Agent-driven heterogeneous device protocol self-adaptation operation and maintenance method and system

Abstract

The application provides an AGENT driven heterogeneous device protocol adaptive operation and maintenance method and system, relates to the technical field of industrial internet, and the method acquires a spatial topological relationship between a first heterogeneous device and a second heterogeneous device; determines a predicted time window in which a target interference event propagates to the second heterogeneous device along the topological relationship based on a running characteristic sequence of the first heterogeneous device; determines a set of restricted channel load constraint parameters within the window based on target interference event attribute information and a basic protocol specification of the second heterogeneous device; generates a protocol reconstruction instruction by an operation and maintenance intelligent agent, and dynamically reconstructs a communication specification of the second heterogeneous device into a restricted matching specification according to the protocol reconstruction instruction, and executes within the predicted time window. The scheme improves the bottom layer transmission reliability under the condition of sudden physical interference, guarantees the complete acquisition of operation and maintenance characteristic data, and further provides high-fidelity bottom layer data support for the local physical degradation trend evaluation and residual life prediction of the device.

Description

Technical Field

[0001] This application relates to the field of industrial internet technology, and more specifically, to an agent-driven heterogeneous device protocol adaptive operation and maintenance method and system. Background Technology

[0002] In industrial internet and process industry sites, there are often heterogeneous devices such as multiple types of sensors, actuators, and edge controllers deployed in parallel. These devices use different underlying communication protocols and data frame formats. The operation and maintenance side needs to continuously collect operational status, degradation symptoms, and diagnostic characteristics to support condition monitoring and lifespan prediction. In actual deployments, operating conditions such as pump start-up and shutdown, valve switching, fluid transient fluctuations, and high-voltage equipment switching can cause electromagnetic interference, structural vibration, or media disturbances, leading to sudden packet loss, retransmission surges, or load limitations in local links within a short period of time. Existing solutions mostly rely on statistical indicators such as RSSI, bit error rate, and packet loss rate for passive rate limiting, retransmission, or simple protocol parameter adjustments. These solutions often fail to match the timing and intensity changes of sudden interference in a timely manner, easily resulting in excessive degradation or delayed adjustments, leading to the loss, truncation, or incomplete acquisition of key operation and maintenance characteristic data during the interference period. Because the transient degradation characteristics hidden inside the equipment are most easily exposed in the instant of a sudden physical impact, the loss of such high-fidelity data will cause the operation and maintenance system to miss the "golden window" for diagnosis, thereby restricting the accuracy of the assessment of the local physical degradation trend of the equipment and the reliability of the remaining life prediction.

[0003] Therefore, the urgent technical problem to be solved is how to improve the reliability of underlying data transmission of heterogeneous equipment in industrial fields when they encounter sudden physical interference, so as to ensure the complete acquisition of operation and maintenance characteristic data, and thus support high-confidence equipment degradation trend analysis and life prediction. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this application provides an agent-driven heterogeneous device protocol adaptive operation and maintenance method and system.

[0005] Firstly, this application provides an agent-driven adaptive operation and maintenance method for heterogeneous device protocols, including:

[0006] Obtain the spatial topology relationship between the first heterogeneous device and the second heterogeneous device in the target operating environment;

[0007] Based on the first operating characteristic sequence of the first heterogeneous device, a prediction time window is determined for the physical propagation of the target interference event to the second heterogeneous device along the spatial topological relationship.

[0008] Based on the attribute information of the target interference event and the basic protocol specifications of the second heterogeneous device, the set of constrained channel load parameters within the prediction time window is determined;

[0009] The operation and maintenance intelligent agent generates protocol reconfiguration instructions based on the set of constrained channel load parameters;

[0010] Based on the protocol reconstruction instruction, the communication protocol of the second heterogeneous device is dynamically reconstructed from the basic protocol protocol to a restricted matching protocol, and the restricted matching protocol is executed within the prediction time window.

[0011] Optionally, the set of constrained channel load parameters includes: upper limit of single frame load length, upper limit of transmission duty cycle, and upper limit of retransmission budget.

[0012] Optionally, the prediction time window for determining the physical propagation of the target interference event to the second heterogeneous device along the spatial topology includes:

[0013] Extract mutation feature vectors from the first running feature sequence and determine the propagation speed parameter and duration parameter;

[0014] Based on the propagation speed parameter, the relative spatial distance between the first heterogeneous device and the second heterogeneous device, and the duration parameter, the arrival timestamp and departure timestamp are calculated, and the prediction time window is constructed from the arrival timestamp and departure timestamp.

[0015] Optionally, the restricted matching specification limits:

[0016] The data frame template corresponding to the upper limit of single frame payload length, the transmission scheduling parameters corresponding to the upper limit of transmission duty cycle, and the error control parameters corresponding to the upper limit of retransmission budget.

[0017] Optionally, a target feature tensor representing local physical degradation is generated and cached locally within the prediction time window; after the prediction time window ends, the target feature tensor is divided into data sub-blocks based on an importance index and encapsulated and sent according to priority, wherein the encapsulation payload of each data sub-block is less than or equal to the upper limit of the single frame payload length.

[0018] Optionally, the prediction time window for determining the physical propagation of the target interference event to the second heterogeneous device along the spatial topology includes:

[0019] Obtain the set of effective propagation path parameters between the first heterogeneous device and the second heterogeneous device in the spatial topology. The set of effective propagation path parameters includes pipe segment structural parameters and fluid medium physical property parameters. The pipe segment structural parameters include pipe diameter, wall thickness, and material elastic modulus. The fluid medium physical property parameters include density and bulk modulus. The set of effective propagation path parameters is used to characterize the effective propagation path between the first heterogeneous device and the second heterogeneous device.

[0020] Calculate the propagation speed parameter of the target interference event on the effective propagation path based on the set of effective propagation path parameters;

[0021] The start time of the target interference event is determined from the first running feature sequence, the arrival timestamp is calculated based on the propagation speed parameter and the propagation distance of the effective propagation path, and the arrival timestamp is used as the starting boundary of the prediction time window.

[0022] Optionally, determining the prediction time window further includes:

[0023] The attenuation coefficient is calculated based on the set of effective propagation path parameters, and the attenuation curve of the disturbance amplitude at the location of the second heterogeneous device over time is predicted based on the attenuation coefficient.

[0024] The time stamp when the decay curve is below the interference threshold is determined as the departure time stamp, and the prediction time window is constructed by the arrival time stamp and the departure time stamp.

[0025] Optionally, the generation of protocol reconfiguration instructions by the operation and maintenance agent based on the constrained channel load constraint parameter set includes:

[0026] Based on the evolution characteristics of the disturbance intensity of the target interference event, the prediction time window is divided into a time slice sequence; the time slice sequence includes a peak abrupt change period and a gradual decay period;

[0027] A dynamic bit width allocation strategy is generated for the time slice sequence, using the upper limit of the single-frame payload length in the set of constrained channel payload constraint parameters as a constraint.

[0028] The dynamic bit-width allocation strategy is encapsulated into the protocol reconfiguration instruction.

[0029] Optionally, the generation of the dynamic bit-width allocation strategy includes:

[0030] The target feature tensor is divided into a high-frequency diagnostic band and a low-frequency base band in the frequency domain;

[0031] During the peak abrupt change period in the time slice sequence, a first quantization bit width is configured for the high-frequency diagnostic frequency band, and the first quantization bit width is greater than the second quantization bit width configured for the low-frequency baseband, and the first quantization bit width is reduced during the gradual decay period;

[0032] The target feature tensor is truncated non-uniformly based on the first quantization bit width and the second quantization bit width, and the truncated result is assembled into a protocol data frame to meet the upper limit of the single frame payload length.

[0033] Secondly, this application provides an agent-driven heterogeneous device protocol adaptive operation and maintenance system, including:

[0034] The acquisition module is used to obtain the spatial topology relationship between the first heterogeneous device and the second heterogeneous device in the target operating environment;

[0035] The processing module is configured to determine, based on the first operational characteristic sequence of the first heterogeneous device, the prediction time window for the physical propagation of the target interference event to the second heterogeneous device along the spatial topology; and to determine the set of constrained channel load parameters within the prediction time window based on the attribute information of the target interference event and the basic protocol specifications of the second heterogeneous device.

[0036] The control module is used to generate a protocol reconfiguration instruction by the operation and maintenance intelligent agent based on the set of constrained channel load constraints; based on the protocol reconfiguration instruction, dynamically reconfigure the communication protocol of the second heterogeneous device from the basic protocol protocol to a constrained matching protocol, and execute the constrained matching protocol within the prediction time window.

[0037] Compared to existing technologies, this application establishes spatial topological associations between heterogeneous devices and utilizes the operational characteristics of upstream devices to determine the time range within which sudden physical interference affects downstream links. This transforms protocol adaptation from traditional post-event response to targeted adjustments for specific time periods. Within this time range, constrained channel load boundaries are formed by combining interference event attributes and basic protocol specifications. The operations and maintenance agent generates protocol reconfiguration instructions and drives the communication protocol to be dynamically reconfigured into a constrained matching protocol. This maintains effective transmission with controlled load, duty cycle, and retransmission budget during link disruption, reducing the impact of sudden packet loss and invalid retransmissions on operations and maintenance data collection. By limiting protocol adjustments to the interference period, unnecessary long-term degradation can be reduced, improving the continuity and completeness of operations and maintenance characteristic data acquisition, and enhancing adaptability and stability for heterogeneous device scenarios. Attached Figure Description

[0038] Figure 1 A flowchart of the heterogeneous device protocol adaptive operation and maintenance method driven by AGENT provided in the embodiments of this application;

[0039] Figure 2 A flowchart illustrating a method for determining a prediction time window provided in an embodiment of this application;

[0040] Figure 3 A flowchart illustrating a method for generating a dynamic bit-width allocation strategy, provided in an embodiment of this application;

[0041] Figure 4 This is a schematic diagram of an AGENT-driven heterogeneous device protocol adaptive operation and maintenance system provided in an embodiment of this application. Detailed Implementation

[0042] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0043] The technical solution of this application can be implemented in various ways, including as a method, system, apparatus, or computer program product. Therefore, specific embodiments of this application can be implemented entirely in hardware, entirely in software, or in a combination of software and hardware. Furthermore, specific embodiments of this application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0044] See Figure 1 The diagram shows a flowchart of an agent-driven heterogeneous device protocol adaptive operation and maintenance method provided in this application embodiment, including steps S101 to S105, wherein:

[0045] S101: Obtain the spatial topology relationship between the first heterogeneous device and the second heterogeneous device in the target operating environment;

[0046] S102: Based on the first operating characteristic sequence of the first heterogeneous device, determine the prediction time window for the physical propagation of the target interference event to the second heterogeneous device along the spatial topology;

[0047] S103: Based on the attribute information of the target interference event and the basic protocol of the second heterogeneous device, determine the set of restricted channel load constraint parameters within the prediction time window;

[0048] S104: The operation and maintenance intelligent agent generates a protocol reconfiguration command based on the set of constrained channel load parameters;

[0049] S105: Based on the protocol reconstruction instruction, dynamically reconstruct the communication protocol of the second heterogeneous device from the basic protocol protocol to a restricted matching protocol, and execute the restricted matching protocol within the prediction time window.

[0050] Regarding the above S101:

[0051] The target operating environment refers to the industrial site environment where heterogeneous devices are actually deployed and perform monitoring, control, or data acquisition tasks. This industrial site environment can include production line workstations, pipe corridors / pipelines, station / cabinet areas, workshop areas, or factory network segments. Heterogeneous devices are device nodes that differ in at least one of the following aspects: hardware form, communication interface, communication protocol, or data sampling characteristics. Heterogeneous devices can include sensor nodes, actuator nodes, edge gateways, industrial switch-side acquisition units, or controllers. The first heterogeneous device and the second heterogeneous device are two target device nodes selected from multiple heterogeneous devices in the target operating environment. The first heterogeneous device provides operational characteristic information related to the target interference event, while the second heterogeneous device is the target device that needs to perform communication protocol reconstructing during the impact of the target interference event.

[0052] For example, the first heterogeneous device is a frequency converter using the industrial Ethernet protocol, and the second heterogeneous device is a vibration sensor using the RS485 protocol.

[0053] It is understandable that the first and second terms mentioned above are only used for distinction and do not require that the physical locations must be adjacent; however, the two have a definable topological order relationship in terms of spatial topology.

[0054] Spatial topology is used to characterize the relative topological positions of the first heterogeneous device and the second heterogeneous device in the target operating environment, and can include the connectivity and topological order between them. The topological order can be represented by upstream / downstream, preceding / following, or sequential relationships along the process flow direction.

[0055] It is important to note that spatial topology is not the same as the IP routing topology in traditional IT network architecture. Instead, it refers to the upstream and downstream physical connection logic of devices in the real physical world along the direction of transmission of physical media (such as pipes, conveyor belts, and mechanical shafts).

[0056] For example, in a scenario where pipe racks or pipelines are deployed along the line, upstream / downstream can correspond to the order of process medium flow; in a production line workstation scenario, preceding / following sequence can correspond to the flow sequence of materials or workpieces; in a station or cabinet scenario, upstream / downstream can correspond to the cascading sequence of power distribution links or control links.

[0057] In practical implementation, spatial topology relationships can be obtained from existing topology mapping information on-site. The sources of this topology mapping information can be varied: for example, equipment installation locations and connection relationships pre-recorded in engineering deployment data or asset ledgers; topology discovery results from on-site network management systems or industrial switching equipment; or device adjacency tables or routing tables maintained in control systems / edge gateways. After acquisition, the first heterogeneous device identifier and the second heterogeneous device identifier can be mapped to the same topology representation structure, and the connectivity and topology order relationships between them can be extracted as the output of this step.

[0058] In some embodiments, spatial topology relationships can be represented by topology relationship description information, which includes: a first heterogeneous device identifier, a second heterogeneous device identifier, a connectivity identifier, and a sequence identifier. For example, the topology relationship description information can be represented as a topology relationship record, with fields including "upstream device ID, downstream device ID, relationship type, and relationship direction"; or as a directed edge in the topology graph, pointing from the first heterogeneous device node to the second heterogeneous device node.

[0059] For example, in the pipe gallery area of ​​a chemical plant, the first heterogeneous device is the edge gateway of the upstream pump station (aggregating operational characteristic data such as pressure / vibration), and the second heterogeneous device is the actuator controller at the downstream valve group. During the commissioning phase, the installation location and connection relationship of the equipment are entered into the asset management system, and the network management platform performs topology discovery on the adjacency relationship of the industrial switch ports. In step S101, the system reads the installation segments and connection relationships of the two devices from the asset management system, and then combines them with the adjacency relationship provided by the network management platform to map the first heterogeneous device and the second heterogeneous device into a "upstream → downstream" topology relationship record; the record is output as spatial topology relationship description information and stored in the local configuration cache for subsequent steps to call.

[0060] Regarding S102 above:

[0061] The first operational characteristic sequence refers to the characteristic time-series data collected or aggregated by the first heterogeneous device, used to characterize the change of its operational status over time. For example, the first operational characteristic sequence may include a time-series sequence of at least one of pressure, flow rate, vibration amplitude, motor current, and valve position change, or a derived characteristic sequence obtained by filtering, differential analysis, and energy statistics of the aforementioned at least one basic measurement. The target interference event refers to a physical disturbance event triggered in the target operating environment by factors such as operating condition switching, media disturbance, or high-voltage switching. This physical disturbance event can propagate downstream along the physical connection logic corresponding to the spatial topology, causing a phased degradation of the communication channel quality during its propagation to the location of the second heterogeneous device. The prediction time window is used to characterize the estimated time range of the target interference event's impact on the second heterogeneous device. The prediction time window includes at least a start boundary and an end boundary, which can be represented by an arrival timestamp and a departure timestamp, for example.

[0062] In practice, the system identifies target interference events based on the first operational feature sequence and determines the start time and scale of the event.

[0063] For example, changes in characteristic quantities representing sudden changes in operating conditions can be detected in the first operating characteristic sequence. When the magnitude or rate of change meets the preset triggering conditions, the corresponding time point is determined as the start time of the target interference event. Based on the duration interval of the change characteristics on the time axis, the duration scale parameter of the target interference event is determined to characterize the time width of the target interference event's impact on downstream.

[0064] Furthermore, to reflect the physical propagation characteristics of the target interference event along spatial topological relationships, the system configures propagation delay parameters or propagation delay mapping relationships bound to the spatial topological relationships. This is used to characterize the time delay experienced by the target interference event as it propagates from the first heterogeneous device to the second heterogeneous device. The propagation delay parameters can be obtained from calibration results during the on-site commissioning phase, empirical configurations recorded in engineering deployment data, or topological association delay tables fitted based on historical operating data. Based on the propagation delay parameters, the system performs time delay correction on the start time of the target interference event to obtain the arrival timestamp of the target interference event to the second heterogeneous device, and determines the departure timestamp by combining the persistence scale parameter. Thus, a prediction time window is constructed from the arrival timestamp and the departure timestamp.

[0065] For example, in a pipe gallery area of ​​a chemical plant, the first heterogeneous device is the edge gateway of the upstream pump station, and the second heterogeneous device is the controller at the downstream valve group. During the commissioning phase, typical start-up and shutdown conditions are recorded on-site, and the propagation delay configuration corresponding to this topology is saved in the asset management system. In step S102, the system detects a sudden change in operating characteristics caused by pump start-up and shutdown in the first operating characteristic sequence, determines this time point as the start time, reads the propagation delay parameters from the propagation delay configuration corresponding to the topology, completes the delay correction, generates a prediction time window, and outputs it in the form of "{second heterogeneous device ID, window start timestamp, window end timestamp}" for subsequent steps to call.

[0066] For example, continuing the scenario of the chemical plant's pipe gallery area described above. The upstream pump station edge gateway (the first heterogeneous device) continuously reports the high-frequency pressure sequence (the first operating characteristic sequence) at the pump outlet. When the pump station performs an emergency pump stop operation, a sharp drop occurs in the pressure sequence, accompanied by a reverse water hammer pressure peak. The system detects that this pressure peak exceeds the safety threshold, determines that a "water hammer impact" target interference event has occurred, and records the occurrence time as T0. The system reads the locally cached spatial topology relationship and delay mapping table, and finds that the empirical propagation delay of the water hammer physical shock wave from the pump station to the valve group section where the downstream actuator controller (the second heterogeneous device) is located is 2.5 seconds, and the empirical oscillation duration is 0.8 seconds. Based on this, it is calculated that the severe accompanying vibration and electrical noise caused by the water hammer impact will reach the downstream node at T0+2.5 seconds and decay completely at T0+3.3 seconds, thus constructing a prediction time window [T0+2.5s, T0+3.3s] as the time reference for subsequent protocol reconstruction.

[0067] Regarding the above S103:

[0068] The attribute information of the target interference event refers to the event profile information used to characterize the impact of the target interference event on the communication link within the prediction time window. For example, the attribute information of the target interference event includes event type label, disturbance intensity level, disturbance duration, and dominant interference mode label. For example, the event type label can be water hammer impact, valve fast closing impact, or electromagnetic pulse interference caused by frequency converter switching; the disturbance intensity level can be divided into high, medium, and low levels according to engineering commissioning practices; the dominant interference mode label can be structural vibration dominant, electromagnetic pulse dominant, or a combination of both.

[0069] The basic protocol specification of the second heterogeneous device refers to the communication protocol and default transmission behavior adopted by the second heterogeneous device before protocol reconfiguration. For example, the basic protocol specification includes the protocol type, basic field organization method, default single-frame service payload length, default transmission rhythm, and default retransmission behavior. For instance, the protocol type can be Modbus RTU, PROFINET RT, or OPC UA; in the Modbus RTU scenario, the basic field organization method includes the device address field, function code field, data field, and check field, and status messages are sent periodically in a polling manner by default; in the OPC UA scenario, data change notifications are transmitted via a session mechanism by default, and a timeout retry policy can be configured. The constrained channel payload constraint parameter set is used to characterize the upper limit constraint imposed on the link payload within the prediction time window. For example, the constrained channel payload constraint parameter set includes the upper limit of single-frame payload length, the upper limit of transmission duty cycle, and the upper limit of retransmission budget.

[0070] In practical implementation, the system pre-establishes an event attribute-protocol constraint mapping library. The mapping library is divided into entries according to event type labels and disturbance intensity levels, and corresponding single frame payload length limit, transmission duty cycle limit and retransmission budget limit are configured for different protocol types.

[0071] In step S103, the system acquires the attribute information of the target interference event and parses out the event type label and disturbance intensity level. It further acquires the basic protocol specification of the second heterogeneous device and determines the protocol type and default transmission behavior. Then, it matches the corresponding entries in the mapping library to obtain the set of constrained channel load parameters within the prediction time window and outputs them.

[0072] For example, continuing with the water hammer impact scenario, when the event type label is water hammer impact and the disturbance intensity level is high, and the protocol type of the second heterogeneous device is Modbus RTU, the mapping library returns a single frame payload length limit of 32 bytes, a transmission duty cycle limit of "only short status frames are allowed to be sent within the prediction time window", and a retransmission budget limit of 1 time, thereby providing constraint inputs for the subsequent operation and maintenance intelligent agent to generate protocol reconfiguration instructions.

[0073] Regarding S104 above:

[0074] In this context, the "operation and maintenance intelligent agent" refers to an intelligent decision-making unit deployed in an edge gateway, field control server, or operation and maintenance platform. This unit generates executable protocol reconfiguration instructions for the second heterogeneous device within a predicted time window, based on a set of constrained channel load parameters. For example, the operation and maintenance intelligent agent can run as a containerized service or process on the pump station's edge gateway; alternatively, it can run as a policy engine module of the operation and maintenance platform on the station control server. The protocol reconfiguration instruction is a message used to trigger the second heterogeneous device to enter the constrained matching protocol. It carries at least the protocol parameterization configuration items corresponding to the constrained channel load constraint parameter set, enabling the second heterogeneous device to generate and send data frames according to the constrained matching protocol.

[0075] In practical implementation, the operation and maintenance intelligent agent generation protocol refactoring instructions can include the following processing:

[0076] First, the operations and maintenance agent reads and parses the set of constrained channel payload parameters to obtain the upper limit of single-frame payload length, the upper limit of transmission duty cycle, and the upper limit of retransmission budget. For example, the upper limit of single-frame payload length can be 32 bytes or 64 bytes; the upper limit of transmission duty cycle can be expressed as "only one service frame is allowed to be transmitted in each transmission cycle within the prediction time window"; the upper limit of retransmission budget can be 0 or 1 times.

[0077] Secondly, the operations and maintenance agent obtains the basic protocol specification profile and device capability profile of the second heterogeneous device, and selects a limited set of matching protocol parameters that can be executed by the second heterogeneous device under the constraints of the basic protocol specification profile. For example, when the protocol type is Modbus RTU, the device capability profile may include the set of available function codes, the allowed range of register addresses, the upper limit of the number of registers that can be read at one time, and the configuration range of the number of application layer retries supported by the device; as another example, when the protocol type is OPC UA, the device capability profile may include available message group identifiers, the configurable range of the release cycle, and the set of message body fields.

[0078] Then, the operations and maintenance agent determines the data frame template parameters and transmission scheduling parameters while meeting the upper limit of the single-frame payload length. It also determines the transmission rhythm within the window while meeting the upper limit of the transmission duty cycle, and determines the error control parameters while meeting the upper limit of the retransmission budget. For example, in a Modbus RTU scenario, the data frame template parameters can be implemented by limiting the number of registers read at once, limiting the set of fields for the reporting code, or limiting the length of the service data fields; the transmission scheduling parameters can be expressed as "adjusting the default 100ms polling period to a 500ms polling period"; and the error control parameters can be expressed as "adjusting the default 2 application layer retries to 1".

[0079] Finally, the operations and maintenance agent encapsulates the restricted matching protocol parameter set into a protocol reconstruction instruction, and binds this instruction to a prediction time window to instruct the second heterogeneous device to execute according to the restricted matching protocol within the prediction time window. For example, the protocol reconstruction instruction can be carried using TLV encoding or JSON encoding, with fields including: target device identifier, window identifier or window start and end timestamps, protocol mode identifier, data frame template identifier, transmission scheduling parameter identifier, and error control parameter identifier; for example, it can be encapsulated as "{deviceId:D2,window:[Ts,Te],mode:restricted,frameTemplateId:T1,period:500ms,payloadMax:32B,retryBudget:1}". Here, deviceId represents the target device identifier, window represents the window start and end timestamps [Ts,Te], mode represents the protocol mode, frameTemplateId represents the data frame template identifier, period represents the transmission period, payloadMax represents the maximum single-frame payload length, and retryBudget represents the maximum retransmission budget.

[0080] In some embodiments, the operation and maintenance intelligent agent includes a policy generation submodule and an instruction orchestration submodule. The policy generation submodule calls a large language model to generate candidate parameterized configurations for restricted matching specifications, and the instruction orchestration submodule performs constraint verification on the candidate parameterized configurations and encapsulates them to form protocol reconstruction instructions.

[0081] For example, the large language model can be a DeepSeek series model, a GPT series model, or a self-built enterprise large language model. The operation and maintenance intelligence agent maintains the rule set and device capability profile locally for constraint verification to avoid the large language model outputting configurations that exceed the device's executable range. The inputs to the policy generation submodule include the attribute information of the target interference event, the prediction time window, the set of constrained channel load constraint parameters, the basic protocol specification profile of the second heterogeneous device, and the device capability profile; the outputs of the policy generation submodule include the data frame template selection result, transmission scheduling parameters, and error control parameters.

[0082] For example, in a water hammer impact scenario, the protocol type of the second heterogeneous device is Modbus RTU, and the device capability profile indicates that "the number of registers read at one time does not exceed 4, and the number of application layer retries can be configured from 0 to 2". The set of constrained channel payload parameters is "maximum single frame payload length = 32 bytes, maximum transmission duty cycle = at most 1 frame sent every 500ms within the window, maximum retransmission budget = 1". The policy generation submodule submits structured prompt information to the large language model. The prompt information includes: the protocol type is Modbus RTU, the candidate frame template set is {full status template, simplified status template, alarm bit only template}, and "the output must be JSON and the fields must include frameTemplateId (data frame template identifier), pollPeriodMs (polling period in milliseconds), payloadMaxBytes (maximum single frame payload length in bytes), retryBudget (maximum retransmission budget), and fields (field set)". The candidate parameterization configuration output by the large language model is as follows: frameTemplateId=simplified state template, fields={valve position register, execution current register}, pollPeriodMs=500, payloadMaxBytes=32, retryBudget=1. The instruction orchestration submodule performs constraint verification on the output. Specifically, the verification field set's corresponding encoding overhead does not exceed the upper limit of single-frame payload length, the verification transmission period meets the upper limit of transmission duty cycle, and the verification retry count does not exceed the upper limit of retransmission budget and falls within the device's configurable range. When the verification passes, the parameterized configuration is encapsulated into a protocol reconstruction instruction and bound to the prediction time window. For example, the encapsulated fields include "deviceId (target device identifier), windowStart (window start timestamp), windowEnd (window end timestamp), mode=restricted (protocol mode is restricted mode), frameTemplateId (data frame template identifier), scheduleParamId (transmission scheduling parameter identifier), errorCtrlParamId (error control parameter identifier)". The protocol reconstruction instruction is then sent to the second heterogeneous device to trigger it to execute the restricted matching protocol within the prediction time window.

[0083] In other embodiments, when the candidate parameterized configurations output by the large language model contain fields that do not meet the constraints, the instruction orchestration submodule executes a deterministic repair strategy to generate alternative configurations that meet the constraints.

[0084] For example, when the candidate field set causes the encoding overhead to exceed the single-frame payload length limit, the instruction orchestration submodule deletes fields from low to high according to the preset field priority until the single-frame payload length limit is met; when the candidate transmission period causes the transmission duty cycle to exceed the limit, the instruction orchestration submodule adjusts the transmission period to the minimum feasible period that meets the duty cycle limit; when the candidate retransmission count exceeds the retransmission budget limit, the instruction orchestration submodule truncates the retransmission count to the retransmission budget limit. The repaired configuration is then encapsulated into protocol reconfiguration instructions.

[0085] Regarding the above S105:

[0086] Dynamic reconfiguration refers to the second heterogeneous device switching online with respect to payload and timing-related configurable items in its protocol stack according to protocol reconfiguration instructions, while maintaining the basic protocol type or without changing the link layer / physical layer configuration within the allowable compatibility range. This allows it to generate and send data frames with parameterized configuration of the restricted matching protocol within the prediction time window. The restricted matching protocol is a protocol execution mode derived from the basic protocol protocol and satisfying the set of restricted channel payload constraint parameters. It at least limits the data frame template (field set and field encoding method), transmission scheduling parameters (transmission period, duty cycle control method, set of frame types allowed to be transmitted within the window), and error control parameters (retransmission budget, acknowledgment policy or redundancy check policy). Executing the restricted matching protocol means that within the prediction time window, the transmission module, encoding module, and retransmission control module of the second heterogeneous device all operate according to the parameterized configuration, and the actual transmitted data frames meet the upper limit of single frame payload length and upper limit of transmission duty cycle, and the retransmission behavior does not exceed the upper limit of retransmission budget.

[0087] In practical implementation, step S105 can be implemented in the following order: instruction reception and verification—window binding—parameter loading—intra-window sending—window end rollback. After receiving the protocol reconstruction instruction, the second heterogeneous device first performs integrity and permission verification on the target device identifier, window start and end timestamps / window identifier, protocol mode identifier, and template / scheduling / error parameter fields. For example, integrity verification may include detection of missing verification fields, verification code verification, or signature verification; permission verification may include agent identity verification based on a whitelist or role verification based on a control domain. After the verification is successful, the second heterogeneous device writes the protocol reconstruction instruction into its local "windowed protocol task table". This task table records at least the window start boundary, window end boundary, and the corresponding restricted matching protocol parameter set, and writes the version number of the task table into the configuration cache.

[0088] To ensure execution within the predicted time window, the second heterogeneous device can employ a triggering mechanism bound to the window. For example, if the second heterogeneous device has a unified clock source, such as time synchronization via NTP / PTP or a station control system, the restricted matching protocol will be triggered when the local clock enters the window's initial boundary. If the second heterogeneous device does not have a unified clock source, the protocol reconfiguration command can carry relative countdown parameters (such as "delay in milliseconds before activation" or "duration duration"). The second heterogeneous device will then generate a local timer based on these parameters and trigger activation and deactivation. After activation, the second heterogeneous device switches from the basic protocol protocol runtime state to the restricted matching protocol runtime state and loads the data frame template, sends scheduling parameters, and error control parameters. For example, loading a data frame template includes: selecting a set of fields and their order from a locally pre-set template table by frameTemplateId, and determining the encoding bit width and scaling factor of each field according to the field type table; loading transmission scheduling parameters includes: setting the minimum transmission interval within the window, the set of frame types allowed to be transmitted (such as only status-type short frames), and the duty cycle control strategy (such as token bucket or fixed-rhythm gating); loading error control parameters includes: setting the upper limit of the number of application layer retries, the acknowledgment waiting timeout threshold, or the redundancy check switch.

[0089] Within the prediction time window, when the second heterogeneous device generates and sends data frames according to the constrained matching protocol, the constraints can be ensured to be met by scheduling first and then framing. Specifically, the transmission scheduling module determines whether data frame transmission is allowed in each scheduling cycle, and if allowed, issues an available payload budget to the framing module. The available payload budget does not exceed the upper limit of the single frame payload length. The framing module writes the service fields into the buffer based on the selected data frame template, and performs boundary checks according to the payload budget during the writing process. If writing a field would cause an over-limit, the lower priority fields are skipped or a shorter alternative template is switched according to the preset order of deletable fields in the template. After encoding, the error control module controls the number of retransmissions according to the upper limit of the retransmission budget, and stops retransmissions when the budget is reached and records the link degradation event log within the window. Thus, the data frames actually sent within the window satisfy the constrained channel payload constraint parameter set in terms of payload length, transmission rhythm, and retransmission behavior.

[0090] After the prediction time window ends, the second heterogeneous device will revert from the restricted matching protocol runtime state to the basic protocol runtime state and clean up the temporary scheduler states bound to the window (such as token bucket count and in-window send count) to avoid cross-window interference.

[0091] For example, rollback may include restoring the default polling cycle, restoring the default field set, restoring the default retry limit, and returning a window execution receipt to the operations and maintenance agent. The receipt includes the target device identifier, window identifier, execution result code, and statistical summary within the window, such as the number of frames sent, packet loss estimate, and retransmission trigger count. In some embodiments, if an exception occurs in the window indicating that the instruction cannot be executed, such as the template not existing locally, the device capability profile not supporting the corresponding function code, or the field encoding overhead not being compressed to the load limit, the second heterogeneous device may refuse execution and return a reason for refusal, while maintaining the basic protocol specification in operation to ensure basic device communication availability.

[0092] For example, in a water hammer impact scenario, the second heterogeneous device uses Modbus RTU as its protocol type, with default behavior of "polling and sending a status message every 100ms, reading 10 holding registers at a time, and setting an application layer retry limit of 2". The protocol reconfiguration instruction specifies "window [T0+2.5s, T0+3.3s], maximum single-frame payload length of 32 bytes, minimum transmission interval within the window of 500ms, maximum retransmission budget of 1, and frameTemplateId = simplified status template (fields = {valve position register, execution current register})". In step S105, when the second heterogeneous device enters the window start boundary, it switches the polling period to 500ms and limits the number of registers read at one time to within 4 to match the load limit. At the same time, it switches the application layer retry limit to 1. Each time transmission is allowed within the window, the device frames and sends according to the simplified state template. If a verification failure occurs, it will retry a maximum of 1 time. After the window ends, it restores the default 100ms polling period and default field set, and reports the window execution receipt for the operation and maintenance platform to archive.

[0093] In summary, this application transforms the propagation of physical disturbances into an alignable predictable time window, and then transforms the link degradation within this window into a parameterizable set of load constraints. Subsequently, the operation and maintenance agent outputs executable, verifiable, and rollback-capable protocol reconfiguration instructions to drive the second heterogeneous device to operate within the window using a restricted matching protocol. This ensures the transmissibility and continuity of critical operation and maintenance data at minimal cost during periods of strong disturbance, and controls online switching risks through verification, capability constraints, acknowledgments, and rollback mechanisms.

[0094] Accordingly, this application can determine the prediction time window of the interference impact based on spatial topology and the first operational characteristic sequence when sudden physical interference in the industrial field causes phased degradation of the communication channel. Within this window, a set of constrained channel load constraint parameters is formed by combining the attribute information of the target interference event with the basic protocol specification of the second heterogeneous device. The operation and maintenance agent generates and issues a protocol reconfiguration command, enabling the second heterogeneous device to switch online to a constrained matching protocol that meets constraints such as load length, transmission duty cycle, and retransmission budget. Thus, under the conditions of load and timing constraints, the complete acquisition and usable transmission of critical status / diagnostic operation and maintenance characteristic data are still prioritized. At the same time, by verifying the integrity and permission of the reconfiguration command, loading parameters and repairing constraints according to the device capability profile, automatically rolling back after the window ends, and abnormal rejection and execution receipt mechanisms, the risk of link instability caused by unexecutable policy outputs or erroneous switching is reduced, and the controllability and on-site stability of the operation and maintenance link under strong disturbance conditions are improved.

[0095] Optional, see Figure 2 The flowchart of a method for determining a prediction time window provided in this application embodiment includes steps S201 to S202, wherein:

[0096] S201: Extract mutation feature vectors from the first running feature sequence and determine the propagation speed parameter and duration parameter;

[0097] S202: Based on the propagation speed parameter, the relative spatial distance between the first heterogeneous device and the second heterogeneous device, and the duration parameter, calculate the arrival timestamp and departure timestamp, and construct the prediction time window from the arrival timestamp and departure timestamp.

[0098] To reduce the risks of jitter in target interference event identification, window boundary drift, and false triggering caused by noise, this embodiment provides a window construction method based on mutation feature vectors, which makes the start and end boundaries of the prediction time window more consistent with the propagation and duration of the target interference event.

[0099] The mutation feature vector refers to the set of features extracted from the first operational feature sequence to characterize the sudden change pattern of the target interference event. For example, the mutation feature vector may include at least one of the following: instantaneous change amplitude, change per unit time, local energy increment, peak sharpness, or the slope of the rising edge of the envelope of the mutation segment. The propagation speed parameter refers to a parameter characterizing the equivalent propagation speed of the target interference event along the spatial topological relationship. For example, the propagation speed parameter may be an "empirical propagation speed level of the shock wave / disturbance front" or a "propagation delay parameter of the corresponding link segment in the topological association delay table." The duration parameter refers to a parameter characterizing the length of the main influence interval of the target interference event at the upstream observation point. For example, the duration parameter may be the duration of the mutation segment, the duration of the envelope being above a threshold, or the duration of the dominant oscillation segment decaying below a threshold.

[0100] In specific implementation, the first running feature sequence is preprocessed to reduce the influence of slowly changing background and measurement noise. For example, preprocessing may include bandpass filtering, moving average detrending, or differencing the sequence to highlight abrupt changes. Subsequently, change detection is performed on the preprocessed sequence within a preset sliding window. For example, the statistical differences (such as mean difference, energy difference) between adjacent windows may be compared, or the rate of change of the sequence may be compared. When the detection result meets a preset trigger condition, the corresponding time point is determined as the abrupt change moment, and a abrupt change feature vector is extracted from the preset neighborhood before and after the abrupt change moment.

[0101] For example, when the first operating characteristic sequence is a "pump outlet high-frequency pressure sequence", a sharp drop and a reverse peak can be detected at the moment of emergency pump stop, and the peak amplitude and rising slope can be extracted as components of the abrupt change feature vector; when the first operating characteristic sequence is a "motor current sequence", the current step amplitude and short-time energy increment can be extracted as components of the abrupt change feature vector.

[0102] Secondly, the propagation speed parameter is determined. For example, the propagation speed parameter can be preferentially read from the propagation delay configuration or topology association delay table bound to the spatial topology relationship; if it is not configured on site, an empirical propagation delay can be formed based on the alignment samples of "upstream mutation time - downstream link degradation time" in historical records, and the sample entries that are similar to the current mutation feature vector can be selected as the basis for the value of the propagation speed parameter.

[0103] For example, in the pipeline corridor scenario, the empirical propagation delay of the "pump station → valve group" link segment can be maintained in the asset management system or local configuration cache; when the mutation feature vector represents a strong water hammer impact, the corresponding shorter propagation delay entry is selected; when the mutation feature vector represents a slow operating condition switch, the corresponding longer propagation delay entry is selected.

[0104] Next, determine the duration parameter. For example, the duration parameter can be determined based on the envelope maintenance interval of the sequence after the mutation: after the mutation time, calculate the short-time amplitude envelope or short-time energy sequence for the preprocessed sequence, and determine the length of the time interval during which it continuously exceeds the influence threshold as the duration parameter; alternatively, the duration parameter can be obtained by taking the dominant oscillation segment after the mutation as the termination condition of "decaying below the threshold". For example, in a water hammer impact scenario, the duration parameter can be the duration for which the oscillation envelope remains above the threshold after the pressure peak; in a frequency converter switching-induced electromagnetic pulse scenario, the duration parameter can be the duration for which the high-frequency noise energy exceeds the limit.

[0105] After obtaining the propagation speed parameters, relative spatial distance, and duration parameters, the system calculates the arrival timestamp and departure timestamp, and constructs a prediction time window using these two parameters. The relative spatial distance refers to the equivalent propagation distance between the first heterogeneous device and the second heterogeneous device in terms of the physical connection logic corresponding to their spatial topology. For example, the relative spatial distance can be the effective length of a pipeline, the workpiece flow distance along a production line station, or the equivalent segment length along a cascaded control link. The relative spatial distance can be read from engineering deployment data, asset ledgers, or topology description information. The arrival timestamp can be obtained by combining the abrupt change time with the propagation speed parameters and relative spatial distance for time delay correction; the departure timestamp can be determined by combining the arrival timestamp with the duration parameter. This results in a prediction time window with the arrival timestamp as the starting boundary and the departure timestamp as the ending boundary, which can be output as "{Second Heterogeneous Device ID, Window Start Timestamp, Window End Timestamp}" for subsequent steps.

[0106] For example, in the water hammer impact scenario, the first operational feature sequence is the pump outlet high-frequency pressure sequence. The system detects a sharp drop accompanied by a reverse peak at time T0, and extracts the peak amplitude and the rising slope to form a sudden change feature vector. Based on the empirical propagation delay entry of the "pump station → valve group" link segment in the local topology association delay table, the propagation delay corresponding to the propagation speed parameter is determined to be 2.5 seconds. At the same time, based on the interval of continuous over-threshold of the oscillation envelope, the duration parameter is determined to be 0.8 seconds. Based on this, the arrival timestamp is 2.5 seconds after T0, and the departure timestamp is 0.8 seconds after the arrival timestamp, thus constructing a prediction time window.

[0107] In some embodiments, the preset triggering conditions can be determined by combining a normal operating condition baseline with a noise margin to reduce the probability of false triggering. For example, historical segments of the first operating characteristic sequence can be collected during non-interference periods as baseline samples, and the distribution of its short-term variation amplitude, rate of change, or short-term energy can be statistically analyzed. The triggering threshold can be set to the high quantile of the baseline distribution or "baseline mean plus noise margin". The noise margin can be determined based on the sensor range, resolution, mechanical resonance background at the installation location, and the intensity of the on-site electromagnetic environment, ensuring that the frequency of triggering conditions being met under normal operating conditions is controlled. For example, taking a high-frequency pressure sequence at the pump outlet as an example, the peak-to-peak pressure difference triggering threshold of adjacent sliding windows can be set to 0.25 MPa, and the "unit time change (slope)" triggering threshold can be set to 0.6 MPa / s. The thresholds can be obtained from statistical analysis of nearly 30 days of normal production data, ensuring that the false triggering frequency is no higher than once per day, and the trigger recall rate in the labeled water hammer operating condition samples is no lower than a preset proportion.

[0108] In some embodiments, the sliding window length and update step size used for change detection can be determined based on the sampling frequency of the first running feature sequence and the typical timescale of the target interference event, to balance detection sensitivity and real-time performance. For example, when the first running feature sequence is a pressure sequence sampled at 1 kHz, the sliding window length can be set to 50 ms to 200 ms, and the update step size can be set to 10 ms to 50 ms; when the first running feature sequence is a vibration RMS sequence sampled at 100 Hz, the sliding window length can be set to 0.5 s to 2 s, and the update step size can be set to 0.1 s to 0.5 s. These values ​​can be set based on observations of the event rise time duration and the dominant oscillation period during the field commissioning phase, ensuring that the abrupt change feature vector covers the event rise time and reflects its suddenness without excessively expanding the window and causing boundary tailing.

[0109] In some embodiments, the duration parameter can be determined by maintaining the excess range of the disturbance envelope, and the impact threshold can be set according to the correspondence between the disturbance amplitude and the communication link degradation index, so that the departure timestamp reflects the actual moment when the link availability is restored. For example, the system can align the changes in disturbance amplitude / energy with communication quality indicators (such as CRC error rate, frame loss rate, retransmission trigger frequency, or RSSI decrease) in historical data, select the disturbance level that causes the communication quality index to fall back to an acceptable range as the impact threshold, and calculate the duration parameter as the length of time the disturbance envelope is continuously higher than the impact threshold. Taking a water hammer impact scenario as an example, if historical alignment shows that when the peak-to-peak pressure oscillation is below 0.08 MPa, the frame drop rate of the second heterogeneous device can be restored to below 1%, then the impact threshold can be set to 0.08 MPa, and the duration parameter can be set to "the duration during which the pressure oscillation envelope is continuously above 0.08 MPa after the peak value". For example, in a pump stop event, the duration parameter can be obtained in the range of 0.8s to 1.2s, and the median or upper quantile value can be taken to leave a safety margin.

[0110] Optionally, within the prediction time window corresponding to the target interference event, the communication channel quality degrades in stages. If the full amount of service data frames is still continuously sent according to the basic protocol specifications, on the one hand, frame loss and retransmission conflicts may further squeeze the limited channel load budget, and on the other hand, key operation and maintenance characteristic data may be interrupted or passively discarded, making it difficult to make subsequent judgments on local physical degradation without continuous evidence.

[0111] Therefore, this embodiment prioritizes the local retention of "local physical degradation features" within the prediction time window, and then transmits them back according to importance after the prediction time window ends, so as to improve the effective acquisition probability of operation and maintenance feature data under the constraint of limited load.

[0112] The target feature tensor refers to the multi-dimensional feature data structure obtained by organizing the multi-source operating features collected by the second heterogeneous device within the prediction time window. It continuously represents the changes in the operating state of the second heterogeneous device at least along the time dimension, and may further include at least one of the sensor channel dimension, feature type dimension, or frequency band dimension.

[0113] For example, in a water hammer impact scenario, the second heterogeneous device is an actuator controller at the valve assembly, and the target feature tensor it collects may include a continuous sampling segment within a window of the valve position change sequence, the execution current sequence, and the valve body vibration RMS sequence; or, for example, when the second heterogeneous device is an RS485 vibration sensor, the target feature tensor may specifically be an acceleration time sequence segment within a window and its short-time energy statistics sequence.

[0114] In practical implementation, after entering the prediction time window, the second heterogeneous device generates a target feature tensor according to a preset sampling strategy and caches it locally. The local cache can be a circular buffer, a segmented file, or a key-value time-series cache, recording the correspondence between "timestamp and feature data" to support retrospective segmentation after the window ends.

[0115] To avoid cache overflow, the caching strategy can be configured based on "upper bound of window duration + feature sampling rate + available storage space", and a safety margin should be reserved to cover sampling jitter or window boundary overflow.

[0116] For example, when the duration of the prediction time window is empirically no more than 1.0s, the vibration sampling rate is 200Hz, and each sampling record includes a timestamp and a triaxial RMS value, the local cache capacity can be configured to be no less than 300 sampling points, and an additional redundant area can be reserved for transition sampling before and after the window boundary, so as to completely preserve the evidence within the window without relying on external links.

[0117] After the prediction time window ends, the second heterogeneous device or operation and maintenance intelligent agent divides the target feature tensor into data sub-blocks based on the importance index and encapsulates and sends them according to priority. The importance index is used to measure the contribution of different feature segments to "local physical degradation discrimination or location". Its setting can be based on the fault sample alignment results during the commissioning and debugging phase, or on the statistical relationship between "abnormality degree and feature sensitivity" in historical operating data, so that when the load is limited, the evidence segments that best support the diagnostic conclusion are prioritized for transmission. For example, the importance index can be at least one of "abnormality score" (such as the degree of deviation from the baseline within the window), "change suddenness score" (such as short-term sudden change energy), or "sensitivity score to remaining lifetime prediction". In the valve actuator scenario, the importance of "sudden increase in execution current segment" can be set higher than that of "slow change in valve position segment" to prioritize the discrimination of jamming or stagnation-type degradation.

[0118] Furthermore, the size of the data sub-block can be set according to the upper limit of the single frame payload length, and after considering fixed overhead such as protocol header and check fields, an available payload budget is allocated to each sub-block so that the encapsulation payload of each data sub-block is less than or equal to the upper limit of the single frame payload length.

[0119] For example, when the maximum length of a single frame payload is 32 bytes and the budget available for the service payload after reserving the header and verification overhead in the current frame template is no more than 24 bytes, the target feature tensor can be divided into multiple data sub-blocks of no more than 24 bytes, and a short sub-block identifier (such as start and end timestamp index, feature type identifier) ​​can be attached to the front of each sub-block so that the receiver can reassemble it; when sending, the sub-blocks are loaded in order of importance from high to low, and the second highest priority sub-block is sent after the high priority sub-block is sent, so as to transmit the key evidence back as much as possible within the limited communication period after the window.

[0120] Optionally, the perturbation intensity of the target interference event within the prediction time window usually exhibits an evolution characteristic of "first sharp and then decaying". If the feature data collected within the window is encapsulated with fixed quantization precision or fixed encoding bit width, two types of problems are likely to occur: First, in the peak abrupt change stage, fixed low-precision quantization will mask the sudden details, resulting in insufficient evidence for identifying the degradation mechanism; Second, in the gradual decay stage, fixed high-precision quantization will unnecessarily consume the payload budget, leading to a decrease in the total amount of effective evidence that can be transmitted back.

[0121] Therefore, in this embodiment, when generating protocol reconfiguration instructions, the operation and maintenance intelligent agent introduces time slices and dynamic bit width allocation that match the evolution of disturbance intensity, so that bit width resources are adaptively allocated to time periods that require more precision over time, thereby improving the diagnostic effectiveness of backhauled data under the constraint of the upper limit of single frame payload length.

[0122] Among them, the disturbance intensity evolution characteristic refers to the characterization information of the intensity of the target interference event changing over time within the prediction time window. It can be obtained from the change trend of the abrupt change amplitude of the first running characteristic sequence, the local envelope strength within the window, or the communication quality degradation index.

[0123] For example, in a water hammer impact scenario, the oscillating envelope of the pump outlet pressure sequence after window alignment can be used as a feature of disturbance intensity evolution; in a high-voltage switching electromagnetic pulse scenario, the short-term statistical value of the CRC error rate on the second heterogeneous device side within the window can be used as a feature of disturbance intensity evolution. A time-slice sequence refers to the division of the prediction time window on the time axis, which includes at least multiple consecutive slices and the start and end boundaries of each slice. The division criteria for the time-slice sequence can be a "segmentation threshold of the relative peak value of the disturbance intensity" or a "turning point of the rate of change of the disturbance intensity," so that the peak abrupt change period can be characterized separately and distinguished from the subsequent gradual decay period.

[0124] For example, the interval where the disturbance intensity is higher than 70% of the peak value and the rate of change is higher than a preset value can be divided into the peak mutation period, and the interval where the disturbance intensity decays from 70% to near the disturbance threshold can be divided into the gradual decay period, forming a time slice sequence such as "[Ts,Ts+0.2s], [Ts+0.2s,Te]"; where Ts represents the start time stamp (i.e., arrival time stamp) of the prediction time window, and Te represents the end time stamp (i.e., departure time stamp) of the prediction time window; the above segmentation threshold can be obtained by statistics on typical water hammer samples during the commissioning and debugging phase, so as to ensure that the peak mutation phase covers the main impact details and the number of slices is not too large.

[0125] The dynamic bit width allocation strategy refers to the strategy of configuring corresponding quantization bit width or coding precision parameters for each slice of the time slice sequence under the constraint of the upper limit of single frame payload length. The setting can be based on the engineering principle of "prioritizing the fidelity of the peak change stage under the same payload budget" and the communication constraint principle of "prioritizing the total amount of backhaul after the window". This allows for higher representation precision during the peak change period and moderate reduction of representation precision during the smooth decay period to save payload.

[0126] For example, when the maximum single-frame payload length is 32 bytes and the device is allowed to send a maximum of 1 frame every 500ms after the window, the operation and maintenance agent can configure a 12-bit quantization width for the peak mutation period and an 8-bit quantization width for the smooth decay period, and form a policy entry together with the time slice boundary; when the number of communication cycles available after the predicted window is small, the bit width of the smooth decay period can be further reduced to 6 bits in exchange for more slice coverage or more feature point backhaul.

[0127] When encapsulating the dynamic bit width allocation strategy into the protocol reconfiguration instruction, the time slice sequence and the corresponding bit width configuration can be written into the instruction payload as parameterized fields, so that the second heterogeneous device can select the corresponding bit width for encoding and loading according to the slice boundary when framing within or after the window.

[0128] For example, the protocol reconstruction instruction can carry fields such as "sliceCount, sliceBoundaryList, bitWidthList, payloadMaxBytes", where sliceBoundaryList records the relative offset or absolute timestamp of each slice, and bitWidthList records the bit width value corresponding to each slice. In the water hammer impact scenario, the instruction can be encapsulated as "window [T0+2.5s, T0+3.3s], slice = [0~0.2s, 0.2~0.8s], bit width = [12, 8], payloadMax = 32B". The second heterogeneous device uses higher bit width encoding for the data segment corresponding to the peak change period and lower bit width encoding for the data segment corresponding to the smooth decay period, so that the loading result of each frame stably meets the upper limit of the single frame payload length and improves the fidelity of key details in the peak change phase.

[0129] Optional, see Figure 3 The flowchart below illustrates a method for generating a dynamic bit-width allocation strategy, as provided in this application embodiment, including steps S301 to S303, wherein:

[0130] S301: Divide the target feature tensor into a high-frequency diagnostic band and a low-frequency base band in the frequency domain;

[0131] S302: During the peak abrupt change period in the time slice sequence, a first quantization bit width is configured for the high-frequency diagnostic frequency band, and the first quantization bit width is greater than the second quantization bit width configured for the low-frequency baseband, and the first quantization bit width is reduced during the smooth decay period;

[0132] S303: Perform non-uniform bit width truncation on the target feature tensor based on the first quantization bit width and the second quantization bit width, and assemble the truncation result into a protocol data frame to meet the upper limit of the single frame payload length.

[0133] In the above text, the target interference event often contains both low-frequency operating condition offsets (such as slow valve position drift and changes in motor load baseline) and high-frequency transient diagnostic components (such as structural vibration spikes caused by water hammer impact and electromagnetic pulses and harmonics caused by high-voltage switching) during the peak abrupt change period. If the bit width is adjusted only along the time dimension, there may still be a problem of uneven distribution of bit width resources in the frequency domain, that is, the low-frequency substrate occupies too much load while the high-frequency diagnostic details are over-compressed.

[0134] Therefore, this embodiment divides the target feature tensor into a high-frequency diagnostic band and a low-frequency base band in the frequency domain, and uses differentiated quantization bit widths for the two types of bands in different slices of the time slice sequence, thereby improving the fidelity of evidence related to transient impacts and local degradation under the constraint of the upper limit of single-frame payload length.

[0135] Among them, the high-frequency diagnostic band refers to the frequency band that is more sensitive to sudden impacts, rapid oscillations, and impulse noise in the frequency domain representation of the target feature tensor; the low-frequency baseline band refers to the frequency band that is more sensitive to the operating condition baseline, slow changing trends, and steady-state components. The frequency band division can be determined based on the sensor bandwidth, sampling rate, typical characteristic frequencies of the measured object, and event type labels of the second heterogeneous device, and can be preset as a device capability profile or scenario configuration item during the commissioning phase.

[0136] For example, in the scenario of joint vibration / current monitoring of valve actuators, the low-frequency base band can correspond to the "low-frequency range dominated by operating condition changes", such as the low-frequency components corresponding to valve position changes and load fluctuations, while the high-frequency diagnostic band can correspond to the "high-frequency range dominated by impact and structural resonance", such as the high-frequency vibration components caused by water hammer impact; in the scenario of electromagnetic interference during high-voltage switching, the high-frequency diagnostic band can correspond to the range dominated by pulses and their high-frequency harmonics, while the low-frequency base band retains the slow drift and baseline of current / voltage.

[0137] For example, the low-frequency baseband can be configured to be in the range of 0 to tens of hertz, and the high-frequency diagnostic band can be configured to be in the range of hundreds of hertz to kilohertz, depending on the sensor sampling rate, installation structure and event type calibration.

[0138] The criteria for setting frequency band boundaries can be: to ensure that the low-frequency baseband can stably characterize the operating condition baseline and is insensitive to transient spikes, to ensure that the high-frequency diagnostic band can cover the main energy concentration area of ​​the impact event and is insensitive to baseline drift, and to ensure that sufficient frequency domain sampling points are retained in each frequency band to support diagnostic discrimination.

[0139] In practical implementation, the operation and maintenance intelligent agent or the second heterogeneous device can perform frequency domain transformation or equivalent filtering decomposition on the target feature tensor in each time slice to obtain the high-frequency diagnostic sub-tensor and the low-frequency basis sub-tensor. For example, the frequency domain partitioning can obtain the spectral coefficient sequence of each frequency band through short-time spectrum analysis, or it can directly output the "low-frequency component sequence" and "high-frequency component sequence" through a set of digital filter banks, and use the two as the frequency domain partitioning result of the target feature tensor.

[0140] Subsequently, during the peak abrupt change period in the time slice sequence, a first quantization bit width is configured for the high-frequency diagnostic band, and this first quantization bit width is made greater than the second quantization bit width configured for the low-frequency baseband; during the gradual decay period, the first quantization bit width is reduced, so that the bit width resources are gradually recovered from the "high-frequency transient details" and released to the subsequent backhaul total or other time slices.

[0141] For example, during the peak abrupt change period, the quantization bit width of the high-frequency diagnostic band can be set to 12 bits, and the low-frequency baseband to 8 bits; during the gradual decay period, the quantization bit width of the high-frequency diagnostic band can be reduced to 8 bits, while the low-frequency baseband can remain at 8 bits or be further reduced to 6 bits. The specific values ​​can be determined based on the upper limit of the single-frame payload length, the number of available transmissions after the window, and empirical diagnostic requirements. This setting can be described as follows: during the peak abrupt change period, the high-frequency diagnostic components contribute more to the determination of the degradation mechanism and have a higher dynamic range, requiring higher quantization accuracy; during the gradual decay period, the energy of the high-frequency components decreases and their marginal contribution to the conclusion decreases, so the bit width can be reduced to obtain more slice coverage or more feature point backhaul.

[0142] When performing non-uniform bit-width truncation, independent quantization configurations and truncation boundaries can be applied to the high-frequency diagnostic band and the low-frequency baseband, respectively. Necessary decoding auxiliary information can be carried during encapsulation to ensure reconfigurability at the receiver. For example, a "larger step size, smaller bit width" quantization can be used for the low-frequency baseband to preserve trend information, while a "smaller step size, larger bit width" quantization can be used for the high-frequency diagnostic band to preserve peak details. During the gradual decay period, the quantization configuration of the high-frequency diagnostic band can be switched to coarser granularity to reduce coding overhead. To meet the single-frame payload length limit, the protocol data frame assembly can adopt a loading order of "first placing the necessary low-frequency baseband fields, then placing the key high-frequency diagnostic fields," and when the payload budget is insufficient, priority should be given to retaining the core frequency points or core statistics of the high-frequency diagnostic sub-tensors that contribute more to diagnosis.

[0143] For example, the data can be organized in the order of "slice identifier - frequency band identifier - bit width identifier - scaling / offset configuration - data payload" within a frame, and the quantization bit width can be written for each frequency band. When the expected load will exceed the upper limit of the single frame load length, low importance frequency points can be pruned according to preset priority within the high frequency diagnostic frequency band, the number of frequency points in the high frequency diagnostic frequency band can be reduced, or the high frequency diagnostic frequency band can be downgraded from full spectrum coefficients to several key frequency band energy statistics to ensure that the encapsulated load of this frame does not exceed the upper limit of the single frame load length.

[0144] For example, in a water hammer impact scenario, the target feature tensor acquired by the second heterogeneous device within the prediction time window includes a time-series segment of valve body vibration within the window. The frequency band division configuration is as follows: the low-frequency baseband is used to characterize the low-frequency components related to the valve position and load baseline, and the high-frequency diagnostic band is used to characterize the transient vibration and local resonance components caused by water hammer impact; the time slice sequence is the peak abrupt change period of 0-0.2s and the gradual decay period of 0.2-0.8s; the maximum length of a single frame payload is 32 bytes. During the peak abrupt change period, the high-frequency diagnostic band is configured with a bit width of 12 bits and the low-frequency baseband with a bit width of 8 bits, and several key frequency band statistics of the high-frequency diagnostic band and trend quantities of the low-frequency baseband are preferentially encapsulated; during the gradual decay period, the bit width of the high-frequency diagnostic band is reduced to 8 bits to reduce payload occupation, so that the payload encapsulated in each frame is stable and does not exceed 32 bytes, and high-frequency diagnostic evidence that is more sensitive to water hammer impact and local degradation of the valve group is preferentially transmitted back within a limited transmission period after the window.

[0145] Optionally, to further reduce the uncertainty of propagation delay parameters depending on empirical configuration or historical fitting, and to make the prediction time window more closely match the physical fact that disturbance events propagate directionally along the fluid medium in the scenario of the pipe gallery / pipeline, this embodiment refines the spatial topology relationship into "effective propagation paths" and uses the structural and medium parameters corresponding to the path as the basis for calculating the propagation speed.

[0146] The effective propagation path parameter set is used to characterize the combination of pipe segments and their equivalent parameters that the target interference event can actually propagate from the location of the first heterogeneous device to the location of the second heterogeneous device. Examples include: pipe segment structural parameters (pipe diameter, wall thickness, material elastic modulus) and fluid medium physical property parameters (density, bulk modulus).

[0147] In practice, the above parameters can be obtained from on-site engineering data and asset ledgers. For example, pipe diameter and wall thickness can be read from P&ID / pipeline isometric drawings, pipe segment numbers and specifications; pipe type can be read from the bill of materials and mapped to the material elastic modulus; and density and bulk modulus can be read from the process media card or DCS / SCADA media configuration. Alternatively, the path segment list can be exported from the digital twin / asset management system of the operation and maintenance platform and cached as "path parameter records" on the edge side. The record fields include "starting device ID, ending device ID, pipe segment list, geometric parameters of each pipe segment, media type and physical property parameters, and total path propagation distance". Based on the effective propagation path parameter set, the propagation velocity parameters can be calculated using an engineering model for one-dimensional transient wave / water hammer propagation. The engineering model considers the combined effects of medium compressibility and pipe wall elasticity on the propagation velocity. To avoid online calculation overhead, the calculation process can also be performed offline as a table lookup or piecewise interpolation: that is, a wave velocity configuration table is established according to the combination of medium type and pipe diameter / wall thickness / material, and the propagation velocity parameters are retrieved according to the effective propagation path parameter set during the operation of steps S102 / S103.

[0148] Subsequently, the start time of the target interference event is determined from the first running characteristic sequence. For example, the first threshold crossing time when the pressure sequence drops sharply and is accompanied by a reverse peak is taken as the start time. The arrival timestamp is calculated based on the propagation speed parameter and the propagation distance of the effective propagation path. The arrival timestamp is then output as the starting boundary of the prediction time window.

[0149] For example, in a water hammer impact scenario, the path parameter record shows: the pipe diameter is 200mm, the wall thickness is 8mm, the material is carbon steel (the elastic modulus can be taken from the material table), the medium is water (the density and bulk modulus can be taken from the medium card), and the total path length is approximately 2500m. Based on this, the system obtains the propagation speed parameter as being on the order of kilometers per second, and determines the arrival timestamp by adding the path propagation delay to the starting time. Thus, the window start point is calibrated based on physical propagation, rather than relying solely on empirical delay constants.

[0150] Optionally, if the start boundary of the prediction time window is determined solely by the arrival timestamp, and the end boundary of the prediction time window is still set using a fixed duration or a rough empirical setting, problems may arise such as the window being too short, leading to uncovered link degradation at the tail, or the window being too long, leading to unnecessary load degradation.

[0151] To this end, this embodiment further determines the attenuation coefficient based on the effective propagation path parameter set, and generates an attenuation curve of the disturbance amplitude at the location of the second heterogeneous device over time based on the attenuation coefficient, thereby determining the time point when the disturbance effect is attenuated to a level that is insufficient to cause communication degradation as the departure timestamp.

[0152] The attenuation curve can be generated from attenuation curve description information. This description information is bound and stored with the effective propagation path parameter set. For example, it can be maintained as an attenuation configuration table by the operation and maintenance platform or edge gateway, indexed by combination keys such as "event type label / disturbance intensity level / medium type / pipe diameter classification / pipe segment material classification / path length classification". The attenuation curve description information can take at least one of the following forms:

[0153] For example, given a discrete set of amplitude sampling points, the set includes multiple time offsets Δt and corresponding relative amplitude coefficients (e.g., recorded as {Δt1:0.8, Δt2:0.5, Δt3:0.2}). After the timestamp is reached, the system looks up the corresponding amplitude in a table according to the time offset and performs linear interpolation between adjacent sampling points to obtain the predicted disturbance amplitude at any time.

[0154] For example, a segmented attenuation parameter set is used to describe multiple attenuation stages such as "peak holding stage + rapid attenuation stage + slow attenuation stage". Each stage records the duration and the relative amplitude coefficient at the end of the stage. The system generates an attenuation curve according to the segmentation rules.

[0155] For example, there is an experience template identifier, which points to a preset attenuation curve template library, such as "high-strength water hammer template", "medium-strength template for fast valve closing" and "short-time electromagnetic pulse template". The system generates an attenuation curve after correcting the template parameters according to the effective propagation path parameter set.

[0156] The interference threshold is used to characterize the engineering boundary at which the communication channel quality of the second heterogeneous device no longer significantly degrades when the disturbance amplitude is below the threshold. The basis for setting it can come from the joint calibration during the commissioning and debugging phase: under typical interference events, disturbance amplitude indicators near the second heterogeneous device are collected synchronously, such as vibration acceleration amplitude, pressure fluctuation amplitude, or electromagnetic noise amplitude in the cabinet, and link quality indicators, such as packet loss rate, bit error rate, retransmission trigger rate, or round-trip delay jitter, to determine when the disturbance amplitude is below a certain amplitude range and the link quality indicators recover to a stable range, and the upper limit of this amplitude range is determined as the interference threshold.

[0157] For example, the minimum disturbance amplitude that causes the packet loss rate to rise above the baseline by more than a preset proportion can be used as the interference threshold, and stored separately according to protocol type. For example, the thresholds for Modbus RTU and PROFINET RT are different. When determining the departure timestamp, the system searches for the time point on the attenuation curve at which the predicted disturbance amplitude first falls below the interference threshold, and uses this as the departure timestamp. A prediction time window is then constructed using the arrival timestamp and the departure timestamp.

[0158] For example, during commissioning or routine inspection, disturbance amplitude indicators and link quality indicators can be jointly calibrated to determine the basis for setting the interference threshold. For instance, several reproducible target interference events can be selected, such as water hammer impact caused by emergency pump shutdown or valve rapid closure impact. Disturbance amplitude indicators and link quality indicators can be simultaneously collected near the second heterogeneous device. The disturbance amplitude indicator can be the RMS value or peak-to-peak value of the triaxial vibration acceleration inside the cabinet, the pressure fluctuation amplitude, or the electromagnetic noise amplitude. The link quality indicator can be at least one of packet loss rate, retransmission trigger rate, and round-trip time jitter. Based on the synchronously collected data, a criterion is determined for the link quality indicator to show a "significant degradation" compared to the baseline before interference. The minimum disturbance amplitude corresponding to meeting this criterion is then determined as the interference threshold. The criterion can be set separately according to protocol type to reflect the robustness differences of different basic protocol specifications under the same disturbance conditions.

[0159] For example, the interference threshold can be set as follows: In the Modbus RTU scenario, the baseline packet loss rate of the stable operating range before interference is no higher than 0.2%, and "the packet loss rate rises to no less than 3% or the retransmission trigger rate is no less than 1 time / second in any statistical period within the window" is used as a significant degradation criterion; In a water hammer impact commissioning, the vibration acceleration RMS measured near the second heterogeneous device decays over time. When the vibration acceleration RMS is higher than 0.35g, the packet loss rate remains between 3% and 6% and a retransmission peak occurs. When the vibration acceleration RMS drops below 0.35g, the packet loss rate falls back to less than 1% and the retransmission trigger rate returns to the baseline range. Based on this, the vibration acceleration RMS corresponding to 0.35g is determined as the interference threshold under this protocol type and written into the threshold configuration item corresponding to "water hammer impact - high intensity level - effective propagation path classification"; In the PROFINET RT scenario, the same calibration process can be used, but due to the different periodic scheduling and confirmation mechanisms, the obtained thresholds can be different and saved independently.

[0160] For example, in a water hammer impact scenario, the attenuation configuration table returns a set of discrete sampling points for the "high-intensity water hammer template". After the arrival timestamp, the system generates an attenuation curve based on the sampling points and interpolation. Combined with the interference threshold, the system calculates the departure timestamp as approximately 0.8 seconds after the arrival timestamp, thereby constructing a prediction time window for determining the end boundary of subsequent protocol reconstruction.

[0161] Based on the same inventive concept, this application also provides an AGENT-driven heterogeneous device protocol adaptive operation and maintenance system corresponding to the AGENT-driven heterogeneous device protocol adaptive operation and maintenance method. Since the principle of the system in this application is similar to the AGENT-driven heterogeneous device protocol adaptive operation and maintenance method described above in this application, the implementation of the system can refer to the implementation of the method, and the repeated parts will not be described again.

[0162] Reference Figure 4 The diagram shown is a schematic of an agent-driven heterogeneous device protocol adaptive operation and maintenance system provided in an embodiment of this application. The system includes:

[0163] The acquisition module 10 is used to acquire the spatial topology relationship between the first heterogeneous device and the second heterogeneous device in the target operating environment;

[0164] Processing module 20 is configured to determine, based on the first operating characteristic sequence of the first heterogeneous device, a prediction time window for the physical propagation of a target interference event to the second heterogeneous device along the spatial topology; and to determine, based on the attribute information of the target interference event and the basic protocol specifications of the second heterogeneous device, a set of constrained channel load parameters within the prediction time window.

[0165] The control module 30 is used to generate a protocol reconfiguration instruction by the operation and maintenance intelligent agent based on the set of constrained channel load constraints; based on the protocol reconfiguration instruction, dynamically reconfigure the communication protocol of the second heterogeneous device from the basic protocol protocol to a constrained matching protocol, and execute the constrained matching protocol within the prediction time window.

[0166] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. An agent-driven adaptive operation and maintenance method for heterogeneous device protocols, characterized in that, include: Obtain the spatial topology relationship between the first heterogeneous device and the second heterogeneous device in the target operating environment; Based on the first operating characteristic sequence of the first heterogeneous device, a prediction time window is determined for the physical propagation of the target interference event to the second heterogeneous device along the spatial topological relationship. Based on the attribute information of the target interference event and the basic protocol specifications of the second heterogeneous device, the set of constrained channel load parameters within the prediction time window is determined; The operation and maintenance intelligent agent generates protocol reconfiguration instructions based on the set of constrained channel load parameters; Based on the protocol reconstruction instruction, the communication protocol of the second heterogeneous device is dynamically reconstructed from the basic protocol protocol to a restricted matching protocol, and the restricted matching protocol is executed within the prediction time window.

2. The AGENT-driven heterogeneous device protocol adaptive operation and maintenance method as described in claim 1, characterized in that, The set of constrained channel payload parameters includes: upper limit of single frame payload length, upper limit of transmission duty cycle, and upper limit of retransmission budget.

3. The AGENT-driven heterogeneous device protocol adaptive operation and maintenance method as described in claim 1, characterized in that, The prediction time window for determining the physical propagation of the target interference event to the second heterogeneous device along the spatial topology includes: Extract mutation feature vectors from the first running feature sequence and determine the propagation speed parameter and duration parameter; Based on the propagation speed parameter, the relative spatial distance between the first heterogeneous device and the second heterogeneous device, and the duration parameter, the arrival timestamp and departure timestamp are calculated, and the prediction time window is constructed from the arrival timestamp and departure timestamp.

4. The AGENT-driven heterogeneous device protocol adaptive operation and maintenance method as described in claim 2, characterized in that, The restricted matching rule limits: The data frame template corresponding to the upper limit of single frame payload length, the transmission scheduling parameters corresponding to the upper limit of transmission duty cycle, and the error control parameters corresponding to the upper limit of retransmission budget.

5. The AGENT-driven heterogeneous device protocol adaptive operation and maintenance method as described in claim 2, characterized in that, Within the prediction time window, a target feature tensor representing local physical degradation is generated and cached locally. After the prediction time window ends, the target feature tensor is divided into data sub-blocks based on an importance index and encapsulated and sent according to priority, wherein the encapsulation payload of each data sub-block is less than or equal to the upper limit of the single frame payload length.

6. The AGENT-driven heterogeneous device protocol adaptive operation and maintenance method as described in claim 1, characterized in that, The prediction time window for determining the physical propagation of the target interference event to the second heterogeneous device along the spatial topology includes: Obtain the set of effective propagation path parameters between the first heterogeneous device and the second heterogeneous device in the spatial topology. The set of effective propagation path parameters includes pipe segment structural parameters and fluid medium physical property parameters. The pipe segment structural parameters include pipe diameter, wall thickness, and material elastic modulus. The fluid medium physical property parameters include density and bulk modulus. The set of effective propagation path parameters is used to characterize the effective propagation path between the first heterogeneous device and the second heterogeneous device. Calculate the propagation speed parameter of the target interference event on the effective propagation path based on the set of effective propagation path parameters; The start time of the target interference event is determined from the first running feature sequence, the arrival timestamp is calculated based on the propagation speed parameter and the propagation distance of the effective propagation path, and the arrival timestamp is used as the starting boundary of the prediction time window.

7. The AGENT-driven heterogeneous device protocol adaptive operation and maintenance method as described in claim 6, characterized in that, Determining the prediction time window further includes: The attenuation coefficient is calculated based on the set of effective propagation path parameters, and the attenuation curve of the disturbance amplitude at the location of the second heterogeneous device over time is predicted based on the attenuation coefficient. The time stamp when the decay curve is below the interference threshold is determined as the departure time stamp, and the prediction time window is constructed by the arrival time stamp and the departure time stamp.

8. The AGENT-driven heterogeneous device protocol adaptive operation and maintenance method according to claim 5, characterized in that, The protocol reconfiguration instruction generated by the operation and maintenance intelligent agent based on the set of constrained channel load parameters includes: Based on the evolution characteristics of the disturbance intensity of the target interference event, the prediction time window is divided into a time slice sequence; the time slice sequence includes a peak abrupt change period and a gradual decay period; A dynamic bit width allocation strategy is generated for the time slice sequence, using the upper limit of the single-frame payload length in the set of constrained channel payload constraint parameters as a constraint. The dynamic bit-width allocation strategy is encapsulated into the protocol reconfiguration instruction.

9. The AGENT-driven heterogeneous device protocol adaptive operation and maintenance method as described in claim 8, characterized in that, The dynamic bit-width allocation generation strategy includes: The target feature tensor is divided into a high-frequency diagnostic band and a low-frequency base band in the frequency domain; During the peak abrupt change period in the time slice sequence, a first quantization bit width is configured for the high-frequency diagnostic frequency band, and the first quantization bit width is greater than the second quantization bit width configured for the low-frequency baseband, and the first quantization bit width is reduced during the gradual decay period; The target feature tensor is truncated non-uniformly based on the first quantization bit width and the second quantization bit width, and the truncated result is assembled into a protocol data frame to meet the upper limit of the single frame payload length.

10. An agent-driven heterogeneous device protocol adaptive operation and maintenance system, used to implement the agent-driven heterogeneous device protocol adaptive operation and maintenance method according to any one of claims 1-9, characterized in that, include: The acquisition module is used to obtain the spatial topology relationship between the first heterogeneous device and the second heterogeneous device in the target operating environment; The processing module is used to determine the prediction time window for the physical propagation of the target interference event to the second heterogeneous device along the spatial topology based on the first operating characteristic sequence of the first heterogeneous device; Based on the attribute information of the target interference event and the basic protocol specifications of the second heterogeneous device, the set of constrained channel load parameters within the prediction time window is determined; The control module is used to generate a protocol reconfiguration instruction by the operation and maintenance intelligent agent based on the set of constrained channel load constraints; based on the protocol reconfiguration instruction, dynamically reconfigure the communication protocol of the second heterogeneous device from the basic protocol protocol to a constrained matching protocol, and execute the constrained matching protocol within the prediction time window.

Images