Mobile phone extreme heat-electricity multi-modal fusion safety protection system based on deep learning

Abstract

The application discloses a mobile phone extreme charging heat-electricity multi-modal fusion safety protection system based on deep learning, relates to the technical field of charging control, and comprises a handshake echo module, which is used for completing the extreme charging handshake of a mobile phone and an adapter, issuing a tentative pulse package by the mobile phone, sequentially outputting the tentative pulse package by the adapter, and forming first round echo data; an instruction coding module, which is used for jointly feeding the first round echo data and a simplified heat zone diagram into a neural echo encoder to generate a safety ripple instruction set and a partition allocation table; and a comparison revision module, which is used for comparing and rewriting the echo index in the first round echo data and an execution reply item by item to generate a revised safety ripple instruction set and partition allocation table. The application utilizes the neural echo encoder to jointly analyze the first round echo data and the simplified heat zone diagram, generates a safety ripple instruction set facing the adapter side and a partition allocation table facing the battery side, and realizes the integrated output of the charging pulse sub-palette shape and the spatial rotation bearing strategy.

Description

Technical Field

[0001] This invention relates to the field of charging control technology, and in particular to a deep learning-based mobile phone extreme charging thermal-electrical multimodal fusion safety protection system. Background Technology

[0002] With the simultaneous increase in smartphone power density and user demand for longer battery life, high-rate, high-current ultra-fast charging solutions have become the industry mainstream. In existing technologies, most ultra-fast charging solutions negotiate voltage and current limits at the protocol layer through a handshake between the adapter and the phone. During charging, they combine overall voltage and current data collected by the battery management unit, along with data from single-point or a few temperature sensors, with strategies such as constant current-constant voltage and segmented current limiting to achieve basic safety protection. Simultaneously, explorations have emerged using electrochemical impedance spectroscopy, charging curve characteristics, or data-driven methods to comprehensively assess lithium deposition risk and consistency degradation, providing a foundation for lifetime prediction and anomaly detection in ultra-fast charging scenarios.

[0003] However, in the context of high-speed mobile phone charging, the instantaneous power is high, space is limited, and thermal coupling is complex. Traditional technologies mainly rely on the average temperature of the battery pack or individual cells, total current, and voltage curves with low time resolution for control. This makes it difficult to reflect local thermal imbalances and electrochemical anomalies under high-speed waveforms in a timely manner. At the same time, existing solutions mostly use fixed safety margins and static limiting, lacking joint modeling and real-time coupling of the fine-grained shape of the charging voltage-current waveform and the spatial distribution of temperature at multiple points in the battery. This makes it difficult to fully exploit the high-speed charging capabilities of mobile phones while ensuring safety, and may lead to conservative protection strategies that affect charging efficiency or safety hazards caused by delayed response in local hot areas. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a deep learning-based mobile phone extreme charging thermal-electric multimodal fusion safety protection system to solve the problem of poor safety protection and delayed response caused by the difficulty in real-time coupling of voltage and current waveforms with multi-point battery temperature during mobile phone extreme charging.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] This invention provides a deep learning-based multi-modal safety protection system for mobile phone extreme charging, encompassing:

[0008] The handshake echo module completes the handshake between the mobile phone and the adapter. After the mobile phone sends out a probe pulse packet, the adapter outputs it in sequence to form the first round of echo data.

[0009] The instruction encoding module sends the first round of echo data and the simplified hot zone diagram into the neural echo encoder to generate a safety ripple instruction set and a partition allocation table.

[0010] The shaping and partitioning module sends the safety ripple instruction set and partition allocation table to the adapter and battery management unit, performs pulse shaping and partition rotation, and generates an execution receipt.

[0011] By comparing and rewriting the echo index and execution receipt in the first round of echo data with the revision module, a revised security ripple instruction set and partition allocation table are generated.

[0012] As a preferred embodiment of the deep learning-based mobile phone extreme charging thermal-electric multimodal fusion safety protection system described in this invention, the probe pulse packet is configured by the mobile phone after the extreme charging handshake is completed, based on the extreme charging capability reported by the adapter and the current state of charge of the battery, and is obtained by combining continuous test pulse segments according to different rise processes and different duty cycles and then encapsulating them.

[0013] As a preferred embodiment of the deep learning-based mobile phone thermal-electrical multimodal fusion safety protection system described in this invention, the specific steps for generating the first round of echo data are as follows:

[0014] After completing the handshake between the phone and the adapter, the phone sends a probe pulse packet to the adapter. The adapter outputs the corresponding Extreme Charge current segment by segment through the reconfigurable secondary filter network according to the test pulse segments in the probe pulse packet.

[0015] During the segmented output of the polar charging current, voltage changes, current changes, and voltage recovery trajectory are recorded synchronously to form continuous trajectory data.

[0016] Select the corresponding extreme charge safety sensitive trajectory segments from the continuous trajectory data, and label each trajectory segment with time information and pulse segment number consistent with the test pulse packet, and organize them into the first round of echo data in chronological order.

[0017] As a preferred embodiment of the deep learning-based mobile phone extreme charging thermal-electric multimodal fusion safety protection system described in this invention, the extreme charging safety sensitive trajectory segments include trajectory segments corresponding to the extreme charging current rising phase, trajectory segments where the extreme charging current is within a preset peak range, and trajectory segments during the voltage recovery phase after extreme charging is paused.

[0018] As a preferred embodiment of the deep learning-based mobile phone thermal-electrical multimodal fusion safety protection system described in this invention, the steps for obtaining the simplified thermal zone diagram are as follows:

[0019] By collecting multi-point temperature data of the battery and hot spot temperature data of the whole machine through temperature sensors deployed at different locations of the battery, a raw dataset of hot zones is formed.

[0020] Based on the original dataset of hot zones, hot zones are divided according to temperature distribution, and each zone is assigned a number and a representative temperature value.

[0021] A simplified hot zone diagram is created by combining the area number of all hot zones, the representative temperature value, and the time index corresponding to the first round of displayed data.

[0022] As a preferred embodiment of the deep learning-based mobile phone extreme charging thermal-electrical multimodal fusion safety protection system of the present invention, the specific steps for generating the safety ripple instruction set and partition allocation table are as follows:

[0023] The neural echo encoder establishes a correspondence between segments and hot zones based on the pulse segment numbers of the first round of echo data and the region numbers of the simplified hot zone diagram, and generates an echo description set.

[0024] Extract deep feature vectors representing the waveform change characteristics of each segment and the temperature response characteristics of each thermal zone from the echo description set;

[0025] Based on deep feature vectors, the pulse sub-palette and the number of continuous segments are determined for each segment, and arranged according to the pulse segment number to form a safe ripple instruction set;

[0026] The rotation order and load ratio of each thermal zone are determined based on the same deep feature vector, and then summarized by region number to form a zone allocation table.

[0027] As a preferred embodiment of the deep learning-based mobile phone thermal-electrical multimodal fusion safety protection system described in this invention, the pulse shaping process includes the following specific steps.

[0028] Read the pulse sub-palette sequence to be output and the number of continuous segments corresponding to each sub-palette from the safety ripple instruction set, and pre-configure the filter branch control sequence of the internal capacitor and inductor switching matrix.

[0029] During the polar charging process, the adapter switches the corresponding capacitor and inductor branches according to the filter branch control sequence, so that the output current in each segment presents the rising, plateauing and intermittent process specified by the safety ripple instruction set, and arranges fixed-duration pauses between consecutive segments according to the arc protection interval to complete the pulse shaping of the polar charging current.

[0030] As a preferred embodiment of the deep learning-based mobile phone thermal-electrical multimodal fusion safety protection system of the present invention, the partitioned rotation bearing specifically includes the following steps.

[0031] Read the rotation sequence and load ratio of each hot zone from the partition allocation table, and generate a partition switch control instruction sequence by combining the number of each pulse segment in the safety ripple instruction set;

[0032] The zonal switch control command sequence switches the conducting battery hot zones before the start of each segment, so that each hot zone distributes the transmission electrode charging current according to the rotation order and load ratio. At the end of the segment, the temperature, voltage and zonal current are collected as execution records, which are summarized with the segment number and hot zone number to form an execution receipt.

[0033] As a preferred embodiment of the deep learning-based mobile phone extreme charging thermal-electrical multimodal fusion safety protection system described in this invention, the specific steps for generating the revised safety ripple instruction set and partition allocation table are as follows:

[0034] Based on the segment number and time information in the first round of echo data, the corresponding segment execution record is retrieved from the execution receipt to form a set of difference descriptions;

[0035] Using the deviation segments and corresponding deviation hot zones identified in the difference description set as the revision objects, the corresponding pulse sub-palettes are replaced in the safety ripple instruction set and the number of continuous segments is adjusted. At the same time, the rotation order and load ratio of the corresponding hot zones are adjusted in the partition allocation table, and the revised safety ripple instruction set and partition allocation table are output.

[0036] As a preferred embodiment of the deep learning-based mobile phone thermal-electric multimodal fusion safety protection system described in this invention, the deviation segment refers to the comparison of the temperature, voltage, and partition current of each hot zone in the segment execution record in the execution receipt with the preset safety range. When any parameter exceeds the corresponding safety range, it is marked as a deviation segment and the corresponding hot zone is marked as a deviation hot zone.

[0037] The beneficial effects of this invention are as follows: After the handshake handshake is completed, the handshake echo module outputs a probe pulse packet with multiple rising edge and duty cycle combinations, and collects and forms the first round of echo data. This allows the charging voltage-current transient response and multi-point battery temperature information to be quantified synchronously, providing high-time-resolution thermal-electrical multimodal basic data for subsequent safety decisions. Using a neural echo encoder, the first round of echo data is jointly analyzed with a simplified thermal zone diagram to generate a safety ripple instruction set for the adapter side and a partition allocation table for the battery side, realizing the shape and spatial rotation of the charging pulse sub-palette. The strategy provides integrated output; it completes pulse shaping under the reconfigurable filter branch and performs hot zone rotation conduction according to the partition allocation table, which improves the power utilization rate of extreme charging while alleviating the problems of local overheating and long-term high stress in a single area; by comparing the echo index in the first round of echo data with the execution receipt item by item, it performs targeted rewriting and adjustment of deviation segments and deviation hot zones, so that the safety ripple instruction set and partition allocation table can adaptively converge in the multi-round extreme charging process, thereby realizing the fine and dynamic optimization of thermal-electrical multimodal safety constraints in the mobile phone extreme charging process, improving the extreme charging safety margin while taking into account charging efficiency. Attached Figure Description

[0038] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0039] Figure 1 This is a schematic diagram of a deep learning-based mobile phone extreme charging thermal-electrical multimodal fusion safety protection system.

[0040] Figure 2 A flowchart for generating the first round of displayed data.

[0041] Figure 3 Flowchart for generating the safety ripple instruction set and partition allocation table.

[0042] Figure 4 A flowchart for revision and implementation feedback. Detailed Implementation

[0043] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0044] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0045] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0046] Reference Figures 1-4 This is one embodiment of the present invention, which provides a deep learning-based mobile phone extreme charging thermal-electrical multimodal fusion safety protection system, including the following steps:

[0047] The handshake echo module completes the handshake between the mobile phone and the adapter. After the handshake is completed, the mobile phone sends out a probe pulse packet, which is then output by the adapter in sequence to form the first round of echo data.

[0048] After completing the handshake between the phone and the adapter, the phone uses the adapter's charging capability information and the battery's current state of charge information obtained during the handshake phase to continuously test pulse segments according to different rise processes and different duty cycles. The continuous test pulse segments are then sequentially packaged into a probe pulse packet, with each test pulse segment assigned a unique pulse segment number in the probe pulse packet.

[0049] The mobile phone sends a test pulse packet to the adapter. The adapter outputs the corresponding polar charging current segment by segment through a reconfigurable secondary filter network according to the preset test pulse segment sequence in the test pulse packet. Before the start of each test pulse segment, the reconfigurable secondary filter network switches the capacitor and inductor combination according to the corresponding pulse segment number, so that the output current in each test pulse segment presents a current waveform that matches the rise process and duty cycle in the test pulse packet. The polar charging current output segment by segment maintains a one-to-one correspondence with the pulse segment number in the test pulse packet on the time axis.

[0050] It should be noted that the preset test pulse segment order is based on the adapter's extreme charging capability information and the battery's current state of charge information obtained during the extreme charging handshake phase, selected in order of increasing detection intensity. Low-stress segments are prioritized to ensure safety, and then gradually transition to high-stress segments to increase the amount of echoed information. For example, segment 1 (rapid rise, duty cycle 20%, frequency 80kHz, segment duration 10ms), segment 2 (slow rise, duty cycle 50%, frequency 80kHz, segment duration 10ms), segment 3 (rapid rise, duty cycle 80%, frequency 100kHz, segment duration 10ms), and segment 4 (pause segment, used for voltage recovery acquisition, duration 5ms).

[0051] During the entire trial period of segmented current output, the mobile phone synchronously collects the voltage and current on the output side of the adapter at a preset sampling period, obtaining a voltage change data sequence and a current change data sequence containing all test pulse segments. After each test pulse segment ends, voltage recovery data is collected, so that the voltage change data sequence, current change data sequence and voltage recovery data together constitute continuous trajectory data.

[0052] It should be noted that the preset sampling period is determined based on the fastest rise time and the highest operating frequency of the test pulse segment, so that the voltage change data sequence and the current change data sequence can cover edge and ripple information. For example, the sampling period is set to be no more than one-tenth of the fastest rise time and no more than one-twentieth of the highest operating frequency period.

[0053] Based on continuous trajectory data, trajectory segments sensitive to extreme charging safety are screened. At each sampling point, a comprehensive change index is constructed using voltage and current changes, expressed as:

[0054]

[0055] in, For the first The comprehensive change index of each sampling point is used to measure the drastic change in the extreme charging process at that sampling point. For the first Voltage sample value at each sampling point For the first Voltage sample value at each sampling point For the first Current sampling value at each sampling point For the first Current sampling value at each sampling point This is a reference value for voltage changes. This is a reference value for current changes.

[0056] Among them, the voltage change reference value and the current change reference value are calibrated based on the safety sensitivity of the extreme charging process, so that the contribution of the voltage change and the current change to the comprehensive change index are comparable. Under the arc protection limit of the mobile phone fast charging connector and the risk boundary of lithium plating in the battery cell, the small voltage step often has high indicative value, and the current step also has significant indicative value for thermal shock and lithium plating. For example, the voltage change reference value is 0.1V and the current change reference value is 1A.

[0057] By comparing the comprehensive change index with the preset change threshold, the continuous sampling point intervals in which the comprehensive change index exceeds the change threshold during the peak charging current rise phase, the continuous sampling point intervals in which the comprehensive change index exceeds the change threshold during the peak charging current within the preset peak range, and the continuous sampling point intervals in which the comprehensive change index exceeds the change threshold during the voltage recovery phase after the peak charging pause are selected; the trajectory segments corresponding to all continuous sampling point intervals are determined as the trajectory segments that are sensitive to peak charging safety.

[0058] It should be noted that the change threshold is set based on the upper limit of voltage change allowed by the connector in a single sampling interval and the upper limit of current change allowed by the cell in a single sampling interval. For example, setting it to 1 can stably capture the rapid transition point during the rising phase of the charging current.

[0059] The preset peak range is based on the target peak current of the corresponding test pulse segment, and a fixed proportion of the peak judgment range is determined to identify the continuous sampling points where the polar charging current is in the peak state, such as [5.4A, 6.0A].

[0060] Based on the safety-sensitive trajectory segments of the extreme charging system, the extreme charging safety-sensitive trajectory segments are paired with their respective test pulse segments according to the time information and pulse segment number corresponding to each extreme charging safety-sensitive trajectory segment. In each pairing, the corresponding voltage change trajectory, current change trajectory, and voltage recovery trajectory are recorded, and all pairing results are summarized and organized into the first round of display data in chronological order.

[0061] The instruction encoding module sends the first round of echo data and the simplified hot zone diagram into the neural echo encoder to generate a safety ripple instruction set and a partition allocation table.

[0062] After generating the first round of display data, a simplified hot zone diagram is constructed based on the battery multi-point temperature data and the overall hot spot temperature data.

[0063] Specifically, the steps for constructing a simplified hot zone diagram are as follows: During the extreme charging test, temperature sensors deployed at different locations on the battery collect multi-point temperature data. Simultaneously, temperature detection units inside the device collect hot spot temperature data near the processor, RF module, and charging interface. The multi-point temperature data of the battery and the hot spot temperature data are matched one-to-one with the time information recorded in the first round of display data to form the original hot zone dataset. Based on the original hot zone dataset, the battery and the entire device are divided into several hot zones according to temperature distribution (e.g., 6, including the hot zone on the upper surface of the battery, the hot zone on the lower surface of the battery, the hot zone on the side of the battery, the hot zone on the motherboard, the hot zone of the RF device, and the hot zone of the charging interface). Each hot zone is assigned a region number and a representative temperature value. The region numbers, representative temperature values, and time indices corresponding to the first round of display data of all hot zones are combined and organized into a simplified hot zone diagram.

[0064] By using the pulse segment numbers in the first round of echo data and the area numbers in the simplified hot zone diagram, a correspondence between segments and hot zones is established. Within the time interval covered by each pulse segment number, the area number in the active state within the corresponding time interval is found according to the simplified hot zone diagram. Each pulse segment number is combined with the corresponding area number and the corresponding extreme charging safety sensitive trajectory segments, voltage change trajectories, current change trajectories, and voltage recovery trajectories. All combinations of pulse segment numbers and all area numbers are recorded and summarized to form an echo description set.

[0065] While forming the echo description set, in order to characterize the response intensity of different pulse segments in different thermal zones, a response index for the segment and thermal zone is constructed based on a comprehensive change index. The expression for the response index is:

[0066]

[0067] in, Number the pulse segments With area code The corresponding segment and thermal response index are used to characterize the pulse segment number. In area code The average intensity of change in To number the pulse segments Within the time range and belonging to the area code The number of sampling points To number the pulse segments Within the time range and belonging to the area code The set of sampling point indices.

[0068] When the first round of echo data and the simplified hot zone diagram are fed into the neural echo encoder, the neural echo encoder takes the extreme charge safety-sensitive trajectory segment, voltage change trajectory, current change trajectory and voltage recovery trajectory corresponding to each pulse segment number in the echo description set as time series input, and takes the region number, representative temperature value and corresponding segment and hot zone response index in the simplified hot zone diagram as hot zone feature input. Based on the network parameters obtained by pre-training, the encoder performs joint forward inference on the two types of inputs to obtain a deep feature vector sequence that characterizes the extreme charge waveform change characteristics of each segment and the temperature response characteristics of each hot zone.

[0069] It should be noted that the neural echo encoder is constructed as follows:

[0070] The neural echo encoder is constructed using a deep neural network structure, including an input layer, a time-series feature extraction layer, a thermal zone feature extraction layer, a feature fusion layer, and an output layer. The input layer receives voltage change trajectories, current change trajectories, and voltage recovery trajectories based on the first round of echo data, as well as region numbers, representative temperature values, and segment and thermal zone response indices based on a simplified thermal zone diagram. The time-series feature extraction layer extracts features from the voltage-current time series using a one-dimensional convolutional network and a recurrent neural network. The thermal zone feature extraction layer performs fully connected mapping and normalization on the representative temperature values ​​and segments and thermal zone response indices for each region number. The feature fusion layer concatenates the time-series features and thermal zone features and obtains a deep feature vector through multi-layer nonlinear mapping. The output layer has two branches: one branch outputs the pulse sub-palette category and the number of continuous segments, and the other branch outputs the rotation order-related scores and load ratio-related parameters for each region number.

[0071] The training process for the neural echo encoder is as follows:

[0072] By collecting a large amount of historical voltage-current trajectory data and battery multi-point temperature data during the extreme charging process of mobile phones, the first round of echo data and simplified hot zone diagram are constructed. At the same time, a target safety ripple instruction set and target partition allocation table are generated for each set of first round of echo data and simplified hot zone diagram. The first round of echo data + simplified hot zone diagram + segment and hot zone response index are used as input samples, and the target pulse sub-palette category and continuous segment number, target rotation order and load ratio are used as supervision labels to form a training dataset.

[0073] The training dataset is fed into the neural echo encoder in batches. Forward propagation is used to obtain parameters related to the predicted pulse sub-palette category, the number of continuous segments, and the rotation order of each region number, as well as the carrying ratio. A joint loss function containing multiple losses is constructed, including a classification loss to constrain the prediction of the pulse sub-palette category, a regression loss to constrain the prediction of the number of continuous segments, and a distribution matching loss to constrain the prediction of the carrying ratio of the region number. The network parameters of the neural echo encoder are iteratively adjusted through the backpropagation algorithm until the joint loss function converges. The trained network parameters are then stored in the phone's memory as the basis for online inference during the Extreme Charge operation.

[0074] Based on the deep feature vector sequence, the neural echo encoder selects one type of pulse sub-palette from the three preset pulse sub-palette sets for each pulse segment number, and determines the number of segments that the corresponding pulse segment number needs to continue in the subsequent maximal charge period based on the degree of change reflected by the deep feature vector. Each pulse segment number is combined with the corresponding pulse sub-palette type and the number of continuous segments to form a segment configuration item, and all segment configuration items are arranged in the order of pulse segment numbers to generate a safe ripple instruction set.

[0075] It should be noted that the preset three types of pulse sub-palettes are based on the waveform characteristics of existing mobile phone extreme charging protocols and the safe operating range of battery cells under different temperature rise conditions. They are obtained by summarizing typical waveforms of the constant current stage, pulse extreme charging stage, and pause / recovery stage. In one embodiment, the three types of pulse sub-palettes include:

[0076] A slow-rise short-peak pulse sub-palette is used to increase the current at a relatively gentle slope and reach the peak current in a short time.

[0077] Flat-top slow-descent pulse sub-palette is used to maintain a near-constant current plateau over a period of time and gradually reduce the current at the tail end;

[0078] A bi-peak intermittent pulse sub-palette is used to output two peak pulses within the same segment and insert an interval between the two peaks.

[0079] Based on the deep feature vectors corresponding to the safety ripple instruction set and the segment and thermal response indicators, the neural echo encoder performs statistics and sorting on the thermal response indicators of each pulse segment number in the corresponding region number for each region number, thereby obtaining the distribution of thermal response indicators of each pulse segment number under the current region number.

[0080] The order in which the region numbers participate in the extreme charge rotation in the next time period is determined based on the magnitude of the hot zone response index. The time ratio of the corresponding pulse sub-palette on different region numbers is determined based on the distribution ratio of the hot zone response index among different region numbers. The rotation order and carrying ratio of each region number are recorded as hot zone configuration items. All hot zone configuration items are summarized and organized in the order of region numbers to generate a zone allocation table.

[0081] The shaping and partitioning module sends the safety ripple instruction set and partition allocation table to the adapter and battery management unit, performs pulse shaping and partition rotation, and generates an execution receipt.

[0082] The safety ripple instruction set and the partition allocation table are indexed and aligned according to the common pulse segment number, and encapsulated into safety encapsulation control data. The safety encapsulation control data includes the pulse sub-palette sequence, the number of continuous segments corresponding to each sub-palette, the arc protection interval parameter, and the rotation order and load ratio of each area number, and is sent to the adapter and battery management unit as the execution entry point.

[0083] After receiving the security encapsulation control data, the adapter reads the pulse sub-palette sequence and the number of continuous segments from the security ripple instruction set. Based on the pulse segment number, it pre-configures the filter branch control sequence in the capacitor and inductor switching matrix so that each pulse segment number corresponds to a set of defined filter branches and switching timing. Before execution, it completes self-test and ready marking to ensure that the filter branch control sequence and the pulse segment number correspond one-to-one.

[0084] When the adapter starts pulse shaping, it switches the corresponding capacitor and inductor combination according to the filter branch control sequence at the beginning of each pulse segment number, so that the output current in the corresponding segment presents the rising process, plateau process and intermittent process given in the safety ripple instruction set; a fixed duration pause (e.g. 100 microseconds) is inserted between adjacent pulse segment numbers according to the arc protection interval parameter to complete the pulse shaping of all pulse segment numbers in the current cycle.

[0085] After receiving the safety encapsulation control data, the battery management unit reads the rotation order and load ratio of each area number from the partition allocation table, and generates a partition switch control command sequence in combination with the pulse segment number, so that each pulse segment number obtains the corresponding conduction area number arrangement and the load time ratio allocation within the segment before the segment starts; the battery management unit sets the partition switch array before segment execution, and switches the conduction area number according to the load ratio during segment execution, so that the charging current is allocated to different area numbers within the segment according to the rotation order and load ratio given by the partition allocation table.

[0086] When each pulse segment number is executed, the adapter records the actual filter branch identifier, peak output current and plateau value used in this segment, and reads the connector status; at the same time, it records the temperature, partition current and terminal voltage of the current conduction area number, and writes the start and end timestamps of the bearings of each area number in the segment into the segment execution record.

[0087] Once all the extreme charging time slots corresponding to the safety encapsulation control data have been executed, the adapter and battery management unit will summarize the execution records of the segments according to the pulse segment number and the region number index, and form an execution receipt.

[0088] By comparing and rewriting the echo index and execution receipt in the first round of echo data with the revision module, a revised security ripple instruction set and partition allocation table are generated.

[0089] Using the time information and pulse segment number in the first round of echo data as the echo index, the corresponding segment execution record is retrieved one by one in the execution receipt. The extreme charging safety sensitive trajectory segment, voltage change trajectory, current change trajectory and voltage recovery trajectory recorded in the first round of echo data for each pulse segment number are matched one-to-one with the filter branch identifier, output current peak value, output current plateau value, conduction area number, temperature, partition current and terminal voltage recorded in the execution receipt according to the pulse segment number and area number, forming a set of reference records.

[0090] The reference record set is compared one by one according to the pre-set safety range. The safety range includes the upper limit of temperature for each hot zone, the upper limit of current for each partition, and the upper and lower limits of terminal voltage. When any record in the reference record set exceeds the corresponding upper limit of temperature, the upper limit of current for a partition, or the upper or lower limit of terminal voltage in a certain hot zone, the pulse segment number corresponding to the record is marked as a deviation segment, and the region number corresponding to the record is marked as a deviation hot zone. All marked deviation segments and corresponding deviation hot zones are summarized according to the correspondence between pulse segment number and region number to form a difference description set.

[0091] It should be noted that the safety range is based on the rated charging temperature window given by the cell manufacturer, the surface temperature allowed by the overall thermal design, the rated continuous current capability of the charging path devices and connectors, and the cell's termination charging voltage and discharge cutoff voltage settings. For example, the upper limit of the battery hot zone temperature is selected as 50°C, the upper limit of the zone current is selected as 7.5A, and the lower limit is 3.0V.

[0092] Each deviation segment and its corresponding deviation hot zone listed in the difference description set is used as the revision object. The corresponding segment configuration item is located in the safety ripple instruction set according to the pulse segment number. For pulse segment numbers that exceed the temperature limit or the zone current limit in the deviation hot zone, the pulse sub-palette originally selected for the corresponding pulse segment number is adjusted to a pulse sub-palette type with a gentler energy release, such as changing from a bi-peak intermittent pulse sub-palette to a flat-top slow-descent pulse sub-palette. The number of continuous segments configured for the corresponding pulse segment number in the subsequent extreme charging period is also reduced.

[0093] For pulse segment numbers that only show overvoltage without significant temperature rise in the deviation hot zone, the pulse sub-palette type is kept unchanged in the safety ripple instruction set. Only the number of continuous segments is shortened or more intermittent segments (e.g., 2) are inserted before and after the corresponding pulse segment number to form a revised safety ripple instruction set. In the revised safety ripple instruction set, all pulse segment numbers remain unchanged, and only the pulse sub-palette type and the number of continuous segments are adjusted accordingly.

[0094] After the revised safety ripple instruction set is determined, based on the deviation hot zone information recorded in the difference description set, the rotation order and load ratio of the corresponding hot zone are located in the partition allocation table according to the area number. For area numbers that repeatedly show deviations in multiple executions, the rotation order of the area number in the partition allocation table is postponed, and the load ratio of the area number under the corresponding pulse segment number is reduced. At the same time, the load ratio is increased accordingly for other area numbers that do not show deviations or have low hot zone response indicators. This allows the partition allocation table to allocate more high-stress segments to lower-risk area numbers while keeping the total rotation participation time basically unchanged, thus forming the revised partition allocation table. The revised partition allocation table is consistent with the revised safety ripple instruction set in terms of pulse segment number and area number.

[0095] In summary, this invention, through a handshake echo module, outputs a probe pulse packet with various rising edge and duty cycle combinations after the handshake is completed, and collects and forms the first round of echo data. This allows the charging voltage-current transient response and multi-point battery temperature information to be quantified synchronously, providing high-temporal-resolution thermal-electrical multimodal fundamental data for subsequent safety decisions. Furthermore, by utilizing a neural echo encoder, the first round of echo data is jointly analyzed with a simplified thermal zone diagram to generate a safety ripple instruction set for the adapter side and a partition allocation table for the battery side, realizing a charging pulse sub-palette shape and spatial rotation load-bearing strategy. The integrated output completes pulse shaping under the reconfigurable filter branch and performs hot zone rotation conduction according to the partition allocation table, which improves the power utilization rate of extreme charging while alleviating the problems of local overheating and long-term high stress in a single area; by comparing the echo index in the first round of echo data with the execution receipt item by item, the deviation segments and deviation hot zones are rewritten and adjusted in a targeted manner, so that the safety ripple instruction set and partition allocation table can adaptively converge in the multi-round extreme charging process, thereby realizing the fine and dynamic optimization of thermal-electric multimodal safety constraints in the mobile phone extreme charging process, improving the extreme charging safety margin while taking into account charging efficiency.

[0096] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A deep learning-based multi-modal safety protection system for mobile phone fast charging (thermal-electrical), characterized in that: include, The handshake echo module completes the handshake between the mobile phone and the adapter. After the mobile phone sends out a probe pulse packet, the adapter outputs it in sequence to form the first round of echo data. The instruction encoding module sends the first round of echo data and the simplified hot zone diagram into the neural echo encoder to generate a safety ripple instruction set and a partition allocation table. The steps for obtaining the simplified thermal zone diagram are as follows: By collecting multi-point temperature data of the battery and hot spot temperature data of the whole machine through temperature sensors deployed at different locations of the battery, a raw dataset of hot zones is formed. Based on the original dataset of hot zones, hot zones are divided according to temperature distribution, and each zone is assigned a number and a representative temperature value. A simplified hot zone diagram is created by combining the area number of all hot zones, the representative temperature value, and the time index corresponding to the first round of displayed data. The specific steps for generating the security ripple instruction set and partition allocation table are as follows. The neural echo encoder establishes a correspondence between segments and hot zones based on the pulse segment numbers of the first round of echo data and the region numbers of the simplified hot zone diagram, and generates an echo description set. Extract deep feature vectors representing the waveform change characteristics of each segment and the temperature response characteristics of each thermal zone from the echo description set; Based on deep feature vectors, the pulse sub-palette and the number of continuous segments are determined for each segment, and arranged according to the pulse segment number to form a safe ripple instruction set; The rotation sequence and load ratio of each thermal zone are determined based on the same deep feature vector, and then summarized by region number to form a zone allocation table; The neural echo encoder is constructed using a deep neural network structure, including an input layer, a time-series feature extraction layer, a thermal zone feature extraction layer, a feature fusion layer, and an output layer. The input layer receives voltage change trajectories, current change trajectories, and voltage recovery trajectories based on the first round of echo data, as well as region numbers, representative temperature values, and segment and thermal zone response indices based on a simplified thermal zone diagram. The time-series feature extraction layer extracts features from the voltage-current time series using a one-dimensional convolutional network and a recurrent neural network. The thermal zone feature extraction layer performs fully connected mapping and normalization on the representative temperature values ​​and segments and thermal zone response indices for each region number. The feature fusion layer concatenates the time-series features and thermal zone features and obtains a deep feature vector through multi-layer nonlinear mapping. The output layer has two branches: one branch outputs the pulse sub-palette category and the number of continuous segments, and the other branch outputs the rotation order-related scores and load ratio-related parameters for each region number. The shaping and partitioning module sends the safety ripple instruction set and partition allocation table to the adapter and battery management unit, performs pulse shaping and partition rotation, and generates an execution receipt. The specific steps for performing pulse shaping are as follows: Read the pulse sub-palette sequence to be output and the number of continuous segments corresponding to each sub-palette from the safety ripple instruction set, and pre-configure the filter branch control sequence of the internal capacitor and inductor switching matrix. During the extreme charging process, the adapter switches the corresponding capacitor and inductor branches according to the filter branch control sequence, so that the output current in each segment presents the rising, plateauing and intermittent process specified by the safety ripple instruction set, and arranges fixed-duration pauses between consecutive segments according to the arc protection interval to complete the pulse shaping of the extreme charging current; The specific steps for the partitioned rotation of bearers are as follows: Read the rotation sequence and load ratio of each hot zone from the partition allocation table, and generate a partition switch control instruction sequence by combining the number of each pulse segment in the safety ripple instruction set; The zonal switch control command sequence switches the conducting battery hot zones before the start of each segment, so that each hot zone distributes the transmission electrode charging current according to the rotation order and load ratio. At the end of the segment, temperature, voltage and zonal current are collected as execution records, which are summarized with the segment number and hot zone number to form an execution receipt. By comparing and rewriting the echo index and execution receipt in the first round of echo data with the revision module, a revised security ripple instruction set and partition allocation table are generated.

2. The deep learning-based mobile phone thermal-electrical multimodal fusion safety protection system as described in claim 1, characterized in that: The test pulse packet is configured by the mobile phone after the Extreme Charge handshake is completed, based on the Extreme Charge capability reported by the adapter and the current state of charge of the battery. It is obtained by combining continuous test pulse segments with different rise processes and different duty cycles and then encapsulating them.

3. The deep learning-based mobile phone thermal-electrical multimodal fusion safety protection system as described in claim 1, characterized in that: The specific steps for generating the first round of echo data are as follows. After completing the handshake between the phone and the adapter, the phone sends a probe pulse packet to the adapter. The adapter outputs the corresponding Extreme Charge current segment by segment through the reconfigurable secondary filter network according to the test pulse segments in the probe pulse packet. During the segmented output of the polar charging current, voltage changes, current changes, and voltage recovery trajectory are recorded synchronously to form continuous trajectory data. Select the corresponding extreme charge safety sensitive trajectory segments from the continuous trajectory data, and label each trajectory segment with time information and pulse segment number consistent with the test pulse packet, and organize them into the first round of echo data in chronological order.

4. The deep learning-based mobile phone thermal-electrical multimodal fusion safety protection system as described in claim 3, characterized in that: The trajectory segments that are sensitive to the safety of the extreme charging include the trajectory segments corresponding to the rising phase of the extreme charging current, the trajectory segments where the extreme charging current is within the preset peak range, and the trajectory segments during the voltage recovery phase after the extreme charging is paused.

5. The deep learning-based mobile phone thermal-electrical multimodal fusion safety protection system as described in claim 1, characterized in that: The specific steps for generating the revised security ripple instruction set and partition allocation table are as follows. Based on the segment number and time information in the first round of echo data, the corresponding segment execution record is retrieved from the execution receipt to form a set of difference descriptions; Using the deviation segments and corresponding deviation hot zones identified in the difference description set as the revision objects, the corresponding pulse sub-palettes are replaced in the safety ripple instruction set and the number of continuous segments is adjusted. At the same time, the rotation order and load ratio of the corresponding hot zones are adjusted in the partition allocation table, and the revised safety ripple instruction set and partition allocation table are output.

6. The deep learning-based mobile phone thermal-electrical multimodal fusion safety protection system as described in claim 5, characterized in that: The deviation segment refers to the comparison of the temperature, voltage, and partition current of each hot zone in the segment execution record in the execution receipt with the preset safety range. When any parameter exceeds the corresponding safety range, it is marked as a deviation segment and the corresponding hot zone is marked as a deviation hot zone.

Images