A federated learning optimization system for low-power positioning on mobile devices

Abstract

This invention relates to the field of power management technology, specifically a federated learning optimization system for low-power positioning of mobile terminals. It acquires multi-dimensional contextual information of wireless communication terminals, including real-time location coordinates, task progress, predetermined time planning, and dynamic environmental parameters. Based on federated learning, training information from multiple wireless communication terminals is used to construct an activity state prediction model. Using the real-time location coordinates and progress status of the wireless communication terminals, the activity state prediction model generates the expected activity state of the terminals. The expected state is compared with the current activity state, and a power optimization decision model is applied to dynamically adjust the power consumption configuration of the wireless communication terminals. Simultaneously, the system monitors the actual operating trajectory and state of the wireless communication terminals for anomalies; if anomalies are detected, correction information is generated to adjust the activity state prediction model and the power optimization decision model to optimize the performance and energy efficiency of the wireless communication terminals.

Description

Technical Field

[0001] This invention relates to the field of power management technology, specifically to a federated learning optimization system for low-power positioning in mobile devices. Background Technology

[0002] In Internet of Things (IoT) applications, ultra-low power consumption is one of the core technological challenges, especially for mobile devices that require long-term deployment and are difficult to charge or replace frequently. In complex environments such as large industrial plants and construction sites, wireless communication terminals are typically deployed on these assets to improve management efficiency and asset security by enabling state awareness of numerous mobile assets. However, these application scenarios place extremely stringent requirements on the battery life of these wireless communication terminals, often requiring performance levels lasting several months or even years.

[0003] Location information acquisition is fundamental to many state-aware applications; however, the positioning process itself is often one of the main sources of energy consumption for mobile devices. Existing positioning technologies, while pursuing accuracy, often come at the cost of high power consumption, making it difficult to meet the ultra-long battery life requirements of the aforementioned scenarios.

[0004] To reduce power consumption, existing technologies typically employ a periodic sleep / wake-up strategy, where the device remains in a low-power sleep state most of the time, only waking up briefly at preset intervals to perform sensing, positioning, and communication tasks. However, this fixed, non-intelligent periodic strategy is often suboptimal: frequent wake-ups still cause unnecessary energy waste when the asset status is stable; while fixed long sleep intervals may lead to delays in information acquisition when the asset status needs close monitoring.

[0005] Federated learning, as a distributed machine learning technique, can train models using data from multiple devices while protecting data privacy. While federated learning can be used to optimize positioning accuracy or learn device behavior patterns, the process itself (communication during local model training and updates) introduces significant energy consumption on mobile devices. Without optimization, frequent federated learning interactions may actually exacerbate the device's energy burden.

[0006] In recent years, monitoring accelerometers to detect movement events and wake up devices, or monitoring battery voltage to induce deep sleep, have been used. However, in the complex scenarios of industrial / construction site asset management, the parameters and events that need to be monitored are multi-dimensional. Furthermore, how to combine the distributed intelligence of federated learning to dynamically and collaboratively optimize triggering conditions and power mode switching strategies to achieve extremely low power consumption while ensuring necessary information perception remains a problem that current technologies have not been able to effectively solve.

[0007] To address this, a federated learning optimization system for low-power positioning on mobile devices is proposed. Summary of the Invention

[0008] This invention provides a federated learning optimization system for low-power positioning on mobile devices. It acquires multi-dimensional contextual information of the wireless communication terminal, including its real-time location coordinates, task progress, predetermined time plan, and dynamic environmental parameters. Based on federated learning, it constructs an activity state prediction model from training information obtained from multiple wireless communication terminals. Using the real-time location coordinates and progress status of the wireless communication terminals, the activity state prediction model generates the terminal's expected activity state. By comparing the expected state with the current activity state, a power optimization decision model is applied to dynamically adjust the power consumption configuration of the wireless communication terminal. Simultaneously, it monitors the actual operating trajectory and state of the wireless communication terminal for anomalies; if anomalies are detected, correction information is generated to adjust the activity state prediction model and the power optimization decision model to optimize the performance and energy efficiency of the wireless communication terminal.

[0009] To achieve the above objectives, the present invention provides the following technical solution:

[0010] A federated learning optimization system for low-power positioning on mobile devices, comprising:

[0011] The context information acquisition module is used to acquire multi-dimensional context information related to the wireless communication terminal. The multi-dimensional context information includes: the real-time location coordinates of the wireless communication terminal, the progress status of the tasks associated with the wireless communication terminal, the predetermined time plan, and the dynamic parameters of the environment in which the wireless communication terminal is located.

[0012] The activity state prediction module is used to construct an activity state prediction model based on federated learning by acquiring training information from multiple wireless communication terminals; and to generate the expected activity state of the wireless communication terminal based on the acquired real-time location coordinates and progress status through the activity state prediction model.

[0013] The power consumption optimization control module is used to adjust the power consumption operation configuration of the wireless communication terminal by comparing the expected activity state of wireless communication with the current activity state and applying a power consumption optimization decision model.

[0014] The status verification and correction module is used to monitor whether there are any abnormalities in the actual operation trajectory and status of the wireless communication terminal by combining the predetermined time plan of the task, environmental dynamic parameters and expected activity status; if an abnormality is detected, correction information is generated to adjust the activity status prediction model and power consumption optimization decision model.

[0015] Preferably, the real-time location coordinates are obtained through wireless network positioning, including latitude and longitude values; the task progress status is obtained through a task management system, and the progress status includes completion percentage, current task stage identifier, and Boolean value; the task's scheduled time plan is obtained through a task scheduling system, including the task's planned start time, end time, and key time nodes; the environmental dynamic parameters are collected through sensors, including physical values ​​and environmental state classifications.

[0016] Preferably, the activity state prediction model includes:

[0017] The input layer is used to receive and process the normalized real-time position coordinates and the progress status;

[0018] The recurrent neural network layer is used to process time-series dependencies and learn the activity patterns of the wireless communication terminal based on historical location and task progress;

[0019] The output layer receives the output from the last hidden layer and generates a probability distribution and specific category label for the expected activity state of the wireless communication terminal in the next time step through an activation function; the expected activity state includes stationary, low-speed movement, high-speed movement, and stationary at the task location.

[0020] Preferably, the power consumption optimization decision model includes a state deviation calculation unit, a trigger condition adjustment unit, a power mode conversion unit, and a strategy collaborative optimization unit;

[0021] The state deviation calculation unit is used to calculate the deviation between the real-time perceived state of the wireless communication terminal and the dynamic expected state, and to generate the degree of deviation of the operating state of the wireless communication terminal.

[0022] The trigger condition adjustment unit dynamically adjusts the operation trigger conditions for positioning and context awareness based on the degree of deviation. Specifically, it adjusts the data acquisition frequency of the wireless network and sensors, the network communication frequency, and the sensitivity threshold for detecting state changes.

[0023] The power mode switching unit comprehensively analyzes the degree of deviation and the adjusted triggering conditions, generates and executes power mode switching instructions, and switches between multiple preset power modes;

[0024] The strategy collaborative optimization unit utilizes a federated learning framework to comprehensively analyze the deviation degree, trigger condition adjustment efficiency, and power mode switching data of multiple wireless communication terminals, and optimizes the trigger condition adjustment strategy and power mode switching logic.

[0025] Preferably, the deviation degree generation process includes:

[0026] The latest positioning data is obtained through the real-time sensors of the wireless communication terminal to calculate the current actual activity state; the expected activity state generated by the activity state prediction model is compared with the real-time inferred current actual activity state to generate the degree of deviation; the comparison method is as follows: if the activity state is a predefined discrete category, then category matching comparison is performed; if the activity state is a continuous value, then the difference measure between the two is calculated.

[0027] The degree of deviation is a Boolean value indicating the consistency between the two, a numerical value quantifying the degree of deviation, and a signal identifying the specific expected state and the current state category.

[0028] Preferably, the specific process for determining whether there are any abnormalities in the actual operating trajectory and status of the wireless communication terminal includes:

[0029] The actual location coordinates acquired by the wireless communication terminal over a period of time are compared with the expected trajectory deduced based on the expected activity state and the predetermined time planning of the task, and the trajectory deviation is calculated.

[0030] The degree of deviation is detected to determine whether there is a persistent or / and significant state mismatch; the current location and activity status of the wireless communication terminal are detected to ensure they match the predetermined time schedule of the task.

[0031] Evaluate whether there is a logical contradiction between the activity state of the wireless communication terminal and the environmental dynamic parameters;

[0032] If any result exceeds the preset threshold or / and violates the preset logic rules, an anomaly is determined.

[0033] Preferably, the specific process of adjusting the activity state prediction model and the power consumption optimization decision model includes:

[0034] When an anomaly is detected, the status verification and correction module generates correction information, including the anomaly type, occurrence time, location, degree of deviation, and relevant context information.

[0035] Adjust the activity state prediction model: Use anomalous event data with correct labels as new training samples to incrementally train the activity state prediction model on the wireless communication terminal locally.

[0036] Adjust the power consumption optimization decision model: Based on the anomaly type in the correction information, adjust the parameters in the trigger condition adjustment unit and the switching logic of the power mode conversion unit.

[0037] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0038] 1. This invention integrates multi-dimensional contextual information such as real-time positioning, task progress, and environmental dynamics from wireless communication terminals. It constructs an activity state prediction model through federated learning and employs recurrent neural networks to mine historical positioning data and task evolution patterns, predicting states such as stationary, slow movement, high-speed sprint, and task pause, and formulating low-power operation strategies. By intelligently adjusting the wireless network and sensor acquisition frequencies, it achieves a balance between power consumption and positioning accuracy, effectively reducing energy consumption, extending device battery life, and adapting to different scenario changes. It exhibits excellent performance in power management and communication stability, enhancing the system's intelligence and adaptability.

[0039] 2. This invention identifies data deviations and logical inconsistencies by integrating task-defined time planning, environmental parameters, and real-time terminal trajectories. It compares the predicted and actual states in real time, and through quantified deviation calculation and category matching, determines whether there are continuous anomalies or sudden malfunctions during terminal operation, helping to quickly identify potential problems. When an operational anomaly is detected, the system generates detailed correction information, covering the anomaly type, occurrence time, location information, and deviation value. It uses incremental training with local data to precisely fine-tune the model, effectively achieving fault self-recovery. This not only improves the accuracy of anomaly identification but also optimizes data transmission and processing flows, ensuring the overall security and stability of the system and enhancing the collaborative decision-making capabilities between terminals.

[0040] 3. This invention is built upon a federated learning framework, enabling multi-terminal data sharing and distributed model training, breaking through the bottleneck of traditional single-machine training and promoting collaborative optimization among nodes. Through cross-terminal information fusion, it can update activity state prediction and power consumption decision models in real time, forming a dynamic feedback loop to ensure that each terminal adaptively adjusts its operating mode according to the actual environment. By fully utilizing the complementary advantages of data between terminals, the power consumption mode and triggering conditions are adjusted in real time through a policy collaborative optimization unit, which not only improves overall energy efficiency but also significantly improves positioning response speed and accuracy. This effectively reduces management costs and improves the overall network operating efficiency and security level. Attached Figure Description

[0041] Figure 1 A schematic diagram of a federated learning optimization system structure for low-power positioning on a mobile device provided by the present invention;

[0042] Figure 2 This is a schematic diagram of the power consumption optimization decision model structure provided in an embodiment of the present invention;

[0043] Figure 3 This is a schematic diagram of a low-power positioning process for mobile devices provided in an embodiment of the present invention. Detailed Implementation

[0044] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0045] Example 1

[0046] Please see Figure 1 This invention provides a federated learning optimization system for low-power positioning on mobile devices, the technical solution of which is as follows:

[0047] A federated learning optimization system for low-power positioning on mobile devices, comprising:

[0048] The context information acquisition module is used to acquire multi-dimensional context information related to the wireless communication terminal. The multi-dimensional context information includes: the real-time location coordinates of the wireless communication terminal, the progress status of the tasks associated with the wireless communication terminal, the predetermined time plan, and the dynamic parameters of the environment in which the wireless communication terminal is located.

[0049] Furthermore, the real-time location coordinates are obtained through wireless network positioning, including latitude and longitude values; the task progress status is obtained through the task management system, and the progress status includes completion percentage, current task stage identifier, and Boolean value; the task's scheduled time plan is obtained through the task scheduling system, including the task's planned start time, end time, and key time nodes; the environmental dynamic parameters are collected through sensors, including physical values ​​and environmental state classifications.

[0050] In this embodiment, by collecting multi-dimensional contextual information of the wireless communication terminal, comprehensive and real-time monitoring of the terminal's status is achieved, providing solid data support for subsequent optimization decisions. Real-time location coordinates, including latitude and longitude values, are obtained using wireless network positioning technology to ensure accurate tracking of the terminal's location. Simultaneously, information such as progress status, completion percentage, task stage identifiers, and predetermined start, end, and key time nodes are collected in real time through the task management and scheduling systems, enabling a comprehensive understanding of task execution. This improves data integrity and accuracy and provides high-quality input data for subsequent modules such as activity status prediction, power consumption optimization, and abnormal status detection. This achieves intelligent and precise system operation and low-power goals, while also improving the overall system's response speed and adaptability, providing technical assurance for the efficient and stable operation of the wireless communication terminal in various complex scenarios.

[0051] The activity state prediction module is used to construct an activity state prediction model based on federated learning by acquiring training information from multiple wireless communication terminals; and to generate the expected activity state of the wireless communication terminal based on the acquired real-time location coordinates and progress status through the activity state prediction model.

[0052] Furthermore, the activity state prediction model includes:

[0053] The input layer is used to receive and process the normalized real-time position coordinates and the progress status;

[0054] The recurrent neural network layer is used to process time-series dependencies and learn the activity patterns of the wireless communication terminal based on historical location and task progress;

[0055] The output layer receives the output from the last hidden layer and generates a probability distribution and specific category label for the expected activity state of the wireless communication terminal in the next time step through an activation function; the expected activity state includes stationary, low-speed movement, high-speed movement, and stationary at the task location.

[0056] In this embodiment, by introducing federated learning and deep recurrent neural networks, multi-terminal collaborative training can be achieved to capture the historical activity patterns of wireless communication terminals, enabling accurate prediction of real-time status. Using normalized real-time location coordinates and progress status as input, effective data preprocessing is first performed through the input layer, followed by in-depth mining of the dependencies between time series using a recurrent neural network layer to accurately learn the movement patterns of the terminal at different task stages. The output layer outputs the state probability distribution and specific category labels for the next time step based on an activation function, covering various activity states such as stationary, low-speed movement, high-speed movement, and stationary at the task location. This not only improves prediction accuracy and response speed but also provides reliable data support for subsequent power consumption optimization and intelligent scheduling. Compared with rule-based prediction methods and single-device machine learning models, the activity state prediction model provided by this invention effectively improves prediction accuracy, state transition sensitivity, and inter-device collaboration capabilities. Rule-based prediction methods include decision tree models and linear regression models, as detailed in Table 1.

[0057] Table 1. Performance Comparison of Activity State Prediction Model and Existing Algorithms

[0058]

[0059] Prediction accuracy is primarily assessed by comparing model predictions with actual state labels on a test dataset. State transition detection sensitivity is measured by determining the time required for the system to successfully detect a change in the actual state of the terminal. Dependence on historical data is evaluated through model structure analysis and experimental verification. Multi-device collaboration capability is assessed based on the implementation and test results of the federated learning framework. Adaptability to environmental changes is assessed by designing environmental change scenarios for testing.

[0060] The power consumption optimization control module is used to adjust the power consumption operation configuration of the wireless communication terminal by comparing the expected activity state of wireless communication with the current activity state and applying a power consumption optimization decision model.

[0061] Furthermore, the power consumption optimization decision model includes a state deviation calculation unit, a trigger condition adjustment unit, a power mode conversion unit, and a strategy collaborative optimization unit, see reference. Figure 2 ;

[0062] The state deviation calculation unit is used to calculate the deviation between the real-time perceived state of the wireless communication terminal and the dynamic expected state, and to generate the degree of deviation of the operating state of the wireless communication terminal.

[0063] The trigger condition adjustment unit dynamically adjusts the operation trigger conditions for positioning and context awareness based on the degree of deviation. Specifically, it adjusts the data acquisition frequency of the wireless network and sensors, the network communication frequency, and the sensitivity threshold for detecting state changes.

[0064] The power mode switching unit comprehensively analyzes the degree of deviation and the adjusted triggering conditions, generates and executes power mode switching instructions, and switches between multiple preset power modes (e.g., deep sleep mode, light hibernation mode, low power consumption sensing mode, high performance activity mode).

[0065] The strategy collaborative optimization unit utilizes a federated learning framework to comprehensively analyze the deviation degree, trigger condition adjustment efficiency, and power mode switching data of multiple wireless communication terminals, and optimizes the trigger condition adjustment strategy and power mode switching logic.

[0066] In this embodiment, intelligent energy management is achieved by comparing the expected activity state of the wireless communication terminal with its current activity state in real time and utilizing a power consumption optimization decision model. First, a state deviation calculation unit accurately measures the difference between the terminal's actual state and its dynamically expected state, providing precise data support for subsequent energy consumption control. Next, a trigger condition adjustment unit dynamically adjusts the data acquisition frequency, network communication frequency, and state change detection sensitivity of the wireless network and sensors based on the degree of deviation, thereby achieving flexible response in different environments. Simultaneously, a power mode switching unit comprehensively evaluates the degree of deviation and the adjusted trigger conditions, intelligently switching between multiple preset power modes such as deep sleep, light hibernation, low-power awareness, and high-performance activity, ensuring that the terminal achieves optimal energy consumption in different scenarios. Finally, a strategy collaborative optimization unit utilizes a federated learning framework to integrate the operating data and adjustment feedback from multiple terminals, continuously optimizing the trigger condition adjustment strategy and power mode switching logic, thereby achieving network-wide collaborative optimization. This not only effectively reduces device energy consumption and extends terminal battery life but also significantly improves the system's adaptive capability and stability.

[0067] Furthermore, the deviation degree generation process includes:

[0068] The latest positioning data is obtained through the real-time sensors of the wireless communication terminal to calculate the current actual activity state; the expected activity state generated by the activity state prediction model is compared with the real-time inferred current actual activity state to generate the degree of deviation; the comparison method is as follows: if the activity state is a predefined discrete category, then category matching comparison is performed; if the activity state is a continuous value, then the difference measure between the two is calculated.

[0069] The degree of deviation is a Boolean value indicating the consistency between the two, a numerical value quantifying the degree of deviation, and a signal identifying the specific expected state and the current state category.

[0070] In this embodiment, the deviation generation process utilizes real-time sensor data from the wireless communication terminal to accurately calculate the current actual activity state and compare it with the expected state generated by the activity state prediction model. This results in a multi-dimensional representation of the differences between the two states, expressed in the form of Boolean consistency, quantization deviation, and specific state category signals. This comparison method, whether for discrete category matching or continuous numerical difference measurement, can quickly capture subtle deviations between states, enabling the system to achieve higher accuracy and response speed in detecting anomalies, adjusting power consumption, and optimizing operational configurations. By identifying potential inconsistencies and anomalies in advance, this mechanism not only improves the reliability of intelligent control of the wireless communication terminal but also effectively reduces energy waste and system failure risks.

[0071] The status verification and correction module is used to monitor whether there are any abnormalities in the actual operation trajectory and status of the wireless communication terminal by combining the predetermined time plan of the task, environmental dynamic parameters and expected activity status; if an abnormality is detected, correction information is generated to adjust the activity status prediction model and power consumption optimization decision model.

[0072] Furthermore, the specific process for determining whether there are any anomalies in the actual operating trajectory and status of a wireless communication terminal includes:

[0073] The actual location coordinates acquired by the wireless communication terminal over a period of time are compared with the expected trajectory deduced based on the expected activity state and the predetermined time planning of the task, and the trajectory deviation is calculated.

[0074] The degree of deviation is detected to determine whether there is a persistent or / and significant state mismatch; the current location and activity status of the wireless communication terminal are detected to ensure they match the predetermined time schedule of the task.

[0075] Evaluate whether there is a logical contradiction between the activity state of the wireless communication terminal and the environmental dynamic parameters;

[0076] If any result exceeds the preset threshold or / and violates the preset logic rules, an anomaly is determined.

[0077] In this embodiment, by comparing the actual operating trajectory and status of the wireless communication terminal in real time, the system's continuous and efficient operation in a dynamic environment is effectively ensured. The actual location data acquired by the terminal is compared with the expected trajectory derived from the anticipated activity status and task timeline, accurately calculating trajectory deviations and detecting the degree of deviation to promptly determine if there are persistent or significant state mismatches. Simultaneously, this module further detects whether the current activity status conforms to the predetermined task timeline and environmental dynamic parameters, assessing the logical consistency between the operating status and the environment. If the detection result exceeds a preset threshold or violates logical rules, correction information is rapidly generated to dynamically adjust the activity status prediction model and power consumption optimization decision model, thereby achieving adaptive optimization. This not only improves the system's sensitivity and response speed to abnormal situations but also significantly reduces energy waste and safety hazards caused by state deviations.

[0078] Furthermore, the specific process of adjusting the activity state prediction model and the power consumption optimization decision model includes:

[0079] When an anomaly is detected, the status verification and correction module generates correction information, including the anomaly type, occurrence time, location, degree of deviation, and relevant context information.

[0080] Adjust the activity state prediction model: Use anomalous event data with correct labels as new training samples to incrementally train the activity state prediction model on the wireless communication terminal locally.

[0081] Adjust the power consumption optimization decision model: Based on the anomaly type in the correction information, adjust the parameters in the trigger condition adjustment unit and the switching logic of the power mode conversion unit.

[0082] In this embodiment, a dynamic adaptive mechanism is introduced to achieve real-time updates and optimizations of the activity state prediction model and the power consumption optimization decision model. When an anomaly is detected, the state verification and correction module can quickly generate correction information containing the anomaly type, occurrence time, location information, deviation degree, and relevant contextual information, providing a precise basis for subsequent model adjustments. By using correctly labeled anomaly event data for local incremental training or fine-tuning, the activity state prediction model can continuously adapt to environmental changes, significantly improving prediction accuracy. Simultaneously, by dynamically adjusting trigger condition parameters and power mode switching logic based on the anomaly type, the power consumption optimization decision model can flexibly switch between multiple preset power modes, effectively reducing energy consumption and system failure risks. This enhances the rapid response and self-healing capabilities of the wireless communication terminal to anomalies and greatly improves the system's stability, energy efficiency, and operational reliability.

[0083] This invention integrates multi-dimensional contextual data acquisition, federated learning, and deep recurrent neural network technologies to achieve real-time monitoring and accurate prediction of the status of wireless terminals. The system automatically calculates the state deviation by comparing the expected activity state with the actual state, and dynamically adjusts the data acquisition frequency, power mode, and trigger conditions to promptly identify and correct anomalies, thereby optimizing power consumption. This mechanism effectively reduces energy consumption, extends battery life, and significantly improves system response speed and adaptability, ensuring stable and efficient operation of the terminal in complex environments. See details... Figure 3 For different activity scenarios, the system of this invention exhibits significant power consumption advantages compared to traditional periodic wake-up solutions. As shown in Table 2, in a stationary state, the average power consumption of this system is only 15.2 mW, a reduction of 80.6% compared to the traditional solution's 78.4 mW, extending battery life by 5.16 times. Even in high-speed mobile scenarios with high resource consumption, this system still achieves a 9.5% reduction in power consumption. In a typical mixed activity mode, the system's average power consumption is 56.4 mW, a reduction of 66.5% compared to the traditional solution's 168.3 mW, extending battery life by nearly 3 times. These data fully demonstrate the significant innovative effect of this invention in power consumption optimization, providing possibilities for the long-term deployment of mobile devices.

[0084] Table 2 Comparison of Power Consumption Optimization Effects

[0085]

[0086] Example 2

[0087] This embodiment introduces a complete implementation scheme for a federated learning optimization system for low-power positioning on mobile devices. This system integrates multi-dimensional contextual information, federated learning, and deep neural network technologies to achieve intelligent power management and precise positioning of wireless communication terminals. Specifically, it includes:

[0088] The context information acquisition module is used to acquire multi-dimensional context information related to the wireless communication terminal. The multi-dimensional context information includes: the real-time location coordinates of the wireless communication terminal, the progress status of the tasks associated with the wireless communication terminal, the predetermined time plan, and the dynamic parameters of the environment in which the wireless communication terminal is located.

[0089] Furthermore, the real-time location coordinates are obtained through wireless network positioning, including latitude and longitude values. Specifically, GPS, WiFi positioning, or cellular network triangulation technologies are used to obtain latitude and longitude coordinate values ​​with an accuracy of ±5 meters. The coordinates are represented using the WGS84 coordinate system in decimal degree format, such as longitude 120.1234° and latitude 30.5678°. The data update frequency is dynamically adjusted according to the terminal's activity status: once every 10 minutes in a stationary state, once every minute in a low-speed moving state, and once every 10 seconds in a high-speed moving state.

[0090] The progress status of the task is obtained through the task management system. The progress status includes the completion percentage, the current task stage identifier, and a Boolean value. Specifically, it includes the completion percentage (represented as a value between 0 and 100, accurate to one decimal place), the current task stage identifier (using an integer ID to identify the current task stage), and a Boolean value (indicating whether the task is active).

[0091] The scheduled time for the task is obtained through the task scheduling system, including the planned start time, end time, and key time nodes, stored as a timestamp array, with each time node accompanied by identification information. The system calculates the proximity D between the current time and the key time node, using the formula: D = |current time - nearest key time node| / total task duration. When D < 0.05, meaning the distance to the key time node is less than 5% of the total duration, the system increases the frequency of status monitoring.

[0092] The environmental dynamic parameters are collected by sensors, including physical values ​​such as temperature, humidity, light intensity, and noise level; and environmental state classifications such as indoor / outdoor, quiet / noisy, and bright / dark. The sensor sampling frequency is 5 minutes / sample under normal conditions, increasing to 30 seconds / sample when a rapid change in the environment is detected. All physical parameters are converted to the [0,1] range after min-max standardization.

[0093] The activity state prediction module is used to build an activity state prediction model based on federated learning by acquiring training information from multiple wireless communication terminals.

[0094] The federated learning framework includes local model training, parameter aggregation, and model update distribution. Local model training involves training a model for each terminal based on local data and calculating parameter gradients. Parameter aggregation includes a central server collecting model parameter updates from each terminal and calculating a weighted average. Model update distribution involves distributing the updated global model parameters back to each terminal. Based on the acquired real-time location coordinates and progress status, the expected activity state of the wireless communication terminal is generated using the activity state prediction model. The optimizer uses Adam with an initial learning rate of 0.001 and a decay rate of 0.95. The batch size is 32, the federated learning aggregation cycle is 48 hours, local training consists of 10 rounds per session, and the global model is updated weekly.

[0095] Furthermore, the activity state prediction model includes:

[0096] The input layer is used to receive and process the normalized real-time position coordinates and the progress status;

[0097] A recurrent neural network layer is used to handle time-series dependencies, learning the activity patterns of the wireless communication terminal based on historical location and task progress. An LSTM structure is employed, including components such as forget gates, input gates, output gates, candidate memory units, memory unit updates, and hidden state calculations. The network structure consists of two LSTM layers, each with 64 neurons, and a time window length of 24, representing consideration of historical data from the previous 24 time points. A dropout rate of 0.2 is used during training to prevent overfitting.

[0098] The output layer receives the output from the last hidden layer and generates a probability distribution and specific category label for the expected activity state of the wireless communication terminal in the next time step through an activation function. The expected activity states include stationary, low-speed movement, high-speed movement, and stationary at the task location. The state determination criteria include: stationary (speed v < 0.5 m / min), low-speed movement (0.5 m / min ≤ v < 30 m / min), high-speed movement (v ≥ 30 m / min), and stationary at the task location (distance d from the location coordinates to the task location < 50 meters and duration t > 3 minutes).

[0099] The power consumption optimization control module is used to adjust the power consumption operation configuration of the wireless communication terminal by comparing the expected activity state of wireless communication with the current activity state and applying a power consumption optimization decision model.

[0100] Furthermore, the power consumption optimization decision model includes a state deviation calculation unit, a trigger condition adjustment unit, a power mode conversion unit, and a strategy collaborative optimization unit;

[0101] The state deviation calculation unit is used to calculate the deviation between the real-time perceived state of the wireless communication terminal and the dynamic expected state, and to generate the degree of deviation of the operating state of the wireless communication terminal.

[0102] The trigger condition adjustment unit dynamically adjusts the operation trigger conditions for positioning and context awareness based on the degree of deviation. Specifically, it adjusts the data acquisition frequency of the wireless network and sensors, the network communication frequency, and the sensitivity threshold for state change detection. When the degree of deviation exceeds the preset threshold for more than 3 cycles, an emergency state is triggered, and all frequencies are increased to the highest level. When the system battery is below 20%, the acquisition frequency is reduced to the lowest level regardless of the degree of deviation.

[0103] The power mode switching unit comprehensively analyzes the degree of deviation and the adjusted triggering conditions, generates and executes power mode switching instructions, and switches between multiple preset power modes. These preset power modes include deep sleep mode (power consumption approximately 10-20mW), light hibernation mode (power consumption approximately 50-80mW), low-power sensing mode (power consumption approximately 100-200mW), and high-performance activity mode (power consumption approximately 300-500mW). The mode switching decision is based on a comprehensive score. :

[0104] ;

[0105] in, , and The weights are respectively the degree of deviation, the urgency of the task, and the battery level. To indicate the degree of deviation, To assess the urgency of the mission, Battery level;

[0106] The strategy collaborative optimization unit utilizes a federated learning framework to comprehensively analyze the deviation degree, trigger condition adjustment efficiency, and power mode switching data of multiple wireless communication terminals, and optimizes the trigger condition adjustment strategy and power mode switching logic.

[0107] Furthermore, the deviation degree generation process includes:

[0108] The latest positioning data is obtained through the real-time sensors of the wireless communication terminal to calculate the current actual activity state; the expected activity state generated by the activity state prediction model is compared with the real-time inferred current actual activity state to generate the degree of deviation; the comparison method is as follows: if the activity state is a predefined discrete category, then category matching comparison is performed; if the activity state is a continuous value, then the difference measure between the two is calculated.

[0109] The degree of deviation is defined as a Boolean value indicating the consistency between the two, a numerical value quantifying the degree of deviation, and a signal identifying the specific expected state and the current state category. In addition, a deviation persistence assessment is performed to distinguish between short-term deviation, medium-term deviation, and long-term deviation, and a weighted deviation degree is calculated.

[0110] The status verification and correction module is used to monitor whether there are any abnormalities in the actual operation trajectory and status of the wireless communication terminal by combining the predetermined time plan of the task, environmental dynamic parameters and expected activity status; if an abnormality is detected, correction information is generated to adjust the activity status prediction model and power consumption optimization decision model.

[0111] Furthermore, the specific process for determining whether there are any anomalies in the actual operating trajectory and status of a wireless communication terminal includes:

[0112] The actual location coordinates acquired by the wireless communication terminal over a period of time are compared with the expected trajectory deduced based on the expected activity state and the predetermined time plan of the task to calculate the trajectory deviation. The trajectory deviation calculation collects the actual location coordinates over a period of time and compares them with the expected trajectory deduced based on the expected activity state and the predetermined time plan of the task to calculate the point-to-point distance deviation, average trajectory deviation, maximum trajectory deviation and standard deviation.

[0113] The degree of deviation is detected to determine whether there is a persistent or / and significant state mismatch. When the calculated period of continuous state mismatch is >20 minutes or the state mismatch ratio is >0.4, it is determined to be a persistent or significant mismatch. The current location and activity status of the wireless communication terminal are checked to see if they match the predetermined time plan of the task. Task time planning consistency detection extracts key time nodes of the current task, checks the consistency between the actual state and the expected state near each key time point, and calculates the task time planning consistency score.

[0114] The environmental logical consistency assessment extracts current environmental dynamic parameters, defines a set of logical rules between activity states and environmental parameters, evaluates rule violations, and calculates an environmental consistency score. A comprehensive anomaly judgment process sets thresholds for trajectory deviation, state mismatch, task time planning consistency, and environmental consistency, comprehensively assessing judgment conditions and quantifying the severity of anomalies.

[0115] If any result exceeds the preset threshold or / and violates the preset logic rules, an anomaly is determined.

[0116] Furthermore, the specific process of adjusting the activity state prediction model and the power consumption optimization decision model includes:

[0117] When an anomaly is detected, the status verification and correction module generates correction information that includes the anomaly type, occurrence time, location, degree of deviation, and relevant context information.

[0118] Adjusting the activity state prediction model:

[0119] Abnormal event data with correct labels (actual state) are used as new training samples to incrementally train the activity state prediction model on the wireless communication terminal locally.

[0120] Through the federated learning mechanism, the anomaly information or the corrected local model parameter update is contributed to the global model aggregation process to improve the robustness and accuracy of the overall model.

[0121] Adjusting the power consumption optimization decision model:

[0122] Based on the abnormal patterns reflected by the correction information (e.g., frequent errors in the prediction model causing the power consumption strategy to fail), adjust the parameters (such as deviation threshold, adjustment range) in the power consumption strategy trigger condition adjustment unit or the switching logic of the power mode intelligent switching unit (e.g., increase the state verification frequency, adopt a more conservative power consumption mode).

[0123] The effect of the adjustment strategy is fed back to the federated collaborative optimization unit to optimize the global power consumption optimization decision strategy under the federated learning framework.

[0124] Table 3 Comparison of Overall System Performance and Resource Consumption

[0125]

[0126] This invention achieves an excellent balance between performance and resource consumption. As shown in Table 3, the system's average positioning accuracy reaches ±8.6 meters, significantly better than the ±18.4 meters of the traditional low-power system A. Simultaneously, the average battery life reaches 84 days, 2.6 times that of the traditional system and 16.8 times that of the high-precision positioning system B (5 days). Although the CPU utilization (4.8%) and memory usage (8.4MB) of this system are slightly higher than those of the traditional low-power system, they are far lower than the resource consumption of the high-precision positioning system. Particularly in terms of data storage requirements, this system requires only 12.6MB per month, while the high-precision system requires 268.4MB. Furthermore, this invention employs an adaptive network communication frequency strategy, averaging 5.2 communications per hour, dynamically adjusting according to actual conditions, avoiding the limitations of a fixed-frequency strategy. Simultaneously, the system response latency is only 740ms, significantly better than the traditional low-power system. The traditional low-power system A uses a low-power microcontroller ESP32C3, while the high-precision positioning system includes Bluetooth-assisted positioning, UWB, and an inertial measurement unit.

[0127] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A federated learning optimization system for low-power positioning on mobile devices, characterized in that, include: The context information acquisition module is used to acquire multi-dimensional context information related to the wireless communication terminal. The multi-dimensional context information includes: the real-time location coordinates of the wireless communication terminal, the progress status of the tasks associated with the wireless communication terminal, the predetermined time plan, and the dynamic parameters of the environment in which the wireless communication terminal is located. The activity state prediction module is used to construct an activity state prediction model based on federated learning by acquiring training information from multiple wireless communication terminals; and to generate the expected activity state of the wireless communication terminal based on the acquired real-time location coordinates and progress status through the activity state prediction model. The power consumption optimization control module is used to adjust the power consumption operation configuration of the wireless communication terminal by comparing the expected activity state of the wireless communication terminal with the current activity state and applying a power consumption optimization decision model. The status verification and correction module is used to monitor whether there are any abnormalities in the actual operation trajectory and status of the wireless communication terminal by combining the predetermined time plan of the task, environmental dynamic parameters and expected activity status; if an abnormality is detected, correction information is generated to adjust the activity status prediction model and power consumption optimization decision model.

2. The federated learning optimization system for low-power positioning on a mobile device according to claim 1, characterized in that: The real-time location coordinates are obtained through wireless network positioning, including latitude and longitude values; the task progress status is obtained through the task management system, and the progress status includes the completion percentage, the current task stage identifier, and a Boolean value; the task's scheduled time plan is obtained through the task scheduling system, including the task's planned start time, end time, and key time nodes; the environmental dynamic parameters are collected through sensors, including physical values ​​and environmental state classifications.

3. The federated learning optimization system for low-power positioning on a mobile device according to claim 1, characterized in that: The activity state prediction model includes: The input layer is used to receive and process the normalized real-time position coordinates and the progress status; The recurrent neural network layer is used to process time-series dependencies and learn the activity patterns of the wireless communication terminal based on historical location and task progress; The output layer receives the output from the last hidden layer and generates a probability distribution and specific category label for the expected activity state of the wireless communication terminal in the next time step through an activation function; the expected activity state includes stationary, low-speed movement, high-speed movement, and stationary at the task location.

4. The federated learning optimization system for low-power positioning on a mobile device according to claim 1, characterized in that: The power consumption optimization decision model includes a state deviation calculation unit, a trigger condition adjustment unit, a power mode conversion unit, and a strategy collaborative optimization unit. The state deviation calculation unit is used to calculate the deviation between the real-time perceived state of the wireless communication terminal and the dynamic expected state, and to generate the degree of deviation of the operating state of the wireless communication terminal. The trigger condition adjustment unit dynamically adjusts the operation trigger conditions for positioning and context awareness based on the degree of deviation. Specifically, it adjusts the data acquisition frequency of the wireless network and sensors, the network communication frequency, and the sensitivity threshold for detecting state changes. The power mode switching unit comprehensively analyzes the degree of deviation and the adjusted triggering conditions, generates and executes power mode switching instructions, and switches between multiple preset power modes; The strategy collaborative optimization unit utilizes a federated learning framework to comprehensively analyze the deviation degree, trigger condition adjustment efficiency, and power mode switching data of multiple wireless communication terminals, and optimizes the trigger condition adjustment strategy and power mode switching logic.

5. The federated learning optimization system for low-power positioning on a mobile device according to claim 4, characterized in that: The deviation generation process includes: The latest positioning data is obtained through the real-time sensors of the wireless communication terminal to calculate the current actual activity state; the expected activity state generated by the activity state prediction model is compared with the real-time inferred current actual activity state to generate the degree of deviation; the comparison method is as follows: if the activity state is a predefined discrete category, then category matching comparison is performed; if the activity state is a continuous value, then the difference measure between the two is calculated. The degree of deviation is a Boolean value indicating the consistency between the two, a numerical value quantifying the degree of deviation, and a signal identifying the specific expected state and the current state category.

6. The federated learning optimization system for low-power positioning on a mobile device according to claim 1, characterized in that: The specific process for determining whether there are any anomalies in the actual operating trajectory and status of a wireless communication terminal includes: The actual location coordinates acquired by the wireless communication terminal over a period of time are compared with the expected trajectory deduced based on the expected activity state and the predetermined time planning of the task, and the trajectory deviation is calculated. The degree of deviation is detected to determine whether there is a persistent or / and significant state mismatch; the current position and state of the wireless communication terminal are detected to see if they match the predetermined time plan of the task. Evaluate whether there is a logical contradiction between the activity state of the wireless communication terminal and the environmental dynamic parameters; If any result exceeds the preset threshold or / and violates the preset logic rules, an anomaly is determined.

7. The federated learning optimization system for low-power positioning on a mobile device according to claim 1, characterized in that: The specific process of adjusting the activity state prediction model and the power consumption optimization decision model includes: When an anomaly is detected, the status verification and correction module generates correction information, including the anomaly type, occurrence time, location, degree of deviation, and relevant context information. Adjust the activity state prediction model: Use anomalous event data with correct labels as new training samples to incrementally train the activity state prediction model on the wireless communication terminal locally. Adjust the power consumption optimization decision model: Based on the anomaly type in the correction information, adjust the parameters in the trigger condition adjustment unit and the switching logic of the power mode conversion unit.

Images