Power consumption optimization method, system and storage medium

Abstract

The application discloses a power consumption optimization method, system and storage medium. The method comprises the following steps: collecting environment parameters of an environment where an Internet of Things communication module is located and running state parameters of the Internet of Things communication module; predicting power consumption requirements of the Internet of Things communication module under different data processing modes based on the environment parameters and the running state parameters; determining task allocation ratios of the Internet of Things communication module under different data processing modes based on the power consumption requirements and current network quality; and adjusting working parameters of the Internet of Things communication module based on the task allocation ratios, so as to realize power consumption optimization of the Internet of Things communication module. The application can comprehensively perceive environment conditions and running conditions of the Internet of Things communication module and adapt to differentiated requirements of different application scenarios. In addition, the Internet of Things communication module can independently make power consumption management decisions according to actual running environments and task requirements, so that adaptive management and optimization of power consumption are realized.

Description

Technical Field

[0001] This application relates to the field of Internet of Things (IoT) technology, specifically to a power consumption optimization method, system, and storage medium. Background Technology

[0002] With the rapid development of IoT technology, the power consumption management of IoT communication modules, as the core bridge connecting devices and networks, has become a key factor restricting the application scope and battery life of IoT devices. Typically, IoT communication modules only offer a limited number of power consumption modes for users to manually select, or employ static threshold triggering strategies based on a single parameter (such as battery voltage) for power adjustment. They cannot autonomously adjust according to the actual operating environment and task requirements, resulting in a lack of deep optimization capabilities based on application scenarios and environmental changes. Summary of the Invention

[0003] This application provides a power consumption optimization method, system, and storage medium, aiming to solve the technical problem that IoT communication modules cannot autonomously adjust according to the actual operating environment and task requirements, resulting in a lack of deep optimization capabilities based on application scenarios and environmental changes.

[0004] On one hand, embodiments of this application provide a power consumption optimization method, the method comprising: collecting environmental parameters of the environment in which the IoT communication module is located and the operating status parameters of the IoT communication module; predicting the power consumption requirements of the IoT communication module under different data processing methods based on the environmental parameters and the operating status parameters; determining the task allocation ratio of the IoT communication module under different data processing methods based on the power consumption requirements and the current network quality; and adjusting the operating parameters of the IoT communication module based on the task allocation ratio to achieve power consumption optimization of the IoT communication module.

[0005] In some embodiments, the data processing method includes at least one of the following: local edge computing method and cloud processing method.

[0006] In some embodiments, determining the task allocation ratio of the IoT communication module under different data processing methods based on the power consumption requirements and the current network quality includes: correcting the first energy consumption prediction value under the cloud processing method according to the current network quality; determining the task allocation ratio based on the corrected first energy consumption prediction value and the second energy consumption prediction value under the local edge computing method; wherein the current network quality is determined based on at least one of the following indicators: signal strength index, signal-to-noise ratio index, link stability index, and transmission delay index.

[0007] In some embodiments, predicting the power consumption requirements of the IoT communication module under different data processing methods based on the environmental parameters and the operating status parameters includes: performing preprocessing on the environmental parameters and the operating status parameters; and predicting the power consumption requirements of the IoT communication module under different data processing methods based on the preprocessed environmental parameters and the preprocessed operating status parameters when the change in the environmental parameters exceeds a preset threshold.

[0008] In some embodiments, predicting the power consumption requirements of the IoT communication module under different data processing methods based on the environmental parameters and the operating state parameters includes: determining the weight values ​​of the environmental parameters and the operating state parameters respectively using a deep learning-based time-series prediction model; and performing data fusion on the environmental parameters and the operating state parameters according to the weight values ​​to determine the power consumption requirements of the IoT communication module under different data processing methods in the future prediction time.

[0009] In some embodiments, the step of fusing the environmental parameters and the operating status parameters according to the weight values ​​to determine the power consumption requirements of the IoT communication module under different data processing methods within a future predicted time includes: fusing the environmental parameters and the operating status parameters according to the weight values ​​to obtain fusion parameters; and determining the power consumption requirements of the IoT communication module under different data processing methods within a future predicted time based on historical power consumption data and the fusion parameters; wherein the historical power consumption data is the actual power consumption of the IoT communication module within a preset time period in the past.

[0010] In some embodiments, adjusting the operating parameters of the IoT communication module based on the task allocation ratio includes: determining the target parameters of the IoT communication module through a reinforcement learning optimizer; and gradually adjusting the operating parameters within multiple preset scheduling cycles based on a preset step size until the target parameters are reached.

[0011] In some embodiments, after adjusting the operating parameters of the IoT communication module, the method further includes: obtaining the actual power consumption corresponding to the adjusted operating parameters; determining a prediction deviation value based on the actual power consumption and the predicted power consumption requirement; updating the model parameters of the time-series prediction model used to predict the power consumption requirement based on the prediction deviation value; and / or updating the policy network parameters of the reinforcement learning optimizer used to adjust the operating parameters.

[0012] On the other hand, embodiments of this application also provide a power consumption optimization system, the system comprising: a multimodal sensing module configured to collect environmental parameters of the environment in which the IoT communication module is located and the operating status parameters of the IoT communication module; a power consumption prediction module configured to predict the power consumption requirements of the IoT communication module under different data processing methods based on the environmental parameters and the operating status parameters; a collaborative decision-making module configured to determine the task allocation ratio of the IoT communication module under different data processing methods based on the power consumption requirements and the current network quality; and a dynamic adjustment module configured to adjust the operating parameters of the IoT communication module based on the task allocation ratio to achieve power consumption optimization of the IoT communication module.

[0013] In some embodiments, the system further includes a neural processing module and a graphics processing module; wherein the power consumption prediction module and / or the collaborative decision-making module operate on the neural processing module; and the image data acquired by the multimodal perception module is processed by the graphics processing module.

[0014] In some embodiments, the system further includes an intelligent power management module; the intelligent power management module is configured to: monitor the power consumption of the hardware modules corresponding to the multimodal sensing module, the power consumption prediction module, the collaborative decision-making module, and the dynamic adjustment module respectively; and adjust the power allocation strategy of each hardware module according to the adjustment instructions of the dynamic adjustment module.

[0015] On the other hand, embodiments of this application also provide a computer-readable storage medium storing a computer program or instructions thereon, wherein the computer program or instructions, when executed by a processor, implement the steps in any of the power optimization methods provided in embodiments of this application.

[0016] This application discloses a power consumption optimization method, system, and storage medium. The method includes: collecting environmental parameters of the environment in which the IoT communication module operates and the operating status parameters of the IoT communication module; predicting the power consumption requirements of the IoT communication module under different data processing methods based on the environmental parameters and operating status parameters; determining the task allocation ratio of the IoT communication module under different data processing methods based on the power consumption requirements and the current network quality; and adjusting the operating parameters of the IoT communication module based on the task allocation ratio to achieve power consumption optimization. By collecting environmental parameters and operating status parameters, the environmental conditions and operating status of the IoT communication module can be comprehensively perceived, enabling power consumption management to adapt to the differentiated needs of different application scenarios, from smart agriculture to industrial automation. By predicting the power consumption requirements under different data processing methods, the limitations of reactive power adjustment based solely on the current state are overcome. By combining the predicted power consumption requirements and the current network quality to determine the task allocation ratio, and then adjusting the operating parameters after determining the task allocation ratio, joint optimization of communication power consumption and computing power consumption is achieved, solving the problem of low overall system energy efficiency caused by separately managing communication power consumption and computing power consumption. Furthermore, through an end-to-end closed-loop process of collecting environmental and operational parameters, predicting power consumption requirements, determining task allocation ratios, and adjusting operating parameters, the IoT communication module can autonomously make power consumption management decisions based on the actual operating environment and task requirements, without the need for manual intervention or pre-configuration of power consumption modes for specific scenarios, thus achieving adaptive power consumption management. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A structural block diagram of a power optimization system provided in an embodiment of this application; Figure 2 A structural block diagram of another power optimization system provided in this application embodiment; Figure 3 A schematic diagram of the hardware architecture of an Internet of Things (IoT) communication module provided in an embodiment of this application; Figure 4 A flowchart illustrating a power consumption optimization method provided in an embodiment of this application; Figure 5 This is a flowchart illustrating another power consumption optimization method provided in an embodiment of this application. Detailed Implementation

[0019] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of 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.

[0020] In the following description, specific embodiments of the invention will be illustrated with reference to steps and symbols performed by one or more computers, unless otherwise stated. Therefore, these steps and operations will be referred to several times as being performed by a computer, and computer execution as referred to herein includes operations by a computer processing unit representing electronic signals of data in a structured format. This operation transforms the data or maintains it at a location in the computer's memory system, which can be reconfigured or otherwise alter the operation of the computer in a manner well known to those skilled in the art. The data structure maintained by the data is the physical location of the memory, which has specific characteristics defined by the data format. However, the principles of the invention described above are not intended to be limiting, and those skilled in the art will understand that many of the steps and operations described below can also be implemented in hardware.

[0021] The terms "module" or "unit" as used herein can be considered as software objects executing on the computing system. The various components, modules, engines, and services described herein can be considered as implementations on the computing system. While the apparatus and methods described herein are preferably implemented in software, they can also be implemented in hardware, both of which are within the scope of this invention.

[0022] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this specification means the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. It should be understood that when an element is “connected” or “coupled” to another element, it may be directly connected or coupled to the other element, or there may be intermediate elements. Furthermore, “connected” or “coupled” as used herein may include wireless connections or wireless coupling. The term “and / or” as used herein includes all or any units and all combinations of one or more associated listed items.

[0023] The distribution scenarios of IoT devices are becoming increasingly diversified, ranging from smart cities to industrial automation, from smart agriculture to telemedicine. Different application scenarios have significantly different power consumption requirements for communication modules. For example, in smart agriculture scenarios, sensor nodes typically need to work continuously for several years in unattended farmland environments, requiring extremely low power consumption. In contrast, in industrial automation scenarios, devices may require high-frequency data transmission and real-time response, making the balance between performance and power consumption even more important.

[0024] In related technologies, power management of IoT communication modules mainly adopts static configuration or simple dynamic adjustment strategies, lacking the ability to deeply optimize according to changes in application scenarios and environments. Traditional modules typically only provide a limited number of power consumption modes for users to manually select, unable to autonomously adjust according to actual operating environments and task requirements. This approach not only increases the complexity of user configuration, but more importantly, it makes it difficult to achieve the optimal power consumption-performance balance under various dynamically changing scenarios.

[0025] The rise of edge computing technology has brought new opportunities to IoT applications, but it has also placed higher demands on the power management of communication modules. According to Gartner's prediction, by 2035, 75% of enterprise-generated data will be processed in edge computing facilities rather than traditional data centers. In edge computing architectures, communication modules not only need to handle data transmission but also need to work collaboratively with local computing resources, making the power balance between communication and computing particularly complex. How to optimize the energy allocation between communication and computing while ensuring real-time data processing has become a key challenge for IoT devices in edge computing scenarios.

[0026] In recent years, artificial intelligence technology has shown great potential in the field of power consumption optimization. Meanwhile, the development of multimodal sensor fusion technology has provided richer data sources for environmental perception. These technological trends offer new ideas for solving the power consumption management problem of IoT communication modules, but currently there is no comprehensive solution that organically combines these technologies and applies them to the power consumption optimization of communication modules.

[0027] The inventors discovered that the IoT communication modules in related technologies have three main technical problems in power consumption management that urgently need to be solved.

[0028] First, the problem of insufficient scenario adaptability is particularly prominent. The modules provided by related technologies struggle to cope with the differentiated power consumption requirements of various application scenarios. Traditional modules typically employ fixed power consumption configuration strategies or offer only a limited number of operating modes for users to choose from, failing to dynamically adjust according to environmental changes and task requirements. For example, in smart agriculture scenarios, sensor nodes may need to remain in deep sleep mode most of the time, only periodically waking up to transmit small amounts of data; while in industrial monitoring scenarios, continuous transmission of high-definition video streams is required, requiring a completely different strategy for balancing power consumption and bandwidth. These significant differences between scenarios make it difficult for statically configured power management solutions to balance efficiency and performance.

[0029] Secondly, the problem of co-optimizing communication and computing power consumption in edge computing scenarios has not yet been effectively solved. With the development of edge computing technology, IoT devices are increasingly involved in local data processing. This means that communication modules not only need to consider transmission power consumption but also need to coordinate with computing units to optimize overall energy consumption. Related technologies often manage communication power consumption and computing power consumption separately, lacking a unified optimization framework, resulting in low overall system energy efficiency. For example, when a module needs to process a large amount of sensor data, whether to perform edge computing locally or transmit the raw data to the cloud for processing directly affects overall power consumption, but related technologies lack intelligent decision-making mechanisms to achieve the optimal choice.

[0030] Furthermore, power management technologies in these fields generally lack intelligence and adaptability, making it difficult to cope with complex and ever-changing real-world application environments. Traditional power management methods are mostly based on simple threshold triggering or time scheduling mechanisms, which cannot learn and adapt to changing environmental patterns. For example, timer-based periodic wake-up strategies may waste energy during periods of low data volume, while failing to respond promptly during data bursts. In addition, these technologies rarely utilize multi-source environmental data for power demand prediction, resulting in power adjustments often lagging behind actual demand changes.

[0031] To address the aforementioned technical issues, this application aims to provide an intelligent adaptive power consumption optimization IoT communication module and its implementation method. By integrating multimodal sensing, AI prediction algorithms, and dynamic collaborative optimization technology, it achieves intelligent power consumption management of the communication module in different application scenarios. In particular, it solves the problem of collaborative optimization of communication and computing in edge computing scenarios, thereby significantly improving the energy efficiency ratio and battery life of IoT devices.

[0032] Figure 1 This is a structural block diagram of a power optimization system 100 provided in an embodiment of this application. (See diagram below.) Figure 1As shown, the power optimization system 100 may include a multimodal sensing module 110, a power prediction module 120, a collaborative decision-making module 130, and a dynamic adjustment module 140. In some embodiments, the power optimization system 100 may be integrated into the Internet of Things (IoT) communication module to dynamically adjust the power consumption configuration of the IoT communication module according to environmental changes and task requirements.

[0033] The multimodal sensing module 110 is configured to collect environmental parameters of the environment in which the IoT communication module is located and the operating status parameters of the IoT communication module. The power consumption prediction module 120 is configured to predict the power consumption requirements of the IoT communication module under different data processing methods based on environmental parameters and operating status parameters. The collaborative decision-making module 130 is configured to determine the task allocation ratio of the IoT communication module under different data processing methods based on power consumption requirements and current network quality. The dynamic adjustment module 140 is configured to adjust the operating parameters of the IoT communication module based on the task allocation ratio in order to optimize the power consumption of the IoT communication module.

[0034] The power optimization system 100 provided in this application embodiment achieves adaptive management of power optimization through the collaborative effect between various modules.

[0035] In some embodiments, such as Figure 2 As shown, the power optimization system 100 further includes a neural processing module 150 and a graphics processing module 160; wherein, the power prediction module 120 and / or the collaborative decision-making module 130 operate on the neural processing module 150; and the image data acquired by the multimodal perception module 110 is processed by the graphics processing module 160. For example, the neural processing module 150 integrates a neural processing unit (NPU) responsible for intelligent reasoning tasks, such as power prediction and collaborative decision-making; the graphics processing module 160 integrates a graphics processing unit (GPU) responsible for image processing tasks, such as processing the image data acquired by the multimodal perception module 110.

[0036] In some embodiments, such as Figure 2As shown, the aforementioned power optimization system 100 further includes an intelligent power management module 170. The intelligent power management module 170 is configured to monitor the power consumption of the hardware modules corresponding to the multimodal sensing module 110, power prediction module 120, collaborative decision-making module 130, and dynamic adjustment module 140, respectively; and adjust the power allocation strategy of each hardware module according to the adjustment instructions from the dynamic adjustment module 140. The intelligent power management module 170 is primarily responsible for providing independent power supply and power consumption monitoring for each of the aforementioned modules, optimizing power allocation, and achieving refined power consumption control.

[0037] As a concrete example, Figure 3 This is a schematic diagram of the hardware architecture of an IoT communication module provided in an embodiment of this application. The aforementioned power optimization system 100 can be integrated inside the IoT communication module to dynamically adjust the power consumption configuration of the IoT communication module according to environmental changes and task requirements. This hardware architecture adopts a layered modular design concept, dividing the system into four core layers: a sensor and peripheral device layer 310, a communication interface layer 320, a core processor layer 330, and a power management layer 340. This layered design achieves functional decoupling and power domain isolation, providing a hardware foundation for refined power management. The layers are connected through a standardized power distribution network, control signal bus, and data bus, achieving functional decoupling and power domain isolation, thus providing a hardware foundation for refined power management.

[0038] like Figure 3 As shown, the hardware architecture mainly includes four layers: sensor and peripheral device layer 310, communication interface layer 320, core processor layer 330, and power management layer 340.

[0039] The sensor and peripheral device layer 310 includes a multimodal sensing module 110, which is primarily responsible for acquiring environmental parameters and operational status parameters, providing input for power consumption prediction. The multimodal sensing module 110 integrates various environmental sensors, including but not limited to: a temperature sensor (measurement range -40°C to +85°C, accuracy ±0.5°C), a humidity sensor (measurement range 0% to 100%RH, accuracy ±3%RH), a light sensor (measurement range 0 to 100,000 lux), a triaxial accelerometer (measurement range ±16g, 16-bit resolution), and a signal strength monitoring circuit. These sensors can be used to collect environmental parameters of the environment in which the IoT communication module operates (such as temperature, humidity, light intensity, motion status, etc.) as well as the module's own operational status parameters (such as signal strength, etc.). The multimodal sensing module 110 can connect to the main controller via an Inter-Integrated Circuit (I2C) bus and a Serial Peripheral Interface (SPI) bus. Sensor data is preprocessed through a dedicated low-power acquisition circuit, supporting programmable sampling frequency and threshold triggering modes. The module's operating current is less than 5mA, and its sleep current is less than 10μA. A specially designed intelligent power management unit for the aforementioned sensors allows unused sensor channels to be individually shut down to reduce static power consumption.

[0040] The communication interface layer 320 mainly includes a multi-mode communication module, responsible for the wireless data transmission function of the IoT communication module. This multi-mode communication module supports various communication standards, including but not limited to: 5G NewRadio Sub-6GHz band (5G NewRadio Sub-6GHz, 5G NR Sub-6GHz), Long Term Evolution-Advanced (LTE-A), Narrow Band Internet of Things (NB-IoT), and Wi-Fi 6 / 6E. It can automatically switch to the optimal communication mode according to the network environment and application requirements, providing diverse wireless connectivity options. Its 5G communication power consumption is 300-800mW, and its NB-IoT communication power consumption is less than 50mW. The RF front-end adopts a reconfigurable design, supporting dynamic adjustment of transmit power, frequency bandwidth, and modulation scheme. A specially designed antenna diversity system can dynamically select the optimal antenna configuration based on signal quality, reducing power waste caused by signal fading.

[0041] The core processor layer 330 mainly includes AI computing units and a storage controller. It serves as the computational hub for power optimization algorithms, running power prediction models and decision-making algorithms, and executing local edge computing tasks. The AI ​​computing unit is based on the Qualcomm QCS8550 chipset, integrating a quad-core Kryo CPU (Central Processing Unit), an Adreno GPU (Graphics Processing Unit), and a 48-top-PS NPU (Neural Processing Unit). The CPU handles control logic tasks and general-purpose computing; the GPU handles image processing tasks (such as processing image data acquired by the multimodal perception module 110); and the NPU handles AI inference tasks (such as running power prediction models and collaborative decision-making algorithms). Its full-load power consumption is less than 8W, and its idle power consumption is less than 500mW. The computing unit supports Dynamic Voltage and Frequency Scaling (DVFS), which adjusts operating parameters in real time according to the computing load. Through a heterogeneous computing architecture, the system allocates different types of tasks to the most suitable processing units, achieving optimal utilization of computing resources. The storage controller includes memory (such as 4GB LPDDR4 memory) and flash memory (such as 128GB UFS flash memory), supporting memory power management technologies (such as partial memory bank hibernation) and low-power modes for storage devices. Its memory standby power consumption is less than 100mW, and flash memory hibernation power consumption is less than 5mW. The storage controller employs intelligent prefetching and caching strategies to reduce power consumption caused by data access, achieving data caching and persistent storage.

[0042] The power management layer 340 mainly comprises an intelligent power management unit and a power management system. The intelligent power management unit is responsible for providing independent power supply and power consumption monitoring for each of the above layers, optimizing power allocation, and achieving fine-grained power consumption control. This intelligent power management unit employs multi-channel adaptive power converters, including 6 DC-DC converters and 12 LDO regulators, providing independent power regulation for different components, achieving a conversion efficiency of 92%-96% and a static power consumption of less than 100mW. The power management system also integrates high-precision current and voltage monitoring circuits (sampling frequency 1kHz, accuracy ±1%) to monitor the power consumption of each module in real time. Based on the monitoring data, the power management system can dynamically adjust the power allocation strategy. For example, it can increase the power supply to the RF module during peak communication periods, increase the CPU power budget during computationally intensive tasks, and shut down unnecessary power channels during idle periods.

[0043] The aforementioned layers can collaborate through three types of connections: a power distribution network, a control signal bus, and a data bus. Specifically, the power management layer 340 supplies power to the core processor layer 330, the communication interface layer 320, and the sensor and peripheral device layer 310 via the power distribution network, and the power supply to each layer can be independently adjusted, achieving fine-grained power consumption control. The core processor layer 330 sends control commands (such as configuring communication parameters and adjusting sensor sampling frequencies) to the communication interface layer 320 and the sensor and peripheral device layer 310 via the control signal bus. Data collected by the sensor and peripheral device layer 310 is transmitted to the core processor layer 330 for processing via the data bus; the processed data from the core processor layer 330 is then transmitted externally via the communication interface layer 320.

[0044] It should be understood that the hardware architecture design provided in this application embodiment places particular emphasis on low-power optimization. All chips can be selected from industrial-grade low-power devices, critical paths employ low-voltage design, and the device automatically enters deep sleep mode in idle state. The printed circuit board (PCB) layout adopts advanced power optimization technologies, including partitioned layout, multi-plane grounding, and optimized power distribution network, reducing unnecessary power loss. The overall size of the IoT communication module is controlled within 33.0mm × 39.0mm × 3.85mm, compatible with existing module packaging standards, facilitating rapid integration into various IoT devices by customers.

[0045] To facilitate better implementation of the power optimization system 100 provided in this application embodiment, this application embodiment also provides a power optimization method based on the power optimization system 100 described above. Figure 4 This is a flowchart illustrating a power optimization method provided in an embodiment of this application. This method can be applied to the power optimization system 100 of the aforementioned IoT communication module. Figure 4 As shown, the power consumption optimization method includes the following steps (steps S110 to S140): Step S110: Collect environmental parameters of the environment where the IoT communication module is located and the operating status parameters of the IoT communication module. In some embodiments, step S110 can be executed by the multimodal sensing module 110.

[0046] Environmental parameters refer to relevant parameters of the external physical environment in which the IoT communication module operates. For example, environmental parameters may include, but are not limited to, at least one of the following: temperature, humidity, light intensity, and motion status. These environmental parameters reflect the current operating environment of the IoT communication module. Operating status parameters refer to the operating status information of the IoT communication module itself. For example, operating status parameters may include, but are not limited to, at least one of the following: communication quality, data traffic, computing load, CPU utilization, memory usage, network traffic, battery voltage, and battery current. These operating status parameters reflect the current workload and resource usage of the IoT communication module.

[0047] By collecting the aforementioned environmental and operational parameters, the system can comprehensively perceive the environmental conditions and operational status of the IoT communication module, enabling the power management solution to adapt to the differentiated needs of various application scenarios, from smart agriculture to industrial automation, and providing a data foundation for subsequent power consumption prediction.

[0048] It should be noted that the specific types of the above-mentioned environmental parameters and operating status parameters are only illustrative examples, and the embodiments of this application do not impose specific limitations on them. In practical applications, more or fewer parameter types can be selected as needed.

[0049] Step S120: Based on environmental parameters and operating status parameters, predict the power consumption requirements of the IoT communication module under different data processing methods. After collecting the environmental parameters and operating status parameters in step S110, the above parameters are input into the prediction model in the power consumption prediction module 120, and the prediction model predicts the power consumption requirements of the IoT communication module under different data processing methods.

[0050] The data processing method refers to the way the IoT communication module processes the collected data. In some embodiments, the data processing method includes at least one of the following: local edge computing and cloud processing. Local edge computing means processing the data locally on the IoT communication module; cloud processing means transmitting the data to a cloud server for processing.

[0051] It is understandable that different data processing methods may have different power consumption requirements. For example, in local edge computing, the main power consumption comes from the computation of computing units (such as CPUs and NPUs) and the energy consumption of transmitting computation results; while in cloud processing, the main power consumption comes from data transmission (such as the transmission power consumption of radio frequency modules) and the computing power consumption of the cloud. Therefore, this application's embodiments, by predicting the power consumption requirements under different data processing methods, overcome the limitations of reactive power adjustment based solely on the current state in related technologies, and can provide a basis for subsequent task allocation decisions.

[0052] In some embodiments of this application, the prediction model can employ a deep learning-based method to predict power consumption requirements. Using a pre-trained deep learning model, based on input environmental parameters and operating state parameters, the model outputs predicted power consumption values ​​for the IoT communication module under different data processing methods. It should be understood that the prediction model can be trained on a large amount of historical data before actual deployment to establish a mapping relationship between environmental parameters, operating state parameters, and power consumption requirements.

[0053] Step S130: Based on power consumption requirements and current network quality, determine the task allocation ratio of the IoT communication module under different data processing methods. After predicting the power consumption requirements under different data processing methods in step S120, the collaborative decision-making module 130, in conjunction with the current network quality, determines the task allocation ratio of the IoT communication module under different data processing methods.

[0054] The task allocation ratio refers to the proportion of tasks to be processed among various data processing methods allocated by the IoT communication module. Current network quality refers to the quality of the network environment in which the IoT communication module is currently located, which can be measured from multiple dimensions, such as signal strength, signal-to-noise ratio, link stability, and transmission latency. The quality of the network directly affects the power consumption and efficiency of data transmission; therefore, network quality factors need to be comprehensively considered when making task allocation decisions. For example, when the network quality is good, the proportion of cloud processing can be appropriately increased to utilize cloud computing capabilities and reduce local computing overhead; when the network quality is poor, the proportion of local processing should be increased to avoid additional power consumption caused by retransmissions and delays.

[0055] In some embodiments of this application, the current network quality can be determined based on at least one of the following indicators: signal strength index, signal-to-noise ratio index, link stability index, and transmission delay index. These indicators can be used individually or in combination to more comprehensively assess the current network condition.

[0056] The task allocation ratio can be determined in various ways. For example, a decision function can be constructed based on power demand prediction results and current network quality, and the optimal task allocation ratio can be determined by optimizing the decision function; alternatively, the task allocation ratio can be determined based on preset decision rules or optimization algorithms. This application does not limit the specific method for determining the task allocation ratio, as long as it can be determined based on power demand prediction results and current network quality.

[0057] Step S140: Based on the task allocation ratio, adjust the operating parameters of the IoT communication module to optimize the power consumption of the IoT communication module. After determining the task allocation ratio in step S130, the dynamic adjustment module 140 adjusts the operating parameters of the IoT communication module based on the task allocation ratio.

[0058] Operating parameters refer to the adjustable parameters of an IoT communication module during operation. In some embodiments of this application, operating parameters may include communication parameters and / or computational parameters. Communication parameters may include, but are not limited to, radio frequency transmission power, modulation method, and channel bandwidth, while computational parameters may include, but are not limited to, CPU operating frequency, CPU core voltage, and data transmission rate. By adjusting the above operating parameters, the IoT communication module can switch between different operating states or adaptively adjust, thereby optimizing power consumption while meeting task processing requirements.

[0059] The specific methods for adjusting the operating parameters can be as follows: when it is determined that many tasks are processed locally, the operating parameters of the computing unit are increased accordingly to provide sufficient computing power; when it is determined that many tasks are processed in the cloud, the transmission power and other parameters of the communication module are adjusted accordingly to optimize data transmission power consumption. The adjustment of operating parameters can be carried out gradually or by switching directly, and this application embodiment does not specifically limit this.

[0060] Through steps S110 to S140 above, the power consumption optimization method provided in this application embodiment can predict power consumption requirements under different data processing methods based on environmental parameters and operating status parameters, determine the task allocation ratio in conjunction with the current network quality, and adjust the working parameters based on the task allocation ratio, thereby achieving power consumption optimization of the IoT communication module. This method can adapt to different application scenarios and environmental changes, without manual intervention or pre-configuration of power consumption modes for specific scenarios, and can achieve adaptive power consumption management.

[0061] In some embodiments, this application also provides a specific method for determining the task allocation ratio. The above step S130 may include the following steps (steps S131 and S132): Step S131: Correct the first energy consumption prediction value under the cloud processing mode according to the current network quality; wherein, the current network quality is determined based on at least one of the following indicators: signal strength index, signal-to-noise ratio index, link stability index, and transmission delay index.

[0062] The collaborative decision-making module 130 first obtains the first energy consumption prediction value under the cloud processing mode output by the power consumption prediction module 120. This first energy consumption prediction value mainly includes two parts: data transmission energy consumption and cloud computing energy consumption. Specifically, data transmission energy consumption refers to the energy consumption required by the IoT communication module to transmit raw data to the cloud server. This energy consumption is estimated in real-time based on a network energy consumption model, taking into account current network conditions (such as signal strength and bandwidth), data volume, and transmission distance. Cloud computing energy consumption refers to the energy consumption required by the cloud server to process the data. This energy consumption is estimated based on the task type (such as image recognition and data aggregation) and computational complexity, using a preset cloud computing energy consumption benchmark model. This benchmark model can be established based on typical power consumption data from mainstream cloud platforms.

[0063] Subsequently, the collaborative decision-making module 130 can correct the aforementioned first energy consumption prediction value based on the current network quality. For example, the current network quality can be characterized by the Network Quality Factor (NQF). The NQF is a dimensionless comprehensive index with a value range of [0, 1]. As a specific example, the NQF can be weighted by the following sub-indicators: the normalized value of the Received Signal Strength Indicator (RSSI), with a weight of 0.3; the normalized value of the Signal-to-Noise Ratio (SNR), with a weight of 0.25; the link stability index calculated based on the packet loss rate, which can be represented by 1 - packet loss rate, with a weight of 0.25; and the normalized value of the Round-Trip Time (RTT), with a weight of 0.2. These sub-indicators can be calculated in real-time based on actual network monitoring data. The specific correction method is: the first energy consumption prediction value E_cloud' under cloud processing mode = E_cloud / NQF. When NQF approaches 1, the network quality is excellent, and the corrected energy consumption is close to the original estimate. When NQF approaches 0, the network quality is extremely poor, and the energy consumption of cloud processing approaches infinity. At this time, the collaborative decision module 130 can automatically prioritize the local edge computing strategy to avoid additional energy consumption caused by network fluctuations.

[0064] Step S132: Determine the task allocation ratio based on the corrected first energy consumption prediction value and the second energy consumption prediction value under the local edge computing method.

[0065] The collaborative decision-making module 130 first obtains the second energy consumption prediction value under the local edge computing mode output by the power consumption prediction module 120. The second energy consumption prediction value includes two parts: local computing energy consumption and result transmission energy consumption.

[0066] Subsequently, the collaborative decision-making module 130 compares the corrected first energy consumption prediction value with the second energy consumption prediction value. The collaborative decision-making module 130 uses processing latency and total energy consumption as optimization objectives, and the task allocation ratio between local edge computing and cloud processing as decision variables. Under the premise of satisfying latency constraints, it selects the task allocation scheme that minimizes the total system energy consumption. For example, the task allocation ratio can be expressed as α: 1-α, where α represents the proportion using local edge computing, 1-α represents the proportion using cloud processing, and the value of α ranges from [0, 1]. When α=1, it means all tasks are processed using local edge computing; when α=0, it means all tasks are processed using cloud processing; and when 0<α<1, it means some tasks are processed using local edge computing and some tasks are processed using cloud processing.

[0067] For example, the collaborative decision-making module 130 uses a multi-objective optimization algorithm to determine the task allocation ratio. In some embodiments of this application, the multi-objective optimization algorithm uses Non-dominated Sorting Genetic Algorithm II (NSGA-II). The NSGA-II algorithm uses the task allocation ratio α as the decision variable, with the dual optimization objectives of minimizing the total system energy consumption and satisfying the processing delay constraint. Through non-dominated sorting and congestion calculation, it selects the optimal task allocation ratio α with the lowest energy consumption under the delay constraint on the Pareto front. It should be noted that the NSGA-II algorithm can trigger offline search during the system initialization phase or when the network quality changes significantly, storing the obtained Pareto front in the storage module in the form of a lookup table; during online decision-making, the collaborative decision-making module 130 quickly matches the corresponding optimal task allocation ratio according to the current system state and network quality through the lookup table. In other embodiments, when computing resources are sufficient, the NSGA-II algorithm can also be executed online in real time. The specific execution method of the multi-objective optimization algorithm is not limited in the embodiments of this application.

[0068] For example, when the IoT communication module receives sensor data or application commands, the task analyzer in the collaborative decision-making module 130 first extracts data features. For example, for video stream data, the task analyzer can analyze its resolution, frame rate, and compression format; for sensor data, the task analyzer can analyze its sampling frequency, data accuracy, and real-time requirements. Then, the energy consumption predictor in the collaborative decision-making module 130 estimates energy consumption based on the above data features and the current system status (including but not limited to battery power, CPU load, network signal strength, etc.). Finally, the decision engine in the collaborative decision-making module 130 uses a multi-objective optimization algorithm (such as the aforementioned NSGA-II algorithm) to select the processing strategy that minimizes the total system energy consumption from the local edge computing method and the cloud processing method, or determines the optimal task allocation ratio between the local edge computing method and the cloud processing method, provided that the latency requirements are met. For example, for high-definition video streams in real-time monitoring scenarios, the collaborative decision-making module 130 can decide to perform edge computing locally, transmitting only key frames and analysis results to the cloud to reduce data transmission volume and power consumption. For non-real-time environmental sensor data, the collaborative decision-making module 130 can choose to upload it in batches to the cloud for processing when network conditions are good, utilizing cloud computing capabilities to reduce local computing overhead. Through these methods, the collaborative decision-making module 130 can flexibly formulate optimal task allocation strategies based on different data types and real-time requirements, achieving deep collaborative optimization of communication and computing.

[0069] In some embodiments, step S120 may include the following steps (steps S121 and S122): Step S121: Perform preprocessing on environmental parameters and operating status parameters.

[0070] After the multimodal sensing module 110 collects environmental parameters and operating status parameters at a preset frequency (e.g., 10Hz), the sensing layer performs preprocessing on the collected raw data. Preprocessing includes, but is not limited to, at least one of the following operations: noise reduction (filtering out noise interference in sensor data), normalization (unifying parameters of different dimensions to the same numerical range), outlier detection (identifying and removing outliers during data acquisition), and feature extraction (extracting key features from the raw data). Through preprocessing, the high-dimensional raw data is converted into low-dimensional feature vectors to improve the computational efficiency of the subsequent power consumption prediction module 120.

[0071] Step S122: When the change in environmental parameters exceeds a preset threshold, predict the power consumption requirements of the IoT communication module under different data processing methods based on the preprocessed environmental parameters and the preprocessed operating status parameters.

[0072] The preset thresholds are dynamically set based on the physical characteristics of each sensor and the application scenario. For example, the thresholds for temperature change are ±0.5°C, humidity change are ±2%RH, light intensity change are ±100 lux, acceleration change is ±0.1g, and signal strength change is ±3dBm. The system also provides an adaptive threshold learning mechanism, which automatically calculates and sets the thresholds for each sensor based on the statistical distribution of environmental data (e.g., mean ± 2 standard deviations) within a preset time after initial deployment (e.g., 48 hours). Subsequent manual fine-tuning can be performed based on a trade-off between actual power consumption and data integrity.

[0073] When the changes in environmental parameters and / or operating status parameters exceed a preset threshold, the preprocessed environmental feature vector is transmitted to the power consumption prediction module 120 via shared memory. The power consumption prediction module 120 then predicts the power consumption requirements of the IoT communication module under different data processing methods based on the preprocessed environmental parameters and operating status parameters.

[0074] In this embodiment, by decoupling hardware sampling from data transmission, uplink transmission is only triggered when data changes exceed a preset threshold, effectively reducing communication overhead between modules and power consumption of the sensing layer while ensuring that environmental changes are not missed.

[0075] In some embodiments, step S120 may include the following steps (steps S123 and S124): Step S123: Determine the weight values ​​of environmental parameters and operating state parameters respectively through a deep learning-based time series prediction model; wherein, the time series prediction model has model parameters.

[0076] Time series prediction models can employ an improved LSTM (Long Short-Term Memory) architecture, which effectively processes time series data and captures the dynamic changes in environmental and operational parameters. These models have model parameters, including network weights and biases, which can be optimized during training using extensive historical data.

[0077] The time-series prediction model consists of an input layer, a feature extraction layer, an attention fusion layer, and a prediction output layer. The input layer receives environmental parameters (such as temperature, humidity, light intensity, and motion status) and operational parameters (such as CPU utilization, memory usage, network traffic, battery voltage, and battery current) from various sensors. The feature extraction layer comprises multiple parallel one-dimensional convolutional neural networks (CNNs), each CNN branch processing one type of sensor data and extracting its time-series features. The attention fusion layer learns the importance weights of different sensor features through a self-attention mechanism and dynamically combines them. The prediction output layer uses a bidirectional LSTM network to predict power consumption requirements for future time periods based on the fused features.

[0078] Time series prediction models can be trained through transfer learning. First, a base model is trained in a laboratory environment using a large amount of labeled data. Then, during actual deployment, the model parameters are continuously optimized through an online learning mechanism. This approach ensures both the model's generalization ability and its adaptability to the characteristics of specific application scenarios.

[0079] Step S124: According to the weight values, perform data fusion on environmental parameters and operating status parameters to determine the power consumption requirements of the IoT communication module under different data processing methods in the future forecast time.

[0080] The attention fusion layer automatically learns the contribution weights of different sensor data to power consumption prediction through a self-attention mechanism. Sensor features that contribute significantly to power consumption prediction are assigned higher weights, while those that contribute less are assigned lower weights. Then, according to these weights, environmental parameters and operating status parameters from different sensors are fused, and the feature vectors of each sensor are weighted and combined into a fused feature vector. Finally, the prediction output layer uses a bidirectional LSTM network to predict the power consumption requirements of the IoT communication module under different data processing methods within a future prediction time (e.g., 5 seconds or 5 minutes). For example, the 5-second prediction is for real-time DVFS parameter adjustment, and the 5-minute prediction is for macro-level decisions regarding communication strategies and task allocation; both constitute a multi-level prediction system. For example, the system predicts the power consumption demand curve for the next 5 minutes or the power consumption demand for the next 5 seconds. The system can adjust the operating parameters of the communication module in advance based on this power consumption demand curve. For instance, if it is predicted that the light intensity will increase significantly in the future, meaning the device may be outdoors in sunlight and solar charging is available, the system can appropriately increase the power budget and enable a higher-performance communication mode. Conversely, when it is predicted that a device will enter a state of motion, such as the transportation process in a logistics tracking scenario, the system can switch to a low-power mobile mode in advance and optimize the radio frequency circuit parameters to adapt to the rapidly changing signal environment.

[0081] In this embodiment, by employing a multimodal sensor fusion deep learning model based on an attention mechanism, the contribution weights of different sensor data to power consumption prediction can be automatically learned. Compared to traditional power consumption prediction methods based on a single parameter, this significantly improves the accuracy of power consumption prediction and provides a more reliable basis for subsequent task allocation decisions. Experiments show that the prediction error of this multimodal fusion model is reduced by 62% compared to the traditional single-parameter model and by 43% compared to the simple weighted fusion model.

[0082] In some embodiments, after adjusting the operating parameters of the IoT communication module, the power consumption optimization method may further include the following steps (steps S151 to S153A): Step S151: Obtain the actual power consumption corresponding to the adjusted operating parameters. The feedback monitoring mechanism feeds back power consumption indicators (actual power consumption values ​​of each module such as CPU, RF, NPU, and total system power consumption) to the decision layer through a shared memory ring buffer. The actual power consumption is monitored and recorded in real time by the intelligent power management module 170 at a sampling frequency of 1kHz.

[0083] Step S152: Determine the prediction deviation value based on the actual power consumption and the predicted power consumption demand. The collaborative decision-making module 130 compares the actual power consumption with the power consumption demand predicted by the power consumption prediction module 120 and calculates the prediction deviation value. The prediction deviation value reflects the prediction accuracy of the power consumption prediction module 120 for power consumption demand. When the deviation is large, it indicates that the prediction model needs to be updated.

[0084] Step S153A: Update the model parameters of the time-series prediction model used to predict power consumption requirements based on the prediction deviation value.

[0085] When the prediction deviation exceeds a preset deviation threshold, the power consumption prediction module 120 triggers a model update process. The power consumption prediction module 120 uses the actual power consumption data and corresponding environmental and operating status parameters from the recent period to update the model parameters (such as network weights and biases) of the time series prediction model through an online learning mechanism.

[0086] For example, the time-series prediction model can be updated asynchronously in the background without affecting the normal operation of the system. After the update is completed, the power consumption prediction module 120 uses the updated model parameters to predict subsequent power consumption requirements. Through the above online update mechanism, the time-series prediction model can continuously adapt to changes in the environment and task requirements, maintaining prediction accuracy.

[0087] In this embodiment, the actual power consumption is obtained through a feedback monitoring mechanism and compared with the predicted value. The prediction model parameters are updated online according to the prediction deviation, realizing closed-loop adaptive optimization. This enables the system to adapt to factors such as equipment aging and environmental changes during long-term use, and maintain optimal power consumption performance throughout its entire life cycle.

[0088] In some embodiments, step S124 above includes the following steps (steps S1241 and S1242): Step S1241: According to the weight values, perform data fusion on the environmental parameters and the operating status parameters to obtain the fused parameters.

[0089] The attention fusion layer weights and combines the original feature vectors of environmental parameters and operating state parameters from different sensors to obtain a comprehensive fusion feature vector (i.e., fusion parameters). The fusion parameters integrate the contribution information of each sensor to power consumption prediction and serve as the input to the subsequent prediction model.

[0090] Step S1242: Based on historical power consumption data and fusion parameters, determine the power consumption requirements of the IoT communication module under different data processing methods within the future predicted time; wherein, the historical power consumption data is the actual power consumption of the IoT communication module within the past preset time.

[0091] In addition to receiving fusion parameters, the input layer of the time-series prediction model also receives historical power consumption data as time-series feature input. The historical power consumption data is the actual power consumption of the IoT communication module within a preset time period in the past. For example, the preset time period in the past can be the past 5 seconds, the past 10 seconds, or the past 1 minute, etc., and this application does not make a specific limitation in this regard.

[0092] Historical power consumption data is monitored and recorded in real time by the intelligent power management module 170. The intelligent power management module 170 continuously monitors the power consumption of each hardware module at a sampling frequency of 1kHz, and records the actual power consumption of each module and the total system power consumption in the storage module. When the power consumption prediction module 120 performs power demand prediction, it reads historical power consumption data from the storage module within a preset time period.

[0093] Time-series forecasting models predict power consumption requirements under different data processing methods within a future forecast period based on historical power consumption data and fused parameters. By incorporating historical power consumption data, time-series forecasting models can capture the trend characteristics of power consumption changes over time. For example, if historical power consumption data shows that power consumption is continuously increasing, even if the parameter values ​​at the current moment have not changed drastically, power consumption may continue to rise in the future, thereby improving the accuracy and reliability of power consumption forecasting.

[0094] In this embodiment, by using historical power consumption data along with environmental parameters and operating status parameters as inputs to the prediction model, the time-series variation pattern of power consumption can be better captured, further improving the accuracy and reliability of power consumption prediction.

[0095] In some embodiments, step S140 may include the following steps (steps S141 and S142): Step S141: Determine the target parameters of the IoT communication module through a reinforcement learning optimizer; wherein, the policy network parameters of the reinforcement learning optimizer are used to determine the target parameters.

[0096] The dynamic adjustment module 140 employs a reinforcement learning-based optimizer to determine the optimal target parameters. The reinforcement learning optimizer learns the optimal parameter adjustment strategy through continuous interaction with the environment.

[0097] The reinforcement learning optimizer takes the current system state as input and outputs a set of target parameter configurations. The current system state includes, but is not limited to, at least one of the following: task type, data volume, network quality, battery level, current CPU frequency, current RF transmit power, and current modulation scheme. The target parameter configurations output by the reinforcement learning optimizer include, but are not limited to, at least one of the following: CPU operating frequency, CPU core voltage, RF transmit power, modulation scheme, and data transmission rate.

[0098] For example, a reinforcement learning optimizer can employ an Asynchronous Advantage Actor-Critic (A3C) architecture to achieve parameter optimization in a continuous action space. The policy network parameters of the reinforcement learning optimizer are used to determine the target parameters; that is, the policy network outputs an action probability distribution or deterministic action value based on the current system state, and the action corresponds to the target value or adjustment amount of each working parameter. Through continuous interaction between the agent and the environment, it autonomously learns the optimal parameter adjustment strategy, enabling the simultaneous adjustment of multiple hardware parameters and solving the problem of limited optimization space in traditional discrete parameter adjustment methods.

[0099] The optimization objective is to minimize the total system power consumption while meeting application performance requirements (such as data transmission rate and latency constraints). Optimization variables include: CPU operating frequency (e.g., 800MHz to 2.8GHz, in 100MHz steps), CPU core voltage (e.g., 0.8V to 1.3V, in 0.05V steps), RF transmit power (e.g., -40dBm to +23dBm, in 1dBm steps), modulation scheme (e.g., QPSK, 16QAM, 64QAM, or 256QAM), and data transmission rate (dynamically adjusted according to the communication standard).

[0100] Step S142: Based on the preset step size, gradually adjust the working parameters within multiple preset scheduling cycles until the target parameters are reached.

[0101] After the reinforcement learning optimizer determines the target parameters, the dynamic adjustment module 140 adopts an incremental adjustment strategy, based on a preset step size, to gradually adjust the working parameters within multiple preset scheduling cycles until the target parameters are reached.

[0102] Incremental adjustment strategy refers to a mechanism where, when power consumption prediction results trigger parameter changes, the system does not directly jump to the target configuration, but rather gradually approaches it in multiple steps with the minimum step size supported by the hardware, completing a smooth transition within multiple scheduling cycles. The dynamic adjustment module 140 can employ a multi-level clock architecture to implement incremental adjustment. For example, the software flow uses a 10ms basic scheduling cycle. The LSTM prediction module predicts future power consumption requirements with a 100ms cycle; the reinforcement learning optimizer calculates the optimal target parameter configuration with a 50ms cycle; and the execution layer uses a 10ms scheduling tick, executing parameter adjustments one step at a time. 10ms is the lowest-level scheduling tick, and 50ms and 100ms are both integer multiples of 10ms, with hierarchical triggering achieved through the RTOS's timer interrupt mechanism. A 50ms task is triggered every 5 10ms ticks, and a 100ms task is triggered every 10 10ms ticks.

[0103] The specific process for incremental adjustment is as follows: Step 1 (Prediction Trigger): The power consumption prediction module 120 predicts future power consumption requirements at a period of 100ms and determines whether parameter adjustments are needed.

[0104] Step 2 (Target Configuration Calculation): The reinforcement learning optimizer calculates the optimal target parameter set based on the prediction results in 50ms cycles.

[0105] Step 3 (Incremental Planning): The dynamic adjustment module 140 calculates the difference between the current parameter and the target parameter, and decomposes the difference into several steps. Each parameter has a preset adjustment step size, such as a step size of 100MHz for CPU frequency, a step size of 0.05V for CPU core voltage, and a step size of 1dBm for RF transmit power.

[0106] Step 4 (Period-by-cycle execution): The dynamic adjustment module 140 performs a parameter adjustment step of one size in every 10ms tick.

[0107] Step 5 (Effect Monitoring and Closed-Loop Correction): The feedback monitoring mechanism evaluates the adjustment effect in real time, and corrects the subsequent step size when the deviation is too large.

[0108] Step 6 (Reaching the Target or Re-predicting): When the parameters are adjusted to reach the target value or when new prediction results change midway, re-plan the incremental path.

[0109] Key mechanisms for smooth transitions include: (1) Early start: Based on the forecast window (e.g., 5 seconds), the incremental process is started before the demand changes, avoiding post-event remediation.

[0110] (2) Synchronous stepping: When multiple parameters need to be adjusted at the same time (for example, when the CPU frequency and RF power need to be increased at the same time), the parameters are increased in parallel according to the same rhythm to avoid local suboptimal state caused by the change of a single parameter first.

[0111] (3) Step size adaptation: The execution layer provides feedback monitoring to evaluate the adjustment effect in real time. If a transient power consumption spike or communication quality deteriorates, the subsequent step size will be automatically reduced.

[0112] (4) Emergency pass-through: In extreme scenarios (such as sudden drop in battery power or network interruption), the incremental strategy can be bypassed and the system can be directly switched to safe mode.

[0113] In this embodiment, the optimal target parameters are determined by a reinforcement learning optimizer, which can adapt to different application scenarios and working conditions and achieve joint optimization of multiple parameters and multiple objectives. The incremental adjustment strategy avoids system instability and communication link drop-off caused by drastic parameter changes, eliminates power supply transient current spikes, and ensures service quality and communication reliability.

[0114] In some embodiments, after adjusting the operating parameters of the IoT communication module, the power consumption optimization method may further include the following steps (steps S151 to S153B): Step S151: Obtain the actual power consumption corresponding to the adjusted operating parameters; Step S152: Determine the prediction deviation value based on the actual power consumption and the predicted power consumption requirement; Step S153B: Update the policy network parameters of the reinforcement learning optimizer used to adjust the working parameters based on the prediction bias value.

[0115] When the prediction deviation exceeds a preset deviation threshold, the dynamic adjustment module 140 triggers the update process of the reinforcement learning optimizer. The feedback period is synchronized with the parameter optimization period (e.g., 50ms). The dynamic adjustment module 140 inputs the feedback signals of actual power consumption and performance indicators to the reinforcement learning optimizer, and updates the policy network parameters through a reinforcement learning algorithm (such as A3C).

[0116] For example, the update process can use feedback signals from actual power consumption and performance metrics to calculate reward values, and update the policy network parameters through gradient ascent, enabling the reinforcement learning optimizer to output better parameter configurations in the future. Performance metrics include communication quality metrics (actual throughput, end-to-end latency, packet loss rate, number of retransmissions) and computational performance metrics (task processing completion time, CPU / GPU / NPU utilization).

[0117] Through the aforementioned online update mechanism, the reinforcement learning optimizer can continuously adapt to changes in the environment and task requirements, constantly optimizing its parameter adjustment strategy. After the update is completed, the dynamic adjustment module 140 uses the updated policy network to determine subsequent target parameters.

[0118] In this implementation, the actual power consumption is obtained through a feedback monitoring mechanism and compared with the predicted value. The policy network parameters of the reinforcement learning optimizer are updated online based on the prediction deviation, achieving closed-loop adaptive optimization. This allows the system to adapt to factors such as equipment aging and environmental changes during long-term use, maintaining optimal power consumption performance throughout its entire lifecycle. Experiments show that after two years of equipment use, the module using this solution can still maintain more than 85% of the initial optimization effect, while traditional solutions typically drop to below 50%.

[0119] As a concrete example, in this embodiment, the software system of the power optimization method adopts a layered architecture design, including three layers: a perception layer, a decision layer, and an execution layer. The perception layer is responsible for data acquisition and preprocessing, serving as the information entry point for the entire power optimization system 100. The decision layer is the core of the system, realizing power prediction and optimization decision-making functions. The execution layer is responsible for translating the optimization strategies formulated by the decision layer into specific hardware control instructions. The layers communicate with each other through standardized interfaces (such as shared memory) to ensure the modularity and scalability of the system. The software process uses a basic scheduling cycle of 10ms to achieve rapid response to environmental changes and dynamic adjustments to task requirements. The system employs a real-time operating system (RTOS) to ensure timely response and execution of critical tasks.

[0120] Figure 5 This is a flowchart illustrating another power optimization method provided in an embodiment of this application. Figure 5 As shown, the power consumption optimization method includes the following steps (steps S501 to S516): The perception layer corresponds to steps S501 to S504 below and is mainly responsible for data acquisition and preprocessing.

[0121] Step S501: System initialization; Start the power optimization system 100 and initialize the configuration of each functional module and parameter. This includes initializing the sensor array of the multimodal perception module 110, configuring the sampling frequency and threshold parameters, loading the model parameters of the pre-trained prediction model, and initializing the network policy parameters of the reinforcement learning optimizer.

[0122] Step S502: Sensor data acquisition; At a sampling frequency of 10Hz, environmental parameters (such as temperature, humidity, light, motion status, etc.) are collected in real time through the sensor array of the multimodal sensing module 110, while the operating status parameters of the Internet of Things communication module (such as CPU utilization, memory usage, network traffic, battery voltage and current, etc.) are detected at the same time.

[0123] Step S503: Data preprocessing; The perception layer performs preprocessing operations such as noise reduction, filtering, and normalization on the collected raw data to remove outliers and noise interference.

[0124] Step S504: Feature Extraction; Extract feature vectors for power consumption prediction from the preprocessed data, including environmental features and operating status features. After feature extraction, the perception layer uses an event-triggered mechanism to decide whether to transmit the data to the decision layer.

[0125] The decision-making layer includes the following steps S505 and S506, which are mainly responsible for power consumption prediction and optimization decisions.

[0126] Step S505: Power consumption demand prediction; Input the extracted feature vector into the power consumption prediction module 120 to predict the power consumption demand of the IoT communication module under different data processing methods (local edge computing method, cloud processing method).

[0127] Step S506: Local processing? Based on the power consumption demand prediction results and the current network quality, determine whether the data to be processed is to be processed locally using edge computing. If yes, proceed to step S507 to enter the local processing path; if no, proceed to step S510 to enter the cloud processing path.

[0128] The execution layer includes steps S507 to S512, which are mainly responsible for parameter configuration and task execution.

[0129] The local processing path includes the following steps S507 to S509: Step S507: Configure edge computing parameters; Configure relevant parameters for local edge computing (such as CPU frequency, core voltage, etc.) according to task requirements and power consumption optimization goals.

[0130] Step S508: Perform local data processing; perform data processing tasks (such as data aggregation, image preprocessing, inference calculation, etc.) locally on the IoT communication module.

[0131] Step S509: Transmit processing results; transmit the locally processed result data (not the original data) to the cloud or other terminals through the communication module.

[0132] The cloud processing path includes the following steps S510 to S512: Step S510: Configure communication transmission parameters; Configure communication parameters (such as radio frequency transmission power, modulation method, channel bandwidth, etc.) according to the current network quality and data transmission requirements.

[0133] Step S511: Transmit data to the cloud; transmit the raw data to the cloud server via wireless network (such as 5G, NB-IoT, Wi-Fi, etc.).

[0134] Step S512: Receive the cloud processing result; receive the cloud server's processing result of the data, and perform subsequent operations as needed.

[0135] The above steps S509 and S512 merge the paths, and then proceed to step S513: Step S513: Parameter effect evaluation; obtain the actual power consumption corresponding to the adjusted working parameters, evaluate the actual operating effect of the current parameter configuration, and calculate the deviation between the actual power consumption and the predicted power consumption.

[0136] Step S514: Is the effect optimal? Determine whether the current parameter configuration has achieved the optimal power consumption effect. If yes, proceed to step S515; if no, proceed to step S516.

[0137] Step S515: Maintain the current configuration; keep the current working parameter configuration unchanged, return to S502, and continue to monitor changes in environment and operating status.

[0138] Step S516: Update and optimize the model; based on the deviation between the actual power consumption and the predicted power consumption, update the model parameters of the prediction model and / or update the policy network parameters of the reinforcement learning optimizer to achieve closed-loop adaptive optimization. After the update is completed, return to S502 to enter the next round of data collection and optimization loop.

[0139] The above steps constitute a complete closed-loop adaptive power optimization process: Perception phase: including steps S501 to S504 above, to complete data acquisition and feature extraction; Prediction phase: Includes step S505 above, to complete the power consumption demand prediction; Decision-making stage: including step S506 above, determining local / cloud processing based on prediction results and network quality; Execution phase: including the aforementioned local processing paths S507 to S509 and / or the aforementioned cloud processing paths S510 to S512, to complete data processing and transmission; Feedback phase: This includes steps S513 to S516 above, completing the effect evaluation and model update.

[0140] The power optimization method and system provided in this application, through an end-to-end closed-loop architecture of multimodal sensing, power prediction, collaborative decision-making, and dynamic adjustment, have the following specific technical effects.

[0141] First, the technical solution provided in this application reduces the power consumption of IoT devices and extends their battery life. This application uses a multimodal sensing module 110 to collect environmental and operational parameters in real time, a power consumption prediction module 120 to perform multimodal fusion prediction, a collaborative decision-making module 130 to determine the task allocation ratio between local edge computing and cloud processing based on the prediction results and current network quality, and finally, a dynamic adjustment module 140 dynamically adjusts the operating parameters of the IoT communication module to minimize power consumption while meeting application requirements. Experimental data shows that in typical IoT application scenarios, the power consumption of the IoT communication module using the technical solution provided in this application is reduced by 40% to 65% compared to traditional modules. For example, in smart agriculture sensor node applications, devices using the technical solution provided in this application extend battery life from 6 months to 22 months compared to traditional modules while maintaining the same data acquisition and transmission performance; in industrial monitoring scenarios, the power consumption of high-definition camera modules using the technical solution provided in this application is reduced by 48%, significantly reducing wiring and power supply costs.

[0142] Secondly, the technical solution provided in this application significantly improves the adaptability of the device in different application scenarios. This application embodiment utilizes the real-time acquisition of environmental and operational parameters by the multimodal sensing module 110, the forward-looking prediction of future power consumption needs by the power consumption prediction module 120, and the adaptive adjustment of operating parameters by the dynamic adjustment module 140. This enables the IoT communication module to intelligently identify application scenarios and automatically adjust power consumption optimization strategies, adapting to diverse working environments without manual intervention. Test results show that the system can automatically identify at least 12 typical IoT application scenarios, including smart agriculture, industrial monitoring, smart metering, logistics tracking, and smart homes, and optimize power consumption configuration for each scenario. For example, in the logistics tracking scenario, when the multimodal sensing module 110 detects continuous motion, the collaborative decision-making module 130 automatically determines the task allocation ratio between local edge computing and cloud processing, and the dynamic adjustment module 140 switches to a low-power mobile mode accordingly, reducing the power consumption of the radio frequency circuit by 55%. In the fixed-installation smart metering scenario, the dynamic adjustment module 140 is optimized to a deep sleep mode, only briefly waking up during data transmission, reducing the average power consumption to below 5μA.

[0143] Third, the technical solution provided in this application optimizes the overall system performance in local edge computing scenarios. This application embodiment, through a collaborative decision-making module 130, determines the task allocation ratio between local edge computing and cloud processing based on predicted power consumption requirements and current network quality, achieving deep collaborative optimization of communication and computing. In edge computing applications, this mechanism improves system response speed by 30% to 60% while reducing data transmission volume by 40% to 75%. For example, in industrial quality inspection scenarios, devices using the technical solution provided in this application embodiment can complete over 90% of preliminary image analysis locally, transmitting only suspicious defect images to the cloud for in-depth detection. This reduces data transmission volume by 85% and shortens response time from 2 seconds in traditional solutions to 0.3 seconds, improving real-time performance. While related edge computing controllers have also attempted to solve similar problems, they have failed to achieve deep collaborative optimization of communication and computing.

[0144] Fourth, the technical solution provided in this application improves the accuracy of power consumption prediction and decision-making. This application's embodiment employs a deep learning-based time-series prediction model through the power consumption prediction module 120. Based on historical power consumption data and environmental feature vectors, it determines the weights of different environmental parameters and operating state parameters, and performs data fusion according to these weights. This automatically learns the contribution weights of different sensor data to power consumption prediction, improving the accuracy of power consumption prediction. Experiments show that the prediction error of the multimodal fusion model is reduced by 62% compared to the traditional single-parameter model and by 43% compared to the simple weighted fusion model, providing a reliable foundation for the accurate decision-making of the collaborative decision-making module 130. Through the closed-loop feedback mechanism of the dynamic adjustment module 140, the deviation between actual power consumption and predicted power consumption is continuously fed back to the power consumption prediction module 120, updating the model parameters of the time-series prediction model online, further improving prediction accuracy and decision-making accuracy.

[0145] Fifth, the technical solution provided in this application provides intelligent and adaptive full lifecycle power consumption management. This application uses the feedback monitoring mechanism of the dynamic adjustment module 140 to obtain the actual power consumption corresponding to the adjusted operating parameters. The collaborative decision-making module 130 determines the prediction deviation value based on the actual power consumption and the predicted power consumption requirement. The power consumption prediction module 120 updates the model parameters of the time-series prediction model and / or updates the policy network parameters of the reinforcement learning optimizer based on the prediction deviation value, achieving continuous learning and model optimization. This allows the system to adapt to long-term factors such as device aging, battery degradation, and environmental changes, maintaining optimal power consumption performance throughout its lifecycle. Static power consumption management schemes in related technologies typically perform a one-time optimization at the time of device delivery. As usage time increases, the optimization effect gradually decreases or even fails due to factors such as battery capacity reduction and component aging. The adaptive learning mechanism provided in this application can periodically update the prediction model to compensate for these long-term changes. Experiments show that after two years of device use, the IoT communication module using the technical solution provided in this application can still maintain more than 85% of the initial optimization effect, while traditional solutions typically drop to below 50%.

[0146] Sixth, the heterogeneous computing architecture and intelligent power management provided in this application embodiment achieve refined power consumption control. This application embodiment uses a neural processing module 150 (NPU) to run a power prediction module 120 and / or a collaborative decision-making module 130, and a graphics processing module 160 (GPU) to process image data collected by a multimodal perception module 110, thus realizing a heterogeneous computing architecture. Compared to traditional single-core CPU processing, this improves the energy efficiency of computing tasks by 2 to 4 times. Simultaneously, the intelligent power management module 170 monitors the power consumption of the hardware modules corresponding to the multimodal perception module 110, power prediction module 120, collaborative decision-making module 130, and dynamic adjustment module 140 in real time, and dynamically adjusts the power allocation strategy of each hardware module according to the adjustment instructions of the dynamic adjustment module 140, achieving independent power supply and refined power consumption control for each hardware module, further reducing the overall system power consumption.

[0147] Seventh, the technical solution provided in this application embodiment achieves rapid deployment and flexible configuration. Through modular system design and standardized interface communication, this application embodiment provides flexible configuration interfaces and development tools, enabling customers to customize power optimization strategies according to their own application characteristics. The sampling frequency of the multimodal sensing module 110, the prediction model parameters of the power prediction module 120, the decision-making strategy of the collaborative decision-making module 130, and the operating parameters of the dynamic adjustment module 140 can all be flexibly configured according to application requirements. Based on a general IoT communication module hardware platform, the technical solution provided in this application embodiment can be quickly integrated into various IoT devices, shortening the product development cycle. The system also provides rich API interfaces and development examples to support secondary development by customers, further expanding power optimization functions. This flexibility enables the technical solution to adapt to a wide range of IoT application scenarios, from low-speed sensor nodes to high-speed video transmission devices, providing customized power optimization solutions.

[0148] This application discloses a power consumption optimization method, system, and storage medium. The method includes: collecting environmental parameters of the environment in which the IoT communication module operates and the operating status parameters of the IoT communication module; predicting the power consumption requirements of the IoT communication module under different data processing methods based on the environmental parameters and operating status parameters; determining the task allocation ratio of the IoT communication module under different data processing methods based on the power consumption requirements and the current network quality; and adjusting the operating parameters of the IoT communication module based on the task allocation ratio to achieve power consumption optimization. By collecting environmental parameters and operating status parameters, the environmental conditions and operating status of the IoT communication module can be comprehensively perceived, enabling power consumption management to adapt to the differentiated needs of different application scenarios, from smart agriculture to industrial automation. By predicting the power consumption requirements under different data processing methods, the limitations of reactive power adjustment based solely on the current state are overcome. By combining the predicted power consumption requirements and the current network quality to determine the task allocation ratio, and then adjusting the operating parameters after determining the task allocation ratio, joint optimization of communication power consumption and computing power consumption is achieved, solving the problem of low overall system energy efficiency caused by separately managing communication power consumption and computing power consumption. Furthermore, through an end-to-end closed-loop process of collecting environmental and operational parameters, predicting power consumption requirements, determining task allocation ratios, and adjusting working parameters, the IoT communication module can autonomously make power consumption management decisions based on the actual operating environment and task requirements, without the need for manual intervention or pre-configuration of power consumption modes for specific scenarios, thus achieving adaptive power consumption management.

[0149] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be performed by instructions, or by instructions controlling related hardware. These instructions can be stored in a computer-readable storage medium and loaded and executed by a processor.

[0150] Therefore, embodiments of this application provide a computer-readable storage medium storing a computer program thereon, which is loaded by a processor to execute the steps described in the above-described method embodiments of this application. For example, the computer program loaded by the processor can execute the steps in any of the above-described power optimization methods.

[0151] For details on the implementation of each of the above operations / steps, please refer to the previous examples, which will not be repeated here.

[0152] The computer-readable storage medium may include: read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0153] Since the computer program stored in the computer-readable storage medium can execute the steps in any of the above method embodiments provided in the embodiments of this application, the beneficial effects that the methods described in any of the above method embodiments can achieve can be realized, as detailed in the preceding embodiments, and will not be repeated here.

[0154] This application also provides a computer program product or computer program that includes computer instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the electronic device to perform the methods provided in the various optional implementations of the above embodiments.

[0155] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0156] The above provides a detailed description of a power optimization method, system, and storage medium provided in the embodiments of this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this application. At the same time, those skilled in the art will recognize that there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A power consumption optimization method, characterized in that, The method includes: Collect environmental parameters of the environment in which the IoT communication module is located, as well as the operating status parameters of the IoT communication module; Based on the environmental parameters and the operating status parameters, predict the power consumption requirements of the IoT communication module under different data processing methods; Based on the power consumption requirements and the current network quality, determine the task allocation ratio of the IoT communication module under different data processing methods; Based on the task allocation ratio, the operating parameters of the IoT communication module are adjusted to optimize the power consumption of the IoT communication module.

2. The method according to claim 1, characterized in that, The data processing method includes at least one of the following: local edge computing method and cloud processing method.

3. The method according to claim 2, characterized in that, The step of determining the task allocation ratio of the IoT communication module under different data processing methods based on the power consumption requirements and the current network quality includes: The first energy consumption prediction value under the cloud processing method is corrected based on the current network quality. The task allocation ratio is determined based on the corrected first energy consumption prediction value and the second energy consumption prediction value under the local edge computing method. The current network quality is determined based on at least one of the following indicators: signal strength index, signal-to-noise ratio index, link stability index, and transmission delay index.

4. The method according to claim 1, characterized in that, The method of predicting the power consumption requirements of the IoT communication module under different data processing methods based on the environmental parameters and the operating status parameters includes: Preprocessing is performed on the environmental parameters and the operating status parameters; When the change in the environmental parameters exceeds a preset threshold, the power consumption requirements of the IoT communication module under different data processing methods are predicted based on the preprocessed environmental parameters and the preprocessed operating status parameters.

5. The method according to claim 1, characterized in that, The method of predicting the power consumption requirements of the IoT communication module under different data processing methods based on the environmental parameters and the operating status parameters includes: The weight values ​​of the environmental parameters and the operating state parameters are determined by a time-series prediction model based on deep learning. Based on the weight values, the environmental parameters and the operating status parameters are fused to determine the power consumption requirements of the IoT communication module under different data processing methods within a future predicted time.

6. The method according to claim 5, characterized in that, The step of fusing the environmental parameters and the operating status parameters according to the weighted values ​​to determine the power consumption requirements of the IoT communication module under different data processing methods within a future predicted time period includes: According to the weight values, the environmental parameters and the operating status parameters are fused to obtain fused parameters; Based on historical power consumption data and the fusion parameters, the power consumption requirements of the IoT communication module under different data processing methods within a future predicted time period are determined; wherein, the historical power consumption data is the actual power consumption of the IoT communication module within a preset time period in the past.

7. The method according to claim 1, characterized in that, The adjustment of the operating parameters of the IoT communication module based on the task allocation ratio includes: The target parameters of the IoT communication module are determined by a reinforcement learning optimizer. Based on a preset step size, the working parameters are gradually adjusted within multiple preset scheduling cycles until the target parameters are reached.

8. The method according to claim 1, characterized in that, After adjusting the operating parameters of the IoT communication module, the method further includes: Obtain the actual power consumption corresponding to the adjusted operating parameters; Based on the actual power consumption and the predicted power consumption requirement, determine the prediction deviation value; Based on the predicted deviation value, update the model parameters of the time-series prediction model used to predict the power consumption requirement, and / or update the policy network parameters of the reinforcement learning optimizer used to adjust the operating parameters.

9. A power consumption optimization system, characterized in that, The system includes: The multimodal sensing module is configured to collect environmental parameters of the environment in which the IoT communication module is located and the operating status parameters of the IoT communication module. The power consumption prediction module is configured to predict the power consumption requirements of the IoT communication module under different data processing methods based on the environmental parameters and the operating status parameters. The collaborative decision-making module is configured to determine the task allocation ratio of the IoT communication module under different data processing methods based on the power consumption requirements and the current network quality. The dynamic adjustment module is configured to adjust the operating parameters of the IoT communication module based on the task allocation ratio, so as to optimize the power consumption of the IoT communication module.

10. The system according to claim 9, characterized in that, The system further includes a neural processing module and a graphics processing module; wherein the power consumption prediction module and / or the collaborative decision-making module run on the neural processing module; and the image data acquired by the multimodal perception module is processed by the graphics processing module.

11. The system according to claim 9, characterized in that, The system also includes an intelligent power management module; The intelligent power management module is configured to: monitor the power consumption of the hardware modules corresponding to the multimodal sensing module, the power consumption prediction module, the collaborative decision-making module, and the dynamic adjustment module; and adjust the power allocation strategy of each hardware module according to the adjustment instructions of the dynamic adjustment module.

12. A computer-readable storage medium, characterized in that, It stores a computer program or instructions that, when executed by a processor, implement the steps in the power optimization method as described in any one of claims 1 to 8.

Images