Interaction method and device of multi-intelligent fusion terminal, equipment and medium

Abstract

The application provides an interaction method and device of a multi-intelligent fusion terminal, equipment and a medium, and belongs to the technical field of data interaction. The method comprises the following steps: when it is detected that the network state of a current terminal is an offline state, predicting the agent capability values of multiple terminals based on a trained agent capability prediction model, and then determining an agent intelligent fusion terminal; sending a broadcast signal to the outside, so that the agent intelligent fusion terminal sends the mark data of the current terminal to a general control. When it is detected that the network state of the current terminal is restored to an online state, sending an agent data request signal to the general control, receiving the agent data list and control instructions sent by the general control, sending a cooperative parameter request signal to multiple terminals, and receiving the model key control parameters during the offline period of the current terminal sent by multiple associated terminals; sending unmarked data to the general control based on the agent data list and the control instructions, and updating the model parameters based on the model key control parameters. The application can improve the efficiency of data interaction.

Description

Technical Field

[0001] This application belongs to the field of data interaction technology, and more specifically, relates to an interaction method, device, equipment, and medium for multi-intelligent fusion terminals. Background Technology

[0002] In existing smart converged terminal application scenarios, multiple smart converged terminals typically need to interact with a central control unit and work collaboratively to achieve various complex tasks and functions. However, in actual operation, the network status of smart converged terminals is not always stable online, and they frequently switch from online to offline states. When a smart converged terminal is offline, its normal data transmission and command interaction with the central control unit are interrupted. This means that data accumulated by the terminal during its offline period cannot be uploaded to the central control unit in a timely manner, and the central control unit cannot issue new commands to the terminal or obtain its relevant status information, severely impacting the efficiency of information interaction between multiple devices. Summary of the Invention

[0003] The purpose of this application is to provide an interaction method, device, equipment, and medium for multiple intelligent fusion terminals, so as to improve the information interaction efficiency of multiple intelligent fusion terminals when switching between online and offline states and reduce the occurrence of malfunctions.

[0004] A first aspect of this application provides an interaction method for multiple intelligent fusion terminals, comprising:

[0005] In response to the detection that the network status of the current intelligent fusion terminal has switched from online to offline, the agent capability values ​​of multiple target intelligent fusion terminals are predicted based on the trained agent capability prediction model. The trained agent capability prediction model is obtained by federated learning collaborative training based on multiple target intelligent fusion terminals and the central control when the current intelligent fusion terminal is online.

[0006] The agent intelligent fusion terminal is determined from multiple target intelligent fusion terminals based on the agent capability value;

[0007] It sends out broadcast signals, which are used to enable the agent intelligent fusion terminal to send the current intelligent fusion terminal's tag data to the central control.

[0008] In response to the detection that the network status of the current intelligent fusion terminal has switched from offline to online, it sends a proxy data request signal to the central control and receives a proxy data list and control instructions generated based on the proxy data request signal from the central control.

[0009] If a proxy data list and control instructions are received from the central control, a collaborative parameter request signal is sent to multiple target intelligent fusion terminals, and the key control parameters of the model during the offline period are received from multiple target intelligent fusion terminals. The proxy data list and control instructions are generated based on the proxy data request signal.

[0010] The system sends unlabeled data to the central controller based on the agent data list and control instructions, and updates the model parameters of the trained agent capability prediction model based on the key control parameters of the model.

[0011] A second aspect of this application provides an interactive device for a multi-intelligent fusion terminal, the device comprising:

[0012] The first detection unit is used to respond to the detection that the network status of the current intelligent fusion terminal has switched from online to offline. Based on the trained agent capability prediction model, it predicts the agent capability values ​​of multiple target intelligent fusion terminals. The trained agent capability prediction model is obtained by federated learning collaborative training based on multiple target intelligent fusion terminals and the central control when the current intelligent fusion terminal is in the online state.

[0013] The first processing unit is used to determine the agent intelligent fusion terminal from multiple target intelligent fusion terminals based on the agent capability value;

[0014] The signal transmitting unit is used to transmit broadcast signals to the outside world. The broadcast signals are used to enable the agent intelligent fusion terminal to send the current intelligent fusion terminal's tag data to the central control.

[0015] The second detection unit is used to send a proxy data request signal to the central control in response to the detection that the network status of the current intelligent fusion terminal has switched from offline to online, and to receive a proxy data list and control instructions generated by the central control based on the proxy data request signal.

[0016] The second processing unit is used to send a collaborative parameter request signal to multiple target intelligent fusion terminals if it receives a proxy data list and control instructions sent by the central control, and to receive the key control parameters of the model during the offline period sent by multiple target intelligent fusion terminals. The proxy data list and control instructions are generated based on the proxy data request signal.

[0017] The model update unit is used to send unlabeled data to the master controller based on the agent data list and control instructions, and to update the model parameters of the trained agent capability prediction model based on the key control parameters of the model.

[0018] A third aspect of this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the steps of the above-described multi-intelligent fusion terminal interaction method.

[0019] A fourth aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described interaction method for a multi-intelligent fusion terminal.

[0020] The beneficial effects of the multi-intelligent fusion terminal interaction method, apparatus, device, and medium provided in this application embodiment are as follows:

[0021] This application effectively improves the operational stability and reliability of various terminal devices under unstable network conditions. When the current intelligent fusion terminal is offline, it can accurately predict the proxy capability values ​​of multiple target associated intelligent fusion terminals and identify the proxy intelligent fusion terminal. This process is based on a federated learning collaborative training model, which fully utilizes the data and computing resources of each target associated intelligent fusion terminal to ensure the accuracy and reliability of the prediction. When the current intelligent fusion terminal returns to online status, it sends a proxy data request signal to the central control, receives the proxy data list and control instructions, and then requests the key model control parameters from multiple target intelligent fusion terminals during the offline period. Finally, it updates the model parameters and sends unlabeled data based on this information. The entire process forms a complete closed loop, enabling the system to maintain data continuity and model stability even when the network status changes frequently. This effectively avoids data loss and model failure caused by offline conditions, ensuring that each terminal device can operate stably and efficiently in different network environments. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 A flowchart illustrating the interaction method of a multi-intelligent fusion terminal provided in an embodiment of this application;

[0024] Figure 2 A structural block diagram of an interactive device for a multi-intelligent fusion terminal provided in an embodiment of this application;

[0025] Figure 3 This is a schematic block diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0026] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0027] To make the objectives, technical solutions, and advantages of this application clearer, the following description will be provided in conjunction with the accompanying drawings and specific embodiments.

[0028] Please refer to Figure 1 , Figure 1 This is a flowchart illustrating an embodiment of the interaction method for multiple intelligent fusion terminals provided in this application. The method is applied to a large-scale terminal interaction system, which includes multiple interconnected intelligent fusion terminals and a central control unit. Each intelligent fusion terminal interacts with the central control unit. The multiple interconnected intelligent fusion terminals include a current intelligent fusion terminal and multiple target intelligent fusion terminals. The current intelligent fusion terminal is any one of the multiple interconnected intelligent fusion terminals. The method is executed by the current intelligent fusion terminal and includes:

[0029] S101: In response to detecting that the network status of the current intelligent fusion terminal has switched from online to offline, predict the agent capability values ​​of multiple target intelligent fusion terminals based on the trained agent capability prediction model.

[0030] Among them, the trained agent capability prediction model is obtained by federated learning collaborative training based on multiple target intelligent fusion terminals and the central control when the current intelligent fusion terminal is online; the multiple target intelligent fusion terminals are all devices that have established an association with the current intelligent fusion terminal, and the agent capability value is used to judge the reliability of the target intelligent fusion terminal in transmitting data on behalf of the current intelligent fusion terminal;

[0031] In this embodiment, the large-scale terminal interaction system includes multiple interconnected intelligent fusion terminals and a central control unit. Each intelligent fusion terminal receives operating data from the electrical equipment in its assigned area and can make preliminary judgments based on this data, such as whether the electrical equipment has malfunctioned, for example, whether a certain electrical equipment is overloaded. If a malfunction is detected, the malfunction information is promptly uploaded to the central control unit. If no malfunction is detected, or only a trend towards a malfunction is detected, all received data is shared with other terminals associated with that intelligent fusion terminal, and all data is simultaneously sent to the central control unit for unified judgment.

[0032] Among them, the distribution area is the smallest power supply management unit in the power system, specifically referring to a specific geographical area covered by a distribution transformer or a distribution substation, such as a community, a village, or a small industrial and commercial area. It is the basic scope for a single intelligent fusion terminal to collect data and monitor equipment.

[0033] The central control unit is the core management and scheduling center of a large-scale terminal interactive system. It can receive equipment operation data uploaded by various intelligent converged terminals and perform unified analysis and judgment. When the central control unit detects that a terminal is offline, it can coordinate with other associated terminals to receive the data collected by the current terminal in the offline state and perform timely analysis and judgment to avoid delays in handling faults.

[0034] When all devices in a large-scale terminal interaction system are online, the current intelligent fusion terminal, multiple target intelligent fusion terminals, and the central control unit collaboratively train the proxy capability prediction model corresponding to each terminal at specific intervals through federated learning, resulting in a trained proxy capability prediction model for each terminal. When some intelligent fusion terminals are offline, these terminals can predict the proxy capability values ​​of other online intelligent fusion terminals based on their respective trained proxy capability prediction models. The proxy capability value is used to determine the reliability of the intelligent fusion terminal in transmitting data; the higher the proxy capability value, the greater the likelihood that it will be able to transmit data for offline intelligent fusion terminals.

[0035] In this embodiment, the training samples for the agent capability prediction model include historical hardware performance data of the intelligent fusion terminal, historical agent task performance data, network environment and topology relationship data, and corresponding agent capability values.

[0036] Historical hardware performance data includes terminal communication module performance data, such as wireless communication signal strength, transmission rate, and packet loss rate; computing and storage capacity data, such as CPU processing speed and remaining memory / storage space; and battery life and stability data, such as device power supply stability and continuous fault-free operation time.

[0037] Historical proxy task performance data includes proxy transmission success rate, transmission latency, and data integrity. Proxy transmission success rate is the percentage of times data has been successfully and completely delivered when replacing other terminals in the past; transmission latency is the average and maximum time it takes for data to travel from reception to forwarding to the central control; data integrity is whether data loss, errors, or tampering occurred during the proxy process.

[0038] Network environment and topology data include the physical distance to the offline intelligent converged terminal, the network link quality with the central control unit, and the network load within the distribution area.

[0039] The training samples mentioned above can be processed locally by each intelligent fusion terminal and the central control unit under the framework of federated learning. They can be trained collaboratively by sharing model parameters, which ultimately enables the trained agent capability prediction model to accurately predict the agent capability values ​​of each target intelligent fusion terminal when the current intelligent fusion terminal is offline.

[0040] S102: Determine the agent intelligent fusion terminal from multiple target intelligent fusion terminals based on the agent capability value.

[0041] In this embodiment, when the network status of the current intelligent fusion terminal changes from online to offline, the current intelligent fusion terminal can obtain the proxy capability value of each target intelligent fusion terminal based on the trained proxy capability prediction model. Based on the proxy capability value, a proxy intelligent fusion terminal can be determined. For example, a proxy capability threshold can be set. When the proxy capability value corresponding to a target intelligent fusion terminal is greater than or equal to the proxy capability threshold, that target intelligent fusion terminal is designated as a proxy intelligent fusion terminal. There can be multiple proxy intelligent fusion terminals. Alternatively, the target intelligent fusion terminal with the highest proxy capability value can be selected as the proxy intelligent fusion terminal.

[0042] S103: Send a broadcast signal to the outside world. The broadcast signal is used to enable the agent intelligent fusion terminal to send the current intelligent fusion terminal's tag data to the central control.

[0043] In this embodiment, after identifying a proxy smart fusion terminal capable of transmitting offline data, the current smart fusion terminal broadcasts information via a broadcast signal. This broadcast signal may carry information indicating that the current smart fusion terminal is offline, the identifier of the selected "proxy smart fusion terminal" (such as a device ID), and an indication that "tagged data" needs to be transmitted. The broadcast signal's transmission range typically covers all target smart fusion terminals associated with the current smart fusion terminal, ensuring that the selected proxy smart fusion terminal can accurately receive and respond to it. In this embodiment, the indication to transmit "tagged data" included in the broadcast signal is used to determine that the proxy smart fusion terminal needs to assist the offline smart fusion terminal in transmitting offline data for a period of time.

[0044] In one embodiment, the current intelligent fusion terminal filters offline data within a preset time period based on preset rules to obtain labeled data every preset time interval. Both the preset time interval and the preset rules can be set based on the historical offline data of the current intelligent fusion terminal. The preset rules may include: types of historically prone-to-fault data, data collection frequency lower than a preset data collection frequency, etc., and the preset time interval can be set based on experience.

[0045] Assuming there is one agent intelligent fusion terminal, the current intelligent fusion terminal can send all tagged data to the agent intelligent fusion terminal, which then sends the tagged data to the central control. Assuming there are at least N agent intelligent fusion terminals (N≥2), the tagged data can be divided into N equal parts, or it can be divided into N parts based on a specific classification rule. The current intelligent fusion terminal can then send each part of the data to a different agent intelligent fusion terminal, and each agent intelligent fusion terminal will then send its respective data to the central control.

[0046] S104: In response to detecting that the network status of the current intelligent fusion terminal has switched from offline to online, send a proxy data request signal to the central control, and receive the proxy data list and control instructions generated by the central control based on the proxy data request signal.

[0047] In this embodiment, the network status of the current intelligent converged terminal switches from online to offline and then back to online, constituting one cycle, or a complete network failure event. In response to detecting a switch from offline to online, the current intelligent converged terminal needs to send the untagged data from the offline state to the central control. Therefore, it first sends a proxy data request signal to the central control. This proxy data request signal is a data reconciliation and demand application signal, its core function being to prepare for the subsequent retransmission of offline untagged data. This proxy data request signal includes the current intelligent converged terminal's identity information, failure cycle information (offline duration), and a confirmation instruction.

[0048] After the current intelligent fusion terminal sends a proxy data request signal to the central control, the central control searches for the received tagging information based on the instruction, identity information, and fault cycle information, and generates a proxy data list and control instructions. The proxy data list typically includes information such as the type, quantity, and timestamp of the tagging information. The control instructions are operation instructions generated by the central control and sent to the current intelligent fusion terminal. The instruction content can be "Please send data that was not sent offline based on the proxy data list" or "Please update model parameters".

[0049] S105: If the agent data list and control instructions sent by the central control are received, a collaborative parameter request signal is sent to multiple target intelligent fusion terminals, and the key control parameters of the model during the offline period are received from multiple target intelligent fusion terminals.

[0050] The agent data list and control instructions are generated based on agent data request signals.

[0051] In this embodiment, if the current intelligent fusion terminal receives the proxy data list and control instructions sent by the central control unit, it can determine the unmarked data and the marked data that failed to be sent in the offline state based on the proxy data list, and send the above data to the central control unit for storage. In addition, the current intelligent fusion terminal also needs to send a collaborative parameter request signal to multiple target intelligent fusion terminals. This signal is used to obtain model adjustment data during the offline period. After receiving this signal, the multiple target intelligent fusion terminals will send their local key model adjustment parameters to the current intelligent fusion terminal. These key model adjustment parameters include weight adjustment parameters, feature mapping parameters, and loss function correction parameters. After receiving the key model adjustment parameters sent by each terminal, the current intelligent fusion terminal can adjust its local model parameters based on this data.

[0052] S106: Send unlabeled data to the central control based on the agent data list and control instructions, and update the model parameters of the trained agent capability prediction model based on the key control parameters of the model.

[0053] In this embodiment, the current intelligent fusion terminal identifies unmarked data and marked data that failed to be sent while offline based on the agent data list, and sends this data to the central control for storage. This operation fills the "data gap" during the offline period of the current intelligent fusion terminal, ensuring the integrity of the electrical equipment data obtained by the central control. In addition, the current intelligent fusion terminal integrates the weight adjustment parameters, feature mapping parameters, and loss function correction parameters sent by multiple target intelligent fusion terminals to obtain effective control parameters. Based on the effective control parameters, the model parameters of the local agent capability prediction model are updated. The updated model parameters are fed back to the central control through federated learning, ensuring that the agent capability prediction model parameters of all terminals in the entire system remain consistent and avoiding prediction deviations due to differences in model parameters.

[0054] As can be seen from the above, the embodiments of this application are applied to large-scale terminal interaction systems, effectively improving the operational stability and reliability of various terminal devices in large-scale terminal interaction systems under unstable network conditions. When the current intelligent fusion terminal is offline, it can accurately predict the proxy capability values ​​of multiple associated intelligent fusion terminals and identify the proxy intelligent fusion terminal. This process is based on a federated learning collaborative training model, making full use of the data and computing resources of each terminal to ensure the accuracy and reliability of the prediction. When the terminal returns to online status, it sends a proxy data request signal to the central control, receives the proxy data list and control instructions, and then requests the key adjustment parameters of the model during the offline period from multiple target intelligent fusion terminals. Finally, it updates the model parameters and sends unlabeled data based on this information. The entire process forms a complete closed loop, enabling the system to maintain data continuity and model stability even when the network state frequently switches, effectively avoiding data loss and model failure caused by offline conditions, and ensuring that various terminal devices in large-scale terminal interaction systems can operate stably and efficiently in different network environments.

[0055] In one embodiment of this application, determining a proxy intelligent fusion terminal from multiple target intelligent fusion terminals based on proxy capability values ​​includes:

[0056] Select at least one terminal whose proxy capability value is greater than or equal to the preset proxy capability value as the initial candidate set;

[0057] Calculate the physical distance and transformer zone overlap between each terminal in the initial candidate set and the current intelligent fusion terminal;

[0058] Based on physical distance and the overlap of distribution transformer areas, agent intelligent fusion terminals are determined from the initial candidate set.

[0059] In this embodiment, the proxy capability value is used to measure the comprehensive capability index of the intelligent fusion terminal when transmitting data on behalf of others, including but not limited to dimensions such as computing power, communication stability, load capacity, and reliability. The higher the proxy capability value, the stronger the proxy capability. The preset proxy capability value is a pre-set threshold used to filter terminals with basic proxy qualifications; terminals with values ​​lower than this value are excluded due to insufficient capability. If the proxy capability value of an intelligent fusion terminal is greater than or equal to the preset proxy capability value, the intelligent fusion terminal is added to the initial candidate set until all intelligent fusion terminals have been evaluated, resulting in the initial candidate set.

[0060] Because the terminals in the initial candidate set only meet basic qualifications and do not consider adaptability, it is necessary to filter the terminals in the initial candidate set based on physical distance and transformer area overlap to obtain proxy smart converged terminals. Physical distance represents the physical distance between the smart converged terminal and the current smart converged terminal; the closer the distance, the better the match. Transformer areas are basic topological units in the power network. High overlap means that the terminals are more interconnected in the power network, and data interaction is more in line with the collaborative logic of the power system, which can reduce the complexity of cross-regional collaboration.

[0061] Therefore, after normalizing the two parameters and performing a weighted sum, a weighted score is obtained. The terminal with the highest weighted score is selected to improve the agent's real-time performance, stability, and collaborative efficiency.

[0062] In one embodiment of this application, after determining the proxy intelligent fusion terminal from multiple target intelligent fusion terminals based on the proxy capability value, the method further includes:

[0063] The distribution transformer data acquired offline are classified, and the comprehensive score corresponding to each type of data is calculated based on the urgency and timeliness of each type of data.

[0064] Data with a comprehensive score greater than the first threshold are identified as labeled data;

[0065] The tagged data is sent to the agent's intelligent fusion terminal.

[0066] In this embodiment, the distribution transformer data represents the power-related data of the distribution transformer area (power supply area) where the current smart converged terminal is located, which may include voltage, current, power, equipment status, fault information, and power load. Different data have different levels of urgency and timeliness. For example, the distribution transformer data can be divided into categories such as real-time operating data (including current and voltage, etc.), equipment status data (including fault status and operating time, etc.), and historical statistical data (including daily and monthly power consumption, etc.).

[0067] For each category of data after classification, a comprehensive score is calculated based on its urgency and timeliness. Urgency reflects the magnitude of the data's impact on the current operating state of the system in which the intelligent fusion terminal is located, while timeliness reflects how quickly the data loses its value over time. This embodiment uses a specific algorithm, such as a weighted average algorithm, to assign different weights to urgency and timeliness, comprehensively considering these two factors to obtain a comprehensive score for each category of data.

[0068] A different first threshold is set for each data category, or the first thresholds for each category are the same. Data with a comprehensive score greater than the first threshold are identified as labeled data in each category. These labeled data are the data that are currently critical to the system's operation and require timely processing.

[0069] In this embodiment, when the current intelligent fusion terminal is detected to be offline, the pre-defined tagged data is sent to the selected agent intelligent fusion terminal every first time interval. This allows important data to be gradually transmitted while offline for subsequent processing and utilization. In addition, the tagged and untagged data are stored in the current intelligent fusion terminal's storage space, such as a memory, so that when the online state is restored, the untransmitted data can be directly sent to the central control.

[0070] As can be seen from the above, this embodiment, by calculating a comprehensive score based on urgency and timeliness and determining the marked data, can prioritize the processing of data that has a significant impact on system operation and is highly time-sensitive. This ensures that critical data is not missed or delayed in offline mode, improving the system's response speed to important data and guaranteeing stable system operation.

[0071] In addition, when the current intelligent converged terminal is offline, the above methods can transmit important data to the agent intelligent converged terminal in a timely manner, so that some terminals can still maintain a certain data processing and transmission capability when the network is unstable or interrupted, which enhances the system's fault tolerance and adaptability to network failures and improves the stability and reliability of the entire large-scale terminal interaction system.

[0072] In one embodiment of this application, the proxy smart fusion terminal includes at least two target smart fusion terminals; before sending the tagging data to the proxy smart fusion terminal, the method further includes:

[0073] Receive attribute information sent by each agent intelligent fusion terminal. The attribute information includes at least one of network stability, remaining storage capacity and historical data transmission success rate.

[0074] Sending the tagged data to the agent intelligent fusion terminal includes:

[0075] The tagged data is divided into at least two data subsets according to the amount of data or the urgency level, and each data subset is sent to its corresponding agent intelligent fusion terminal.

[0076] The priority ranking of each agent's intelligent converged terminal is determined based on attribute information;

[0077] Based on priority sorting, the corresponding data subset is sent to each agent intelligent fusion terminal in sequence at every first time interval. The time point for sending data to the agent intelligent fusion terminal with the highest priority is the first time interval starting from the detection of the offline state, and the time points for sending data to the agent intelligent fusion terminals with subsequent priorities are successively delayed by a preset time difference.

[0078] For example, consider three data subsets A, B, and C, and three agent intelligent converged terminals D1, D2, and D3. The first time interval is 10 minutes, with a preset time difference of 2 minutes. The priority order of the three intelligent converged terminals is D1 > D2 > D3. Ten minutes after a current intelligent converged terminal switches to offline mode, data subset A is sent to terminal D1; 12 minutes later, data subset B is sent to terminal D2; and 14 minutes later, data subset C is sent to terminal D3.

[0079] In this embodiment, the attribute information of the agent intelligent converged terminal includes at least one of network stability, remaining storage capacity, and historical data transmission success rate. Network stability reflects the reliability of the network environment in which the agent intelligent converged terminal resides and can be measured by indicators such as network packet loss rate and network latency fluctuations. High network stability means a lower probability of data interruption or errors during transmission. Remaining storage capacity refers to the amount of free space currently available for storing data on the agent intelligent converged terminal. Sufficient remaining storage capacity ensures that received tagged data has a place to be stored, avoiding data loss due to insufficient storage space. Historical data transmission success rate refers to the ratio of the number of times the agent intelligent converged terminal successfully transmitted data to the total number of transmissions over a past period. A high historical data transmission success rate indicates that the terminal has high reliability and stability in data transmission.

[0080] If the proxy intelligent fusion terminal includes the above three attribute information, then the three attribute information can be assigned values ​​respectively and then weighted and summed to obtain the final evaluation score. The priority order of each proxy intelligent fusion terminal is determined according to the evaluation score from largest to smallest. This priority order is used to guide the corresponding transmission relationship of data subsets in the following text. The marked data is divided into at least two data subsets according to the data volume or urgency. The current intelligent fusion terminal sends different data subsets to the proxy intelligent fusion terminals respectively. For example, the data subset with the largest data volume is sent to the proxy intelligent fusion terminal with the highest evaluation score / highest priority.

[0081] In this embodiment, at each first time interval, the current intelligent fusion terminal needs to filter and mark the data acquired within the first time interval, and then send the marked data sequentially to the proxy intelligent fusion terminal for data transmission. This embodiment, by sequentially delaying data transmission based on time points, minimizes the possibility of all proxy intelligent fusion terminals simultaneously competing for the Bluetooth transmission channel, thus avoiding channel congestion. In environments where Bluetooth signals are susceptible to interference, time-division transmission reduces the likelihood of signal collisions, lowers the risk of data transmission failure, and improves the overall data transmission success rate.

[0082] In one embodiment of this application, updating the model parameters of a trained agent capability prediction model based on key model control parameters includes:

[0083] The key control parameters of the model sent by multiple target intelligent fusion terminals are classified to obtain the weight adjustment parameters, feature mapping parameters and loss function correction parameters;

[0084] For each type of parameter, based on the model contribution of each intelligent fusion terminal in historical collaborative training, the differentiated weights for sending this type of parameter by different intelligent fusion terminals are determined.

[0085] Based on the key control parameters of the model sent by different intelligent fusion terminals and the differentiated weights corresponding to each type of parameter, the parameters of the same type are weighted and fused to obtain the fusion results corresponding to each type of parameter.

[0086] Based on the fusion results corresponding to various parameters, the model parameters of the trained agent capability prediction model are updated using the network layers corresponding to the trained agent capability prediction model.

[0087] In this embodiment, key model control parameters refer to parameters that have a significant impact on the performance of the trained agent capability prediction model. These parameters are used to optimize the model's prediction effect on agent capability and involve adjustments at different levels of the model. Key model control parameters include weight adjustment parameters, feature mapping parameters, and loss function correction parameters. Weight adjustment parameters are a type of parameter used to adjust the weights of each network layer in the agent capability prediction model. Weights determine the importance of different input features or neurons in the model's computation. Adjusting the weights can change the model's data processing method and prediction results. Feature mapping parameters represent parameters related to feature mapping operations in the model. Feature mapping is the process of transforming input data from one feature space to another. These parameters control the transformation method and rules, affecting the model's extraction and representation of data features. Loss function correction parameters are parameters used to correct the loss function of the agent capability prediction model. The loss function measures the difference between the model's prediction results and the actual results. Correcting the loss function can guide the model towards more accurate training and optimization.

[0088] In this embodiment, different intelligent fusion terminals contribute differently to the model during historical collaborative training. Some terminals contribute significantly to improving model performance, while others contribute less. This embodiment analyzes historical collaborative training data to evaluate the model contribution of each terminal, and then assigns differentiated weights to the same type of key model control parameters sent by each terminal based on their contribution.

[0089] This embodiment uses the key model control parameters sent by different intelligent fusion terminals and the differentiated weights corresponding to each type of parameter to perform weighted fusion on parameters of the same type, thereby obtaining the fusion results corresponding to each type of parameter. For example, for loss function correction parameters, weighted fusion is performed on this type of parameter to obtain the fusion result:

[0090] .

[0091] Where n represents the number of other intelligent converged terminals besides the current intelligent converged terminal, and j represents the j-th intelligent converged terminal. This represents the correction value of the loss function for the j-th intelligent fusion terminal. This represents the differentiated weight of the j-th intelligent fusion terminal.

[0092] In one embodiment, based on the fusion results corresponding to various parameters, the model parameters of the trained agent capability prediction model are updated using the network layer corresponding to the trained agent capability prediction model, including:

[0093] The fusion results corresponding to various parameters are subjected to hierarchical sensitivity verification. Based on the parameter sensitivity threshold of each network layer of the proxy capability prediction model, abnormal parameter components that exceed the parameter sensitivity threshold are screened out. The parameter sensitivity threshold is determined based on the influence of historical model adjustment, which is the influence of parameter adjustment of each network layer in historical training on the model prediction accuracy.

[0094] For abnormal parameter components, a parameter correction factor is generated based on the local interaction data characteristics of the current intelligent fusion terminal in offline state; the parameter correction factor is positively correlated with the distribution deviation of the local interaction data characteristics.

[0095] By concatenating the abnormal parameter components of the same type after correction based on the parameter correction factor with the normal parameter components that do not exceed the parameter sensitivity threshold, the parameter set to be updated for each network layer is obtained.

[0096] Based on the parameter set to be updated for each network layer, the model parameters of the trained agent capability prediction model are updated using the network layer corresponding to the trained agent capability prediction model.

[0097] In this embodiment, the object of the layered sensitivity verification is the fusion result corresponding to each type of parameter in the key control parameters of the model, that is, the parameter adjustment amount. "Layered" refers to performing sensitivity analysis on the key control parameters of the model layer by layer according to the network layer structure of the agent capability prediction model.

[0098] Agent capability prediction models typically consist of multiple network layers with different functions, such as input layers, convolutional layers, pooling layers, and fully connected layers. Each network layer plays a unique role in the prediction process and exhibits significantly different sensitivities to parameter changes. For example, convolutional layers are responsible for extracting features from images or data; even small changes in their parameters can significantly impact the accuracy of feature extraction, thus affecting the overall predictive performance of the model. Pooling layers, on the other hand, are primarily used for dimensionality reduction and extracting key features, and their sensitivity to parameter changes may be relatively low. Therefore, applying a uniform sensitivity standard to the parameters of all network layers can lead to the neglect of changes in parameters of certain critical network layers or oversensitivity to parameters of non-critical network layers, thereby affecting the model's optimization performance. By implementing hierarchical sensitivity verification, more reasonable and accurate parameter verification standards can be developed based on the characteristics of different network layers, improving the effectiveness of model parameter updates.

[0099] After obtaining the fusion results corresponding to each type of parameter, they are classified according to their respective network layers. Then, for each network layer, the fusion result of the layer's parameters is compared with a pre-set parameter sensitivity threshold, and abnormal parameter components that exceed the parameter sensitivity threshold are filtered out.

[0100] In this embodiment, for abnormal parameter components, a parameter correction factor is generated based on the local interaction data characteristics of the current intelligent fusion terminal in its offline state. The abnormal parameter components are then corrected based on this parameter correction factor. The parameter correction factor is positively correlated with the distribution deviation of the local interaction data characteristics. The distribution deviation of the local interaction data characteristics represents the difference in distribution between the local interaction data characteristics generated when the current intelligent fusion terminal is offline and the "standard data characteristics" used during model training. The greater the difference, the larger the value of the parameter correction factor.

[0101] In this embodiment, abnormal parameter components of the same type, corrected by parameter correction factors, are concatenated with normal parameter components that do not exceed the parameter sensitivity threshold to obtain the parameter set to be updated for each network layer. Based on the parameter set to be updated for each network layer, the model parameters of the trained agent capability prediction model are updated using the network layer corresponding to the trained agent capability prediction model.

[0102] As can be seen from the above, this embodiment uses hierarchical sensitivity verification to screen outlier parameter components, which can accurately locate parameters that have a significant impact on the model's prediction accuracy and improve the targeting of parameter adjustments. It generates parameter correction factors based on local interactive data characteristics, allowing for flexible correction of outlier parameters in combination with actual conditions, thus enhancing the model's adaptability. By concatenating the corrected outlier parameters with normal parameters and updating the model parameters, it retains valid information while optimizing the outlier parts, helping to improve the prediction accuracy and stability of the proxy capability prediction model, enabling it to make more accurate predictions in different scenarios.

[0103] In one embodiment of this application, the parameter adjustment of each network layer during historical training includes the number of parameter adjustments and the adjustment magnitude of the parameters of the l-th layer each time;

[0104] Based on historical models, the impact is adjusted to determine the parameter sensitivity threshold, including:

[0105] The parameter sensitivity threshold is determined based on the following formula:

[0106]

[0107] in, The parameter sensitivity threshold of the l-th network layer. These are the weighting coefficients. This refers to the number of parameter adjustments made to the l-th network layer during historical training. Let be the change in model prediction accuracy after the i-th adjustment of the parameters of the l-th layer. Let be the adjustment range of the parameters of the ith layer. denoted as the standard deviation of the parameter adjustment magnitude of the l-th layer network during historical training.

[0108] Corresponding to the interaction method of the multi-intelligent fusion terminal in the above embodiment, Figure 2 This is a structural block diagram of an interactive device for multiple intelligent fusion terminals provided in an embodiment of this application. The device is applied to a large-scale terminal interaction system, which, in addition to this device, also includes multiple interconnected intelligent fusion terminals and a central control unit. Each intelligent fusion terminal interacts with the central control unit. The multiple interconnected intelligent fusion terminals include a current intelligent fusion terminal and multiple target intelligent fusion terminals, where the current intelligent fusion terminal is any one of the multiple interconnected intelligent fusion terminals. For ease of explanation, only the parts relevant to the embodiment of this application are shown. (Reference) Figure 2 The interactive device 20 of the multi-intelligent fusion terminal includes: a first detection unit 21, a first processing unit 22, a signal transmission unit 23, a second detection unit 24, a second processing unit 25, and a model update unit 26.

[0109] The first detection unit 21 is used to predict the proxy capability values ​​of multiple target intelligent fusion terminals based on a trained proxy capability prediction model in response to the detection that the network status of the current intelligent fusion terminal has switched from online to offline.

[0110] Among them, the trained agent capability prediction model is obtained by federated learning collaborative training based on multiple target intelligent fusion terminals and the central control when the current intelligent fusion terminal is online.

[0111] The first processing unit 22 is used to determine the agent intelligent fusion terminal from multiple target intelligent fusion terminals based on the agent capability value;

[0112] The signal transmitting unit 23 is used to transmit broadcast signals to the outside world. The broadcast signals are used to enable the agent intelligent fusion terminal to send the current intelligent fusion terminal's tag data to the central control.

[0113] The second detection unit 24 is used to send a proxy data request signal to the central control in response to the detection that the network status of the current intelligent fusion terminal has switched from offline to online, and to receive a proxy data list and control instructions generated by the central control based on the proxy data request signal.

[0114] The second processing unit 25 is used to send a collaborative parameter request signal to multiple target intelligent fusion terminals if it receives a proxy data list and control instructions sent by the central control, and to receive the key control parameters of the model during the offline period sent by multiple target intelligent fusion terminals. The proxy data list and control instructions are generated based on the proxy data request signal.

[0115] The model update unit 26 is used to send unlabeled data to the master controller based on the agent data list and control instructions, and to update the model parameters of the trained agent capability prediction model based on the key control parameters of the model.

[0116] In one embodiment of this application, the first processing unit 22 is specifically used for:

[0117] Select at least one terminal whose proxy capability value is greater than or equal to the preset proxy capability value as the initial candidate set;

[0118] Calculate the physical distance and transformer zone overlap between each terminal in the initial candidate set and the current intelligent fusion terminal;

[0119] Based on physical distance and the overlap of distribution transformer areas, agent intelligent fusion terminals are determined from the initial candidate set.

[0120] In one embodiment of this application, after determining the agent intelligent fusion terminal from multiple target intelligent fusion terminals based on the agent capability value, the interaction device 20 of the multiple intelligent fusion terminals further includes a third processing unit.

[0121] The third processing unit is used to classify the distribution transformer data of this area acquired in the offline state, and calculate the comprehensive score corresponding to each type of data based on the urgency and timeliness of each type of data.

[0122] Data with a comprehensive score greater than the first threshold are identified as labeled data;

[0123] The tagged data is sent to the agent's intelligent fusion terminal.

[0124] In one embodiment of this application, the proxy smart fusion terminal includes at least two target smart fusion terminals; before sending the tag data to the proxy smart fusion terminal, the interaction device 20 of the multiple smart fusion terminals further includes a receiving unit;

[0125] The receiving unit is used to receive attribute information sent by each agent intelligent fusion terminal. The attribute information includes at least one of network stability, remaining storage capacity, and historical data transmission success rate.

[0126] The third processing unit is specifically used for:

[0127] The tagged data is divided into at least two data subsets according to the amount of data or the urgency level, and each data subset is sent to its corresponding agent intelligent fusion terminal.

[0128] The priority ranking of each agent's intelligent converged terminal is determined based on attribute information;

[0129] Based on priority sorting, the corresponding data subset is sent to each agent intelligent fusion terminal in sequence at every first time interval. The time point for sending data to the agent intelligent fusion terminal with the highest priority is the first time interval starting from the detection of the offline state, and the time points for sending data to the agent intelligent fusion terminals with subsequent priorities are successively delayed by a preset time difference.

[0130] In one embodiment of this application, the model update unit 26 is specifically used for:

[0131] The key control parameters of the model sent by multiple target intelligent fusion terminals are classified to obtain the weight adjustment parameters, feature mapping parameters and loss function correction parameters;

[0132] For each type of parameter, based on the model contribution of each intelligent fusion terminal in historical collaborative training, the differentiated weights for sending this type of parameter by different intelligent fusion terminals are determined.

[0133] Based on the key control parameters of the model sent by different intelligent fusion terminals and the differentiated weights corresponding to each type of parameter, the parameters of the same type are weighted and fused to obtain the fusion results corresponding to each type of parameter.

[0134] Based on the fusion results corresponding to various parameters, the model parameters of the trained agent capability prediction model are updated using the network layers corresponding to the trained agent capability prediction model.

[0135] In one embodiment of this application, the model update unit 26 is specifically used for:

[0136] The fusion results corresponding to various parameters are subjected to hierarchical sensitivity verification. Based on the parameter sensitivity threshold of each network layer of the proxy capability prediction model, abnormal parameter components that exceed the parameter sensitivity threshold are screened out. The parameter sensitivity threshold is determined based on the influence of historical model adjustment, which is the influence of parameter adjustment of each network layer in historical training on the model prediction accuracy.

[0137] For abnormal parameter components, a parameter correction factor is generated based on the local interaction data characteristics of the current intelligent fusion terminal in offline state; the parameter correction factor is positively correlated with the distribution deviation of the local interaction data characteristics.

[0138] By concatenating the abnormal parameter components of the same type after correction based on the parameter correction factor with the normal parameter components that do not exceed the parameter sensitivity threshold, the parameter set to be updated for each network layer is obtained.

[0139] Based on the parameter set to be updated for each network layer, the model parameters of the trained agent capability prediction model are updated using the network layer corresponding to the trained agent capability prediction model.

[0140] In one embodiment of this application, the parameter adjustment of each network layer during historical training includes the number of parameter adjustments and the adjustment magnitude of the parameters of the l-th layer each time;

[0141] Model update unit 26 is specifically used to: determine the parameter sensitivity threshold based on the following formula:

[0142]

[0143] in, The parameter sensitivity threshold of the l-th network layer. These are the weighting coefficients. This refers to the number of parameter adjustments made to the l-th network layer during historical training. Let be the change in model prediction accuracy after the i-th adjustment of the parameters of the l-th layer. Let be the adjustment range of the parameters of the ith layer. denoted as the standard deviation of the parameter adjustment magnitude of the l-th layer network during historical training.

[0144] See Figure 3 , Figure 3 This is a schematic block diagram of an electronic device provided according to an embodiment of this application. Figure 3The electronic device 300 in this embodiment may include one or more processors 301, one or more input devices 302, one or more output devices 303, and one or more memories 304. The processors 301, input devices 302, output devices 303, and memories 304 communicate with each other via a communication bus 305. The memories 304 store computer programs, including program instructions. The processors 301 execute the program instructions stored in the memories 304. Specifically, the processors 301 are configured to invoke the program instructions to perform the functions of the units in the above-described device embodiments, for example... Figure 2 The functions of the first detection unit 21, the first processing unit 22, the signal sending unit 23, the second detection unit 24, the second processing unit 25, and the model update unit 26 are shown.

[0145] It should be understood that, in the embodiments of this application, the processor 301 may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0146] Input device 302 may include a touchpad, a fingerprint sensor (for collecting the user's fingerprint information and fingerprint orientation information), a microphone, etc., and output device 303 may include a display (LCD, etc.), a speaker, etc.

[0147] The memory 304 may include read-only memory and random access memory, and provides instructions and data to the processor 301. A portion of the memory 304 may also include non-volatile random access memory.

[0148] In specific implementations, the processor 301, input device 302, and output device 303 described in the embodiments of this application can execute the implementation method described in the interaction method of the multi-intelligent fusion terminal provided in the embodiments of this application, or they can execute the implementation method of the electronic device described in the embodiments of this application, which will not be repeated here.

[0149] In another embodiment of this application, a computer-readable storage medium is provided. This computer-readable storage medium stores a computer program, which includes program instructions. When executed by a processor, the program instructions implement all or part of the processes in the methods described above. Alternatively, the computer program can instruct related hardware to complete the process. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include any entity or device capable of carrying computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0150] The computer-readable storage medium can be an internal storage unit of the electronic device in any of the foregoing embodiments, such as a hard disk or memory of the electronic device. The computer-readable storage medium can also be an external storage device of the electronic device, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc., provided on the electronic device. Furthermore, the computer-readable storage medium can include both internal and external storage units of the electronic device. The computer-readable storage medium is used to store computer programs and other programs and data required by the electronic device. The computer-readable storage medium can also be used to temporarily store data that has been output or will be output.

[0151] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this application.

[0152] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the electronic devices and units described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0153] In the several embodiments provided in this application, it should be understood that the disclosed electronic devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces or units, or it may be an electrical, mechanical, or other form of connection.

[0154] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of the embodiments of this application, depending on actual needs.

[0155] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0156] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. An interaction method for multiple intelligent fusion terminals, characterized in that, include: In response to the detection that the network status of the current intelligent fusion terminal has switched from online to offline, the agent capability values ​​of multiple target intelligent fusion terminals are predicted based on the trained agent capability prediction model. The prediction model is obtained by federated learning collaborative training based on the multiple target intelligent fusion terminals and the central control when the current intelligent fusion terminal is in the online state. The agent intelligent fusion terminal is determined from the plurality of target intelligent fusion terminals based on the agent capability value; The system sends a broadcast signal to the outside world, which enables the agent intelligent fusion terminal to send the current tag data of the intelligent fusion terminal to the central control. In response to detecting that the network status of the current intelligent fusion terminal has switched from offline to online, a proxy data request signal is sent to the central control, and a proxy data list and control instructions generated based on the proxy data request signal are received from the central control; If the proxy data list and control instructions sent by the central control are received, a collaborative parameter request signal is sent to the multiple target intelligent fusion terminals, and the key control parameters of the model during the offline period are received from the multiple target intelligent fusion terminals. The proxy data list and control instructions are generated based on the proxy data request signal. Based on the agent data list and control instructions, unlabeled data is sent to the master controller, and the key control parameters of the model sent by multiple target intelligent fusion terminals are classified to obtain weight adjustment parameters, feature mapping parameters and loss function correction parameters. For each type of parameter, based on the model contribution of each intelligent fusion terminal in historical collaborative training, the differentiated weights for sending this type of parameter by different intelligent fusion terminals are determined. Based on the key control parameters of the model sent by different intelligent fusion terminals and the differentiated weights corresponding to each type of parameter, the parameters of the same type are weighted and fused to obtain the fusion results corresponding to each type of parameter. The fusion results corresponding to various parameters are subjected to hierarchical sensitivity verification. Based on the parameter sensitivity threshold of each network layer of the proxy capability prediction model, abnormal parameter components that exceed the parameter sensitivity threshold are screened out. The parameter sensitivity threshold is determined based on the historical model adjustment influence degree, which is the influence degree of the adjustment of parameters of each network layer in historical training on the model prediction accuracy. For abnormal parameter components, a parameter correction factor is generated based on the characteristics of local interactive data in the current offline state of the intelligent fusion terminal; The parameter correction factor is positively correlated with the distribution deviation of the local interactive data features; By concatenating the abnormal parameter components of the same type after correction based on the parameter correction factor with the normal parameter components that do not exceed the parameter sensitivity threshold, the parameter set to be updated for each network layer is obtained. Based on the parameter set to be updated for each network layer, the model parameters of the trained agent capability prediction model are updated using the network layer corresponding to the trained agent capability prediction model.

2. The method as described in claim 1, characterized in that, The step of determining the agent intelligent fusion terminal from the plurality of target intelligent fusion terminals based on the agent capability value includes: At least one terminal whose proxy capability value is greater than or equal to a preset proxy capability value is selected as an initial candidate set; Calculate the physical distance and transformer zone overlap between each terminal in the initial candidate set and the current intelligent fusion terminal; Based on the physical distance and the overlap of the distribution transformer area, the agent intelligent fusion terminal is determined from the initial candidate set.

3. The method as described in claim 1, characterized in that, After determining the proxy intelligent fusion terminal from the plurality of target intelligent fusion terminals based on the proxy capability value, the method further includes: The distribution transformer data acquired offline are classified, and the comprehensive score corresponding to each type of data is calculated based on the urgency and timeliness of each type of data. Data with a comprehensive score greater than the first threshold are identified as labeled data; The tagged data is sent to the agent intelligent fusion terminal.

4. The method as described in claim 3, characterized in that, The proxy intelligent fusion terminal includes at least two target intelligent fusion terminals; before sending the tagging data to the proxy intelligent fusion terminal, the method further includes: Receive attribute information sent by each agent intelligent fusion terminal, the attribute information including at least one of network stability, remaining storage capacity and historical data transmission success rate; The step of sending the labeled data to the agent intelligent fusion terminal includes: The labeled data is divided into at least two data subsets according to the amount of data or the urgency level, and each data subset is sent to its corresponding agent intelligent fusion terminal. The priority ranking of each agent intelligent fusion terminal is determined based on the attribute information; Based on the priority sorting, a corresponding subset of data is sent to each agent intelligent fusion terminal in sequence at each first time interval. The time point for sending data to the agent intelligent fusion terminal with the highest priority is the first time interval starting from the detection of the offline state, and the time points for sending data to the agent intelligent fusion terminals with subsequent priorities are successively delayed by a preset time difference.

5. The method as described in claim 1, characterized in that, The parameter adjustments for each network layer during historical training include the number of parameter adjustments and the magnitude of each adjustment to the parameters of layer l. The sensitivity threshold of the parameter is determined based on the adjustment of the impact based on the historical model, including: The parameter sensitivity threshold is determined based on the following formula: in, The parameter sensitivity threshold of the l-th network layer. These are the weighting coefficients. This refers to the number of parameter adjustments made to the l-th network layer during historical training. Let be the change in model prediction accuracy after the i-th adjustment of the parameters of the l-th layer. Let be the adjustment range of the parameters of the ith layer. denoted as the standard deviation of the parameter adjustment magnitude of the l-th layer network during historical training.

6. An interactive device for a multi-intelligent fusion terminal, characterized in that, include: The first detection unit is used to respond to the detection that the network status of the current intelligent fusion terminal has switched from online to offline, and to predict the proxy capability values ​​of multiple target intelligent fusion terminals based on a trained proxy capability prediction model. The prediction model is obtained by federated learning collaborative training based on multiple target intelligent fusion terminals and the central control when the current intelligent fusion terminal is in the online state. The multiple target intelligent fusion terminals are all devices that are associated with the current intelligent fusion terminal; The first processing unit is used to determine the agent intelligent fusion terminal from the plurality of target intelligent fusion terminals based on the agent capability value of the plurality of target intelligent fusion terminals; A signal transmitting unit is used to transmit broadcast signals to the outside world. The broadcast signals are used to enable the agent intelligent fusion terminal to send the current intelligent fusion terminal's tag data to the central control. The second detection unit is used to send a proxy data request signal to the central control unit in response to the detection that the network status of the current intelligent fusion terminal has switched from offline to online, and to receive a proxy data list and control instructions generated by the central control unit based on the proxy data request signal. The second processing unit is configured to, if it receives the proxy data list and control instructions sent by the central control, send a collaborative parameter request signal to multiple target intelligent fusion terminals, and receive the key control parameters of the model during the offline period sent by multiple target intelligent fusion terminals, wherein the proxy data list and control instructions are generated based on the proxy data request signal; The model update unit is used to send unlabeled data to the master controller based on the agent data list and control instructions, and to classify the key model control parameters sent by multiple target intelligent fusion terminals to obtain weight adjustment parameters, feature mapping parameters and loss function correction parameters. For each type of parameter, based on the model contribution of each intelligent fusion terminal in historical collaborative training, the differentiated weights for sending this type of parameter by different intelligent fusion terminals are determined. Based on the key control parameters of the model sent by different intelligent fusion terminals and the differentiated weights corresponding to each type of parameter, the parameters of the same type are weighted and fused to obtain the fusion results corresponding to each type of parameter. The fusion results corresponding to various parameters are subjected to hierarchical sensitivity verification. Based on the parameter sensitivity threshold of each network layer of the proxy capability prediction model, abnormal parameter components that exceed the parameter sensitivity threshold are screened out. The parameter sensitivity threshold is determined based on the historical model adjustment influence degree, which is the influence degree of the adjustment of parameters of each network layer in historical training on the model prediction accuracy. For abnormal parameter components, a parameter correction factor is generated based on the characteristics of local interactive data in the current offline state of the intelligent fusion terminal; The parameter correction factor is positively correlated with the distribution deviation of the local interactive data features; By concatenating the abnormal parameter components of the same type after correction based on the parameter correction factor with the normal parameter components that do not exceed the parameter sensitivity threshold, the parameter set to be updated for each network layer is obtained. Based on the parameter set to be updated for each network layer, the model parameters of the trained agent capability prediction model are updated using the network layer corresponding to the trained agent capability prediction model.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1 to 5.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 5.

Images