer system, including: a memory storing a computer program; A processor is used to load the computer program to implement the cross-system collaborative control method based on digital threads as described above.

[0006] The cross-system collaborative control method based on digital threads provided by the present invention realizes real-time data transmission and instruction interaction between multiple business systems by constructing a cross-system interactive architecture, collects dynamic operation feature data based on the architecture, and uses a multimodal state fusion algorithm to generate a cross-system collaborative control logic, so that the collaborative control logic can comprehensively reflect the operating status and correlation relationship of different business systems, avoid the limitations of a single system control logic, and is conducive to improving the accuracy of cross-system collaborative control; the cross-system collaborative control logic is converted into standardized control instructions and distributed to the corresponding business system for execution, and the state feedback characteristics during the control process are collected, and the standardized instructions are used to ensure that each business system is in control of the control logic. Consistent execution, while capturing dynamic changes in the collaborative control process in real time through state feedback features, providing a reliable basis for subsequent strategy adjustments; dynamic deviation analysis of state feedback features is performed on the time series feature migration algorithm to generate collaborative control strategy adjustment parameters, and dynamic trend analysis of current deviations is achieved in combination with historical collaborative control feature laws, which improves the comprehensiveness of deviation analysis and the adaptability of adjustment parameters; collaborative control strategy adjustment parameters are synchronized to each business system through a cross-system interaction architecture to achieve dynamic optimization of cross-system collaborative control, and through the close connection of the entire process, promote efficient cooperation of each business system in the collaborative control process, thereby improving the overall effectiveness of cross-system collaborative control. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Figure 1 This is a flowchart of a cross-system collaborative control method based on digital threads provided by an embodiment of the present invention.

[0008] Figure 2 It is a schematic diagram of the composition of a computer system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0009] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0010] See also Figure 1 , Figure 1A flowchart of a cross-system collaborative control method based on a digital thread provided in an embodiment of the present invention, which can be executed by a computer system, includes the following steps: Step S100: Build a cross-system interaction architecture and establish real-time data transmission and instruction interaction between multiple business systems through a distributed node communication protocol.

[0011] The cross-system interaction architecture is an architectural system used to connect and interact with multiple business systems, providing a unified communication and collaboration platform for each business system. The distributed node communication protocol is a protocol used for communication between nodes in a distributed system, ensuring accurate and efficient data transmission between different nodes. Real-time data transmission refers to the transmission of data to the target system as it is generated, ensuring its timeliness and validity. Command interaction allows business systems to send and receive control commands to each other, enabling coordinated control.

[0012] When building a cross-system interaction architecture, first analyze the network topology, data interface types, and communication requirements of each business system. Select an appropriate distributed node communication protocol, such as MQTT (Message Queuing Telemetry Transport). Then, deploy the corresponding communication nodes in each business system. These nodes can be hardware devices or software programs. Through these communication nodes, each business system can perform real-time data transmission and command exchange according to the rules of the distributed node communication protocol.

[0013] For example, in an enterprise information environment encompassing production, logistics, and sales systems, the production system needs to transmit production progress data to the logistics and sales systems in real time, enabling the logistics system to schedule shipments and the sales system to stay informed of product availability. Simultaneously, the logistics and sales systems can also send production adjustment instructions to the production system. By building a cross-system interaction architecture and adopting the MQTT protocol, efficient and stable real-time data transmission and instruction exchange can be achieved between these systems.

[0014] Step S200: Collect dynamic operation characteristic data of each business system based on the cross-system interaction architecture and generate cross-system collaborative control logic.

[0015] Dynamic operational characteristic data refers to the various characteristic data exhibited by each business system during operation, which can reflect the system's operating status and performance. Cross-system collaborative control logic is a set of logical rules used to coordinate the work of multiple business systems. It can rationally allocate resources and schedule tasks based on the operating status and needs of each system, thereby achieving cross-system collaboration.

[0016] When collecting dynamic running feature data of each business system based on the cross-system interaction architecture, distributed collection nodes in the cross-system interaction architecture are used to collect data of each business system at a preset time interval. These distributed collection nodes can be hardware devices such as sensors and data collection cards, or data collection programs running in each business system. The collected dynamic running feature data includes system task queue features, resource occupation sequence features, and interaction response time length features.

[0017] The system task queue features refer to the length of the task queue, the distribution of task types, and the task waiting time in the business system, which can reflect the task load of the system. The resource occupation sequence features refer to the occupation of various resources (such as CPU, memory, network bandwidth, etc.) by the system within a period of time, which can reflect the resource utilization efficiency of the system. The interaction response time length features refer to the response time of the system when interacting with other systems, which can reflect the response speed and interaction performance of the system. After collecting the dynamic running feature data, the data is processed and analyzed to generate cross-system collaborative control logic. This process needs to consider factors such as the correlation between business systems, resource competition, and task coupling degree.

[0018] As an implementation, in step S200, dynamic running feature data of each business system is collected based on the cross-system interaction architecture, and cross-system collaborative control logic is generated. Specifically, it can be implemented as steps S210-S250 as follows: Step S210: Collect dynamic running feature data of each business system through distributed collection nodes in the cross-system interaction architecture at a preset time interval. The dynamic running feature data includes system task queue features, resource occupation sequence features, and interaction response time length features.

[0019] The distributed collection nodes are nodes in the cross-system interaction architecture for collecting data of each business system. They are distributed in each business system and can collect system running data in real time and accurately. The preset time interval refers to the time interval for collecting data.

[0020] The collection of system task queue features can be realized by monitoring the related information of the task queue in the business system. For example, for a task queue in an operating system, the length of the task queue, the priority of the task, and the waiting time of the task can be obtained through system calls. The collection of resource occupation sequence features can be realized by system monitoring tools. For example, in a Linux system, tools such as the top command and the vmstat command can be used to obtain the occupation of resources such as CPU, memory, and disk I / O. The collection of interaction response time length features can be realized by sending test messages between systems and recording the sending time and receiving time of the messages.

[0021] Step S220: performing spatiotemporal alignment processing on the dynamic operation feature data, mapping the feature data of different business systems to a unified time coordinate system, and generating a spatiotemporal alignment feature set.

[0022] Spatiotemporal alignment refers to processing the feature data collected by different business systems at different times to make them consistent in time, thereby enabling effective comparison and analysis. The unified time coordinate system is a reference coordinate system used to unify the time of different business systems. It can ensure that the feature data of each system are comparable in time. The spatiotemporal alignment feature set is a feature data set obtained after spatiotemporal alignment, which contains the feature data of each business system in the unified time coordinate system. Because the clocks of different business systems may have deviations and the time intervals for data collection may also be different, it is necessary to perform spatiotemporal alignment processing on the dynamic operation feature data. This process mainly includes steps such as extracting timestamp information, constructing a time reference axis, interpolation processing, dimension splicing, and normalization processing.

[0023] As an implementation method, step S220 performs spatiotemporal alignment processing on the dynamic operation feature data, maps the feature data of different business systems to a unified time coordinate system, and generates a spatiotemporal alignment feature set. Specifically, the process can be implemented as follows: steps S221 to S225: Step S221: extracting the timestamp information in the dynamic operation characteristic data, and determining the collection time and data sampling period of each business system characteristic data.

[0024] Timestamp information refers to the time stamp included in dynamic operation feature data, recording the specific time when the data was collected. The collection moment refers to the specific point in time when the data was collected, and the data sampling period refers to the time interval between two consecutive data collections. When collecting dynamic operation feature data, each data point is accompanied by a timestamp. By analyzing this timestamp information, the collection moment of each business system's feature data can be determined. Furthermore, by analyzing the timestamp differences between adjacent data points, the data sampling period can be calculated.

[0025] Step S222: Based on the collection time and the data sampling period, a time base axis is constructed. The minimum time unit of the time base axis is the common divisor of the sampling periods of each business system.

[0026] The time base axis is a reference axis used to unify the time across different business systems. The minimum time unit is the common divisor of the sampling periods of each business system. By constructing a time base axis, the characteristic data of different business systems can be unified in time.

[0027] Calculate the common divisor of the sampling period of each business system and use this common divisor as the minimum time unit of the time base axis. Then, determine the corresponding position on the time base axis based on the collection time of the characteristic data of each business system.

[0028] Step S223: interpolate the dynamic operation characteristic data of each business system, supplement the characteristic data of the missing moments on the time reference axis, and generate a characteristic sequence with equal time intervals.

[0029] Interpolation involves inserting new data points between known data points to fill in missing data. A feature series with equal time intervals means that, after interpolation, the feature data of each business system has the same temporal interval. Because different business systems have different sampling periods, there may be some missing moments on the time axis. Interpolation can be used to fill in the corresponding feature data at these missing moments. Common interpolation methods include linear interpolation and polynomial interpolation.

[0030] Step S224: Dimensionally splice the time interval feature sequences of each business system according to the system identifier to obtain a three-dimensional feature tensor including the system dimension, time dimension and feature dimension, and use the three-dimensional feature tensor as the spatiotemporal alignment feature set.

[0031] A system identifier uniquely identifies each business system, such as the system name or system number. Dimensional concatenation involves concatenating feature sequences from different business systems along the system dimension to form a higher-dimensional feature data structure. A three-dimensional feature tensor encompasses system, time, and feature dimensions, and comprehensively represents the feature data of each business system at different times.

[0032] Arrange the time-interval feature sequences of each business system by system ID and then concatenate them along the system dimension. For example, consider three business systems A, B, and C, whose time-interval feature sequences are FA, FB, and FC, respectively. Concatenate FA, FB, and FC along the system dimension to obtain a three-dimensional feature tensor, where the first dimension represents the system dimension, the second dimension represents the time dimension, and the third dimension represents the feature dimension.

[0033] Step S225: normalize the three-dimensional feature tensor to eliminate the dimensional differences of feature data of different business systems, and keep the temporal variation trend of each feature sequence unchanged during the normalization process.

[0034] Normalization is the process of processing the characteristic data of different business systems so that they have the same dimension and range, thereby enabling effective comparison and analysis. Dimensional differences refer to the different units and magnitudes of characteristic data from different business systems, which may affect the analysis and processing results of the data. Viable normalization methods include maximum and minimum value normalization and Z-score normalization. Taking maximum and minimum value normalization as an example, for each feature dimension in the three-dimensional feature tensor, its maximum and minimum values ​​are calculated, and then the minimum value is subtracted from each data point and divided by the difference between the maximum and minimum values ​​to obtain the normalized data. During the normalization process, it is necessary to ensure that the temporal trend of each feature sequence remains unchanged, that is, the relative size relationship of the data remains unchanged.

[0035] Step S230: Based on the spatiotemporal alignment feature set, the association dependency features between the business systems are extracted through the attention mechanism. The association dependency features represent the resource competition relationship and task coupling degree of different business systems during the task execution process.

[0036] The attention mechanism is used to automatically focus on important information within large amounts of data, assigning different attention weights based on the importance of the data. Correlation dependency features refer to the characteristics of the correlation and dependency between different business systems, reflecting the mutual influence of each system during task execution. Resource competition refers to the competition between different business systems when using shared resources, and task coupling refers to the degree of correlation between tasks in different business systems. The attention mechanism can automatically identify the correlation between feature data from each business system in a spatiotemporal alignment feature set, thereby extracting correlation dependency features. This process primarily includes steps such as time window segmentation, similarity weight calculation, weighted aggregation, feature sequence splicing, and filtering.

[0037] As an implementation method, step S230 extracts the correlation dependency features between the business systems based on the spatiotemporal alignment feature set through the attention mechanism, which can be specifically implemented as follows: steps S231 to S235: Step S231: dividing the spatiotemporal alignment feature set into multiple time window feature subsets according to the time dimension, each time window feature subset contains feature data of a preset number of consecutive time steps.

[0038] A time window is a fixed-length time period divided along the time dimension. A time window feature subset is a collection of feature data within a time window. The preset number refers to the number of consecutive time steps contained in each time window. The spatiotemporal alignment feature set is segmented along the time dimension based on the preset time window size.

[0039] Step S232: For each time window feature subset, the similarity weight between the feature data of different business systems is calculated through the self-attention mechanism. The similarity weight represents the degree of correlation between the feature data of two business systems in the same time window.

[0040] A similarity weight is a numerical value used to indicate the closeness of the correlation between the feature data of two business systems. A larger weight indicates a closer correlation. For each feature subset in a time window, the feature data of different business systems is input into the self-attention mechanism for calculation. The self-attention mechanism calculates the similarity score between the feature data and then normalizes the score to obtain the similarity weight. Similarity calculation methods include cosine similarity and Euclidean distance.

[0041] Step S233: Based on the similarity weight, perform weighted aggregation on the feature data of each business system to generate an interaction correlation vector between the business systems. The dimension of the interaction correlation vector is consistent with the number of business systems.

[0042] Weighted aggregation involves weighted summation of the feature data of each business system based on similarity weights, resulting in a comprehensive feature representation. The interaction correlation vector is a vector used to represent the interaction relationship between business systems. Its dimension is consistent with the number of business systems.

[0043] Step S234: The interactive correlation vectors are spliced ​​in the order of the time windows to obtain a correlation dependency feature sequence, where each element in the correlation dependency feature sequence corresponds to a business system correlation relationship within a time window.

[0044] The interaction correlation vectors generated within each time window are arranged in the order of the time windows and then concatenated along the time dimension to obtain a correlation dependency feature sequence. Each element in this sequence corresponds to a business system correlation relationship within a time window.

[0045] Step S235: Filter the association dependency feature sequence to eliminate the interference of short-term fluctuations on the association relationship extraction, wherein the filtering algorithm is a sliding average algorithm, and the window size is a preset ratio of the number of time windows.

[0046] Filtering is the process of processing data to remove noise and short-term fluctuations, resulting in smoother data. The sliding average algorithm smoothes data by calculating the average value of the data within a fixed window. The window size refers to the length of the window used in the sliding average algorithm, and the preset ratio refers to the ratio of the preset window size to the number of time windows.

[0047] According to a preset proportion of the number of time windows, the window size of the moving average algorithm is determined. Then, the sliding average calculation is performed on the associated dependency feature sequence. For example, assuming that the number of time windows is 100 and the preset proportion is 0.1, the window size is 10. For each element in the associated dependency feature sequence, the average value of the 10 elements before and after the element is calculated as the filtered value of the element.

[0048] Step S240: input the associated dependency features into the pre-trained collaborative logic generation network, the collaborative logic generation network including a feature encoding layer, a dependency reasoning layer, and a logic output layer, the feature encoding layer being used to convert the associated dependency features into high-dimensional semantic vectors, the dependency reasoning layer being used to perform node relationship reasoning on the high-dimensional semantic vectors through a graph neural network, and the logic output layer being used to generate an initial framework of the cross-system collaborative control logic based on the reasoning result.

[0049] The pre-trained collaborative logic generation network is a neural network model that is pre-trained and can generate the cross-system collaborative control logic according to the input associated dependency features. The feature encoding layer is the first layer of the collaborative logic generation network, and its function is to convert the associated dependency features into high-dimensional semantic vectors for subsequent processing and analysis. The dependency reasoning layer is the second layer of the collaborative logic generation network, and it performs node relationship reasoning on the high-dimensional semantic vectors through a graph neural network to mine the association relationships between the business systems. The logic output layer is the last layer of the collaborative logic generation network, and it generates an initial framework of the cross-system collaborative control logic based on the reasoning result of the dependency reasoning layer.

[0050] The associated dependency features are input into the feature encoding layer of the collaborative logic generation network, which can adopt structures such as a fully connected layer or a convolutional layer to map the associated dependency features to a high-dimensional space and obtain high-dimensional semantic vectors. Then, the high-dimensional semantic vectors are input into the dependency reasoning layer, which adopts a graph neural network (GNN), such as a Graph Convolutional Network (GCN), to perform node relationship reasoning on the high-dimensional semantic vectors. The graph neural network regards each business system as a node and regards the association relationships between them as edges, and obtains the dependency relationships between the business systems by propagating and aggregating the information of the nodes and edges. Finally, the logic output layer generates an initial framework of the cross-system collaborative control logic according to the reasoning result of the dependency reasoning layer, and the framework contains basic rules such as task allocation and resource scheduling between the business systems.

[0051] As an implementation manner, the pre-training process of the collaborative logic generation network can be implemented as steps S2401-S2405 as follows: Step S2401: construct a training data set containing historical collaborative control cases, each case containing business system dynamic running feature data and corresponding optimal collaborative control logic.

[0052] Historical collaborative control cases refer to actual cases accumulated during the past cross-system collaborative control process. These cases contain the dynamic operational characteristics of business systems and the corresponding optimal collaborative control logic. The training dataset is used to train the collaborative logic generation network and contains a large number of historical collaborative control cases.

[0053] Collect cross-system collaborative control cases from the past period and organize and annotate each case. For example, record the dynamic operational characteristics of each business system in each case, such as system task queue characteristics, resource usage sequence characteristics, and interactive response duration characteristics. Also, record the corresponding optimal collaborative control logic, such as task allocation plans and resource scheduling strategies. These cases will form a training dataset.

[0054] Step S2402: Divide the training data set into a training set and a validation set, and use the cross-validation method to perform model training.

[0055] Cross-validation is a method used to evaluate model performance and select optimal model parameters. It divides a dataset into multiple subsets and uses different subsets for training and validation. The training set is the dataset used to train the model, and the validation set is the dataset used to verify the model's performance.

[0056] The training dataset is divided into a training set and a validation set according to a certain ratio, for example, an 8:2 ratio. Then, the cross-validation method is used for model training.

[0057] Step S2403: Initialize the weight parameters of the collaborative logic generation network, set the weights of the feature encoding layer and the dependency reasoning layer, and set the bias of the logic output layer.

[0058] Weight parameters are learnable parameters in a neural network model that determine the model's performance and behavior. The weights of the feature encoding layer and the dependency inference layer refer to the connection weights between neurons in these two layers, while the bias of the logistic output layer refers to the bias value of the neurons in that layer.

[0059] Before training the collaborative logic generation network, its weight parameters need to be initialized. This can be done by using a random initialization method, such as using a Gaussian distribution to randomly initialize the weight parameters. At the same time, the weights of the feature encoding layer and dependency inference layer, as well as the bias of the logic output layer, are set.

[0060] Step S2404: Using the dynamic operation feature data in the training set as input, generate the predicted collaborative control logic through network forward propagation, and calculate the edit distance loss between the predicted logic and the true optimal logic.

[0061] Network forward propagation refers to the process of passing input data through the various layers of a neural network to generate output results. Predicted collaborative control logic refers to the collaborative control logic generated by the collaborative logic generation network based on the input dynamic operating characteristic data. Edit distance loss refers to the degree of difference between the predicted collaborative control logic and the true optimal logic. It can be obtained by calculating the edit distance between the two logics.

[0062] The dynamic operational feature data from the training set is fed into the collaborative logic generation network. The predicted collaborative control logic is generated through forward propagation through the network. The edit distance loss between the predicted logic and the true optimal logic is then calculated. The edit distance can be calculated using methods such as the Levenshtein distance.

[0063] Step S2405: Use the optimizer to minimize the edit distance loss, set the initial learning rate, decay the learning rate according to the preset rules, and stop training when the validation set loss no longer decreases.

[0064] An optimizer is an algorithm used to adjust the weight parameters of a neural network model, aiming to minimize the loss function. The initial learning rate is the learning rate used by the optimizer at the beginning of training, which determines the step size for updating the weight parameters. The default rule is a pre-defined learning rate decay rule, such as fixed-rate decay or exponential decay.

[0065] Select an optimizer, such as the Adam optimizer, to minimize the edit distance loss. Set an initial learning rate, such as 0.001. Then, decay the learning rate according to a pre-set rule, for example, by 0.9 times the original rate every 10 epochs. During training, continuously iteratively update the model's weights until the validation set loss no longer decreases.

[0066] Step S250: Perform conflict detection on the initial framework to identify resource allocation conflicts and task execution timing conflicts in the collaborative control logic, adjust the task priority and resource allocation ratio in the initial framework based on the conflict detection results, and obtain the final cross-system collaborative control logic.

[0067] Conflict detection involves examining the collaborative control logic to identify resource allocation conflicts and task execution sequence conflicts. A resource allocation conflict occurs when the demand for the same resource by multiple business systems exceeds the available resource within the collaborative control logic. A task execution sequence conflict occurs when different tasks within the collaborative control logic have conflicting execution times. For example, task A must be completed before task B, but the logic schedules task B to execute first.

[0068] When detecting conflicts within the initial framework, you can use a rule-matching approach. For example, define a series of resource allocation rules and task execution sequence rules, then match the initial framework against these rules to identify conflicts. Detected resource allocation conflicts can be resolved by adjusting resource allocation ratios, such as reducing the amount of resources allocated to certain business systems. Task execution sequence conflicts can be resolved by adjusting task priorities, such as increasing the priority of certain tasks to prioritize their execution.

[0069] For example, in a collaborative control logic involving multiple production tasks, it was discovered that both Task A and Task B required the use of the same equipment, and the equipment's availability could not meet the needs of both tasks, resulting in a resource allocation conflict. The conflict was resolved by adjusting the resource allocation ratios between Task A and Task B, for example, by reducing the equipment usage time for Task A and increasing it for Task B. At the same time, it was discovered that Task C must be completed before Task D, but the logic scheduled Task D for execution first, resulting in a task execution timing conflict. The conflict was resolved by increasing Task C's priority, allowing it to execute first. After conflict detection and adjustment, the final cross-system collaborative control logic was obtained.

[0070] Step S300: Convert the cross-system collaborative control logic into standardized control instructions, distribute them to the corresponding business systems, execute real-time collaborative control, and collect status feedback features during the control process.

[0071] Standardized control instructions are those with a unified format and specifications that can be accurately recognized and executed by various business systems. Real-time collaborative control refers to the real-time coordination and control of task execution between business systems to achieve cross-system collaboration. State feedback features refer to the operational status characteristic data provided by each business system during the control process, which can reflect the actual operation of the system.

[0072] Converting cross-system collaborative control logic into standardized control instructions requires parsing and processing the logic, converting rules such as task allocation and resource scheduling into specific control instructions. The standardized control instructions are then distributed to the corresponding business systems via a cross-system interaction architecture. Upon receiving the control instructions, each business system executes the corresponding operation and collects real-time status feedback characteristics during the control process.

[0073] As an implementation method, step S300 converts the cross-system collaborative control logic into standardized control instructions, distributes them to the corresponding business systems, and performs real-time collaborative control. State feedback features during the control process are collected. Specifically, the following steps S310 to S340 can be implemented: Step S310: Analyze the task execution process and resource allocation rules in the cross-system collaborative control logic to determine the business system identifier and operation type corresponding to each control node.

[0074] Parsing refers to the analysis and processing of cross-system collaborative control logic to extract key information. The task execution process refers to the task execution sequence and steps specified in the collaborative control logic. The resource allocation rule refers to the resource allocation plan specified in the collaborative control logic. The control node refers to a specific execution point in the collaborative control logic, and each control node corresponds to a specific operation. The business system identifier is an identifier used to uniquely identify each business system, and the operation type refers to the type of operation performed by the control node, such as data reading and writing, task scheduling, etc.

[0075] Parse the cross-system collaborative control logic to identify the task execution process and resource allocation rules. For each control node, determine its corresponding business system identifier and operation type. For example, in a collaborative control logic, it is stipulated that task A is executed by business system A, task B is executed by business system B, and task A requires the use of certain resources. By parsing this logic, determine the business system identifier and operation type corresponding to each control node. For example, the control node of task A corresponds to business system A, and the operation type is task execution; the control node of resource allocation corresponds to the resource management system, and the operation type is resource allocation.

[0076] Step S320: Based on the preset instruction conversion template, the task execution process is converted into a structured control instruction. The structured control instruction includes an instruction header, an instruction body and a check code. The instruction header stores the business system identifier and instruction priority, the instruction body includes specific operation parameters and execution time range, and the check code is used to ensure the integrity of instruction transmission.

[0077] The preset instruction conversion template is a predefined template used to convert task execution processes into structured control instructions, specifying the format and content of the instructions. Structured control instructions are control instructions with a fixed format and structure that can be accurately recognized and executed by various business systems. The instruction header is part of the structured control instruction and contains information such as the business system identifier and instruction priority. The instruction body is the core of the structured control instruction and contains information such as specific operating parameters and execution time range. The checksum is a code used to verify the integrity of the instruction transmission and can be calculated by analyzing the instruction content.

[0078] According to the preset instruction conversion template, each control node in the task execution flow is converted into a corresponding structured control instruction. The business system identifier and the instruction priority are stored in the instruction header. For example, the identifier of the business system A and the priority of the control instruction of the business system A are stored in the instruction header of the control instruction of the business system A. The specific operation parameters and the execution time range are stored in the instruction body. For example, the operation parameter of the task A is the task input data, and the execution time range is from 10:00:00 to 10:30:00. The check code of the instruction is calculated through the cyclic redundancy check (CRC) algorithm, and is added to the instruction.

[0079] As an implementation, in step S320, the task execution flow is converted into a structured control instruction based on the preset instruction conversion template. Specifically, the following steps S321-S324 can be implemented: In step S321, the preset instruction template library is called, and the corresponding instruction template is matched according to the operation type in the task execution flow. The instruction template library includes three basic templates of data read-write type, calculation scheduling type and resource allocation type.

[0080] The preset instruction template library is a library for storing various instruction templates, which includes instruction templates of different operation types. The operation type refers to the type of operation performed by each control node in the task execution flow, such as data read-write, calculation scheduling, resource allocation, etc. According to the operation type in the task execution flow, the corresponding instruction template is selected from the instruction template library. For example, for data read-write operation, the data read-write type instruction template is selected; for calculation scheduling operation, the calculation scheduling type instruction template is selected; for resource allocation operation, the resource allocation type instruction template is selected.

[0081] In step S322, the parameter constraint conditions in the task execution flow are analyzed, and the parameter constraint conditions are mapped to the variable fields of the instruction template. The parameter constraint conditions include resource usage upper limit, task execution deadline and data transmission bandwidth limit.

[0082] The parameter constraint condition refers to the limitation condition imposed on the operation parameter during the task execution process, such as resource usage upper limit, task execution deadline, data transmission bandwidth limit, etc. The variable field refers to the field in the instruction template that can be filled according to the specific situation.

[0083] The parameter constraint conditions in the task execution flow are analyzed and mapped to the variable fields of the instruction template. For example, in the resource allocation type instruction template, there is a variable field for indicating the resource allocation amount. According to the resource usage upper limit in the task execution flow, the upper limit value is filled into the variable field of the instruction template.

[0084] Step S323: Perform syntax check on the mapped instruction template to check whether the parameter format complies with the business system interface specification. If there is a format error, an error prompt is returned and the conversion process is terminated. If the format is correct, a unique instruction identifier is generated. The unique instruction identifier is generated based on the business system identifier, the instruction generation timestamp and a random number combination, and is used to track the execution status of the instruction during the cross-system collaborative control process.

[0085] Syntax validation checks the instruction template to ensure that the parameter format complies with the business system interface specifications. Business system interface specifications refer to the business system's requirements for the format and content of input instructions. The instruction unique identifier uniquely identifies each instruction and helps track the execution status of the instruction during cross-system collaborative control.

[0086] The mapped instruction template is syntax-checked to verify that the parameter format complies with the business system interface specification. For example, it checks whether the resource allocation amount is a positive integer and whether the task execution deadline conforms to the date and time format. If there are format errors, an error message is returned and the conversion process is terminated. If the format is correct, a unique instruction identifier is generated based on the business system identifier, the instruction generation timestamp, and a random number.

[0087] Step S324: Assemble the instruction unique identifier, business system identifier, operation parameters and check code into a structured control instruction in a set order, wherein the check code is calculated by a cyclic redundancy check algorithm.

[0088] The instruction unique identifier, business system identifier, operation parameters, and checksum are assembled in a set order to form a structured control instruction. The cyclic redundancy check (CRC) algorithm is a commonly used checksum algorithm that calculates the checksum based on the instruction content.

[0089] For example, the order of settings is instruction unique identifier, business system identifier, operation parameter, and check code. The instruction unique identifier "SystemA_20240101100000_123", business system identifier "SystemA", operation parameter "{resource_amount:100,task_deadline:'2024-01-0111:00:00'}", and check code "ABCD" calculated by the cyclic redundancy check algorithm are assembled into the structured control instruction "SystemA_20240101100000_123_SystemA_{resource_amount:100,task_deadline:'2024-01-0111:00:00'}_ABCD".

[0090] Step S330: The structured control instructions are routed to the corresponding nodes according to the business system identifiers through the instruction distribution module of the cross-system interaction architecture. The load balancing algorithm is used to allocate the instruction transmission channels during the routing process. After receiving the structured control instructions, each business system executes the corresponding operations and collects the status feedback data of the control process in real time. The status feedback data includes the task completion progress, the actual resource usage and the system response delay.

[0091] The command distribution module distributes control commands within the cross-system interaction architecture, routing commands to corresponding nodes based on business system identifiers. The load balancing algorithm rationally distributes commands based on the load of each channel, improving the efficiency and reliability of command transmission. Status feedback data refers to the operational status data provided by each business system during the control process, reflecting the actual system operation.

[0092] Through the instruction distribution module of the cross-system interaction architecture, structured control instructions are routed to the corresponding nodes according to the business system identifier. During the routing process, a load balancing algorithm is used to allocate instruction transmission channels to ensure efficient and reliable instruction transmission. After receiving the structured control instructions, each business system performs the corresponding operation and collects status feedback data of the control process in real time. For example, for a task execution instruction, the business system executes the task and collects status feedback data such as task completion progress, actual resource usage, and system response delay in real time.

[0093] As an implementation method, in step S330, the load balancing algorithm allocates the instruction transmission channel, which can be specifically implemented as follows: steps S331 to S335: Step S331: monitor the current load rate of each transmission channel in the cross-system interaction architecture in real time. The load rate is defined as the ratio of the channel's used bandwidth to the total bandwidth.

[0094] Real-time monitoring refers to the real-time monitoring and updating of the load status of each transmission channel in the cross-system interaction architecture. The load rate is the ratio of a channel's used bandwidth to its total bandwidth, reflecting the channel's load level. By deploying monitoring devices or programs within the cross-system interaction architecture, the used and total bandwidth of each transmission channel can be monitored in real time. The load rate of each channel is calculated. For example, if Channel A's used bandwidth is 10 Mbps and its total bandwidth is 100 Mbps, then Channel A's load rate is 10%.

[0095] Step S332: Classify the structured control instructions to be transmitted into first-category instructions, second-category instructions, and third-category instructions according to instruction data volume characteristics, and the classification is based on the statistical interval of historical instruction data volume distribution.

[0096] The instruction data volume characteristic refers to the amount of data contained in structured control instructions. The statistical interval of historical instruction data volume distribution refers to the different data volume intervals divided according to the distribution of instruction data volume over a period of time.

[0097] Perform statistical analysis on historical instruction data volumes and classify them into different data size ranges. For example, instructions with a data size of less than 1KB are classified as first-category instructions, instructions with a data size between 1KB and 10KB are classified as second-category instructions, and instructions with a data size greater than 10KB are classified as third-category instructions. Based on these classification criteria, the structured control instructions to be transmitted are categorized.

[0098] Step S333: Set channel load rate thresholds for different types of instructions. In the classification results, instructions with larger data volume characteristics correspond to lower thresholds to ensure that instructions with larger data volume characteristics use low-load channels first.

[0099] The channel load threshold is the upper limit of the channel load ratio set for different types of instructions. Instructions with larger data volumes are assigned lower thresholds to ensure that these instructions are prioritized over lower-load channels, improving instruction transmission efficiency.

[0100] Step S334: traverse all transmission channels and select channels whose load rates are lower than the corresponding threshold and whose bandwidths meet the instruction transmission requirements as candidate channels.

[0101] Traverse all transmission channels and check whether the load rate of each channel is lower than the threshold of the corresponding instruction type and whether the bandwidth meets the instruction transmission requirements. If the conditions are met, the channel is selected as a candidate channel.

[0102] Step S335: If there are multiple candidate channels, the average historical transmission delay of each channel is calculated, and the channel with the smallest delay is selected to assign the instruction. If there is no candidate channel, the instruction is added to the waiting queue and retried periodically.

[0103] The historical average transmission delay refers to the average transmission delay of the channel over the past period of time, which can reflect the transmission performance of the channel. The waiting queue is a queue used to store instructions that cannot be assigned to a channel temporarily.

[0104] If there are multiple candidate channels, the average historical transmission delay of each channel is calculated, and the channel with the smallest delay is selected to assign the instruction. If there are no candidate channels, the instruction is added to the waiting queue and the channel assignment is retried periodically, for example, every 1 minute.

[0105] Step S340: Extract features from the state feedback data, decompose the system response delay data using the wavelet transform algorithm to obtain frequency domain features, calculate the rate of change features of the task completion progress through statistical analysis, generate the distribution features of the actual resource occupancy through histogram statistics, and fuse the frequency domain features, rate of change features, and distribution features to obtain the state feedback features.

[0106] Feature extraction involves extracting key features from state feedback data that reflect the system's operating status. Wavelet transform algorithms decompose time-domain signals into frequency-domain signals, thereby obtaining the signal's frequency-domain characteristics. Statistical analysis involves analyzing data to obtain statistical characteristics, such as mean, variance, and rate of change. Histogram statistics involve grouping data to obtain distributional characteristics.

[0107] When extracting features from state feedback data, the system response delay data is first decomposed using a wavelet transform algorithm to obtain frequency domain features. For example, the system response delay data is treated as a time domain signal and decomposed into components of different frequencies using a wavelet transform algorithm to extract frequency domain features. Next, statistical analysis is performed to calculate the rate of change of task completion progress, for example, calculating the rate of change of task completion progress over a period of time. Next, histogram statistics are used to generate the distribution features of actual resource usage. For example, the actual resource usage is divided into different intervals, and the amount of resource usage within each interval is counted to obtain the distribution features of the actual resource usage. Finally, the frequency domain features, rate of change features, and distribution features are integrated to obtain the state feedback features.

[0108] As an implementation method, step S340 performs feature extraction on the state feedback data, decomposing the system response delay data using a wavelet transform algorithm to obtain frequency domain features, which can be specifically implemented as follows: steps S341 to S345: Step S341: collecting system response delay data within a preset time period to obtain a one-dimensional time series, where the length of the time series is the product of the sampling frequency and the preset time.

[0109] The preset time period refers to the pre-set time range for collecting system response delay data, such as one minute. The sampling frequency refers to the frequency at which data is collected within the preset time period, such as once per second. A one-dimensional time series refers to a sequence of system response delay data arranged in chronological order.

[0110] During a preset time period, the system response delay data is collected at a sampling frequency to obtain a one-dimensional time series. For example, if the preset time period is 1 minute and the sampling frequency is 1 time per second, 60 system response delay data are collected to form a one-dimensional time series with a length of 60.

[0111] Step S342: performing multi-layer wavelet decomposition on the one-dimensional time series based on a preset wavelet basis function to obtain an approximate component and multiple detail components, wherein the approximate component represents the trend characteristics of the system response delay, and the detail component represents the fluctuation characteristics of different frequency bands.

[0112] Wavelet basis functions are the basis functions used in wavelet transform algorithms and determine their properties and effectiveness. Multi-layer wavelet decomposition involves performing multiple wavelet decompositions on a one-dimensional time series to obtain frequency domain features at different scales. Approximate components, the low-frequency components obtained after wavelet decomposition, characterize the trend of system response delay. Detail components, the high-frequency components obtained after wavelet decomposition, characterize the fluctuations of system response delay across different frequency bands.

[0113] Select a preset wavelet basis function, such as the Daubechies wavelet basis function, and perform multi-layer wavelet decomposition on the one-dimensional time series. For example, a three-layer wavelet decomposition yields one approximate component and three detail components. The approximate component reflects the overall trend of the system response delay, such as whether it is increasing or decreasing. The detail component reflects the fluctuations in the system response delay across different frequency bands. For example, high-frequency fluctuations may indicate sudden interference.

[0114] Step S343: Perform Fourier transform on each detail component to convert the time domain signal into a frequency domain signal, and calculate the power spectrum density of each frequency domain signal. The power spectrum density is used to characterize the energy distribution of different frequency components.

[0115] The Fourier transform is a mathematical transformation method used to convert a time-domain signal into a frequency-domain signal. It decomposes the signal into sine and cosine components of different frequencies. The power spectral density (PSD) is the power distribution of different frequency components in a frequency-domain signal. It reflects the energy distribution of the signal at different frequencies.

[0116] Perform a Fourier transform on each detail component to convert it from a time-domain signal to a frequency-domain signal. Then, calculate the power spectral density of each frequency-domain signal. For example, perform a Fourier transform on the detail component using a fast Fourier transform (FFT) algorithm to obtain a frequency-domain signal. Then, calculate the power spectral density by taking the square of the modulus of the frequency-domain signal.

[0117] Step S344: extract the peak frequency, center frequency and bandwidth parameters from the power spectrum density, and combine the peak frequency, center frequency and bandwidth parameters to obtain frequency domain features. The dimension of the frequency domain features is consistent with the number of decomposition layers.

[0118] The peak frequency is the frequency with the highest power in the power spectral density (PSD), reflecting the dominant frequency component of the signal. The center frequency is the frequency at the center of the PSD, reflecting the average frequency of the signal. The bandwidth parameter is the width of the power distribution in the PSD, reflecting the frequency range of the signal.

[0119] Extract the peak frequency, center frequency, and bandwidth parameters from the power spectral density. For example, the peak frequency is obtained by finding the maximum value in the power spectral density; the center frequency is obtained by calculating the weighted average of the power spectral density; and the bandwidth parameter is obtained by calculating the frequency range within the power spectral density where the power exceeds a certain threshold. These parameters are combined to form frequency domain features, whose dimensionality corresponds to the number of decomposition layers. For example, if a three-layer wavelet decomposition is performed, the dimensionality of the frequency domain features is 3.

[0120] Step S345: normalize the frequency domain features, scale each parameter value to a preset numerical range, and use a maximum and minimum value normalization method as the normalization formula.

[0121] Normalization involves processing the parameter values ​​in frequency domain features to have the same dimension and range, enabling effective comparison and analysis. The maximum-minimum normalization method is a commonly used normalization method. It scales data values ​​to a preset range by subtracting the minimum value from the data value and dividing it by the difference between the maximum and minimum values.

[0122] Frequency domain features are normalized using the maximum and minimum value normalization method. For example, the preset value interval is [0, 1]. For each parameter in the frequency domain feature, its maximum and minimum values ​​are calculated. Then, the minimum value is subtracted from each parameter value, and the result is divided by the difference between the maximum and minimum values ​​to obtain the normalized parameter value.

[0123] Step S400: Perform dynamic deviation analysis on the state feedback characteristics to generate collaborative control strategy adjustment parameters.

[0124] Dynamic deviation analysis compares and analyzes state feedback characteristics with historical baseline characteristics to identify deviations and changing trends. Collaborative control strategy adjustment parameters, such as resource allocation ratios and task execution priorities, are used to adjust collaborative control strategy parameters based on the results of dynamic deviation analysis.

[0125] The state feedback features are compared and analyzed with the baseline feature set from the historical collaborative control process, and the deviation between the two is calculated. Through the analysis of the deviation, the trend of the deviation is predicted, and based on the preset deviation threshold, the type and magnitude of the collaborative control parameters that need to be adjusted are determined.

[0126] As an implementation method, step S400 performs dynamic deviation analysis on the state feedback characteristics to generate collaborative control strategy adjustment parameters, which can be specifically implemented as follows: steps S410 to S450: Step S410: Time-series alignment of the state feedback features with a benchmark feature set in a historical collaborative control process, where the benchmark feature set is an average value of the state feedback features collected in historical successful cases.

[0127] Temporal alignment aligns the state feedback features with the baseline feature set to ensure temporal comparability. Historical success cases refer to past successes in cross-system collaborative control, and the baseline feature set is the average value of the state feedback features collected in these cases.

[0128] To time-series align state feedback features with a baseline feature set, the baseline feature set must first be processed, including historical case retrieval, outlier detection, statistical calculation, and length adjustment. The adjusted state feedback features are then aligned with the baseline feature set using a dynamic time warping algorithm.

[0129] As an implementation method, step S410 is to perform time-series alignment of the state feedback feature with the reference feature set in the historical collaborative control process, which can be specifically implemented as follows: steps S411 to S415: Step S411: Retrieve historical cases of the same type as the current collaborative control task from the historical database, extract the state feedback features of each historical case, and obtain an original benchmark feature set.

[0130] The historical database is used to store past cross-system collaborative control cases. It contains detailed information about each case, including state feedback features. The original baseline feature set refers to the state feedback feature set of historical cases of the same type as the current collaborative control task, retrieved from the historical database.

[0131] Based on the type of collaborative control task, relevant historical cases are retrieved from the historical database. For example, if the collaborative control task is production task scheduling, all historical cases of production task scheduling are retrieved from the historical database. State feedback features of these cases are extracted to form the original baseline feature set.

[0132] Step S412: Perform outlier detection on the original benchmark feature set, use statistical methods to identify and eliminate feature samples that deviate from the normal range, and obtain a purified benchmark feature set.

[0133] Outlier detection refers to identifying data points in a dataset that deviate from the normal range. These data points may be caused by measurement error, system failure, or other reasons. Statistical methods refer to statistical analysis methods used to detect outliers, such as those based on standard deviation or quartiles. The purified baseline feature set is the baseline feature set obtained after outlier detection and removal.

[0134] Statistical methods are used to detect outliers in the original baseline feature set. For example, using the standard deviation method, the mean and standard deviation of each feature dimension are calculated. Feature samples that deviate from the mean by more than three standard deviations are identified as outliers and removed. After outlier detection and removal, the purified baseline feature set is obtained.

[0135] Step S413: Calculate the mean value and standard deviation of each time step in the purified benchmark feature set to generate a statistical benchmark feature containing a mean sequence and a standard deviation sequence.

[0136] The mean is the sum of a set of data divided by the number of data points, and it reflects the central tendency of the data. The standard deviation measures the degree of deviation of a set of data from its mean and reflects the degree of dispersion of the data. The mean sequence is the sequence consisting of the mean values ​​of each time step in the purified baseline feature set, and the standard deviation sequence is the sequence consisting of the standard deviations of each time step in the purified baseline feature set. Statistical baseline features are feature sets that include both mean and standard deviation sequences, and they more comprehensively reflect the state feedback characteristics of the historical collaborative control process.

[0137] Calculate the mean and standard deviation for each time step in the cleansed baseline feature set to generate a mean sequence and a standard deviation sequence. For example, for each time step in the cleansed baseline feature set, calculate the mean and standard deviation of all feature samples at that time step. The mean values ​​for each time step are combined into a mean sequence, and the standard deviation values ​​for each time step are combined into a standard deviation sequence. Combine the mean and standard deviation sequences to obtain the statistical baseline feature.

[0138] Step S414: Adjust the time series length of the current state feedback feature to be consistent with the statistical benchmark feature. If the current sequence length is shorter, perform interpolation and completion; if the current sequence length is longer, perform downsampling.

[0139] Time series length adjustment refers to adjusting the time series length of the current state feedback feature to be consistent with the statistical baseline feature to facilitate subsequent time series alignment and comparative analysis. Interpolation refers to inserting new data points into the time series to supplement missing data. Downsampling refers to selecting some data points from the time series to reduce the amount of data.

[0140] Compare the time series length of the current state feedback feature with the time series length of the statistical benchmark feature. If the current series length is shorter, use an interpolation method, such as linear interpolation, to complete it. If the current series length is longer, use a downsampling method, such as average downsampling, to process it.

[0141] Step S415: align the adjusted state feedback features with the mean sequence of the statistical benchmark features using a dynamic time warping algorithm to minimize the cumulative distance between the two sequences and complete nonlinear time alignment.

[0142] The Dynamic Time Warping algorithm minimizes the cumulative distance between two sequences by finding the optimal matching path between them. Cumulative distance is the sum of the distances between points on the matching path between the two sequences in the Dynamic Time Warping algorithm. Nonlinear time alignment aligns two sequences in time, allowing for different time scales between the sequences.

[0143] The mean sequence of the adjusted state feedback features and the statistical benchmark features is input into the dynamic time warping algorithm. The algorithm seeks the optimal matching path to minimize the cumulative distance between the two sequences. For example, the dynamic time warping algorithm can be implemented using dynamic programming to calculate the cumulative distance matrix between the two sequences. The path with the minimum cumulative distance is then found from the matrix to complete the nonlinear time alignment.

[0144] Step S420: Calculate the dimension-by-dimension deviation value between the aligned state feedback feature and the reference feature set to obtain a deviation feature sequence, where each element in the deviation feature sequence represents the degree of feature deviation at the corresponding time step.

[0145] The dimension-by-dimension deviation value refers to the difference between the aligned state feedback features and the baseline feature set in each feature dimension. The deviation feature sequence is a sequence of dimension-by-dimension deviation values, which can reflect the degree of feature deviation at each time step.

[0146] For the aligned state feedback features and baseline feature set, the difference between them is calculated on each feature dimension to obtain a dimension-by-dimension deviation value. The dimension-by-dimension deviation values ​​at each time step are combined to obtain a deviation feature sequence. For example, for a state feedback feature and baseline feature set containing three feature dimensions, the deviation value of each feature dimension is calculated at each time step, resulting in a deviation feature sequence with a length equal to the number of time steps, where each element contains three deviation values.

[0147] Step S430: Trend prediction of the deviation feature sequence is performed through the long short-term memory network. The long short-term memory network includes an input layer, a hidden layer and an output layer. The deviation feature sequence is received through the input layer, the hidden layer memorizes the long-term dependency through the gating mechanism, and the output layer predicts the deviation trend of the future preset time step.

[0148] Long Short-Term Memory (LSTM) networks are capable of processing long-term dependencies in sequential data. The input layer, the first layer of the LSTM, receives input data, specifically a sequence of deviation features. The hidden layer, the core layer of the LSTM, memorizes long-term dependencies through a gating mechanism. The output layer, the final layer of the LSTM, predicts deviation trends at predetermined future time steps based on the state of the hidden layer.

[0149] The deviation feature sequence is fed into the input layer of the LSTM network, which then passes the data to the hidden layer. The hidden layer uses a gating mechanism, including input, forget, and output gates, to control the inflow, outflow, and retention of information, thereby memorizing long-term dependencies. Based on the state of the hidden layer, the output layer predicts the deviation trend for a preset time step in the future.

[0150] As an implementation method, in step S430, the hidden layer memorizes the long-term dependency through a gating mechanism, which can be specifically implemented as follows: steps S431 to S435: Step S431: The input ratio of the feature information at the current moment is adjusted through the input control mechanism, and the input control weight is calculated based on the deviation feature at the current moment and the hidden state at the previous moment to control the degree of inclusion of new information.

[0151] The input control mechanism is the mechanism used in the hidden layer to control the proportion of feature information input at the current moment. The input control weight is the weight used to control the degree of incorporation of new information. It is calculated based on the deviation characteristics of the current moment and the hidden state of the previous moment.

[0152] Based on the current deviation features and the hidden state at the previous moment, a fully connected layer calculates the input control weights. The input control weights are a vector ranging from [0, 1], where each element represents the proportion of new information incorporated in the corresponding feature dimension. For example, an input control weight vector of [0.2, 0.5, 0.8] indicates that the proportion of new information incorporated in the first feature dimension is 20%, the proportion of new information incorporated in the second feature dimension is 50%, and the proportion of new information incorporated in the third feature dimension is 80%.

[0153] Step S432: The retention ratio of historical memory information is adjusted through the forgetting control mechanism, and the forgetting control weight is calculated based on the deviation characteristics of the current moment and the hidden state of the previous moment to control the retention degree of historical memory.

[0154] The forgetting control mechanism is a mechanism used in the hidden layer to control the proportion of historical memory information retained. The forgetting control weight is a weight used to control the degree of historical memory retention. It is calculated based on the deviation characteristics of the current moment and the hidden state of the previous moment.

[0155] Based on the current moment's deviation features and the previous moment's hidden state, a fully connected layer calculates the forgetting control weight. The forgetting control weight is a vector ranging from [0, 1], where each element represents the historical memory retention ratio for the corresponding feature dimension. For example, a forgetting control weight vector of [0.9, 0.7, 0.3] indicates that the historical memory retention ratio for the first feature dimension is 90%, the historical memory retention ratio for the second feature dimension is 70%, and the historical memory retention ratio for the third feature dimension is 30%.

[0156] Step S433: Based on the input control weight and the forgetting control weight, the cell state of the hidden layer is updated, and the new information after input control and the historical memory information after forgetting control are integrated to obtain the cell state at the current moment.

[0157] The cell state is a key state variable in the hidden layer of a long-short-term memory network, used to store historical memory information. New information after input control refers to the deviation feature information at the current moment after being processed by the input control mechanism. Historical memory information after forgetting control refers to the cell state information at the previous moment after being processed by the forgetting control mechanism.

[0158] The current deviation features are weighted according to the input control weights to obtain the new information after input control. The cell state at the previous moment is weighted according to the forgetting control weights to obtain the historical memory information after forgetting control. The new information after input control and the historical memory information after forgetting control are added to obtain the current cell state.

[0159] Step S434: The output ratio of the current memory information is adjusted through the output control mechanism. The output control weight is calculated based on the cell state at the current moment and the hidden state at the previous moment to control the output degree of the current cell state.

[0160] The output control mechanism is the mechanism in the hidden layer that controls the output ratio of the current memory information. The output control weight is the weight used to control the output degree of the current cell state. It is calculated based on the current cell state and the hidden state at the previous moment.

[0161] Based on the current cell state and the hidden state at the previous moment, a fully connected layer calculates the output control weight. The output control weight is a vector ranging from [0, 1], where each element represents the output ratio of the current cell state for the corresponding feature dimension. For example, the output control weight vector is [0.6, 0.4, 0.8], indicating that the output ratio of the current cell state for the first feature dimension is 60%, the output ratio of the current cell state for the second feature dimension is 40%, and the output ratio of the current cell state for the third feature dimension is 80%.

[0162] Step S435: Convert the cell state after output control to the hidden state at the current moment to remember the long-term dependency in the deviation feature sequence.

[0163] The cell state after output control refers to the cell state at the current moment after being processed by the output control mechanism. The hidden state at the current moment refers to the hidden state of the long short-term memory network at the current moment, which is used to remember the long-term dependencies in the bias feature sequence.

[0164] The cell state after output control is transformed through an activation function (such as the tanh function) to obtain the hidden state at the current moment. This hidden state serves as the input for the next moment's calculation and participates in subsequent computations, helping the LSTM network to memorize long-term dependencies in the deviation feature sequence. Through this gating mechanism, the LSTM network can effectively handle long-term dependencies in sequence data, making the prediction of deviation trends more accurate and reliable.

[0165] Step S440: Based on the deviation trend and the preset deviation threshold, determine the type of collaborative control parameters that need to be adjusted. The parameter types include resource allocation ratio, task execution priority, and data transmission bandwidth.

[0166] The deviation trend is the deviation change at a preset future time step predicted by the long short-term memory network. It reflects the development direction and degree of the deviation between the state feedback characteristics and the baseline characteristics. The preset deviation threshold is a pre-set limit value used to determine whether the deviation needs to be adjusted. Different deviation thresholds correspond to different collaborative control parameter adjustment strategies. Collaborative control parameter types refer to the types of parameters that can be adjusted during the cross-system collaborative control process. For example, the resource allocation ratio determines the amount of resources that each business system can obtain, the task execution priority affects the order in which tasks are executed, and the data transmission bandwidth is related to the rate and efficiency of data transmission between systems.

[0167] Determining the type of collaborative control parameters that need to be adjusted based on the deviation trend and the preset deviation threshold requires comprehensive consideration of multiple factors and a series of analysis steps.

[0168] As an implementation method, step S440 determines the type of collaborative control parameter that needs to be adjusted based on the deviation trend and the preset deviation threshold, which can be specifically implemented as follows: steps S441 to S445: Step S441: Preset multiple deviation threshold intervals, each threshold interval corresponding to a different parameter adjustment level.

[0169] The purpose of presetting multiple deviation threshold intervals is to determine different adjustment strategies more carefully according to the size of the deviation. Different deviation threshold intervals represent different degrees of severity of the deviation, and the parameter adjustment level corresponding to each threshold interval reflects the adjustment intensity required for the severity of the deviation. For example, three deviation threshold intervals can be preset: the interval less than the first threshold is a mild deviation interval, corresponding to a lower parameter adjustment level, such as only slight parameter fine-tuning is required; the interval between the first threshold and the second threshold is a moderate deviation interval, corresponding to a medium parameter adjustment level, and a certain degree of parameter adjustment may be required; the interval greater than the second threshold is a severe deviation interval, corresponding to a higher parameter adjustment level, and a larger adjustment of the parameters is required.

[0170] Step S442: Match each predicted deviation value in the deviation trend with a preset deviation threshold interval to determine the parameter adjustment level for each time step.

[0171] For each predicted deviation value in the deviation trend predicted by the LSTM network, it is compared with the preset deviation threshold interval to determine which threshold interval the deviation value belongs to, thereby determining the corresponding parameter adjustment level. For example, if a predicted deviation value is less than the first threshold, the corresponding parameter adjustment level for that time step is the mild adjustment level; if the predicted deviation value is between the first and second thresholds, the corresponding parameter adjustment level is the moderate adjustment level; if the predicted deviation value is greater than the second threshold, the corresponding parameter adjustment level is the severe adjustment level. In this way, the corresponding parameter adjustment level is determined for each time step, which facilitates further analysis.

[0172] Step S443: Count the occurrence frequencies of the adjustment levels of each parameter type within a preset time window, and determine the parameter type with the highest frequency as the main adjustment target.

[0173] The preset time window is a fixed time range used to count the frequency of each parameter type's adjustment level. The number of times each parameter type's different adjustment levels appear within this time window is counted. For example, for the resource allocation ratio parameter type, the frequency of each adjustment level occurring within the preset time window is counted for light, moderate, and heavy, respectively. These frequencies are then added together to obtain the total frequency for that parameter type. This statistics is performed for all parameter types (resource allocation ratio, task execution priority, and data transmission bandwidth). The total frequencies of each parameter type are then compared, and the parameter type with the highest frequency is identified as the primary adjustment target. This is because this parameter type requires adjustment most frequently within the preset time window and is likely the primary factor causing the deviation, so it is adjusted first.

[0174] Step S444: Analyze the historical adjustment effect data of the main adjustment object. If the deviation reduction rate after parameter adjustment in the historical adjustment effect data is greater than a preset threshold, retain the parameter type; otherwise, replace it with a second high-frequency parameter type.

[0175] Historical adjustment effect data refers to the change in the deviation between the state feedback characteristics and the baseline characteristics after the parameters of the main adjustment object were adjusted in the past. The deviation reduction rate refers to the proportion of the deviation reduction after the parameter adjustment. For example, if the deviation before adjustment is 10 and the deviation after adjustment is 5, the deviation reduction rate is (10-5) / 10=50%. The preset threshold is a pre-set boundary value used to determine whether the historical adjustment effect is effective. If the historical adjustment effect data of the main adjustment object shows that the deviation reduction rate after its parameter adjustment is greater than the preset threshold, it means that the adjustment of this parameter type is effective and can reduce the deviation, so this parameter type is retained as the main adjustment object. Conversely, if the deviation reduction rate is less than or equal to the preset threshold, it means that the adjustment effect of this parameter type is not good, and it is necessary to replace it with a less frequently used parameter type as the main adjustment object and try again.

[0176] Step S445: perform correlation analysis on the determined parameter types to determine whether there are associated parameters that need to be collaboratively adjusted. If so, add the associated parameters to the adjustment list to obtain a final collaborative control parameter type set.

[0177] Correlation analysis refers to analyzing whether there is a mutual correlation and influence relationship between the determined parameter type and other parameter types. In cross-system collaborative control, there may be close connections between certain parameter types. For example, the adjustment of the resource allocation ratio may affect the task execution priority and data transmission bandwidth. By performing a correlation analysis on the determined parameter type, it is determined whether there are other parameter types that are associated with it and need to be adjusted at the same time. If there are such associated parameters, add them to the adjustment list. The final adjustment list constitutes the final collaborative control parameter type set, which contains all collaborative control parameter types that need to be adjusted so that these parameters can be adjusted accordingly in subsequent steps.

[0178] Step S450: For each parameter type that needs to be adjusted, calculate the adjustment amplitude value. The size of the adjustment amplitude value is positively correlated with the slope of the deviation trend. The adjustment direction is determined according to the positive or negative sign of the deviation value. The parameter type and the corresponding adjustment amplitude value are combined into the collaborative control strategy adjustment parameter.

[0179] The adjustment amplitude value refers to the degree of adjustment determined for each type of collaborative control parameter that needs to be adjusted. The slope of the deviation trend reflects the rate at which the deviation changes over time. The larger the slope, the more drastic the deviation change, and accordingly, a larger adjustment amplitude is required to correct the deviation. Therefore, the size of the adjustment amplitude value is positively correlated with the slope of the deviation trend. The adjustment direction is determined by the sign of the deviation value. If the deviation value is positive, it means that the value of the state feedback feature is greater than the value of the baseline feature. At this time, the adjustment direction should be to reduce the parameter value; if the deviation value is negative, it means that the value of the state feedback feature is less than the value of the baseline feature. The adjustment direction should be to increase the parameter value.

[0180] For each parameter type requiring adjustment (such as resource allocation ratio, task execution priority, and data transmission bandwidth), the adjustment amplitude is first calculated based on the slope of the deviation trend. For example, a linear function relationship is established: adjustment amplitude = slope × proportional coefficient (the proportional coefficient is a pre-set constant that controls the range of the adjustment amplitude). The adjustment direction is then determined based on the sign of the deviation. Finally, the parameter type and the corresponding adjustment amplitude are combined to form the collaborative control strategy adjustment parameters. For example, for the resource allocation ratio parameter type, if the calculated adjustment amplitude is 10%, and the deviation value indicates an increase in the adjustment direction, the resource allocation ratio portion of the collaborative control strategy adjustment parameters can be expressed as "Resource allocation ratio: increase by 10%." By combining all parameter types requiring adjustment and their adjustment amplitudes in this manner, the complete collaborative control strategy adjustment parameters are obtained.

[0181] Step S500: Synchronize the collaborative control strategy adjustment parameters to each business system through the cross-system interaction architecture.

[0182] The cross-system interaction architecture is the infrastructure for data transmission and communication between multiple business systems. It provides a channel and mechanism for synchronizing collaborative control strategy adjustment parameters. Synchronizing these collaborative control strategy adjustment parameters to each business system ensures that each system promptly adjusts its operating status and behavior based on the new adjustment parameters, thereby achieving cross-system collaborative control optimization. This process requires a series of operations, including encoding, grouping, transmitting, and synchronizing the collaborative control strategy adjustment parameters.

[0183] As an implementation method, step S500, synchronizing the collaborative control strategy adjustment parameters to each business system through the cross-system interaction architecture, can be specifically implemented as follows: steps S510 to S540: Step S510: Encode the cooperative control strategy adjustment parameter, and convert the parameter type and adjustment amplitude value into a binary data stream using a variable-length coding method.

[0184] Variable-length coding is a coding method that assigns codes of varying lengths based on the characteristics and frequency of the data. It flexibly allocates code lengths based on the specific parameter type and adjustment amplitude, effectively reducing the storage space and transmission bandwidth requirements for the encoded data. The parameter types (such as resource allocation ratios, task execution priorities, and data transmission bandwidth) and adjustment amplitude values ​​(such as specific adjustment percentages or numerical values) in the collaborative control strategy adjustment parameters are converted using variable-length coding to produce a binary data stream. For example, the resource allocation ratio parameter type and its adjustment amplitude value of 10% are converted into corresponding binary code combinations according to variable-length coding rules. After all parameter types and adjustment amplitude values ​​that require adjustment undergo this coding conversion, a complete binary data stream is formed.

[0185] Step S520: The encoded binary data stream is grouped by business system through the synchronization control module of the cross-system interaction architecture, and each group of data contains all parameter information that needs to be adjusted for the business system.

[0186] The synchronization control module is a module specifically used to implement data synchronization functions in the cross-system interaction architecture. It is responsible for processing and grouping the encoded binary data stream. According to the needs of each business system, the encoded binary data stream is grouped according to the business system. Each business system corresponds to a set of data, which contains information about all the types of collaborative control parameters that need to be adjusted by the business system and their adjustment amplitude values. For example, for business system A, it needs to adjust the resource allocation ratio and task execution priority. Then, when grouping, the encoded data of the resource allocation ratio and task execution priority related to business system A are combined into a set of data and sent specifically to business system A. Such grouping can ensure that each business system only receives adjustment information related to itself, thereby improving the efficiency and pertinence of data transmission.

[0187] Step S530: Use asynchronous transmission mode to send packet data to the corresponding business system. If the recipient does not return confirmation information within the preset time, the data is resent. After each business system receives the parameter adjustment data, it decodes and updates the local collaborative control strategy. Atomic operations are used during the update process to ensure the consistency of strategy replacement and avoid intermediate states.

[0188] Asynchronous transfer mode is a data transmission mode, in which the sender can continue to perform other operations without waiting for the response of the receiver after sending data, thereby improving the efficiency of data transmission. The grouped data is sent to the corresponding business system through the cross-system interaction architecture. At the same time, a timer is started, and a preset time is set. If the receiver returns the confirmation information containing the correct identification within the preset time, it means that the data transmission is successful; if no confirmation information is received within the preset time, it means that there is a loss or error in the data transmission process, and the data is re-sent until the confirmation information is received.

[0189] After each business system receives the parameter adjustment data, it first performs decoding processing to convert the binary data stream into readable collaborative control parameter type and adjustment amplitude value information. Then, the local collaborative control strategy is updated according to this information. During the updating process, atomic operations are used. Atomic operations are indivisible operations that are either completely completed or not completed at all, which can ensure that there is no intermediate state of partial update when updating the local collaborative control strategy, and ensure the consistency and integrity of the strategy replacement. For example, when updating the resource allocation ratio and task execution priority, either both parameters are successfully updated or none of them is updated, avoiding the inconsistent situation of one parameter being updated and the other parameter not being updated.

[0190] As an implementation, in step S530, the asynchronous transmission mode is used to send the grouped data to the corresponding business system, which can be implemented as steps S531-S535 as follows: Step S531: Assign a transmission sequence number to each grouped data, and the sequence number is incremented in the sending order, which is used to identify the order of data transmission.

[0191] The transmission sequence number is a unique number used to distinguish different grouped data and identify their sending order. An incremental transmission sequence number is assigned to each grouped data according to the sending order, for example, the sequence number of the first grouped data is 1, the sequence number of the second grouped data is 2, and so on. This sequence number plays an important role in the data transmission process, and the receiver can determine the integrity and order of the data according to the sequence number to ensure that the received data is arranged in the correct order. At the same time, in the case of data retransmission, the sequence number is also helpful to distinguish between newly sent data and retransmitted data.

[0192] Step S532: Combine the grouped data, transmission sequence number, and checksum into a transmission frame, and send it to the target business system through the cross-system interaction architecture.

[0193] A checksum is a value used to verify data integrity. It is calculated by performing a specific calculation on packet data. The packet data, transmission sequence number, and checksum are combined to form a transmission frame. A transmission frame is the basic unit of data transmission, containing the data itself and related information to ensure the accuracy and integrity of the data. The transmission frame is sent to the target business system via the cross-system interaction architecture. During the transmission process, the cross-system interaction architecture selects the appropriate transmission channel and protocol based on the network status and configuration, ensuring that the transmission frame reaches the target business system efficiently and reliably.

[0194] Step S533: Start the timer. If a confirmation frame containing the same transmission sequence number is received from the target service system within a preset timeout period, the data transmission is considered successful. Otherwise, the packet data is marked as to be retransmitted.

[0195] The preset timeout is a pre-set limit used to determine if a data transmission has timed out. A timer is started, waiting for the target service system to return an acknowledgment frame. An acknowledgment frame is a response message sent by the target service system after successfully receiving a transmission frame. It contains the same transmission sequence number as the transmitted transmission frame. If an acknowledgment frame containing the same transmission sequence number is received within the preset timeout, the target service system has successfully received the packet data and the data transmission is successful. If no such acknowledgment frame is received within the preset timeout, the packet data is marked as pending for retransmission, allowing for subsequent retransmission.

[0196] Step S534: For the packet data marked as to be retransmitted, a backoff algorithm is used to calculate the retransmission delay time. When the number of retransmissions reaches a preset upper limit and still fails, a link failure alarm is triggered and the backup transmission channel is switched.

[0197] A backoff algorithm is used to prevent network congestion caused by multiple simultaneous data retransmissions. It calculates retransmission delays based on specific rules. For packets marked for retransmission, the backoff algorithm calculates the delay for the next retransmission. For example, a common backoff algorithm increases the delay as the number of retransmissions increases, preventing network congestion caused by large amounts of simultaneous data retransmissions when network problems occur.

[0198] The preset upper limit is the maximum number of retransmissions. If the number of retransmissions reaches the preset limit and no confirmation frame is received from the target service system, the current transmission channel may be faulty. This triggers a link failure alarm, notifying the system administrator or relevant monitoring system. Simultaneously, the system switches to the backup transmission channel and retryes to ensure continued data transmission.

[0199] Step S535: After all packet data transmission is completed, a synchronization completion instruction is sent to each service system, instructing each system to simultaneously take effect on the new collaborative control strategy.

[0200] Once all packet data has been successfully transmitted and confirmed by the target business system, a synchronization completion command is sent to each business system. This command notifies each business system that all parameter adjustment data has been transmitted and that the new collaborative control strategy can be implemented simultaneously. This ensures that each business system begins operating according to the new collaborative control strategy at the same time, achieving cross-system collaborative control synchronization and improving overall system performance and stability.

[0201] Step S540: Collect the system operation status data after parameter adjustment, calculate the collaborative control efficiency index before and after strategy optimization, and complete the dynamic optimization if the efficiency index improvement is greater than the preset threshold. Otherwise, return to the adjustment parameter generation step to recalculate the adjustment amplitude value.

[0202] System operating status data refers to the actual operational status of each business system after parameter adjustments, such as task completion time, resource utilization, and system response time. The collaborative control efficiency index is a quantitative metric used to measure the effectiveness of cross-system collaborative control. It comprehensively considers multiple factors, such as the effective utilization of resources and the efficiency of task execution. By collecting system operating status data after parameter adjustments, the collaborative control efficiency index after strategy optimization is calculated and compared with the index before strategy optimization.

[0203] The preset threshold is a pre-set limit used to determine whether the collaborative control strategy optimization is effective. If the calculated improvement in the collaborative control efficiency index is greater than the preset threshold, it indicates that the collaborative control effect of the system has been significantly improved by adjusting the collaborative control parameters, and the dynamic optimization process has been completed. If the improvement is less than or equal to the preset threshold, it indicates that the current collaborative control strategy adjustment has not achieved the expected effect, and it is necessary to return to the adjustment parameter generation step (step S400), re-perform dynamic deviation analysis based on the new system operating status data, calculate the adjustment amplitude value, and try to adjust the collaborative control strategy again until the improvement in the collaborative control efficiency index exceeds the preset threshold. This continuous dynamic adjustment and optimization can ensure that cross-system collaborative control always maintains a high level of efficiency.

[0204] It can be understood that, in the above introduction of the embodiments of the present application, various algorithms involved, such as the Euclidean distance algorithm, the cosine distance algorithm, the interpolation algorithm, the backoff algorithm and the like, can be known from the related content in the prior art, and in order to save space, they are not expanded too much in the embodiments of the present application. In addition, those skilled in the art can supplement details according to common knowledge in the art when implementing the scheme of the present application, for example, according to common knowledge in the art, normalization can be used to eliminate the dimensional conflict before feature fusion, interpolation can be used to eliminate the dimensional difference, historical data, experience or business scenario requirements can be combined to reasonably set the threshold, a general model training method can be used to train the model, the number of layers in the model structure can be set based on actual needs, the activation function can be selected, and the like. The present application no longer introduces the redundant implementation process in too much detail.

[0205] Please refer to Figure 2 , Figure 2 A structural schematic diagram of a computer system provided by the embodiments of the present application is shown in the figure, which at least includes a processor 101, a communication interface 102 and a memory 103. The processor 101, the communication interface 102 and the memory 103 can be connected through a bus or other means. The processor 101 (or Central Processing Unit, CPU) is the computing core and control core of the computer system, which can parse various instructions in the computer system and process various data of the computer system. The communication interface 102 can optionally include a standard wired interface, a wireless interface (such as WI-FI, a mobile communication interface, etc.), and can be used for transmitting and receiving data under the control of the processor 101; the communication interface 102 can also be used for the transmission and interaction of data within the computer system. The memory 103 is a memory device in the computer system, used for storing programs and data. It can be understood that the memory 103 here can include the built-in memory of the computer system, and of course can also include the extended memory supported by the computer system. The memory 103 provides a storage space that stores the operating system of the computer system, and the present application does not limit this.

[0206] In one embodiment, the processor 101 executes the computer program in the memory 103 to perform the cross-system collaborative control method based on digital threads provided by the embodiments of the present application.

Claims

1. A cross-system collaborative control method based on digital thread, characterized in that: include: Build a cross-system interaction architecture and establish real-time data transmission and instruction interaction between multiple business systems through distributed node communication protocols; Based on the cross-system interaction architecture, dynamic operation characteristic data of each business system is collected to generate cross-system collaborative control logic; Convert the cross-system collaborative control logic into standardized control instructions, distribute them to the corresponding business systems, and execute real-time collaborative control, collecting state feedback features during the control process; Performing dynamic deviation analysis on the state feedback characteristics to generate collaborative control strategy adjustment parameters; The collaborative control strategy adjustment parameters are synchronized to each business system through the cross-system interaction architecture.

2. The method according to claim 1, wherein The step of collecting dynamic operation characteristic data of each business system based on the cross-system interaction architecture and generating cross-system collaborative control logic includes: Collect dynamic operation characteristic data of each business system at preset time intervals through the distributed collection nodes in the cross-system interaction architecture; Performing spatiotemporal alignment processing on the dynamic operation feature data, mapping the feature data of different business systems to a unified time coordinate system, and generating a spatiotemporal alignment feature set; Based on the spatiotemporal alignment feature set, the association dependency features between the business systems are extracted through the attention mechanism. The association dependency features represent the resource competition relationship and task coupling degree of different business systems during task execution. The associated dependency features are input into a pre-trained collaborative logic generation network, which includes a feature encoding layer, a dependency reasoning layer, and a logic output layer. The associated dependency features are converted into high-dimensional semantic vectors through the feature encoding layer. The dependency reasoning layer performs node relationship reasoning on the high-dimensional semantic vectors through a graph neural network. The logic output layer generates an initial framework of the cross-system collaborative control logic based on the reasoning results. Conflict detection is performed on the initial framework to identify resource allocation conflicts and task execution timing conflicts in the collaborative control logic. Based on the conflict detection results, the task priorities and resource allocation ratios in the initial framework are adjusted to obtain the final cross-system collaborative control logic.

3. The method according to claim 2, wherein The performing of spatiotemporal alignment processing on the dynamic operation feature data, mapping the feature data of different business systems to a unified time coordinate system, and generating a spatiotemporal alignment feature set includes: Extracting the timestamp information from the dynamic operation characteristic data to determine the collection time and data sampling period of each business system characteristic data; Based on the collection time and data sampling period, a time base axis is constructed, wherein the minimum time unit of the time base axis is a common divisor of the sampling periods of each business system; Interpolation processing is performed on the dynamic operation characteristic data of each business system, and characteristic data of missing moments are supplemented on the time base axis to generate a characteristic sequence with equal time intervals; Perform dimension splicing on the feature sequences of equal time intervals of each business system according to the system identifier to obtain a three-dimensional feature tensor including the system dimension, the time dimension and the feature dimension, and use the three-dimensional feature tensor as the spatiotemporal alignment feature set; The process of extracting correlation dependency features between business systems based on the spatiotemporal alignment feature set through an attention mechanism includes: Splitting the spatiotemporal alignment feature set into a plurality of time window feature subsets according to the time dimension, each time window feature subset comprising feature data of a preset number of consecutive time steps; For each feature subset of a time window, the similarity weight between feature data of different business systems is calculated through the self-attention mechanism. The similarity weight represents the degree of correlation between feature data of two business systems in the same time window. Based on the similarity weight, weighted aggregation is performed on the feature data of each business system to generate an interaction correlation vector between the business systems, where the dimension of the interaction correlation vector is consistent with the number of business systems; The interactive correlation vectors are spliced ​​in the order of time windows to obtain a correlation dependency feature sequence, wherein each element in the correlation dependency feature sequence corresponds to a business system correlation relationship within a time window; The association dependency feature sequence is filtered to eliminate the interference of short-term fluctuations on the association relationship extraction.

4. The method according to claim 1, wherein The cross-system collaborative control logic is converted into standardized control instructions, distributed to the corresponding business system and performs real-time collaborative control, and the state feedback characteristics during the control process are collected, including: Analyze the task execution process and resource allocation rules in the cross-system collaborative control logic to determine the business system identifier and operation type corresponding to each control node; Based on a preset instruction conversion template, the task execution process is converted into a structured control instruction. The structured control instruction includes an instruction header, an instruction body, and a check code. The instruction header stores the business system identifier and instruction priority, the instruction body contains specific operation parameters and execution time range, and the check code is used to ensure the integrity of instruction transmission; The structured control instructions are routed to the corresponding nodes according to the business system identifiers through the instruction distribution module of the cross-system interaction architecture. During the routing process, a load balancing algorithm is used to allocate instruction transmission channels. After receiving the structured control instructions, each business system executes the corresponding operation and collects status feedback data of the control process in real time. The status feedback data includes task completion progress, actual resource usage, and system response delay. Feature extraction is performed on the state feedback data, and the system response delay data is decomposed by a wavelet transform algorithm to obtain frequency domain features. The change rate features of the task completion progress are calculated through statistical analysis, and the distribution features of the actual resource occupancy are generated through histogram statistics. The frequency domain features, the change rate features and the distribution features are fused to obtain the state feedback features.

5. The method according to claim 4, wherein The step of converting the task execution process into structured control instructions based on a preset instruction conversion template includes: Calling a preset instruction template library to match the corresponding instruction template according to the operation type in the task execution process. The instruction template library contains three basic templates: data reading and writing, calculation scheduling, and resource allocation. Parsing parameter constraints in the task execution process and mapping the parameter constraints to variable fields of the instruction template, wherein the parameter constraints include resource usage upper limit, task execution deadline, and data transmission bandwidth limit; Perform syntax check on the mapped instruction template to check whether the parameter format complies with the business system interface specification. If there is a format error, an error prompt is returned and the conversion process is terminated. If the format is correct, a unique instruction identifier is generated based on the business system identifier, the instruction generation timestamp, and a random number. The instruction unique identifier, the business system identifier, the operation parameter and the check code are assembled into a structured control instruction in a set order.

6. The method according to claim 4, wherein The feature extraction of the state feedback data and the decomposition of the system response delay data by a wavelet transform algorithm to obtain frequency domain features include: Collecting system response delay data within a preset time period to obtain a one-dimensional time series, wherein the length of the time series is the product of the sampling frequency and the preset time; Performing multi-layer wavelet decomposition on the one-dimensional time series based on a preset wavelet basis function to obtain an approximate component and multiple detail components, wherein the approximate component represents the trend characteristics of the system response delay, and the detail components represent the fluctuation characteristics of different frequency bands; Performing Fourier transform on each of the detail components to convert the time domain signal into a frequency domain signal, and calculating the power spectral density of each frequency domain signal, wherein the power spectral density is used to characterize the energy distribution of different frequency components; Extracting the peak frequency, center frequency, and bandwidth parameters from the power spectrum density, and combining the peak frequency, center frequency, and bandwidth parameters to obtain the frequency domain features, where the dimension of the frequency domain features is consistent with the number of decomposition layers; The frequency domain features are normalized and the parameter values ​​are scaled to a preset numerical range.

7. The method according to claim 1, wherein The performing of dynamic deviation analysis on the state feedback characteristics to generate collaborative control strategy adjustment parameters includes: Time-series alignment of the state feedback features with a benchmark feature set from a historical collaborative control process, where the benchmark feature set is an average value of the state feedback features collected from historical successful cases; Calculating the dimension-by-dimension deviation between the aligned state feedback features and the reference feature set to obtain a deviation feature sequence, wherein each element in the deviation feature sequence represents the degree of feature deviation at the corresponding time step; The deviation feature sequence is trend predicted by a long short-term memory network, wherein the long short-term memory network comprises an input layer, a hidden layer, and an output layer. The input layer receives the deviation feature sequence, the hidden layer memorizes the long-term dependency through a gating mechanism, and the output layer predicts the deviation trend of a preset time step in the future; Based on the deviation trend and a preset deviation threshold, determining the type of collaborative control parameters that need to be adjusted, the parameter types including resource allocation ratio, task execution priority, and data transmission bandwidth; For each parameter type that needs to be adjusted, the adjustment amplitude value is calculated. The size of the adjustment amplitude value is positively correlated with the slope of the deviation trend. The adjustment direction is determined according to the positive or negative deviation value. The parameter type and the corresponding adjustment amplitude value are combined into the collaborative control strategy adjustment parameter.

8. The method according to claim 7, wherein The step of performing time-series alignment on the state feedback feature and the benchmark feature set in the historical collaborative control process includes: Retrieve historical cases of the same type as the current collaborative control task from the historical database, extract the state feedback features of each historical case, and obtain the original benchmark feature set; Performing outlier detection on the original benchmark feature set, using statistical methods to identify and eliminate feature samples that deviate from a normal range, to obtain a purified benchmark feature set; Calculate the mean and standard deviation of each time step in the purified benchmark feature set to generate a statistical benchmark feature containing a mean sequence and a standard deviation sequence; Adjust the time series length of the current state feedback feature to be consistent with the statistical benchmark feature; The dynamic time warping algorithm is used to align the mean sequence of the adjusted state feedback features with the statistical benchmark features, minimizing the cumulative distance between the two sequences and completing nonlinear time alignment. The determining of the type of collaborative control parameter that needs to be adjusted based on the deviation trend and a preset deviation threshold includes: Multiple deviation threshold intervals are preset, and each threshold interval corresponds to a different parameter adjustment level; Matching each predicted deviation value in the deviation trend with a preset deviation threshold interval to determine the parameter adjustment level for each time step; Count the frequency of occurrence of adjustment levels for each parameter type within a preset time window, and determine the parameter type with the highest frequency as the main adjustment target; Analyze the historical adjustment effect data of the main adjustment objects. If the deviation reduction rate after parameter adjustment in the historical adjustment effect data is greater than the preset threshold, retain the parameter type; otherwise, replace it with a less frequent parameter type. Perform correlation analysis on the determined parameter types to determine whether there are associated parameters that need to be coordinated and adjusted. If so, add the associated parameters to the adjustment list to obtain the final set of coordinated control parameter types.

9. The method according to claim 1, wherein The step of synchronizing the collaborative control strategy adjustment parameters to each business system through the cross-system interaction architecture to complete dynamic optimization of cross-system collaborative control includes: Encoding the collaborative control strategy adjustment parameters, and converting the parameter type and adjustment amplitude value into a binary data stream using a variable-length encoding method; The coded binary data stream is grouped by business system through the synchronization control module of the cross-system interaction architecture, and each group of data contains all parameter information that needs to be adjusted for the business system; The asynchronous transmission mode is used to send packet data to the corresponding business system. If the receiver does not return a confirmation message within the preset time, the data is resent. After receiving the parameter adjustment data, each business system decodes it and updates the local collaborative control strategy. Collect the system operation status data after parameter adjustment, calculate the collaborative control efficiency index before and after strategy optimization, and complete the dynamic optimization if the improvement in efficiency index is greater than the preset threshold. Otherwise, return to the adjustment parameter generation step to recalculate the adjustment amplitude value.

10. A computer system, characterized in that: include: a memory storing a computer program; A processor, configured to load the computer program to implement the cross-system collaborative control method based on digital threads as described in any one of claims 1 to 9.

Images