A Mobile Phone Extreme Charging Power Adjustment Method Based on Multimodal Sensing and Bidirectional Protocol

Abstract

This invention provides a method for adjusting the power of mobile phone fast charging based on multimodal sensing and a bidirectional protocol, belonging to the field of power adjustment technology. It includes: establishing a bidirectional communication connection between the charging device and the mobile phone; collecting multidimensional state data during the charging process using a distributed sensing layout based on the bidirectional communication connection, preprocessing and fusing features through an NPU-accelerated dynamic weight fusion model to obtain multidimensional features; constructing an adaptive learning model, fusing multidimensional features and interactive data, and using GPU acceleration for precise power adjustment to dynamically adjust the baseline value of the fast charging power; establishing a power adjustment model, combining multidimensional features and interactive data for precise power adjustment, and outputting precise power adjustment commands; monitoring the effect in real time based on the command adjustment results, and dynamically optimizing the power adjustment strategy. Through cross-device collaboration based on a bidirectional communication protocol, it solves the power mismatch and potential safety hazards caused by the fragmentation of fast charging protocols between devices from different manufacturers.

Description

Technical Field

[0001] This invention relates to the field of power regulation technology, and in particular to a method for regulating the power of mobile phone fast charging based on multimodal sensing and bidirectional protocol. Background Technology

[0002] As the demands for both mobile phone battery capacity and high-power charging increase simultaneously, traditional high-power charging control methods have significant technical limitations: they rely solely on voltage and current parameters of the charging circuit to adjust power, ignoring environmental factors, device status, and grid fluctuations, which can easily lead to falsely advertised power output. In traditional high-power charging systems, the phone and charger only transmit power through the charging cable, lacking two-way data interaction—the charger cannot obtain the phone's battery status in real time, and the phone cannot perceive the charger's grid compatibility, resulting in lagging power adjustment. In extreme cases, this may trigger over-temperature protection shutdown, affecting the charging experience. Most high-power charging solutions use fixed power steps and do not dynamically adjust according to real-time scenarios. Traditional fast charging solutions often rely on preset, relatively fixed power curves, which cannot sensitively respond to real-time changes in internal and external conditions during charging. Summary of the Invention

[0003] This invention provides a method for adjusting the charging power of mobile phones based on multimodal sensing and bidirectional protocols, in order to overcome the deficiencies in the prior art.

[0004] On one hand, the present invention provides a method for adjusting the charging power of a mobile phone based on multimodal sensing and bidirectional protocols, comprising:

[0005] S1: Establish a two-way communication connection between the charging device and the mobile phone, automatically identify and match the fast charging protocol through the multi-protocol compatibility module, and synchronize the device parameters and initial status data of both parties through encryption and authentication.

[0006] S2: Based on the bidirectional communication connection, multi-dimensional state data during the charging process is collected using a distributed sensor layout. After filtering and anti-interference processing, the data is preprocessed and fused using an NPU-accelerated dynamic weight fusion model to obtain multi-dimensional features.

[0007] S3: Construct an adaptive learning model, integrate multi-dimensional features and interactive data, accelerate precise power adjustment through GPU, and dynamically adjust the extreme charging power benchmark value.

[0008] S4: Establish a power regulation model, combine multi-dimensional features and interactive data to perform precise power regulation, and output precise power regulation commands.

[0009] S5: Monitors the effect in real time based on the adjustment results of the command, establishes a closed-loop feedback mechanism through a two-way protocol, and dynamically optimizes the power adjustment strategy.

[0010] According to the mobile phone fast charging power adjustment method based on multimodal sensing and bidirectional protocol provided by the present invention, the specific steps for establishing a bidirectional communication connection between the charging device and the mobile phone in step S1 are as follows:

[0011] S11: After the charging device is physically connected to the mobile phone, it automatically scans and identifies the fast charging protocol types supported by the charger according to the multi-protocol compatibility module.

[0012] S12: Based on the identified protocol type, the mobile device initiates a protocol handshake request with the charging device, and completes device authentication and security key negotiation through an encryption authentication algorithm.

[0013] S13: After completing the encryption authentication, both parties synchronize their respective device parameters and initial status data through a mutually agreed protocol format.

[0014] S14: Verify the synchronized device parameters and initial state data. After confirming the data integrity and compatibility, establish a stable two-way communication link to lay the foundation for data interaction during the subsequent charging process.

[0015] According to the mobile phone charging power adjustment method based on multimodal sensing and bidirectional protocol provided by the present invention, in step S2, the specific steps for preprocessing multidimensional state data and connecting it with the vehicle are as follows:

[0016] S21: Collect multi-dimensional raw data during the charging process based on the established two-way communication connection.

[0017] S22: The multidimensional raw data is subjected to anti-interference processing through a preset filtering network to filter out noise caused by electromagnetic interference and obtain purified raw data.

[0018] S23: Perform time synchronization and format standardization calibration on the original purification data to obtain standard original data.

[0019] S24: Input the standard raw data into the NPU-accelerated dynamic weight fusion model. The model dynamically allocates the weights of each dimension of data according to the charging stage, completes the multi-level fusion of the feature layer and the decision layer, and outputs a high-dimensional feature set after fusion processing.

[0020] According to the mobile phone extreme charging power adjustment method based on multimodal sensing and bidirectional protocol provided by the present invention, the specific steps for dynamically adjusting the extreme charging power reference value in step S3 are as follows:

[0021] S31: Initialize the adaptive learning model based on the fusion features and real-time interactive data.

[0022] S32: Utilize the dynamic weight mechanism of the hidden layer of the adaptive learning model to perform feature recombination and correlation analysis on the input multidimensional features and interactive data to obtain high-dimensional feature vectors.

[0023] S33: Input high-dimensional feature vectors into the GPU-accelerated inference engine, quickly solve the power adjustment parameters through parallel computing, and generate a preliminary adjustment scheme by combining the extreme charging power benchmark value.

[0024] S34: Verify the safety boundaries of the preliminary adjustment plan and revise it based on historical adjustment effect data.

[0025] S35: Outputs the final precise power adjustment command, dynamically updates the extreme charging power reference value, and feeds back the adjustment result to the two-way communication link.

[0026] According to the mobile phone fast charging power adjustment method based on multimodal sensing and bidirectional protocol provided by the present invention, the specific steps for initializing the adaptive learning model in step S31 are as follows:

[0027] S311: Receive the fused features and real-time interaction data output from the preceding steps as the initial input dataset for the adaptive learning model.

[0028] S312: Perform format conversion and dimension alignment on the initial input dataset, convert multidimensional features into tensor formats that the model can recognize, and map bidirectional protocol interaction data into standard feature vectors.

[0029] S313: Initialize the weight parameters and bias terms of the input layer of the adaptive learning model.

[0030] S314: Based on the multi-channel processing mechanism of the input layer of the adaptive learning model, an independent processing channel is allocated for multi-dimensional features and bidirectional protocol interaction data to perform preliminary data parsing and feature extraction, and output the intermediate feature vector after processing by the input layer.

[0031] According to the mobile phone fast charging power adjustment method based on multimodal sensing and bidirectional protocol provided by the present invention, the specific steps for generating the preliminary adjustment scheme in step S33 are as follows:

[0032] S331: Converts the recombined feature vector into a GPU parallel data format, loads it into the GPU-accelerated inference engine, and calls the preset parallel computing kernel to output a parallel feature vector that can be directly computed by the GPU.

[0033] S332: The inference engine performs high-dimensional space mapping and parameter solving on parallel feature vectors through multi-threaded parallel computation to obtain power adjustment parameters.

[0034] S333: The extreme charging power reference value is used as a constraint condition and fused with the power adjustment parameters to obtain the preliminary adjustment scheme of the current power reference framework.

[0035] According to the mobile phone fast charging power adjustment method based on multimodal sensing and bidirectional protocol provided by the present invention, the specific steps for precise power adjustment in step S4 are as follows:

[0036] S41: Establish a power regulation model and initialize the core parameters of the power regulation model based on the high-dimensional feature vector.

[0037] S42: Bind the initial power regulation model with multi-dimensional features and bidirectional protocol interaction data to construct a model input and output mapping framework.

[0038] S43: The feature fusion module of the power regulation model performs hierarchical fusion of the input multidimensional features and interactive data, and mines the implicit correlation between features through nonlinear transformation to generate a fused feature matrix.

[0039] S44: Calculate the preliminary power regulation parameters based on the inference of the model driven by the fusion feature matrix.

[0040] S45: Verify the accuracy of the initial adjustment parameters, compare the deviation between the actual charging effect and the model prediction result, correct the model parameters through the backpropagation mechanism, and output a precise power adjustment command.

[0041] The mobile phone fast charging power adjustment method based on multimodal sensing and bidirectional protocol provided by the present invention includes the following specific steps in step S43 for generating the feature fusion matrix:

[0042] S431: Extract the multidimensional features and bidirectional protocol interaction data to be fused based on the model input-output mapping relationship framework.

[0043] S432: Perform time-axis alignment processing on the extracted multidimensional features and interaction data.

[0044] S433: Based on the correlation strength of features in the mapping relationship framework, group the highly correlated multidimensional features with the interaction data to obtain feature combination units.

[0045] S434: Apply a nonlinear transformation algorithm to each feature combination unit, and mine the implicit correlation between features within the combination unit through matrix operations of the multilayer perceptron to obtain the associated feature vector.

[0046] S435: Integrate the associated feature vectors according to the logical structure of the mapping relationship framework to obtain the fused feature matrix.

[0047] According to the mobile phone fast charging power adjustment method based on multimodal sensing and bidirectional protocol provided by the present invention, the specific steps for calculating the preliminary power adjustment parameters in step S44 are as follows:

[0048] S441: Perform data transformation on the fused feature matrix.

[0049] S442: A GPU-accelerated model inference engine that takes the transformed fused feature matrix as input to the engine and performs high-dimensional spatial mapping on the matrix through multi-threaded parallel computation to generate intermediate calculation results that reflect the relationship between features and power regulation.

[0050] S443: The intermediate calculation results generated by the inference engine are fused with the extracted power regulation rules, and the range of power regulation parameters that satisfy the rule constraints is solved by a linear programming algorithm.

[0051] S444: Select the values ​​that best match the balance between efficiency and safety within the range of power regulation parameters to obtain the preliminary power regulation parameters.

[0052] According to the mobile phone fast charging power adjustment method based on multimodal sensing and bidirectional protocol provided by the present invention, the specific steps of dynamically optimizing the power adjustment strategy in step S5 are as follows:

[0053] S51: Adjusts the phone's charging power according to precise power adjustment commands and monitors power data in real time.

[0054] S52: Compare and analyze the monitored power data with the expected target value in the regulation command, and calculate the power deviation and temperature control deviation.

[0055] S53: Based on the power deviation and temperature control deviation analysis results, obtain the current hardware load and safety margin data of both devices.

[0056] S54: Dynamically adjust the core parameters of the power regulation strategy based on hardware load, safety margin data, and cross-device data fed back by the bidirectional protocol to obtain an optimized power regulation strategy.

[0057] The mobile phone fast charging power adjustment method based on multimodal sensing and bidirectional protocol provided by this invention achieves the following beneficial effects:

[0058] This invention achieves true dynamic and precise power regulation, finding the optimal balance between extreme charging speed and long-term battery health. This method utilizes a distributed sensing layout to collect far more data than conventional solutions, including battery core temperature, interface temperature, ambient temperature, real-time voltage and current, cable impedance, battery health, and even user behavior such as screen status and processor load. This multi-dimensional state data, after filtering and anti-interference processing, is integrated by an NPU-accelerated dynamic weighted fusion model. This means the system doesn't treat all data equally but intelligently determines which factor is most critical at a specific charging stage. When the battery level is high or the battery temperature begins to rise, the model's weights dynamically tilt towards temperature characteristics, proactively reducing power in advance to avoid overheating, thus protecting the battery chemistry and effectively delaying battery capacity degradation. This real-time, proactive adjustment avoids the lag of traditional solutions that passively reduce speed only after overheating, achieving a balance between speed and safety.

[0059] This invention completely solves the power mismatch and potential safety hazards caused by the fragmentation of fast charging protocols between different manufacturers' devices through cross-device collaboration based on a bidirectional communication protocol. This deep dialogue mechanism brings multiple benefits: the multi-protocol compatible module can automatically scan and match the optimal fast charging protocol, so users don't need to care whether the charger supports PD, QC, or other proprietary protocols. Through encrypted handshakes and digital certificate verification, the risk of combustion and explosion caused by inferior or counterfeit chargers fraudulently obtaining high power output is eliminated from the source. The system can sense the loss of components such as data line impedance in real time and fine-tune the output voltage accordingly to compensate for line loss, ensuring that electrical energy is efficiently and completely delivered to the phone battery, rather than being wasted on the charging cable.

[0060] This invention, through an adaptive learning model and closed-loop feedback mechanism, enables the system to continuously optimize. Power regulation is driven by an inference engine running efficiently on a GPU. This model can perform deep correlation analysis on high-dimensional features after dynamic weight reorganization, uncovering potential dangerous patterns such as a slight voltage increase accompanied by a rapid temperature rise, and intervening in advance. The system monitors the results of each power adjustment, compares the expected target with the actual effect, and automatically corrects the model parameters through a backpropagation mechanism. The system increasingly understands the characteristics of this specific mobile phone battery, thereby formulating a more personalized charging strategy that fits individual differences, achieving adaptive optimization that becomes smarter with use.

[0061] In summary, this invention, through the deep integration of multimodal sensing and bidirectional protocols, constructs a next-generation intelligent charging system that offers more comprehensive perception, more intelligent decision-making, closer collaboration, and faster response. It not only achieves precise dynamic adjustment of charging power technically, but also fundamentally improves the user experience, eliminating the need for users to weigh charging speed, device safety, and lifespan, truly realizing a balance of efficiency, safety, and convenience. Attached Figure Description

[0062] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0063] Figure 1 This is a flowchart illustrating the mobile phone fast charging power adjustment method based on multimodal sensing and bidirectional protocol provided in an embodiment of the present invention.

[0064] Figure 2 This is a schematic diagram of the process for dynamically adjusting the extreme charging power reference value provided in an embodiment of the present invention;

[0065] Figure 3 This is a schematic diagram of the process for precise power adjustment by combining multi-dimensional features and interactive data, provided by an embodiment of the present invention. Detailed Implementation

[0066] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0067] The following is combined Figures 1-3 This invention describes a mobile phone fast charging power adjustment method based on multimodal sensing and bidirectional protocol.

[0068] like Figures 1-3 As shown in the embodiment of the present invention, the mobile phone fast charging power adjustment method based on multimodal sensing and bidirectional protocol includes:

[0069] S1: Establish a two-way communication connection between the charging device and the mobile phone, automatically identify and match the fast charging protocol through the multi-protocol compatibility module, and synchronize the device parameters and initial status data of both parties through encryption and authentication.

[0070] The phone and charger establish a real-time communication link through the Extreme Charge protocol, which supports bidirectional communication. The communication frequency is set to 500ms / time, enabling bidirectional data interaction between the phone and charger, rather than traditional one-way command transmission.

[0071] The bidirectional communication protocol is an extension of the USB-PD (USB Power Transfer) protocol. It adds a scene feature data field and a power adjustment command field to the original protocol frame structure. The scene feature data field is used to transmit the key feature quantification values ​​output by the charging scene recognition module, while the power adjustment command field is used to transmit the power adjustment requirements generated by the mobile phone or the power response information from the charger.

[0072] The interaction process of the two-way communication protocol is as follows: Every 100ms, the mobile phone sends a protocol frame containing current scene feature data and the target power requirement to the charger. Upon receiving the protocol frame, the charger extracts the scene feature data using its built-in protocol parsing unit and performs a power compatibility assessment based on its own output capabilities. If the charger can meet the mobile phone's target power requirement, it returns a power confirmation command within 50ms. If the charger cannot meet the target power requirement due to its own hardware limitations or safety protection mechanisms, it returns the maximum output power and the reason for the limitation, and waits for a second negotiation from the mobile phone. This two-way real-time interaction ensures that the mobile phone and charger can dynamically match the optimal charging power based on the current charging scenario, avoiding power mismatch problems caused by one-way command transmission.

[0073] S11: After the charging device is physically connected to the mobile phone, the multi-protocol compatibility module automatically scans and identifies the types of fast charging protocols supported by the charger. Once the charging device and mobile phone are connected via a physical interface, the mobile phone immediately activates its built-in multi-protocol compatibility module. This module sends a series of standardized query signals to the charging device through a preset protocol detection mechanism, matching the characteristic responses of mainstream fast charging protocols one by one. By analyzing the signal characteristics fed back by the device, the module can quickly identify all fast charging protocol types supported by the charger, solving the problem of protocol fragmentation between different manufacturers. For example, if it detects that the charger simultaneously supports USB-PD and proprietary SCP (Secure Copy) protocols, the module will prioritize marking the more compatible protocol as the initial negotiation target.

[0074] S12: Based on the identified protocol type, the mobile device and the charging device initiate a protocol handshake request, completing device authentication and security key negotiation through an encryption authentication algorithm. Based on the identified protocol type, the mobile device initiates a handshake request to the charging device according to the corresponding protocol's security specifications. This process uses an asymmetric encryption algorithm to generate a temporary session key and verifies the legitimacy of both devices through digital certificates—the mobile phone verifies whether the charger is a certified and legitimate product, while the charger confirms whether the mobile phone has the corresponding power receiving capability. This step effectively prevents malicious devices from fraudulently obtaining high power output by forging protocols, blocking the security risks that counterfeit chargers may pose at the source. Only after both parties have successfully authenticated their identities and reached a consensus on the key can the subsequent data interaction phase begin.

[0075] S13: After encryption authentication is completed, both parties synchronize their respective device parameters and initial status data using a mutually agreed-upon protocol format. Following encryption authentication, both parties begin bidirectional synchronization of core information based on the agreed-upon protocol format. The charging device sends its own hardware parameters, such as maximum output voltage / current, supported power adjustment levels, and over-temperature protection thresholds. The mobile phone provides battery status data, such as current battery percentage, real-time cell temperature, cycle count, and maximum tolerable input power. This data is transmitted using encryption to ensure it cannot be intercepted or tampered with by third parties. If the phone's battery health is below 80%, it will proactively notify the charger to reduce the maximum supported power to avoid overcharging and discharging damage to the battery.

[0076] S14: Verify the synchronized device parameters and initial state data. After confirming data integrity and compatibility, establish a stable two-way communication link to lay the foundation for subsequent data interaction during charging. After the mobile phone receives all synchronized data, it checks data integrity and verifies parameter compatibility according to the verification mechanism. If data is missing or parameters are mismatched, the system will trigger a negotiation retry mechanism to resynchronize and adjust parameters. If the verification passes, a stable two-way communication link is formally established, and the communication frequency, data format, and abnormal interruption handling mechanism for subsequent real-time interaction are agreed upon, providing a reliable communication foundation for dynamic power adjustment during charging.

[0077] S2: Based on the bidirectional communication connection, multi-dimensional state data during the charging process is collected using a distributed sensor layout. After filtering and anti-interference processing, preprocessing and feature fusion are completed through a dynamic weight fusion model accelerated by NPU.

[0078] S21: Based on the established two-way communication connection, real-time status data during the charging process is synchronously collected from multiple types of sensors such as voltage, current, and temperature in the distributed sensing layout. Real-time status data includes: environmental and equipment status data, charging accessory data, battery health data, and user behavior data.

[0079] Environmental and equipment status data: Battery core temperature T and charging interface temperature T are collected via temperature sensors. i Ambient temperature T e The real-time voltage V across the battery is collected using a voltage sensor. b Charger output voltage V c The real-time current I in the charging circuit is collected by a current sensor. c .

[0080] Charging accessory data: Charger model and rated power (P) are read via the protocol chip of the USB and Type-C interfaces. max Supported charging protocols are checked, and it's also detected to determine if the accessories meet the current high-power charging requirements. The data cable's resistance is also considered. The calculation formula is expressed as follows:

[0081] Battery health data: Read the battery cycle count (Nc), current capacity degradation rate, and current remaining battery charge (SOC) stored in the battery management system (BMS), establish a battery health status assessment model, and determine the upper limit of the battery's tolerance to high power using the following formula:

[0082]

[0083] Wherein, Pd is the maximum power that the battery can withstand.

[0084] User behavior data: Collect user operation behavior during the charging process through the system background, including screen on / off status, CPU utilization Uc, GPU load rate Ug, and calculate the user behavior's demand offset ΔP for charging power.

[0085] All collected data is transmitted to the data preprocessing module via the phone's internal bus for outlier filtering and data standardization, resulting in a standardized multimodal dataset D={T, T...}. i , Te, Vb, Ic, Rc, Nce, Cd, SOC, S, Uc, Ug}.

[0086] S22: The multidimensional raw data is subjected to anti-interference processing through a preset filtering network to filter out noise caused by electromagnetic interference and obtain purified raw data.

[0087] S23: Perform time synchronization and format standardization calibration on the filtered basic data, detect missing values ​​on the denoised data, use linear interpolation to fill in discrete missing data, and use trend interpolation based on battery charging and discharging characteristics to fill in continuous missing data, so as to obtain standard original data and ensure the consistency of data from different sensors in terms of time dimension and data format.

[0088] S24: Input standard raw data into an NPU (Neural Processing Unit) accelerated dynamic weight fusion model. The model dynamically allocates the weights of each dimension of data according to the charging stage, completes the multi-level fusion of the feature layer and the decision layer, and outputs a high-dimensional feature set after fusion processing.

[0089] S3: Construct an adaptive learning model, integrate multi-dimensional features and interactive data, and accelerate precise power adjustment through GPU (graphics processing unit) to dynamically adjust the extreme charging power benchmark value.

[0090] S31: The adaptive learning model is initialized based on the fused features and real-time interaction data. The model input layer receives multi-dimensional features and bidirectional protocol interaction data. The adaptive learning model is then established based on the fused features and real-time interaction data. The model input layer employs a multi-channel design, specifically adapted for different types of data features. For example, continuous value processing channels are configured for physical quantity features such as temperature and current, while classification feature channels are configured for discrete information such as protocol interaction status. This design ensures accurate parsing of multi-dimensional data at the input stage, laying the foundation for subsequent processing. Simultaneously, the model loads initial parameters formed during historical training, ensuring the continuity of the adjustment process.

[0091] S311: Based on the fused features and real-time interactive data, serve as the initial input dataset for the adaptive learning model.

[0092] S312: Perform format conversion and dimension alignment on the input dataset, convert multidimensional features into tensor formats that the model can recognize, and map bidirectional protocol interaction data into standardized feature vectors to ensure uniform adaptation of different types of data at the input layer.

[0093] S313: Based on the transformed standardized data, initialize the weight parameters and bias terms of the model input layer. The initial values ​​of the weight parameters refer to the feature importance distribution formed by historical training, and the bias terms are calibrated according to the initial state of the current charging scenario, specifically the initial battery temperature and charge.

[0094] S314: Based on the multi-channel processing mechanism of the model input layer, independent processing channels are allocated for multi-dimensional features and bidirectional protocol interaction data. Specifically, the continuous feature channel processes physical quantities such as temperature and current, while the discrete feature channel processes protocol status identifiers. This completes the preliminary parsing and feature extraction of the data, outputs the intermediate feature vector processed by the input layer, and passes it to the model hidden layer to provide basic data support for subsequent dynamic weight allocation and feature recombination.

[0095] S32: Utilizing the dynamic weighting mechanism of the model's hidden layer, feature recombination and correlation analysis are performed on the input multidimensional features and interactive data, focusing on capturing the influence weights of different features at each stage of charging. After the model enters the hidden layer processing stage, the core lies in achieving intelligent feature recombination and correlation analysis through the dynamic weighting mechanism. This mechanism automatically adjusts the influence weights of each feature according to the real-time charging scenario: for example, in low-temperature environments, the system increases the weight of the current stability feature to avoid current fluctuations caused by increased internal resistance at low temperatures. When the temperature approaches the safety threshold, the weight of the temperature feature is significantly strengthened to prioritize battery safety. Simultaneously, the hidden layer uses an attention mechanism to uncover potential correlations between features, specifically identifying dangerous combination patterns such as voltage surges accompanied by rapid temperature increases, providing more targeted feature support for subsequent adjustments.

[0096] S321: Based on the dynamic weighting mechanism of the hidden layer, activate the corresponding weight allocation strategy based on the current charging stage. Specifically, for the low-battery fast charging stage and the high-battery trickle charging stage, assign initial dynamic weights to different intermediate feature vectors. These vectors already contain multi-dimensional features that have been preliminarily analyzed.

[0097] S322: Deep correlation analysis of feature vectors is performed through attention mechanism to calculate the synergistic influence coefficient between features, specifically the correlation score between temperature rise and current fluctuation, and the feature weights are dynamically adjusted according to the correlation strength.

[0098] S323: Based on the adjusted feature weights, multidimensional features are reorganized, and high-weight features are combined with strongly correlated features to form a targeted feature subset. The high-dimensional feature vector after reorganization and weight optimization is output, which clearly reflects the influence priority of different features in the current charging stage.

[0099] S33: The recombined feature vector is input into the GPU-accelerated inference engine, which quickly solves for the power adjustment parameters through parallel computing, and generates a preliminary adjustment scheme based on the extreme charging power benchmark value. The recombined high-dimensional feature vector is input into the GPU-accelerated inference engine, which leverages its powerful parallel computing capabilities to quickly complete complex mathematical calculations. The inference process combines the current extreme charging power benchmark value with a preset adjustment algorithm to solve for the optimal power parameters and generate a preliminary adjustment scheme. The scheme specifically includes detailed parameters such as voltage fine-tuning amplitude, current limit threshold, and power maintenance duration, ensuring that the adjustment can both approach the maximum charging efficiency and adapt to real-time state changes.

[0100] S331: Receives the recombined feature vector output from the model's hidden layer. This vector already contains key features after dynamic weight allocation and correlation analysis, specifically the influence weights of temperature and current at different charging stages and the correlation patterns between features. It converts the recombined feature vector into a parallel data format that can be efficiently processed by the GPU and loads it into the GPU-accelerated inference engine, while simultaneously invoking a pre-defined parallel computing kernel.

[0101] S332: The inference engine uses multi-threaded parallel computing to perform high-dimensional space mapping and parameter solving on feature vectors, quickly generating basic parameters related to power regulation, specifically voltage regulation coefficient, current limiting ratio, and power step value.

[0102] S334: Introduce the extreme charging power reference value determined in the previous steps, use it as a constraint condition and integrate it with the solved basic parameters to form a preliminary set of adjustment parameters that conforms to the current power reference framework.

[0103] S335: Generates a preliminary adjustment scheme based on the parameter set, which includes specific execution instructions. It clarifies details such as voltage adjustment range, current limit threshold, and adjustment execution timing, providing the original scheme basis for subsequent safety verification.

[0104] S34: Perform safety boundary verification on the preliminary adjustment plan and revise it based on historical adjustment effect data. After the preliminary adjustment plan is generated, the system immediately performs a dual verification mechanism: on the one hand, it performs hardware safety boundary verification, comparing the voltage, current, and other parameters in the plan with the device's preset protection thresholds to ensure that all parameters are within the safe range. On the other hand, it calls the historical adjustment effect database, finds historical cases in similar scenarios through similarity matching, analyzes possible deviations in the plan, and corrects the plan accordingly. In a specific embodiment, if the preliminary plan suggests increasing the current to 6.2A, but historical data shows that this current is prone to overheating under the current battery health state, it will be automatically corrected to 5.8A, finding a balance between efficiency and safety.

[0105] S35: Outputs the final precise power adjustment command, dynamically updates the Extreme Charge power reference value, and simultaneously feeds the adjustment result back to the two-way communication link, providing a basis for the next round of status interaction. The final determined precise power adjustment command is converted into a hardware-executable signal, and the Extreme Charge power reference value is dynamically updated. The adjustment result is synchronously fed back to the two-way communication link, allowing the charging device to know the changes in the power demand on the mobile phone in real time, and providing the latest adjustment basis for the next round of status interaction.

[0106] S4: Establish a power regulation model and combine multi-dimensional features and interactive data to achieve precise power regulation.

[0107] S41: Based on the high-dimensional feature vector after feature recombination and correlation analysis, initialize the core parameters of the power regulation model and determine the model input dimension and output target.

[0108] S42: Bind the initialized model with the multi-dimensional features and bidirectional protocol interaction data output from the previous steps to construct a model input-output mapping framework.

[0109] The system receives the initialized power regulation model and simultaneously aggregates multi-dimensional feature data and bidirectional protocol interaction data to form a dataset to be bound.

[0110] The dataset to be bound is classified and format adapted: continuous data in multidimensional features is converted into floating-point tensors that the model can recognize, and discrete state in bidirectional protocol interaction data is encoded into one-hot vectors to ensure that the data format is completely matched with the model input dimension.

[0111] Based on the multi-channel design of the model input layer, the adapted continuous feature data is assigned to the numerical feature channel, and the discrete interactive data is assigned to the state feature channel, thus completing the physical binding between the data and the model input layer and establishing the correspondence between the original data and the input layer nodes.

[0112] Based on the model's output target, we define the association rules between input features and output targets: for example, the combination of input features with temperature > 40℃ and current > 6A corresponds to an output target with a power reduction of 5W, and the protocol negotiation supports 100W and battery health > 90% corresponds to an output target with power approaching the 100W upper limit, thus forming a preliminary mapping logic.

[0113] The effectiveness of association rules is verified by using historical adjustment data: historical data from similar scenarios are selected as input to the model, and the output target is checked to see if it is consistent with the actual adjustment needs. If the deviation exceeds the threshold, the association rules are corrected, and finally a stable input feature → output target mapping relationship framework is constructed.

[0114] S43: Utilize the feature fusion module of the power regulation model to perform hierarchical fusion of the input multidimensional features and interactive data, and mine the implicit correlations between features through nonlinear transformation to generate a fused feature matrix.

[0115] The model input-output mapping relationship framework is received from the previous steps. Based on this framework, the multidimensional features to be fused and the bidirectional protocol interaction data are extracted, and the corresponding position of each data in the mapping relationship is clarified.

[0116] Based on the first-layer fusion mechanism of the feature fusion module, the extracted multidimensional features and interactive data are aligned with the time axis to unify data with different sampling frequencies to the same time granularity, ensuring the consistency of data in the time dimension.

[0117] Based on time alignment, a second layer of feature combination is performed. According to the correlation strength of features in the mapping relationship framework, highly correlated multidimensional features and interactive data are grouped to form several feature combination units.

[0118] A nonlinear transformation algorithm is applied to each feature combination unit, and the implicit correlation between features within the combination unit is mined through matrix operations of the multilayer perceptron, generating an intermediate feature vector that reflects the correlation relationship.

[0119] All intermediate feature vectors are integrated according to the logical structure of the mapping relationship framework to form a fused feature matrix that includes the time dimension and the feature association dimension, providing structured input for the model's subsequent inference calculation.

[0120] S44: Using the fusion feature matrix to drive the inference calculation of the model, combined with the preset power regulation rules, the preliminary power regulation parameters are solved.

[0121] S441: Convert the fused feature matrix according to the input format requirements of the model inference engine to ensure that each feature dimension in the matrix is ​​accurately matched with the computational unit of the inference engine.

[0122] S442: Based on the GPU-accelerated model inference engine, the transformed fused feature matrix is ​​input into the engine. The matrix is ​​then mapped to a high-dimensional space through multi-threaded parallel computation to generate intermediate calculation results that reflect the relationship between features and power regulation.

[0123] S443: Call the preset power regulation rule library, extract core rules related to intermediate calculation results such as safety threshold constraints and efficiency optimization objectives, and use them as constraints for parameter solving. Merge the intermediate calculation results generated by the inference engine with the extracted power regulation rules, and use a linear programming algorithm to solve for the power regulation parameter range that satisfies the rule constraints.

[0124] Within the range of the calculated parameters, and considering the basic requirements of the current charging stage, the values ​​that best match the balance between efficiency and safety are selected and determined as the initial power adjustment parameters.

[0125] S45: Verify the accuracy of the initial adjustment parameters, compare the deviation between the actual charging effect and the model prediction results, correct the model parameters through the backpropagation mechanism, and finally output a precise power adjustment command to achieve dynamic power adjustment.

[0126] Based on the initial adjustment parameters, these parameters are sent to the charging control system to drive the hardware to perform the charging operation, while simultaneously collecting actual charging effect data such as power, temperature, and current during the actual charging process.

[0127] Extract the model prediction results when the initial adjustment parameters are generated, compare them item by item with the collected actual charging effect data, and calculate core deviation indicators such as power deviation and temperature deviation.

[0128] Based on the deviation index, and according to the backpropagation mechanism, the gradient change of parameters in each layer of the model is calculated with the deviation value as the guide, and the direction of parameter correction is determined.

[0129] Based on the gradient change and correction direction, the model's weight parameters and bias terms are adjusted to make the model's prediction results closer to the actual charging effect and reduce the deviation.

[0130] The corrected model is used to recalculate the current charging state and generate new power adjustment parameters, which are then output as precise power adjustment commands to achieve dynamic adaptation and adjustment during the charging process.

[0131] S5: Monitor the effect in real time based on the adjustment results, establish a closed-loop feedback mechanism through a two-way protocol, dynamically optimize the power adjustment strategy, and achieve cross-device collaborative optimization.

[0132] It receives precise power adjustment commands from the preceding steps and, based on the effect monitoring module, collects key charging data in real time after adjustment, including actual charging power, battery temperature, current and voltage change trends, and equipment operating status.

[0133] The actual monitoring data collected is compared and analyzed with the expected target values ​​in the adjustment instructions. Core indicators such as power deviation and temperature control deviation are calculated to determine whether the adjustment effect meets the preset optimization target.

[0134] Based on the deviation analysis results, the system synchronously feeds back monitoring data and deviation information to the charging device and the mobile phone through the established two-way communication protocol, initiates a cross-device status interaction request, and obtains the current hardware load and safety margin data of both devices.

[0135] By combining cross-device data fed back from the two-way protocol, the core parameters of the power regulation strategy are dynamically adjusted, such as the step size of the power reference value and the response sensitivity of the temperature protection threshold, to ensure that the strategy is adapted to the current collaborative state of the devices.

[0136] The optimized power regulation strategy is transformed into new execution instructions, which are then synchronized to the charging device and the mobile phone via a two-way protocol to complete the closed-loop execution feedback. At the same time, the data of this optimization process is recorded to provide a basis for subsequent strategy iterations.

[0137] When an abnormal scenario is detected, a multi-level emergency adjustment mechanism is used to implement current limiting or power outage protection based on the severity of the abnormality, and the abnormality log is synchronized to support subsequent optimization.

[0138] In summary, this embodiment provides a method for adjusting the fast charging power of mobile phones based on multimodal sensing and bidirectional protocols. Through cross-device collaboration based on bidirectional communication protocols, it solves the power mismatch and potential safety hazards caused by the fragmentation of fast charging protocols between different manufacturers. The method of this embodiment, through the deep integration of multimodal sensing and bidirectional protocols, constructs a next-generation intelligent charging system with more comprehensive perception, more intelligent decision-making, closer collaboration, and faster response. It not only achieves precise dynamic adjustment of charging power technically, but also fundamentally improves the user experience, eliminating the need for users to weigh charging speed, device safety, and lifespan, truly achieving a balance of efficiency, safety, and convenience. This sets a new benchmark for the future development of high-power fast charging technology, possessing significant practical value and broad application prospects.

[0139] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments or some parts of embodiments.

[0140] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. These modifications or substitutions do not cause the essence of the corresponding technical solutions to depart from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for mobile phone fast charging power adjustment based on multi-modal sensing and two-way protocol, characterized in that, Comprise: S1: Establish a two-way communication connection between the charging device and the mobile phone, automatically identify and match the fast charging protocol through the multi-protocol compatible module, and synchronize the device parameters and initial state data of both parties through encryption authentication; S2: According to the two-way communication connection, use distributed sensing layout to collect multi-dimensional state data during charging, after filtering and anti-interference processing, use NPU accelerated dynamic weight fusion model for preprocessing and feature fusion, and get multi-dimensional features; S3: Build an adaptive learning model, fuse multi-dimensional features and interaction data, and realize accurate power regulation through GPU acceleration, dynamically adjust the extreme charging power reference value; S4: Establish a power regulation model, combine multi-dimensional features and interaction data for accurate power regulation, and output accurate power regulation instructions; S41: Establish a power regulation model, initialize the core parameters of the power regulation model according to the high-dimensional feature vector; S42: Bind the initialized power regulation model with multi-dimensional features and two-way protocol interaction data, and build a model input and output mapping relationship framework; S43: Use the feature fusion module of the power regulation model to perform hierarchical fusion on the input multi-dimensional features and interaction data, mine the implicit correlation between features through nonlinear transformation, and generate a fusion feature matrix; S431: Extract the multi-dimensional features and two-way protocol interaction data to be fused according to the model input and output mapping relationship framework; S432: Time axis alignment processing is performed on the extracted multi-dimensional features and interaction data; S433: According to the correlation strength of the features in the mapping relationship framework, group the multi-dimensional features and interaction data with high correlation degree to obtain feature combination units; S434: Apply nonlinear transformation algorithm to each feature combination unit, mine the implicit correlation between features in the combination unit through matrix operation of multilayer perceptron, and obtain a correlation feature vector; S435: Integrate the correlation feature vector according to the logical structure of the mapping relationship framework to obtain a fusion feature matrix; S44: Drive the inference calculation of the power regulation model according to the fusion feature matrix to obtain preliminary power regulation parameters; S45: Accuracy verification is performed on the preliminary regulation parameters, the deviation between the actual charging effect and the model prediction result is compared, the model parameters are corrected through the back propagation mechanism, and accurate power regulation instructions are output; S5: Real-time monitoring effect according to instruction regulation result, establish closed loop feedback mechanism through two-way protocol, dynamically optimize power regulation strategy.

2. The method for mobile phone extreme charging power adjustment based on multi-modal sensing and bidirectional protocol according to claim 1, characterized in that, In step S1, the specific steps of establishing a two-way communication connection between the charging device and the mobile phone are as follows: S11: After the charging device and the mobile phone are physically connected, automatically scan and identify the fast charging protocol type supported by the charger according to the multi-protocol compatible module; S12: Based on the identified protocol type, the mobile phone initiates a protocol handshake request with the charging device, and completes device identity verification and security key negotiation through an encryption authentication algorithm; S13: After completing the encryption authentication, both parties synchronize their respective device parameters and initial state data through the agreed protocol format; S14: Verify the synchronized device parameters and initial state data, confirm the data integrity and compatibility, and establish a stable two-way communication link to lay the foundation for subsequent data interaction during charging.

3. The method of claim 1, wherein, In step S2, the specific steps of preprocessing the multi-dimensional state data are: S21: Collecting multi-dimensional raw data during the charging process based on the established two-way communication connection; S22: Anti-interference processing of the multi-dimensional raw data through a pre-set filter network, filtering out noise caused by electromagnetic interference to obtain purified raw data; S23: Time synchronization and format standardization calibration of the purified raw data to obtain standard raw data; S24: Inputting the standard raw data into the NPU-accelerated dynamic weight fusion model, and the model dynamically allocates the weight of each dimension data according to the charging stage to complete multi-level fusion of the feature layer and the decision layer, and output high-dimensional feature set after fusion processing.

4. The method for mobile phone extreme charging power adjustment based on multi-modal sensing and bidirectional protocol according to claim 1, characterized in that, In step S3, the specific steps of dynamically adjusting the extreme charging power reference value are: S31: Initializing the adaptive learning model according to the fusion features and real-time interaction data; S32: Using the dynamic weight mechanism of the hidden layer of the adaptive learning model to perform feature reorganization and correlation analysis on the input multi-dimensional features and interaction data to obtain a high-dimensional feature vector; S33: Inputting the high-dimensional feature vector into the GPU-accelerated inference engine, quickly solving the power adjustment parameters through parallel calculation, and combining the extreme charging power reference value to generate a preliminary adjustment scheme; S34: Safety boundary verification of the preliminary adjustment scheme, and correction combined with historical adjustment effect data; S35: Outputting the final precise power adjustment instruction, dynamically updating the extreme charging power reference value, and feeding back the adjustment result to the two-way communication link.

5. The method for mobile phone extreme charging power adjustment based on multi-modal sensing and bidirectional protocol according to claim 1, characterized in that, In step S31, the specific steps of initializing the adaptive learning model are: S311: Receiving the fusion features and real-time interaction data output by the previous step as the initial input data set of the adaptive learning model; S312: Format conversion and dimension alignment of the initial input data set, converting multi-dimensional features into a tensor format recognizable by the model, and mapping two-way protocol interaction data into a standard feature vector; S313: Initializing the weight parameters and bias terms of the input layer of the adaptive learning model; S314: According to the multi-channel processing mechanism of the input layer of the adaptive learning model, independent processing channels are allocated for multi-dimensional features and two-way protocol interaction data for preliminary analysis and feature extraction of data, and an intermediate feature vector processed by the input layer is output.

6. The method for mobile phone extreme charging power regulation based on multi-modal sensing and bidirectional protocol according to claim 1, characterized in that, In step S33, the specific steps of generating a preliminary adjustment scheme are: S331: Converting the reorganized feature vector into a GPU parallel data format, loading it into a GPU-accelerated inference engine, and calling a pre-set parallel computing kernel to output a parallel feature vector that can be directly operated by the GPU; S332: The inference engine performs high-dimensional space mapping and parameter solving on the parallel feature vector through multi-thread parallel calculation to obtain power adjustment parameters; S333: Fusing the extreme charging power reference value as a constraint condition with the power adjustment parameters to obtain a preliminary adjustment scheme of the current power reference framework.

7. The method of claim 1, wherein, In step S44, the specific steps of calculating the preliminary power adjustment parameters are: S441: Data conversion of the fusion feature matrix; S442: using a GPU-accelerated model inference engine, inputting the converted fused feature matrix into the engine, performing high-dimensional space mapping on the matrix through multi-thread parallel computing, and generating intermediate calculation results reflecting the correlation between features and power adjustment; S443: performing fusion operation on the intermediate calculation results generated by the inference engine and the extracted power adjustment rules, and solving the power adjustment parameter range satisfying the rule constraints through a linear programming algorithm; S444: selecting the value closest to the efficiency and safety balance target in the power adjustment parameter range to obtain the preliminary power adjustment parameter.

8. The method for mobile phone extreme charging power regulation based on multi-modal sensing and bidirectional protocol according to claim 1, characterized in that, In step S5, the specific steps of dynamically optimizing the power adjustment strategy are as follows: S51: adjust the phone extreme charging power according to the precise power adjustment instruction, and monitor the power data in real time; S52: compare and analyze the monitored power data with the expected target value in the adjustment instruction, and calculate the power deviation and temperature control deviation; S53: based on the power deviation and temperature control deviation analysis result, obtain the current hardware load and safety margin data of the two devices; S54: dynamically adjust the core parameters of the power adjustment strategy according to the hardware load, safety margin data and cross-device data feedback of the two-way protocol to obtain the optimized power adjustment strategy.

Images