Low-power-consumption adaptive broadcast adjustment method and system

Abstract

The invention discloses a low-power-consumption adaptive broadcast adjustment method and system, and the method comprises the steps: building a physical model based on the physical characteristics of equipment, and simulating the influence of a broadcast interval and a scanning interval on energy consumption and connection time delay; generating a training data set based on the physical model, and training a proxy prediction model, so as to predict an optimal broadcast interval and a scanning interval according to the service priority, the electric quantity state, the connection time period and the environment equipment density; monitoring the operation state of the equipment, and correcting the input parameters of the proxy prediction model according to the battery attenuation factor and the antenna aging factor; triggering the updating of the proxy prediction model according to the broadcast packet loss rate and the electric quantity change; the connection time period is predicted based on the historical connection data, the input weight of the proxy prediction model is adjusted to adapt to different connection load scenes, and the connection reliability and scene adaptability are improved.

Description

Technical Field

[0001] This invention relates to the field of Bluetooth device communication technology, and in particular to a low-power adaptive broadcast adjustment method and system. Background Technology

[0002] In recent years, with the rapid development of IoT technology, the application of Bluetooth Low Energy (BLE) devices in smart homes, industrial sensing, medical monitoring, and other fields has continued to expand. Achieving a dynamic balance between device power consumption and communication performance has become a core challenge in ensuring the long-term stable operation of the system.

[0003] Existing technologies typically employ fixed-interval strategies or single-factor adjustment mechanisms: some solutions simplify control logic by pre-setting fixed broadcast / scan intervals, which can reduce basic power consumption but cannot respond to service fluctuations and environmental changes; other solutions adjust the broadcast interval unidirectionally based on remaining battery power (e.g., extending the interval when the battery is below a threshold), which can delay battery depletion but ignores the impact of multi-dimensional factors such as service priority and time-of-day connection patterns on real-time communication quality. Fixed-parameter strategies are prone to packet collisions or connection timeouts during peak connection periods, while generating unnecessary power consumption during idle periods; single-factor adjustment ignores service requirements (e.g., urgent instructions require low latency) and environmental interference (e.g., device density fluctuations), leading to delayed response or communication interruptions for critical tasks. The scanning end cannot predict changes in the broadcast end's status, causing a mismatch between the scanning window and the broadcast packet transmission timing, significantly reducing device discovery efficiency. Summary of the Invention

[0004] The technical problem solved by this invention is that in existing Bluetooth communication technology, fixed parameter strategies are prone to data packet collisions or connection timeouts during peak connection periods, while generating unnecessary energy consumption and lacking the ability to detect device aging during idle periods.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a low-power adaptive broadcast adjustment method, the method comprising:

[0006] Constructing a physical model: Establishing state transition equations based on the physical characteristics of the device, simulating the impact of broadcast interval Tb and scan interval T on the device's energy state E and connection delay state, and generating a training dataset by traversing the parameter space for pre-training the surrogate model;

[0007] Constructing and training the proxy model: A lightweight LSTM network is used as the proxy model. The input layer receives the service priority P, power status B, connection time period H, environmental device density and connection latency. The output layer predicts the optimal broadcast interval Tb and scan interval Ts. Energy conservation is introduced in the training process, and the model parameters are optimized through the loss function.

[0008] Damage perception correction: Real-time monitoring of battery degradation factor and antenna aging factor. When the battery degradation factor or antenna aging factor exceeds the preset threshold, the power status and environmental equipment density are corrected respectively, and the corrected parameters are input into the surrogate model to recalculate the optimal interval.

[0009] Implement closed-loop update: Closed-loop update is achieved through data generation, model fine-tuning and emergency triggering mechanism. The physical model generates a new training dataset every 24 hours based on the latest device status. The proxy model updates weights by incremental learning. The proxy model is retrained in real time based on broadcast packet loss rate and sudden power changes.

[0010] Perform self-attention weight allocation: Construct an input vector containing information such as business priority P, battery status B, and connection time period H, calculate the attention distribution through the weight matrix, and output a weight vector that satisfies the normalization constraint;

[0011] Construct a mapping behavior table: Define a comprehensive scoring function that includes time-period enhancement factors, and generate an optimal interval mapping table by traversing the state space through a proxy model. The mapping table includes the comprehensive scores of different intervals and the corresponding broadcast intervals and scanning intervals.

[0012] Predicting connection periods: Collect the connection count sequence of the last K days per hour, calculate the peak probability at time t using the LSTM peak probability prediction network, make time period marking decisions based on the peak probability, and combine the above steps to achieve low-power adaptive broadcast adjustment.

[0013] In a preferred embodiment of the low-power adaptive broadcast adjustment method described in this invention, the construction of the physical model includes:

[0014] A state transition equation based on the physical characteristics of the device is established to simulate the impact of broadcast interval and scan interval on device power consumption and connection latency. The mathematical expression of the state transition equation is as follows:

[0015]

[0016] Where E represents the energy state, δ represents the connection delay state, A and B represent coefficient matrices, ε represents the random perturbation term, and T b Ts represents the broadcast interval, and Ts represents the scan interval;

[0017] A training dataset containing multiple sets (Tb, Ts, E, δ) is generated by traversing the parameter space, and the training dataset is used to pre-train the surrogate model.

[0018] In a preferred embodiment of the low-power adaptive broadcast adjustment method described in this invention, the training of the proxy model includes:

[0019] A lightweight LSTM network is used, and the input layer receives the service priority P, power status B, connection time period H, environmental device density ρ, and connection delay Δt.

[0020] The output layer predicts the optimal broadcast interval Tb and scan interval Ts;

[0021] The training process introduces physical constraints, including energy conservation constraints and time delay constraints:

[0022] The mathematical expression for the energy conservation constraint is:

[0023] in, This represents the energy consumption of the k-th broadcast transmission. N represents the power consumption received during the k-th scan.

[0024] Energy efficiency coefficient

[0025] The mathematical expression for the time delay constraint is: δ≤δ max (P);

[0026] Where, δ max (P) represents the maximum allowable latency linked to the service priority P;

[0027] The loss function is defined as:

[0028]

[0029] Where α, β, and γ are penalty coefficients, and L represents the loss function value. Indicates the predicted broadcast interval. T represents the actual target broadcast interval. s p T represents the predicted scan interval. s t Represents the actual target scanning interval, and ReLU represents the corrected linear unit function.

[0030] As a preferred embodiment of the low-power adaptive broadcast adjustment method described in this invention, the damage sensing correction mechanism includes:

[0031] Real-time monitoring of equipment damage factors, including battery degradation factor and antenna aging factor;

[0032] The battery degradation factor is calibrated using the open-circuit voltage-capacity mapping curve.

[0033] Calculate the antenna aging factor based on the signal strength attenuation rate;

[0034] When the battery degradation factor is greater than 0.2, the battery status is corrected using a battery level correction formula. The mathematical expression of the battery level correction formula is as follows:

[0035] B'=B·ek b ·t op ;

[0036] Where B' represents the corrected actual available power, K b Represents the battery degradation rate coefficient, t op This represents the cumulative operating time of the equipment, where e represents the natural constant.

[0037] When the antenna aging factor is greater than 0.15, the environmental equipment density is corrected using the environmental equipment density correction formula. The mathematical expression of the environmental equipment density correction formula is as follows:

[0038] ρ'=ρ·(1-0.1·k a );

[0039] Where ρ' represents the corrected environmental equipment density, ρ represents the original environmental equipment density, and K a Indicates the antenna aging factor;

[0040] The corrected battery degradation factor and antenna aging factor are input into the surrogate model to recalculate the optimal interval.

[0041] As a preferred embodiment of the low-power adaptive broadcast adjustment method described in this invention, the closed-loop update mechanism includes a data generation stage, a model fine-tuning stage, and an emergency triggering mechanism.

[0042] During the data generation phase, the physical model generates a new training dataset every 24 hours based on the latest device status;

[0043] During the model fine-tuning phase, the proxy model uses incremental learning to update the weights;

[0044] When the emergency trigger mechanism is activated, if the broadcast packet loss rate is greater than 15% or the power fluctuation ΔB is greater than 20%, the proxy model will be retrained immediately.

[0045] As a preferred embodiment of the low-power adaptive broadcast adjustment method described in this invention, the self-attention weight allocation algorithm specifically includes:

[0046] Constructing input vectors

[0047] Calculate the attention distribution using a triple matrix:

[0048]

[0049] Output weight vector [W] P W B W H Satisfying the normalization constraint: WP +W B +W H =1;

[0050] Among them W Q W K W V ∈R 5×5 Let d represent the learnable parameter matrix. k W represents the scaling factor. P W represents the attention weight that indicates business priority. B The attention weight W represents the battery status. H This indicates the attention weight during peak hours.

[0051] As a preferred embodiment of the low-power adaptive broadcast adjustment method described in this invention, the construction of the mapping behavior table includes:

[0052] Define the comprehensive scoring function:

[0053]

[0054] Where μ represents the time-period enhancement factor and S represents the overall score value;

[0055] By traversing the state space using a proxy model, an optimal interval mapping table is generated. The optimal interval mapping table includes the comprehensive score of different intervals, the corresponding broadcast interval, and the scanning interval.

[0056] In a preferred embodiment of the low-power adaptive broadcast adjustment method described in this invention, the prediction of the connection period includes:

[0057] Collect the hourly connection count sequence {c1, ..., c24} for the most recent K days;

[0058] Predicting peak probability using an LSTM network:

[0059]

[0060] Where σ represents the Sigmoid function, W i ∈R 16×24 Let W0 ∈ R be the input weight matrix. 1×16 P represents the output weight matrix. p (t) represents the peak probability at time t. Represents the input vector;

[0061] Time period marking decision:

[0062] In a preferred embodiment of the low-power adaptive broadcast adjustment method described in this invention, the prediction result of the connection period is used to correct the input weights of the proxy model, specifically:

[0063] When the peak probability P is obtained through the LSTM peak probability prediction network p When (t) is greater than 0.5, in the input layer of the surrogate model, the feature weights corresponding to the connection time period H are increased to 1.5 times that of the off-peak time period;

[0064] When P p If (t) is less than 0.5, the feature weights corresponding to the connection period H remain at the baseline value.

[0065] A low-power adaptive broadcast regulation system includes a dual-model prediction system, a dynamic regulation system, a time-period adaptation system, and an update coordination system;

[0066] The dual-model prediction system is used to construct a physical model and a proxy model. The physical model establishes a state transition equation based on the physical characteristics of the device to simulate the impact of broadcast and scan intervals on energy consumption and latency. The proxy model adopts a lightweight LSTM network to receive inputs such as service priority and power status and predict the optimal broadcast and scan intervals. At the same time, energy conservation and latency constraints are introduced during training.

[0067] The dynamic adjustment system is used to monitor the battery degradation factor and antenna aging factor in real time. When the preset threshold is exceeded, the power status and environmental equipment density are corrected, and the corrected parameters are fed back to the core modeling and prediction system to recalculate the optimal interval.

[0068] The time period adaptation system is used to collect historical connection count sequences, calculate the peak probability of the time period through the LSTM peak probability prediction network and make labeling decisions, and adapt the input weights of the physical model and the proxy model based on the decision results, so that the optimal interval prediction matches the needs of peak and non-peak scenarios.

[0069] The update collaboration system is used to generate new training datasets at regular intervals, and uses incremental learning to update the model weights of the physical model and the agent model. It triggers immediate retraining when the broadcast packet loss rate or power fluctuation exceeds the limit.

[0070] The beneficial effects of this invention are as follows: By accurately simulating the physical characteristics of device energy consumption and latency using a physical model, and combining this with a proxy model for multi-dimensional dynamic prediction of service demands, intelligent decision-making for broadcast parameters in complex scenarios is achieved. The physical model generates predicted broadcast loss values ​​based on device physical layer parameters (signal strength, channel attenuation, environmental interference), providing underlying hardware constraints for the system. The proxy model, through a lightweight LSTM network, integrates logical layer parameters such as service priority, power status, and connection time periods to output a joint predicted value for the optimal broadcast interval and scan interval. The model dynamically integrates these parameters using a self-attention weight allocation algorithm, enabling the system to perceive environmental changes and fluctuations in service demands in real time, automatically adjusting the broadcast strategy. While ensuring the core objective of low power consumption, this improves connection reliability and scenario adaptability.

[0071] By real-time monitoring of battery degradation factors and antenna aging factors, combined with open-circuit voltage-capacity mapping curves and signal strength attenuation analysis, accurate diagnosis and parameter correction of equipment physical condition damage are achieved. When the battery degradation factor exceeds the threshold, the power state mapping relationship is dynamically corrected; when the antenna aging factor exceeds the standard, the environmental equipment density sensing algorithm is adaptively adjusted to ensure the physical authenticity of the input parameters of the proxy model. The damage sensing mechanism is combined with a closed-loop update triggered every 24 hours. The physical model generates a new training dataset, the proxy model updates weights through incremental learning, and the emergency triggering mechanism initiates immediate retraining when the packet loss rate or power fluctuation exceeds the standard, forming a continuously optimized self-evolving system. This system can overcome performance degradation problems caused by long-term use such as equipment aging and environmental degradation, enabling the system to maintain stable low power consumption and connection efficiency throughout its entire life cycle, significantly reducing maintenance costs. Attached Figure Description

[0072] Figure 1 This is a schematic diagram of the basic process of a low-power adaptive broadcast adjustment method provided in one embodiment of the present invention. Detailed Implementation

[0073] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0074] Example, refer to Figure 1 As an embodiment of the present invention, a low-power adaptive broadcast adjustment method is provided, the method comprising:

[0075] Constructing a physical model: Establishing state transition equations based on the physical characteristics of the device, simulating the impact of broadcast interval Tb and scan interval T on the device's energy state E and connection delay state, and generating a training dataset by traversing the parameter space for pre-training the surrogate model;

[0076] Constructing and training the proxy model: A lightweight LSTM network is used as the proxy model. The input layer receives the service priority P, power status B, connection time period H, environmental device density and connection latency. The output layer predicts the optimal broadcast interval Tb and scan interval Ts. Energy conservation is introduced in the training process, and the model parameters are optimized through the loss function.

[0077] Damage perception correction: Real-time monitoring of battery degradation factor and antenna aging factor. When the battery degradation factor or antenna aging factor exceeds the preset threshold, the power status and environmental equipment density are corrected respectively, and the corrected parameters are input into the surrogate model to recalculate the optimal interval.

[0078] Implement closed-loop update: Closed-loop update is achieved through data generation, model fine-tuning and emergency triggering mechanism. The physical model generates a new training dataset every 24 hours based on the latest device status. The proxy model updates weights by incremental learning. The proxy model is retrained in real time based on broadcast packet loss rate and sudden power changes.

[0079] Perform self-attention weight allocation: Construct an input vector containing information such as business priority P, battery status B, and connection time period H, calculate the attention distribution through the weight matrix, and output a weight vector that satisfies the normalization constraint;

[0080] Construct a mapping behavior table: Define a comprehensive scoring function that includes time-period enhancement factors, and generate an optimal interval mapping table by traversing the state space through a proxy model. The mapping table includes the comprehensive scores of different intervals and the corresponding broadcast intervals and scanning intervals.

[0081] Predicting connection periods: Collect the connection count sequence of the last K days per hour, calculate the peak probability at time t using the LSTM peak probability prediction network, make time period marking decisions based on the peak probability, and combine the above steps to achieve low-power adaptive broadcast adjustment.

[0082] In this embodiment, a state transition equation is established based on the physical characteristics of the device to simulate the impact of broadcast interval and scan interval on the device's energy state and connection latency state. A training dataset is generated by traversing the parameter space to pre-train the proxy model. A lightweight LSTM network is used as the proxy model, which receives inputs such as service priority, battery status, connection time period, environmental device density, and connection latency, and outputs the optimal broadcast interval and scan interval. Energy conservation constraints are introduced during training, and model parameters are optimized through a loss function. The battery degradation factor and antenna aging factor are monitored in real time. When they exceed the preset threshold, the battery status and environmental device density are corrected respectively and fed back to the proxy model to recalculate the optimal interval. A closed-loop update is achieved by periodically generating new training datasets through the physical model, updating weights through incremental learning of the proxy model, and combining an instant retraining mechanism triggered by broadcast packet loss rate or sudden changes in battery status. An input vector containing information such as service priority, battery status, and connection time period is constructed. The attention distribution is calculated using a matrix, and a weight vector satisfying normalization constraints is output. A comprehensive scoring function with a time period enhancement factor is defined. The proxy model traverses the state space to generate an optimal interval mapping table containing comprehensive scores for different intervals and their corresponding intervals. Historical hourly connection count sequences are collected, and the peak probability of the network calculation time period is predicted using LSTM peak probability. Based on this, the input weights of the proxy model are adjusted to adapt to the time period requirements. This enables low-power adaptive broadcast adjustment, dynamically balancing device energy consumption and communication performance. While ensuring low power consumption, it improves connection reliability and scenario adaptability, accurately detects and corrects parameter deviations caused by device aging, and overcomes the impact of environmental changes through a continuous optimization mechanism. This allows the system to maintain stable low-power characteristics and connection efficiency throughout its entire lifecycle, reducing maintenance costs.

[0083] The construction of the physical model includes:

[0084] A state transition equation based on the physical characteristics of the device is established to simulate the impact of broadcast interval and scan interval on device power consumption and connection latency. The mathematical expression of the state transition equation is as follows:

[0085]

[0086] Where E represents the energy state, δ represents the connection delay state, A and B represent coefficient matrices, ε represents the random perturbation term, and T b Ts represents the broadcast interval, and Ts represents the scan interval;

[0087] A training dataset containing multiple sets (Tb, Ts, E, δ) is generated by traversing the parameter space, and the training dataset is used to pre-train the surrogate model.

[0088] In this embodiment, a state transition equation is established based on the physical characteristics of the device to simulate the impact of broadcast interval and scan interval on device power consumption and connection latency. A training dataset containing multiple sets of broadcast interval, scan interval, energy state and connection latency is generated by traversing the parameter space for pre-training the surrogate model. This provides the surrogate model with underlying data support that fits the physical characteristics of the device, making the pre-training of the surrogate model more accurately reflect the physical laws. This lays a reliable foundation for the subsequent prediction of the optimal broadcast interval and scan interval, improves the system's control accuracy of energy consumption and connection latency, and helps to achieve a dynamic balance between low power consumption and communication performance.

[0089] The training of the proxy model includes:

[0090] A lightweight LSTM network is used, and the input layer receives the service priority P, power status B, connection time period H, environmental device density ρ, and connection delay Δt.

[0091] The output layer predicts the optimal broadcast interval Tb and scan interval Ts;

[0092] The training process introduces physical constraints, including energy conservation constraints and time delay constraints:

[0093] The mathematical expression for the energy conservation constraint is:

[0094] in, This represents the energy consumption of the k-th broadcast transmission. N represents the power consumption received during the k-th scan.

[0095] Energy efficiency coefficient

[0096] The mathematical expression for the time delay constraint is: δ≤δ max (P);

[0097] Where, δ max (P) represents the maximum allowable latency linked to the service priority P;

[0098] The loss function is defined as:

[0099]

[0100] Where α, β, and γ are penalty coefficients, and L represents the loss function value. Indicates the predicted broadcast interval. T represents the actual target broadcast interval. s p T represents the predicted scan interval. s t Represents the actual target scanning interval, and ReLU represents the corrected linear unit function.

[0101] In this embodiment, a lightweight LSTM network is used to construct the proxy model. Its input layer receives multi-dimensional scenario parameters such as service priority P, power status B, connection time period H, environmental device density, and connection latency. The output layer directly predicts the optimal broadcast interval Tb and scan interval Ts. During training, two major physical constraints are introduced: energy conservation constraints and latency constraints. The energy conservation constraint ensures that the model's calculation of energy consumption conforms to physical laws by associating the energy consumption of the k-th broadcast transmission, the energy consumption of the k-th scan reception, and the energy utilization coefficient N, thus avoiding unrealistic low-energy consumption predictions. The latency constraint ensures that the latency requirements of high-priority services are not ignored by using the maximum allowable latency linked to the service priority P. At the same time, by including the difference between the predicted broadcast interval and the actual target broadcast interval, and the difference between the predicted scan interval and the actual target scan interval, combined with a loss function using a penalty coefficient and the ReLU function, targeted penalties are applied to the prediction deviation (especially for deviations that violate constraints), thereby optimizing the model parameters. It can accurately predict the optimal broadcast and scanning intervals by proxy models, while meeting the latency requirements of different service priorities and strictly following the physical laws of energy consumption. It effectively balances the low power consumption requirements of devices with the stability of communication performance, improves the adaptability of models in complex scenarios, and provides accurate decision-making basis that conforms to actual physical constraints for low power adaptive broadcast adjustment.

[0102] The damage perception correction mechanism includes:

[0103] Real-time monitoring of equipment damage factors, including battery degradation factor and antenna aging factor;

[0104] The battery degradation factor is calibrated using the open-circuit voltage-capacity mapping curve.

[0105] Calculate the antenna aging factor based on the signal strength attenuation rate;

[0106] When the battery degradation factor is greater than 0.2, the battery status is corrected using a battery level correction formula. The mathematical expression of the battery level correction formula is as follows:

[0107] B'=B·ek b ·t op ;

[0108] Where B' represents the corrected actual available power, K b Represents the battery degradation rate coefficient, t op This represents the cumulative operating time of the equipment, where e represents the natural constant.

[0109] When the antenna aging factor is greater than 0.15, the environmental equipment density is corrected using the environmental equipment density correction formula. The mathematical expression of the environmental equipment density correction formula is as follows:

[0110] ρ'=ρ·(1-0.1·k a );

[0111] Where ρ' represents the corrected environmental equipment density, ρ represents the original environmental equipment density, and K a Indicates the antenna aging factor;

[0112] The corrected battery degradation factor and antenna aging factor are input into the surrogate model to recalculate the optimal interval.

[0113] In this embodiment, two damage factors of the device, namely the battery attenuation factor and the antenna aging factor, are monitored in real time. The battery attenuation factor is accurately calibrated using an open-circuit voltage-capacity mapping curve, while the antenna aging factor is calculated based on the signal strength attenuation rate. When the battery attenuation factor exceeds a preset threshold, the battery status is corrected using a power correction formula that correlates the battery attenuation rate coefficient, the device's cumulative operating time, and the natural constant to obtain the actual usable power. When the antenna aging factor exceeds a preset threshold, the environmental device density is corrected using an environmental device density correction formula that correlates the original environmental device density and the antenna aging factor. The corrected battery status and environmental device density are then input into the proxy model to recalculate the optimal broadcast interval and scanning interval. This enables accurate perception and parameter correction of the device's physical state damage, dynamically compensating for parameter deviations caused by battery attenuation and antenna aging, ensuring the physical authenticity of the input parameters to the proxy model. This makes the calculation of the optimal interval more closely match the actual state of the device, effectively avoiding uncontrolled energy consumption or communication quality degradation due to device aging, and ensuring that the system maintains low power consumption characteristics and stable connection efficiency during long-term use.

[0114] The closed-loop update mechanism includes a data generation phase, a model fine-tuning phase, and an emergency triggering mechanism.

[0115] During the data generation phase, the physical model generates a new training dataset every 24 hours based on the latest device status;

[0116] During the model fine-tuning phase, the proxy model uses incremental learning to update the weights;

[0117] When the emergency trigger mechanism is activated, if the broadcast packet loss rate is greater than 15% or the power fluctuation ΔB is greater than 20%, the proxy model will be retrained immediately.

[0118] In this embodiment, technical optimization is achieved through the synergistic effect of a closed-loop update mechanism and the correction of proxy model input weights based on connection time period prediction results: In the closed-loop update mechanism, during the data generation phase, the physical model generates a new training dataset every 24 hours based on the latest device status, providing the model with fresh data support that fits the current device status; during the model fine-tuning phase, incremental learning enables the proxy model to update weights while retaining historical learning results, achieving continuous adaptation of the model to gradual changes in device status; the emergency triggering mechanism immediately triggers retraining of the proxy model when the broadcast packet loss rate is greater than 15% or the power fluctuation is greater than 20%, quickly responding to sudden situations such as communication quality degradation or power abnormalities, and avoiding continuous performance deterioration. Simultaneously, the connection time period prediction results dynamically correct the proxy model input weights using Pp(t) obtained from the LSTM peak probability prediction network. When Pp(t) is greater than 0.5, the feature weights of the connection time period H are increased to 1.5 times that of the off-peak period, making the model focus more on the impact of time period factors on connection demand during peak periods; when Pp(t) is less than 0.5, the baseline weights are maintained, balancing the effects of other factors such as power status and service priority. By implementing a closed-loop update mechanism to enable regular model evolution and emergency repairs, and by dynamically adjusting weights to make model decisions fit the characteristics of different time periods, the proxy model can maintain accurate prediction capabilities in complex scenarios such as gradual changes in device status, sudden anomalies, and time period fluctuations. This ensures that the calculation of the optimal broadcast interval and scanning interval always adapts to actual needs, thereby improving system communication stability and scenario adaptability while ensuring low power consumption, and achieving efficient operation throughout the entire life cycle.

[0119] The self-attention weight allocation algorithm specifically includes:

[0120] Constructing input vectors

[0121] Calculate the attention distribution using a triple matrix:

[0122]

[0123] Output weight vector [W] P W B W H Satisfying the normalization constraint: W P +W B +W H =1;

[0124] Among them W Q W K W V ∈R 5×5 Let d represent the learnable parameter matrix. k W represents the scaling factor. P W represents the attention weight that indicates business priority. BThe attention weight W represents the battery status. H This indicates the attention weight during peak hours.

[0125] In this embodiment, technical optimization is achieved through the synergistic effect of a closed-loop update mechanism and a self-attention weight allocation algorithm: In the closed-loop update mechanism, during the data generation phase, the physical model generates a new training dataset every 24 hours based on the latest device status, providing basic data that fits the current state for model updates; during the model fine-tuning phase, incremental learning enables the proxy model to update weights based on the new dataset while retaining historical learning results, achieving continuous model adaptation; the emergency trigger mechanism immediately triggers retraining of the proxy model when the broadcast packet loss rate exceeds the threshold or the battery level changes beyond the threshold, quickly responding to sudden state changes. The self-attention weight allocation algorithm first constructs an input vector containing information such as service priority, battery status, and connection time period, then calculates the attention distribution through the triple learnable parameter matrices WQ, WK, and WV, and after processing with the scaling factor dk, outputs a weight vector that satisfies the normalization constraint WP+WB+WH=1 (where WP, WB, and WH are the attention weights for service priority, battery status, and peak time period, respectively), dynamically adjusting the influence of each factor on the decision. It enables the model to be continuously optimized in dynamically changing device states and environments, ensuring model adaptability through timed updates and emergency adjustments. At the same time, it uses attention mechanisms to accurately focus on key factors, improving the targeting and accuracy of optimal broadcast interval and scan interval prediction, thereby achieving high efficiency and robustness of low-power adaptive broadcast adjustment.

[0126] The construction of the mapping behavior table includes:

[0127] Define the comprehensive scoring function:

[0128]

[0129] Where μ represents the time-period enhancement factor and S represents the overall score value;

[0130] By traversing the state space using a proxy model, an optimal interval mapping table is generated. The optimal interval mapping table includes the comprehensive score of different intervals, the corresponding broadcast interval, and the scanning interval.

[0131] In this embodiment, a comprehensive scoring function S, incorporating time-period enhancement factors, is defined to quantitatively evaluate device state performance, thereby measuring the overall effectiveness of broadcast and scanning strategies under different states. A proxy model then traverses various possible state spaces (covering multiple dimensions such as service priority, power status, and environmental device density) to calculate the comprehensive score for each state. Different comprehensive score intervals are associated with the corresponding optimal broadcast interval Tb and scanning interval Ts to generate an optimal interval mapping table. The comprehensive scoring function highlights the impact of time periods on strategy effectiveness through the time-period enhancement factors, ensuring that the scoring results align with the communication needs of different time periods. The proxy model's traversal of the state space comprehensively covers various scenarios, guaranteeing the integrity and applicability of the mapping table. The final mapping table allows the system to quickly obtain the optimal interval by querying the corresponding score interval without performing complex calculations in real time, significantly improving adjustment efficiency. Simultaneously, it allows the selection of broadcast and scanning strategies to more accurately match time-period characteristics and multi-dimensional states, further optimizing the dynamic balance between low power consumption and communication performance.

[0132] The prediction of the connection period includes:

[0133] Collect the hourly connection count sequence {c1, ..., c24} for the most recent K days;

[0134] Predicting peak probability using an LSTM network:

[0135]

[0136] Where σ represents the Sigmoid function, W i ∈R 16×24 Let W0 ∈ R be the input weight matrix. 1×16 P represents the output weight matrix. p (t) represents the peak probability at time t. Represents the input vector;

[0137] Time period marking decision:

[0138] In this embodiment, basic data on historical connection patterns are obtained by collecting the connection count sequence {c1, ..., c24} for the past K days per hour. Then, an LSTM peak probability prediction network is used to construct an input vector from this sequence. The input vector is processed through an input weight matrix and combined with the output weight matrix. The peak probability Pp(t) at time t is obtained by mapping with a Sigmoid function (converting the network output into a probability value between 0 and 1, quantifying the likelihood that time t is a peak period). Finally, based on the calculated peak probability, a time period labeling decision is made (e.g., determining whether time t is a peak period based on a probability threshold). The LSTM network excels at capturing long-term dependencies in time-series data and can effectively uncover the changing patterns of historical hourly connection counts. The Sigmoid function ensures that the output peak probability has a clear probabilistic meaning, facilitating intuitive judgment of time period characteristics. Through this prediction mechanism, the system can accurately predict the connection busyness at different times, providing a time period basis for subsequent dynamic adjustment of broadcast and scan intervals. This makes the adjustment strategy more aligned with actual connection needs, ensuring communication efficiency during peak periods and reducing energy consumption during idle periods, further optimizing the balance between low power consumption and communication performance.

[0139] The prediction results for the connection period are used to correct the input weights of the surrogate model, specifically:

[0140] When the peak probability P is obtained through the LSTM peak probability prediction network p When (t) is greater than 0.5, in the input layer of the surrogate model, the feature weights corresponding to the connection time period H are increased to 1.5 times that of the off-peak time period;

[0141] When P p If (t) is less than 0.5, the feature weights corresponding to the connection period H remain at the baseline value.

[0142] In this embodiment, the peak probability Pp(t) obtained from the LSTM peak probability prediction network is used to dynamically adjust the input weights of the proxy model: when Pp(t) is greater than 0.5, the feature weights corresponding to the connection period H in the input layer of the proxy model are increased to 1.5 times that of the off-peak period; when Pp(t) is less than 0.5, the feature weights of the connection period H remain at the baseline value, and the peak probability Pp(t) serves as a quantitative basis for judging the characteristics of the period, directly guiding the adjustment of the feature weights. The weights of the connection period H are increased during peak periods, so that the proxy model focuses more on the impact of period factors on connection demand when predicting the optimal broadcast interval and scan interval, in order to adapt to the communication efficiency guarantee under the high connection demand during peak periods; the baseline weights are maintained during off-peak periods, thus avoiding overemphasis on period factors and allowing other factors such as power status and service priority to participate in decision-making more reasonably. Through this dynamic weight adjustment, the importance of the input features of the proxy model can be adaptively adjusted with the characteristics of the period, so that the prediction results can more accurately match the differentiated needs of peak and off-peak periods, further optimizing the dynamic balance between low power consumption goals and communication performance, and improving the system's adaptability in different periods.

[0143] A low-power adaptive broadcast regulation system includes a dual-model prediction system, a dynamic regulation system, a time-period adaptation system, and an update coordination system;

[0144] The dual-model prediction system is used to construct a physical model and a proxy model. The physical model establishes a state transition equation based on the physical characteristics of the device to simulate the impact of broadcast and scan intervals on energy consumption and latency. The proxy model adopts a lightweight LSTM network to receive inputs such as service priority and power status and predict the optimal broadcast and scan intervals. At the same time, energy conservation and latency constraints are introduced during training.

[0145] The dynamic adjustment system is used to monitor the battery degradation factor and antenna aging factor in real time. When the preset threshold is exceeded, the power status and environmental equipment density are corrected, and the corrected parameters are fed back to the core modeling and prediction system to recalculate the optimal interval.

[0146] The time period adaptation system is used to collect historical connection count sequences, calculate the peak probability of the time period through the LSTM peak probability prediction network and make labeling decisions, and adapt the input weights of the physical model and the proxy model based on the decision results, so that the optimal interval prediction matches the needs of peak and non-peak scenarios.

[0147] The update collaboration system is used to generate new training datasets at regular intervals, and uses incremental learning to update the model weights of the physical model and the agent model. It triggers immediate retraining when the broadcast packet loss rate or power fluctuation exceeds the limit.

[0148] By accurately simulating the physical characteristics of device energy consumption and latency using a physical model, and combining this with a proxy model for multi-dimensional dynamic prediction of service demands, intelligent decision-making for broadcast parameters in complex scenarios is achieved. The physical model generates broadcast loss predictions based on device physical layer parameters (signal strength, channel attenuation, environmental interference), providing underlying hardware constraints for the system. The proxy model, through a lightweight LSTM network, integrates logical layer parameters such as service priority, power status, and connection time periods to output a joint prediction of the optimal broadcast interval and scan interval. The model dynamically integrates these parameters using a self-attention weight allocation algorithm, enabling the system to perceive environmental changes and fluctuations in service demands in real time, automatically adjusting the broadcast strategy. This ensures low power consumption while improving connection reliability and scenario adaptability.

[0149] By real-time monitoring of battery degradation factors and antenna aging factors, combined with open-circuit voltage-capacity mapping curves and signal strength attenuation analysis, accurate diagnosis and parameter correction of equipment physical condition damage are achieved. When the battery degradation factor exceeds the threshold, the power state mapping relationship is dynamically corrected; when the antenna aging factor exceeds the standard, the environmental equipment density sensing algorithm is adaptively adjusted to ensure the physical authenticity of the input parameters of the proxy model. The damage sensing mechanism is combined with a closed-loop update triggered every 24 hours. The physical model generates a new training dataset, the proxy model updates weights through incremental learning, and the emergency triggering mechanism initiates immediate retraining when the packet loss rate or power fluctuation exceeds the standard, forming a continuously optimized self-evolving system. This system can overcome performance degradation problems caused by long-term use such as equipment aging and environmental degradation, enabling the system to maintain stable low power consumption and connection efficiency throughout its entire life cycle, significantly reducing maintenance costs.

[0150] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0151] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A low-power adaptive broadcast adjustment method, characterized in that, The methods include: Constructing a physical model: Establishing state transition equations based on the physical characteristics of the device, simulating the impact of broadcast interval Tb and scan interval T on the device's energy state E and connection delay state, and generating a training dataset by traversing the parameter space for pre-training the surrogate model; Constructing and training the proxy model: A lightweight LSTM network is used as the proxy model. The input layer receives the service priority P, power status B, connection time period H, environmental device density and connection latency. The output layer predicts the optimal broadcast interval Tb and scan interval Ts. Energy conservation is introduced in the training process, and the model parameters are optimized through the loss function. Damage perception correction: Real-time monitoring of battery degradation factor and antenna aging factor. When the battery degradation factor or antenna aging factor exceeds the preset threshold, the power status and environmental equipment density are corrected respectively, and the corrected parameters are input into the surrogate model to recalculate the optimal interval. Implement closed-loop update: Closed-loop update is achieved through data generation, model fine-tuning and emergency triggering mechanism. The physical model generates a new training dataset every 24 hours based on the latest device status. The proxy model updates weights by incremental learning. The proxy model is retrained in real time based on broadcast packet loss rate and sudden power changes. Perform self-attention weight allocation: Construct an input vector containing information such as business priority P, battery status B, and connection time period H, calculate the attention distribution through the weight matrix, and output a weight vector that satisfies the normalization constraint; Construct a mapping behavior table: Define a comprehensive scoring function that includes time-period enhancement factors, and generate an optimal interval mapping table by traversing the state space through a proxy model. The mapping table includes the comprehensive scores of different intervals and the corresponding broadcast intervals and scanning intervals. Predicting connection periods: Collect the connection count sequence of the last K days per hour, calculate the peak probability at time t using the LSTM peak probability prediction network, make time period marking decisions based on the peak probability, and combine the above steps to achieve low-power adaptive broadcast adjustment.

2. The low-power adaptive broadcast adjustment method as described in claim 1, wherein the construction of the physical model includes: A state transition equation based on the physical characteristics of the device is established to simulate the impact of broadcast interval and scan interval on device power consumption and connection latency. The mathematical expression of the state transition equation is as follows: Where E represents the energy state, δ represents the connection delay state, A and B represent coefficient matrices, ε represents the random perturbation term, and T b Ts represents the broadcast interval, and Ts represents the scan interval; A training dataset containing multiple sets (Tb, Ts, E, δ) is generated by traversing the parameter space, and the training dataset is used to pre-train the surrogate model.

3. The low-power adaptive broadcast adjustment method as described in claim 2, characterized in that: The training of the proxy model includes: A lightweight LSTM network is used, and the input layer receives the service priority P, power status B, connection time period H, environmental device density ρ, and connection delay Δt. The output layer predicts the optimal broadcast interval Tb and scan interval Ts; The training process introduces physical constraints, including energy conservation constraints and time delay constraints: The mathematical expression for the energy conservation constraint is: in, This represents the energy consumption of the k-th broadcast transmission. N represents the power consumption received during the k-th scan. Energy efficiency coefficient The mathematical expression for the time delay constraint is: δ≤δ max (P); Where, δ max (P) represents the maximum allowable latency linked to the service priority P; The loss function is defined as: Where α, β, and γ are penalty coefficients, and L represents the loss function value. Indicates the predicted broadcast interval. Indicates the actual target broadcast interval. Indicates the predicted scan interval. Represents the actual target scanning interval, and ReLU represents the corrected linear unit function.

4. The low-power adaptive broadcast adjustment method as described in claim 3, characterized in that: The damage perception correction mechanism includes: Real-time monitoring of equipment damage factors, including battery degradation factor and antenna aging factor; The battery degradation factor is calibrated using the open-circuit voltage-capacity mapping curve. Calculate the antenna aging factor based on the signal strength attenuation rate; When the battery degradation factor is greater than 0.2, the battery status is corrected using a battery level correction formula. The mathematical expression of the battery level correction formula is as follows: B'=B·e-k b ·t op ; Where B' represents the corrected actual available power, K b Represents the battery degradation rate coefficient, t op This represents the cumulative operating time of the equipment, where e represents the natural constant. When the antenna aging factor is greater than 0.15, the environmental equipment density is corrected using the environmental equipment density correction formula. The mathematical expression of the environmental equipment density correction formula is as follows: ρ'=ρ·(1-0.1·k a ); Where ρ' represents the corrected environmental equipment density, ρ represents the original environmental equipment density, and K a Indicates the antenna aging factor; The corrected battery degradation factor and antenna aging factor are input into the surrogate model to recalculate the optimal interval.

5. The low-power adaptive broadcast adjustment method as described in claim 4, characterized in that: The closed-loop update mechanism includes a data generation phase, a model fine-tuning phase, and an emergency triggering mechanism. During the data generation phase, the physical model generates a new training dataset every 24 hours based on the latest device status; During the model fine-tuning phase, the proxy model uses incremental learning to update the weights; When the emergency trigger mechanism is activated, if the broadcast packet loss rate is greater than 15% or the power fluctuation ΔB is greater than 20%, the proxy model will be retrained immediately.

6. The low-power adaptive broadcast adjustment method as described in claim 5, characterized in that: The self-attention weight allocation algorithm specifically includes: Constructing input vectors Calculate the attention distribution using a triple matrix: Output weight vector [W] P W B W H Satisfying the normalization constraint: W P +W B +W H =1; Among them W Q W K W V ∈R 5×5 Let d represent the learnable parameter matrix. k W represents the scaling factor. P W represents the attention weight that indicates business priority. B The attention weight W represents the battery status. H This indicates the attention weight during peak hours.

7. The low-power adaptive broadcast adjustment method as described in claim 6, characterized in that: The construction of the mapping behavior table includes: Define the comprehensive scoring function: Where μ represents the time-period enhancement factor and S represents the overall score value; By traversing the state space using a proxy model, an optimal interval mapping table is generated. The optimal interval mapping table includes the comprehensive score of different intervals, the corresponding broadcast interval, and the scanning interval.

8. The low-power adaptive broadcast adjustment method as described in claim 7, characterized in that: The prediction of the connection period includes: Collect the hourly connection count sequence {c1, ..., c24} for the most recent K days; Predicting peak probability using an LSTM network: Where σ represents the Sigmoid function, W i ∈R 16×24 Let W0 ∈ R be the input weight matrix. 1×16 P represents the output weight matrix. p (t) represents the peak probability at time t. Represents the input vector; Time period marking decision:

9. The low-power adaptive broadcast adjustment method as described in claim 8, characterized in that: The prediction results for the connection period are used to correct the input weights of the surrogate model, specifically: When the peak probability P is obtained through the LSTM peak probability prediction network p When (t) is greater than 0.5, in the input layer of the surrogate model, the feature weights corresponding to the connection time period H are increased to 1.5 times that of the off-peak time period; When P p If (t) is less than 0.5, the feature weights corresponding to the connection period H remain at the baseline value.

10. A low-power adaptive broadcast modulation system, comprising a low-power adaptive broadcast modulation method as described in any one of claims 1-9, characterized in that: It includes a physical modeling module, a prediction module, a state monitoring module, an update module, and a time period adaptation module; The physical modeling module is used to construct physical models and generate training datasets; The prediction module is used to train and run the proxy prediction model to output the optimal broadcast interval and scan interval; The condition monitoring module is used to detect the battery degradation factor and antenna aging factor and correct the prediction input parameters. The update module is used to update the agent prediction model according to a preset time or abnormal conditions. The time period adaptation module is used to predict connection time periods based on historical connection data and adjust the input weights of the proxy prediction model.

Images