Shared charging equipment instruction control system and method based on multi-modal signal fusion

Abstract

The invention discloses a shared charging equipment instruction control system and method based on multi-mode signal fusion, particularly relates to the field of shared charging control, solves the protocol out-of-step overshoot risk, takes a handshake millisecond window as a switching-in port, and synchronously checks a digital message and a physical transient state through surge energy and thermal inertia double indexes. Locking an instruction missing risk before a driving link; the residual error-message joint judgment captures hidden abnormity in real time at a power issuing node, a rollback channel is rapidly pulled back to a safe power grade, and brand-channel mapping logic is triggered to re-plan output resources; a threshold bin dynamically tightens a threshold according to a target-actual measurement power difference, so that prediction and actual measurement are gradually fitted, and new prior is written back to a forward inference link; high-voltage overshoot and power failure of the whole row are avoided, a billing curve is completely matched with a charging curve, high turnover of a port can be kept in a passenger flow peak, the safety is integrally improved, the service life of hardware is prolonged, and the user experience is improved.

Description

Technical Field

[0001] This invention relates to the field of shared charging control, and more specifically, to a shared charging equipment command control system and method based on multimodal signal fusion. Background Technology

[0002] Shared charging devices are widely deployed in commercial open spaces. Users complete payment through a touch interface. The main controller reads the terminal handshake message and allocates the appropriate power level, and then sends voltage commands step by step through the power drive link. The micro-vibration, bus voltage fluctuation and port thermal gradient at the moment of plugging and unplugging are simultaneously written into the acquisition channel, accumulating multi-source information for subsequent safety judgment and billing verification.

[0003] Currently, shared charging devices rely solely on digital messages to determine the fast charging version during protocol negotiation, failing to perform cross-modal consistency checks on bus voltage change rate and port temperature rise trends. When the terminal and device are running different versions of the power supply protocol, the main controller mistakenly assumes the handshake is complete and fails to detect the loss of the voltage step command in the drive link. High voltage is then directly applied to the bus, causing power overshoot, triggering multi-port overcurrent protection, and initiating a re-handshake cycle after power failure. This results in screen flickering, charging progress resetting to zero, and billing misalignment. Smartphones also lose continuous power, significantly impacting the overall user experience.

[0004] To address the aforementioned problems, a technical solution is provided. Summary of the Invention

[0005] To overcome the aforementioned deficiencies in the prior art, embodiments of the present invention provide a shared charging device command control system and method based on multimodal signal fusion. Using a handshake millisecond window as the entry point, it synchronously verifies digital messages and physical transients through both surge energy and thermal inertia exponential checks, locking the risk of command omissions before the drive link. Residual-message joint judgment captures hidden anomalies in real time at the power distribution node, quickly pulling back to a safe power level via the fallback channel and triggering brand-channel mapping logic to re-plan output resources. The threshold bin dynamically tightens the threshold based on the difference between the target and measured power, ensuring that predictions and measurements match successively, and writes back new priors in the forward inference stage, thereby solving the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A command control method for shared charging devices based on multimodal signal fusion includes the following steps: S1: When the handshake is triggered, multimodal data is collected synchronously, key features are extracted after time alignment, and analysis is performed to capture signs of lost instructions and mark handshake anomalies; S2: Perform self-supervised reconstruction on the unmarked anomaly segment to obtain the residual sequence, and then input the residual sequence and the protocol verification result into the change point judgment process. If both give an anomaly judgment at the same time, issue a rollback command immediately. S3: The rollback command cancels the current voltage step and switches back to the previous stable power level. At the same time, it rematches the output power and protocol according to the brand-channel occupancy table and writes the abnormal characteristics into the threshold bin. S4: During the power stabilization phase, the threshold bin is iteratively updated by comparing the difference between the target power and the measured power, and then synchronously transmitted back to the self-supervised reconstruction process to shorten the convergence time of the next round of judgment and suppress power overshoot caused by protocol heterogeneity in the long term.

[0007] In a preferred embodiment, step S1 includes the following: Multimodal data includes message streams, bus voltage curves, and port thermal gradients; Time alignment is achieved by using interpolation methods to align message streams, bus voltage curves, and port thermal gradients onto a unified time axis.

[0008] In a preferred embodiment, step S1 further includes the following: Key features include the instantaneous energy surge index and the thermal inertia index; Differentiate the bus voltage curve within the handshake trigger window, calculate the square integral of the voltage change rate to obtain the energy surge increment, and then normalize it using the median value of the window integral of the most recent 100 frames to generate the instantaneous surge energy index; calculate the first derivative of the port thermal image gradient on the time axis to obtain the heating rate, and then perform exponential decay convolution on the heating rate to generate the thermal inertia index.

[0009] In a preferred embodiment, step S1 further includes the following: The instantaneous energy surge index and thermal inertia index are input into the radial basis regression model to generate a multi-source consistency coefficient. When the multi-source consistency coefficient is lower than a preset threshold, it is marked as a handshake abnormality; when the multi-source consistency coefficient is not lower than the preset threshold, it is marked as a handshake normality.

[0010] In a preferred embodiment, step S2 includes the following: For segments not marked as anomalous, a variational autoencoder is applied to reconstruct time-aligned multimodal data. The absolute difference between the original multimodal data and the reconstructed data is calculated to generate a residual sequence. Check bits or error detection codes are parsed from the message stream to generate protocol verification results.

[0011] In a preferred embodiment, step S2 further includes the following: The absolute value of the difference between the residual value at the current time and the residual value at the previous time is calculated for the residual sequence. Then, it is divided by the sum of the residual value at the previous time and a small positive number to generate the residual mutation index. If the residual mutation index exceeds the preset threshold, it is marked as a change point. When the protocol verification result at the change point is a failure value, it is determined to be a synchronization anomaly and a rollback command is issued to cancel the current voltage step command and switch to the previous stable power level.

[0012] In a preferred embodiment, step S3 includes the following: The rollback command parses the current voltage step value to generate the voltage value after cancellation, and extracts the value of the previous stable power level from the historical power record to generate the power value after switching.

[0013] In a preferred embodiment, step S3 further includes the following: The system queries the brand-channel occupancy table to extract terminal brand and channel occupancy rate to generate a rematched output power. It selects a compatible protocol based on the terminal brand and updates the output power to the rematched output power. It extracts the residual sequence and protocol verification results to generate anomaly feature vectors to generate an updated threshold bin.

[0014] In a preferred embodiment, step S4 includes the following: During the power stabilization phase, the target power is compared with the measured power to generate a power difference. The ratio of the power difference to the target power is adjusted by the update rate to generate a threshold update amount. The threshold update amount is then combined with the abnormal feature vector and added to the current threshold bin content to generate an updated threshold bin. The updated threshold bin is then incorporated into the loss function of the self-supervised reconstruction model. The final loss function value is generated by summing the reconstruction error and the threshold bin adjusted residual sequence.

[0015] A command control system for shared charging equipment based on multimodal signal fusion includes: Data fusion module: synchronously collects message stream, bus voltage curve and port thermal image gradient and completes time alignment, extracts instantaneous surge energy index and thermal inertia index and calculates multi-source consistency coefficient. If the multi-source consistency coefficient is lower than the preset threshold, it marks the handshake as abnormal and outputs the marking result. Anomaly detection module: Performs self-supervised reconstruction on segments without handshake anomalies to generate residual sequences. Then, sends the residual sequences and protocol verification results into the change point judgment process. If the change point judgment process and the handshake anomaly marking indicate an anomaly simultaneously, a rollback instruction is generated. Power rollback module: The rollback command cancels the current voltage step and switches back to the previous stable power level. At the same time, it rematches the output power and protocol according to the brand-channel occupancy table and writes the abnormal characteristics into the threshold bin. Threshold calibration module: During the power stabilization phase, the threshold bin is iteratively updated by comparing the difference between the target power and the measured power, and then synchronously transmitted back to the self-supervised reconstruction process to accelerate the next round of judgment convergence and continuously suppress power overshoot caused by protocol heterogeneity.

[0016] The technical effects and advantages of the shared charging equipment command control system and method based on multimodal signal fusion of this invention are as follows: This invention first places the message, bus voltage change, and port temperature rise rate onto the same time coordinate in the first few milliseconds of the handshake. It extracts the instantaneous voltage surge energy index and thermal inertia index, and uses radial basis function regression to quickly compare the trends of these two indices. If a discrepancy is detected between the digital handshake and the physical transient, a "red light" is activated before the power command is actually issued. Then, segments still considered normal are handed over to self-supervised reconstruction. The residuals are calculated and simultaneously fed into the change point model along with the protocol self-check results. If both pieces of evidence simultaneously indicate an anomaly, a rollback action is immediately triggered, pulling the impending voltage surge back to a previous stable level and reassigning idle ports using brand-channel mapping to avoid user queuing. After power recovery, the system compares the target output with the actual output, converting the difference into a new threshold and writing it into the database. The next handshake round directly uses this updated threshold for reassessment, resulting in faster and more accurate judgment. The three-step process integrates "early warning, instant error correction, and continuous calibration" to prevent high-voltage overcharging and power outages, ensure that the billing curve matches the charging curve perfectly, maintain high port turnover during peak hours, and improve overall safety, hardware lifespan, and user experience. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the command control method for shared charging devices based on multimodal signal fusion according to the present invention.

[0018] Figure 2 This is a schematic diagram of the structure of the command control system for shared charging equipment based on multimodal signal fusion according to the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Example 1: Figure 1 The present invention provides a command control method for shared charging devices based on multimodal signal fusion, comprising: S1: When the handshake is triggered, multimodal data is collected synchronously. After time alignment, key features are extracted and analyzed to capture signs of lost instructions and mark handshake anomalies.

[0021] S2: Perform self-supervised reconstruction on the unmarked anomaly fragment to obtain the residual sequence, and then input the residual sequence and the protocol verification result into the change point judgment process. If both give an anomaly judgment at the same time, issue a rollback command immediately.

[0022] S3: The rollback command cancels the current voltage step and switches back to the previous stable power level. At the same time, it rematches the output power and protocol according to the brand-channel occupancy table and writes the abnormal characteristics into the threshold bin.

[0023] S4: During the power stabilization phase, the threshold bin is iteratively updated by comparing the difference between the target power and the measured power, and then synchronously transmitted back to the self-supervised reconstruction process to shorten the convergence time of the next round of judgment and suppress power overshoot caused by protocol heterogeneity in the long term.

[0024] In shared charging device usage scenarios, the handshake phase is the core link in charging protocol negotiation and power allocation, directly affecting the safety and stability of the charging process. Traditional methods rely solely on digital messages to determine the fast charging version, ignoring physical voltage and temperature changes. This leads to the potential loss of voltage step commands in the power drive link when protocol versions are mismatched, causing issues such as power overshoot, multi-port overcurrent protection, and power-down re-handshake, severely impacting user experience. This invention proposes a command control method based on multi-modal signal fusion. By simultaneously acquiring message streams, bus voltage curves, and port thermal gradients during handshake triggering, and fusing digital signals with physical transient features, it aims to capture signs of command loss within a millisecond-level window, ensuring command integrity during the handshake phase and providing a reliable basis for subsequent power allocation. This solution addresses the practical problem of heterogeneous protocols in shared charging devices, overcoming the shortcomings of traditional methods in cross-modal consistency verification.

[0025] S1.1 Handshake Triggering and Data Acquisition.

[0026] When a user initiates the charging operation via the touch interface or inserts the terminal into the charging device, a handshake signal is detected, and the handshake phase begins. During this phase, three types of data are simultaneously collected: message stream, bus voltage curve, and port thermal gradient.

[0027] The message stream records the digital communication content between the terminal and the charging device, including key information for the handshake protocol negotiation, such as the fast charging version and power requirements. By analyzing the message stream, the specific progress of the protocol negotiation and the details of the instruction transmission can be obtained, providing indispensable digital evidence for determining whether the handshake process is normal.

[0028] The bus voltage curve reflects the real-time changes in the output voltage of the charging equipment. During the handshake phase, the execution of the voltage step command directly affects the dynamic performance of the bus voltage. By monitoring this curve, it is possible to verify whether the voltage command is executed correctly, and thus identify whether there are any command losses or execution anomalies, providing an important reference for physical-level anomaly detection.

[0029] Port thermal gradients record the temperature change trend of the charging port and are closely related to power output and current magnitude. During charging, abnormal heating or power overshoot is often accompanied by abnormal temperature rise. By analyzing port thermal gradients, the stability of the handshake phase can be assessed from the perspective of thermal effects, providing supplementary physical evidence for detecting potential anomalies.

[0030] These three data points provide comprehensive information from both digital signal and physical transient perspectives: the message flow focuses on the logical correctness of protocol negotiation, while the bus voltage curve and port thermal gradient reflect the physical effects of instruction execution. Combined, these three data points not only characterize the handshake phase from multiple dimensions but also ensure a high degree of consistency between digital instructions and physical execution through cross-modal consistency checks, thereby significantly improving the accuracy of anomaly detection.

[0031] S1.2 Time Alignment.

[0032] To ensure the consistency of message streams, bus voltage curves, and port thermal gradients across time, time synchronization processing is performed on these three types of data. Specifically, within the time window following the handshake signal trigger, a unified timestamp is assigned to each data set to ensure that message content, voltage values, and temperature distribution data at the same moment correspond to each other. The time synchronization processing first records the acquisition time point for each type of data, and then adjusts the data points using interpolation methods to align them onto a common time axis.

[0033] S1.3 Extract key features and analyze them to capture signs of lost instructions and mark handshake anomalies.

[0034] The handshake phase involves complex interactions between digital signals and physical transients. Traditional methods rely solely on message flow to determine the protocol negotiation result, neglecting the verification of physical layer execution. This leads to the difficulty in timely detection of lost instructions or execution anomalies. In the high-dynamic scenarios of shared charging devices, the handshake process is typically completed within a millisecond window. Anomalies (such as voltage surges or abnormal temperature increases) are fleeting, requiring the rapid capture of these transient characteristics using quantitative indicators. This method introduces key features, including the instantaneous surge energy index and the thermal inertia index, based on the characteristics of voltage and temperature changes, aiming to provide sensitive and reliable anomaly detection methods to compensate for the shortcomings of traditional methods in physical layer analysis.

[0035] The instantaneous surge energy index quantifies the intensity and duration of voltage surges by calculating the square integral of the rate of change of bus voltage. During the handshake phase, the execution of a voltage step command causes a significant change in bus voltage, and this index can sensitively reflect this surge characteristic. When a command is lost or its execution is abnormal, the voltage change pattern deviates from expectations, and the instantaneous surge energy index will exceed the normal range, thus providing a rapid and accurate basis for anomaly detection.

[0036] The thermal inertia index quantifies the hysteresis and persistence of temperature changes by performing exponentially decaying convolution on the rate of temperature change at the port. Because temperature exhibits a certain inertia due to the influence of heat conduction and heat capacity, this index can capture the dynamic trend of temperature changes. When abnormal heating or power overshoot occurs, the thermal inertia index will deviate from its normal value, promptly revealing potential risks.

[0037] The instantaneous surge energy index and thermal inertia index are designed for high sensitivity to voltage surges and temperature change trends, respectively, enabling rapid detection of anomalies during the brief handshake phase. They provide complementary information based on different physical quantities: the instantaneous surge energy index focuses on the execution effect of voltage commands, while the thermal inertia index focuses on abnormal thermal effects. By comprehensively analyzing these two indices, the reliability and robustness of anomaly detection can be effectively improved, avoiding misjudgments that may result from a single index, and ensuring a rapid response in the event of command loss or execution anomalies.

[0038] S1.3-1 Calculate the instantaneous energy surge index.

[0039] The instantaneous surge energy index is used to quantify the degree of abrupt change in bus voltage during the handshake phase. Its calculation process involves several steps. First, the bus voltage curve is differentiated within the time window following the handshake signal trigger, i.e., the change in voltage value at each moment relative to the previous moment is calculated, resulting in a voltage change rate sequence. Next, each value in the voltage change rate sequence is squared to highlight the contribution of the abrupt change. These squared values ​​are then accumulated within the time window to obtain an energy surge increment reflecting the cumulative effect of voltage abrupt changes. To ensure comparability, the time windows of the most recent one hundred historical data sets are selected, and the accumulated sum of the squared voltage change rate values ​​within each window is calculated. The median value is then extracted as a normalization benchmark. Finally, the energy surge increment of the current time window is divided by this median value to obtain the dimensionless instantaneous surge energy index. The instantaneous surge energy index can sensitively reflect the execution status of voltage step commands. When command loss leads to abnormal voltage changes, the index will deviate from the normal range, thus providing crucial information for anomaly detection.

[0040] S1.3-2 Calculate the thermal inertia index.

[0041] The thermal inertia index is used to quantify the hysteresis characteristics of port temperature changes, and its calculation process also involves multiple steps. First, the first derivative of the port thermal image gradient is calculated on the time axis, that is, the change in temperature value at each moment relative to the previous moment is calculated, resulting in a heating rate sequence that reflects the instantaneous trend of temperature change. Next, an exponentially decaying convolution is applied to the heating rate sequence. Specifically, the heating rate sequence is multiplied by a weight function that decays exponentially with time and then summed to obtain the thermal inertia quantification value. This convolution process emphasizes the persistence and hysteresis of temperature changes and is more sensitive to short-term abnormal fluctuations. If a command loss causes an abnormal heating pattern in the port temperature, the thermal inertia quantification value will deviate from the expected range. Finally, the thermal inertia quantification value is output as the thermal inertia index for subsequent analysis. The thermal inertia index provides a supplementary perspective for anomaly detection from the angle of temperature change, enhancing the comprehensiveness of multimodal analysis.

[0042] S1.3-3 generates multi-source consistency coefficients.

[0043] The multi-source consistency coefficient is used to comprehensively evaluate the consistency between voltage and temperature changes, and its generation depends on the processing of a radial basis function (RBF) kernel regression model. The generation process first uses the instantaneous surge energy index and thermal inertia index as input features, providing them to the RBF kernel regression model. The RBF kernel regression model is a nonlinear regression method based on kernel functions, capable of capturing the complex relationship between input features and the output target. In this invention, the model's goal is to generate a continuous value reflecting the consistency of multimodal signals, namely the multi-source consistency coefficient, based on the instantaneous surge energy index and thermal inertia index.

[0044] The construction process of the radial basis function (RBF) kernel regression model is as follows: First, the model uses a Gaussian kernel function as the basis function. The Gaussian kernel function calculates the similarity of features based on the distance between the input features and the basis function centers. The selection of the basis function centers is determined by cluster analysis of the instantaneous energy surge index and thermal inertia index of historical data. Specifically, the historical data is divided into several groups, and the center value of each group is used as the basis function center. The clustering process is completed by calculating the distance between each data point and grouping them. The output of each basis function center is calculated by the Gaussian kernel function to generate a set of intermediate results. Then, the model generates the final output, i.e., the multi-source consistency coefficient, by linearly combining these intermediate results. The weights and bias parameters of the linear combination are determined through the training process.

[0045] Training data sources and parameter calculations: Training data is derived from historical handshake records, including multiple sets of instantaneous surge indices, thermal inertia indices, and corresponding multi-source consistency coefficient reference values. These reference values ​​can be continuous values ​​representing the degree of consistency, or classification labels indicating whether the handshake was normal. The goal of the training process is to minimize the error between the model's predicted values ​​and the reference values. Specifically, first, the predicted values ​​generated by inputting the instantaneous surge indices and thermal inertia indices of each training sample into the model are calculated. Then, the sum of squares of the differences between the predicted values ​​and the reference values ​​is calculated as the error index. By adjusting the weights and biases, the error index is minimized. The adjustment method can be iterative, updating the parameters each time according to the direction of error change, until the error converges. After training, the model can generate accurate multi-source consistency coefficients based on new instantaneous surge indices and thermal inertia indices, which are used to evaluate the consistency of multimodal signals and message flows.

[0046] The generation process of the multi-source consistency coefficient makes full use of the complementary information of voltage and temperature data, and improves the accuracy of anomaly detection through nonlinear mapping.

[0047] S1.3-4 indicates a handshake error.

[0048] The handshake phase is assessed for anomalies based on the multi-source consistency coefficient. A pre-defined threshold is used, determined by the distribution of multi-source consistency coefficients in historical data of normal and abnormal handshakes. If the calculated multi-source consistency coefficient is below this threshold, it indicates inconsistencies between the message flow and the physical transient characteristics of voltage and temperature, suggesting a risk of command loss; therefore, it is marked as a handshake anomaly. If the multi-source consistency coefficient is above or equal to the threshold, the handshake process is considered normal, and command transmission is complete. The marking results guide subsequent power allocation decisions, ensuring timely intervention when anomalies are detected to avoid power overshoot or power-off restart issues. This assessment method improves the reliability of handshake anomaly detection by comprehensively analyzing multi-mode signals.

[0049] Step S1 involves synchronously acquiring message streams, bus voltage curves, and port thermal gradients upon handshake signal triggering, and performing time alignment processing to ensure data consistency across the time dimension. Subsequently, by calculating the instantaneous surge energy index and thermal inertia index, voltage spikes and temperature hysteresis characteristics are quantified respectively. Then, a radial basis function regression model is used to generate multi-source consistency coefficients, comprehensively evaluating the consistency of multi-modal signals. Finally, a handshake anomaly is determined based on a threshold. This process is completed quickly, achieving accurate detection of command loss signs and providing technical support for the stable operation of shared charging equipment in heterogeneous protocol environments.

[0050] Step S1 uses multimodal data such as message streams, bus voltage curves, and port thermal gradients, combined with instantaneous surge power index and thermal inertia index, to calculate the multi-source consistency coefficient. This initial detection of command loss during the handshake phase is then performed, and handshake anomalies are flagged. However, even if the handshake phase is not flagged as anomaly, potential risks may still exist during power allocation and command execution, such as the undetected loss of voltage step commands in the drive link, or asynchrony between protocol negotiation and physical response. If these hidden anomalies are not detected in time, they may lead to power overshoot or power loss. To address this issue, step S2 introduces further monitoring mechanisms for segments not flagged as anomalies to ensure the safety and stability of the charging process.

[0051] S2.1 Self-supervised reconstruction generates residual sequences.

[0052] To detect potential hidden anomalies in unlabeled segments, self-supervised learning is used to reconstruct multimodal data, generating residual sequences to capture anomalous signs deviating from normal patterns. The multimodal data includes the message stream, bus voltage curve, and port thermal gradients acquired in step S1, which have been time-aligned to form a joint representation. Specifically, a variational autoencoder is applied to the time-aligned multimodal data to learn normal data patterns and generate reconstructed data.

[0053] Variational autoencoders are models that compress input data into a latent representation using an encoder and then reconstruct the original data using a decoder, thus capturing the essential characteristics of the data distribution.

[0054] The process of calculating the residual sequence involves subtracting the predicted data reconstructed by the variational autoencoder from the joint representation of the original multimodal data and taking the absolute value of the result. The joint representation of the original multimodal data includes both digital message and physical quantity information, while the predicted data reconstructed by the variational autoencoder is an expected value generated based on the normal pattern. The residual sequence represents the absolute difference between the two. By learning the inherent distribution of normal data, the variational autoencoder can generate reconstructed data that closely matches the normal pattern. The residual sequence amplifies anomalous parts that deviate from the normal pattern, allowing even minor instruction execution deviations or physical response anomalies to manifest in the residuals, thus providing sensitive indicators for subsequent detection.

[0055] Obtaining the S2.2 protocol verification result.

[0056] To verify the integrity of command transmission at the digital level, protocol verification results are extracted from the message stream as an auxiliary basis for judging the residual sequence. The message stream has been collected and time-aligned in step S1.

[0057] Parse the checksum or error detection code in the message stream to generate a protocol verification result: if the verification passes, the protocol verification result is a pass value; if the verification fails, the protocol verification result is a failure value. The generation process of the protocol verification result ensures consistency with the timeline of the residual sequence. The protocol verification result directly reflects the correctness of message transmission and negotiation, and can identify lost or corrupted digital instructions. By combining it with the residual sequence, anomalies can be verified from both digital and physical dimensions, ensuring comprehensive detection.

[0058] S2.3 Change Point Judgment Process.

[0059] By jointly analyzing the residual sequence and protocol verification results, potential synchronization anomalies can be identified through change point detection, thereby ensuring the accuracy and timeliness of anomaly judgment.

[0060] Change point detection is performed on the residual sequence, and the residual mutation index is calculated: the absolute value of the difference between the residual value at the current time and the residual value at the previous time is taken as the numerator; a small positive number is added to the residual value at the previous time to obtain the denominator, so as to avoid the denominator being zero; the numerator is divided by the denominator to obtain the residual mutation index, which represents the relative change of the residual.

[0061] If the residual mutation index exceeds a preset threshold, that moment is marked as a change point. The protocol verification result at the change point is then checked: if the protocol verification result is a failure, it is determined to be a synchronization anomaly. The residual mutation index captures mutation points in the residual sequence through relative change rates, enabling rapid location of physical-level anomalies. Combined with digital verification of the protocol verification results, a determination is triggered only when anomalies occur simultaneously at both the digital and physical levels.

[0062] S2.4 issues a rollback command.

[0063] When a synchronization anomaly is detected, a rollback measure is immediately taken to prevent the risk of power overshoot or power loss and to ensure the safety of the charging process.

[0064] If a synchronization anomaly is detected, a rollback command is issued: the current voltage step command is revoked, and the power is switched to the previous stable power level. The power level information is determined based on the data collected in step S1 and the negotiation results. The rollback command, by revoking the anomaly command and restoring the system to a verified stable state, can quickly interrupt the anomaly's development and prevent the high-voltage busbar or multi-port overcurrent protection from being triggered. This rapid response mechanism minimizes risks and ensures continuous power supply to equipment and users.

[0065] Step S2, for segments not marked as abnormal, generates a residual sequence through self-supervised reconstruction. Combined with protocol verification results and a change point judgment process, it enables real-time monitoring of hidden anomalies. When the residual mutation index indicates a change point and protocol verification fails, it is determined to be a synchronization anomaly, and a rollback command is issued to cancel the voltage step and switch back to a stable power level. This process effectively captures instruction execution deviations or protocol asynchrony issues not detected during the handshake phase, preventing power overshoot and power loss risks, and providing crucial protection for the safe operation of shared charging equipment.

[0066] Step S2 generates a residual sequence through self-supervised reconstruction and detects hidden anomalies in the change point judgment process by combining the protocol verification results. When a synchronization anomaly is determined, a rollback command is issued. However, the execution of the rollback command not only requires timely removal of potential risks but also optimization of the subsequent charging process through resource reallocation and experience accumulation to avoid repeated overcharging caused by protocol heterogeneity. To solve this problem, step S3 introduces a multi-level response mechanism for the rollback command to ensure power stability and system adaptability.

[0067] S3.1 Cancel the current voltage step.

[0068] To interrupt the abnormal voltage loading process, the system receives a rollback command from step S2 and immediately cancels the current voltage step to prevent high voltage from being directly applied to the bus. The rollback command includes an identifier of the synchronization anomaly and the specific parameters of the current voltage step.

[0069] The current voltage step value in the rollback command is analyzed to generate a cancellation signal. The process of calculating the voltage value after cancellation involves subtracting the voltage step increment from the current voltage step value. The current voltage step value is the actual voltage extracted from the rollback command, and the voltage step increment represents the voltage change amplitude before the anomaly. The voltage value after cancellation ensures consistency with the bus voltage curve. By directly subtracting the voltage step increment, the cancellation calculation accurately restores the voltage to the pre-anomaly state, avoiding overshoot propagation. The voltage value after cancellation quickly stabilizes the bus, reducing the risk of power outages.

[0070] S3.2 Switch back to the previous stable power setting.

[0071] To restore the charging process to a safe state, switch to the previous stable power level based on the voltage value after cancellation.

[0072] The previous stable power level value is extracted from historical power records, and the following switching method is applied: the value of the previous stable power level is multiplied by the quotient of the canceled voltage value and the corresponding voltage value of the previous stable power level. Here, the previous stable power level value is the verified stable power level, the canceled voltage value is the voltage calculated in step S3.1, and the voltage value corresponding to the previous stable power level is the corresponding voltage in historical records. The switched power value maintains a consistent power-to-voltage ratio. This switching method ensures power continuity through voltage ratio adjustment, preventing oscillations caused by sudden changes. The switched power value maintains charging continuity, improving the user experience.

[0073] S3.3 Rematch output power and protocol.

[0074] To optimize resource allocation, output power and protocol are rematched based on the brand-channel occupancy table, which records the correspondence between terminal brand, channel occupancy rate, and supported protocols.

[0075] The brand-channel occupancy table is queried to extract the terminal brand and channel occupancy rate. The process of calculating the matched power involves multiplying the switched power value by one and subtracting the channel occupancy rate, then adding the minimum safe power multiplied by the channel occupancy rate. The switched power value is obtained from step S3.2, the channel occupancy rate is a dimensionless proportional value recorded in the table, the minimum safe power is the minimum power level determined based on the terminal brand, and the re-matched output power is the final adjusted power. The matched power is weighted and balanced between the current power and the minimum power using occupancy rate to ensure efficient resource utilization. The re-matched output power and protocol adapt to heterogeneous environments, suppressing repeated anomalies.

[0076] S3.4 Write to the threshold bin.

[0077] To accumulate experience with anomalies, anomaly features are written into the threshold bin. These anomaly features include residual sequences of synchronization anomalies and protocol verification results.

[0078] A threshold bin is a database structure used to store and dynamically update anomaly detection thresholds and related feature vectors. It records anomaly feature vectors captured during the handshake and power allocation phases, such as combinations of residual sequences and protocol verification results. Its function is to continuously optimize anomaly detection standards in heterogeneous protocol environments by accumulating anomaly features and adjusting thresholds in combination with update coefficients. This ensures that the system can adapt to the dynamic changes of different terminals and protocols, thereby improving the accuracy and stability of anomaly judgment. The threshold bin is updated by weighted accumulation of anomaly feature vectors to generate an updated threshold bin to support adaptive adjustments in subsequent steps.

[0079] Extract the residual sequence and protocol verification results from step S2 to generate an anomaly feature vector. Updating the threshold bin involves adding the current threshold bin content to the result of an update coefficient multiplied by the anomaly feature vector. Here, the current threshold bin content is the stored threshold vector, the update coefficient is a dimensionless parameter controlling the write intensity, and the anomaly feature vector is a quantized representation of the anomaly information. The updated threshold bin is the newly accumulated threshold vector. The update incorporates anomaly features through weighted accumulation, achieving dynamic threshold adjustment. The updated threshold bin provides an optimization basis for subsequent steps, suppressing protocol heterogeneity issues in the long term.

[0080] The update coefficient is a dimensionless parameter used in the threshold bin update process to adjust the writing intensity of abnormal feature vectors, ensuring the gradualness and stability of threshold adjustment. Its function is to achieve dynamic optimization of the threshold by controlling the accumulation magnitude of abnormal information, avoiding inaccurate system response caused by excessive or insufficient updates, thereby maintaining the long-term reliability of shared charging devices in heterogeneous protocol environments. The acquisition method is usually based on empirical tuning or historical data analysis. For example, in reinforcement learning-like applications, it is preset as a hyperparameter and determined iteratively through cross-validation to balance update speed and system stability.

[0081] Step S3 cancels the current voltage step and switches back to the previous stable power level by receiving a rollback command. At the same time, it rematches the output power and protocol according to the brand-channel occupancy table and writes the abnormal characteristics into the threshold bin. This process effectively responds to synchronization anomalies, restores the charging stability state, optimizes resource allocation, and accumulates experience to prevent future overcharging and power loss, ensuring the reliable operation of shared charging equipment in heterogeneous protocol scenarios.

[0082] Step S3 performs a rollback operation, canceling the abnormal voltage step and switching back to the stable power level. Simultaneously, the abnormal characteristics are written into the threshold bin, laying the foundation for subsequent optimization. However, to address the long-term need to suppress power overshoot in heterogeneous protocol environments, relying solely on the static information of the initial threshold bin is insufficient to adapt to dynamically changing power allocation deviations. Therefore, step S4 introduces a comparative analysis of the target power and the measured power during the power stabilization phase. By iteratively updating the threshold bin and synchronously transmitting it back to the self-supervised reconstruction process, dynamic optimization of the threshold and continuous improvement of anomaly detection are achieved.

[0083] S4.1 Compare the difference between the target power and the measured power.

[0084] During the power stabilization phase, to quantify the deviation between the negotiated protocol results and the actual execution results, it is necessary to calculate the difference between the target power and the measured power. The target power is the expected output power determined based on the brand-channel occupancy table and the protocol negotiation, while the measured power is the actual output power collected by the power sensor during the stabilization phase. This comparison reflects the accuracy of power allocation and provides data support for subsequent threshold adjustments.

[0085] The specific process for calculating the power difference is as follows: subtract the measured power value from the target power value; the result is called the power difference. A positive power difference indicates that the measured power is lower than the target power, while a negative value indicates that the measured power exceeds the target power. By using the power difference as a quantitative indicator, accurate input data can be provided for the optimization of the threshold bin, thereby improving the targeting of anomaly detection.

[0086] S4.2 Iteratively update the threshold bin.

[0087] Based on the calculated power difference, the threshold bin needs to be iteratively updated to adapt to dynamic changes in power allocation requirements. The threshold bin stores anomaly features and threshold information. Its update process normalizes the power difference and combines it with anomaly feature vectors to ensure that the sensitivity and robustness of anomaly detection are gradually optimized.

[0088] The process of updating the threshold bin consists of two steps. First, the threshold update amount is calculated by dividing the power difference by the target power to obtain a dimensionless ratio reflecting the relative magnitude of the deviation. Then, this ratio is multiplied by a preset update rate to obtain the threshold update amount. The update rate is a dimensionless parameter used to control the step size of the threshold adjustment to avoid instability due to excessively fast adjustment or sluggish response due to excessively slow adjustment. Next, the threshold bin content is updated by adding the threshold information currently stored in the threshold bin to the product of the threshold update amount and the abnormal feature vector to obtain the updated threshold bin content.

[0089] By normalizing the power difference and combining it with anomaly feature vectors, the threshold bin can be progressively adjusted to a state that better reflects the actual power distribution. This gradual adjustment enhances the ability to identify abnormal fluctuations in heterogeneous protocol environments, while avoiding the problems of overly sensitive detection or slow response caused by a single adjustment.

[0090] S4.3 synchronously transmits back to the self-supervised refactoring process.

[0091] To improve the anomaly detection accuracy in the self-supervised reconstruction process in step S2, the updated threshold bins from step S4.2 need to be synchronously fed back to the self-supervised reconstruction model as prior information to optimize the reconstruction process. The self-supervised reconstruction model generates reconstruction data by analyzing multimodal data. Incorporating the updated threshold bins into its loss function allows the reconstruction process to focus more on anomaly-related changes, thereby accelerating the convergence of anomaly detection.

[0092] The loss function of a self-supervised reconstruction model consists of two parts.

[0093] The first part is the sum of squares of the differences between the original multimodal data and the reconstructed data, representing the reconstruction error. The original multimodal data includes message flow, bus voltage curve and port thermal gradient, while the reconstructed data is generated by a self-supervised reconstruction model.

[0094] The second part involves multiplying the updated threshold bin by the residual sequence, and then multiplying by an adjustment coefficient. This adjustment coefficient represents the moderating effect of the threshold bin on the residuals. The residual sequence is the difference between the original multimodal data and the reconstructed data, and the adjustment coefficient is a dimensionless parameter used to balance the weight of the reconstruction error and the threshold bin's influence. Adding these two parts yields the final loss function value, which guides the optimization of the self-supervised reconstruction model.

[0095] The self-supervised reconstruction model is built upon a variational autoencoder framework. This framework compresses the original multimodal data into a low-dimensional latent representation through an encoder, and then reconstructs the original data through a decoder, achieving self-supervised learning. The specific construction process includes: First, designing an encoder network that uses a convolutional neural network or fully connected layers to process the fused representation of the message stream, bus voltage curve, and port thermal gradient, outputting mean and variance parameters to sample the latent vector; second, constructing a decoder network that upsamples or deconvolves the latent vector to generate reconstructed data; finally, adding the reconstruction loss and the KL divergence loss to form a basic loss function, used to train the model to capture normal data distribution. For example, in a shared charging device scenario, the encoder can process 100 frames of time-aligned data, with a latent dimension of 32, ensuring the model efficiently reconstructs normal signals within a millisecond-level handshake window.

[0096] The model's parameter settings need to be optimized according to the specific application, mainly including the learning rate, batch size, latent space dimension, and training epochs. For example, the learning rate is usually set to 0.001 to balance convergence speed and stability; the batch size is 64 to facilitate noise smoothing in gradient descent; the latent space dimension is 16 to 64, adjusted according to data complexity to avoid underfitting or overfitting; and the training epochs are 1000 to prevent overtraining through an early stopping mechanism. Furthermore, the adjustment coefficient is set to 0.1 in the loss function to fine-tune the impact of the threshold bin on the residual weights. For example, if the power difference is large, the adjustment coefficient can be dynamically increased to 0.5 to enhance attention to abnormal residuals and ensure that the model quickly adapts to power fluctuations in heterogeneous protocol environments.

[0097] By incorporating the updated threshold bin into the loss function, the self-supervised reconstruction model can prioritize residual changes related to anomalous features during the optimization process. This is because the threshold bin contains anomalous pattern information accumulated from power difference iterations, which makes the reconstructed data more accurately simulate the distribution of normal multimodal data, thereby amplifying the performance of residual sequences that hide anomalous features. In the change point judgment process in step S2, this enhanced residual sequence provides more significant abrupt change signals, reduces noise interference, and improves the accuracy of anomaly detection. At the same time, the dynamic adjustment of the threshold bin accelerates model convergence, shortens the number of reconstruction iterations, achieves faster response time, and ensures timely capture of instruction loss or power overshoot signs in heterogeneous protocol environments.

[0098] In step S4, during the power stabilization phase, the threshold bin is iteratively updated by calculating the difference between the target power and the measured power, and the updated result is synchronously transmitted back to the self-supervised reconstruction process, optimizing the accuracy and efficiency of anomaly detection. This process enables dynamic adjustment of the threshold bin, allowing anomaly detection to gradually adapt to the power fluctuation characteristics in a heterogeneous protocol environment, while effectively suppressing power overshoot, providing technical support for the stable operation of shared charging equipment and improving user experience.

[0099] Example 2: Figure 2 The present invention provides a command control system for shared charging devices based on multimodal signal fusion, comprising: Data fusion module: synchronously collects message stream, bus voltage curve and port thermal image gradient and completes time alignment, extracts instantaneous surge energy index and thermal inertia index and calculates multi-source consistency coefficient. If the multi-source consistency coefficient is lower than the preset threshold, it marks the handshake as abnormal and outputs the marking result. Anomaly detection module: Performs self-supervised reconstruction on segments without handshake anomalies to generate residual sequences. Then, sends the residual sequences and protocol verification results into the change point judgment process. If the change point judgment process and the handshake anomaly marking indicate an anomaly simultaneously, a rollback instruction is generated. Power rollback module: The rollback command cancels the current voltage step and switches back to the previous stable power level. At the same time, it rematches the output power and protocol according to the brand-channel occupancy table and writes the abnormal characteristics into the threshold bin. Threshold calibration module: During the power stabilization phase, the threshold bin is iteratively updated by comparing the difference between the target power and the measured power, and then synchronously transmitted back to the self-supervised reconstruction process to accelerate the next round of judgment convergence and continuously suppress power overshoot caused by protocol heterogeneity.

[0100] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0101] It should be noted that the system of the present invention can be deployed on the device itself to realize embedded applications, or it can run on a PC or other terminal with a user interface, thereby meeting a variety of hardware environments and usage requirements.

[0102] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

[0103] It should be noted that, in this document, the use of relational terms such as "first" and "second" is merely to distinguish one entity or operation from another, and does not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0104] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A command control method for shared charging equipment based on multi-modal signal fusion, characterized in that, Including the following steps: S1: When the handshake is triggered, multimodal data is collected synchronously, key features are extracted after time alignment, and analysis is performed to capture signs of lost instructions and mark handshake anomalies; S2: Perform self-supervised reconstruction on the unmarked anomaly segment to obtain the residual sequence, and then input the residual sequence and the protocol verification result into the change point judgment process. If both give an anomaly judgment at the same time, issue a rollback command immediately. S3: The rollback command cancels the current voltage step and switches back to the previous stable power level. At the same time, it rematches the output power and protocol according to the brand-channel occupancy table and writes the abnormal characteristics into the threshold bin. S4: During the power stabilization phase, the threshold bin is iteratively updated by comparing the difference between the target power and the measured power, and then synchronously transmitted back to the self-supervised reconstruction process to shorten the convergence time of the next round of judgment and suppress power overshoot caused by protocol heterogeneity in the long term.

2. The command control method for shared charging equipment based on multi-modal signal fusion according to claim 1, characterized in that, Step S1 includes the following: Multimodal data includes message streams, bus voltage curves, and port thermal gradients; Time alignment is achieved by using interpolation methods to align message streams, bus voltage curves, and port thermal gradients onto a unified time axis.

3. The command control method for shared charging equipment based on multi-modal signal fusion according to claim 1, characterized in that, Step S1 also includes the following: Key features include the instantaneous energy surge index and the thermal inertia index; Differentiate the bus voltage curve within the handshake trigger window, calculate the square integral of the voltage change rate to obtain the energy surge increment, and then normalize it using the median value of the window integral of the most recent 100 frames to generate the instantaneous surge energy index; calculate the first derivative of the port thermal image gradient on the time axis to obtain the heating rate, and then perform exponential decay convolution on the heating rate to generate the thermal inertia index.

4. The command control method for shared charging equipment based on multi-modal signal fusion according to claim 1, characterized in that, Step S1 also includes the following: The instantaneous energy surge index and thermal inertia index are input into the radial basis regression model to generate a multi-source consistency coefficient. When the multi-source consistency coefficient is lower than a preset threshold, it is marked as a handshake abnormality; when the multi-source consistency coefficient is not lower than the preset threshold, it is marked as a handshake normality.

5. The command control method for shared charging equipment based on multi-modal signal fusion according to claim 4, characterized in that, Step S2 includes the following: For segments not marked as anomalous, a variational autoencoder is applied to reconstruct time-aligned multimodal data. The absolute difference between the original multimodal data and the reconstructed data is calculated to generate a residual sequence. Check bits or error detection codes are parsed from the message stream to generate protocol verification results.

6. The command control method for shared charging equipment based on multi-modal signal fusion according to claim 5, characterized in that, Step S2 also includes the following: The absolute value of the difference between the residual value at the current time and the residual value at the previous time is calculated for the residual sequence. Then, it is divided by the sum of the residual value at the previous time and a small positive number to generate the residual mutation index. If the residual mutation index exceeds the preset threshold, it is marked as a change point. When the protocol verification result at the change point is a failure value, it is determined to be a synchronization anomaly and a rollback command is issued to cancel the current voltage step command and switch to the previous stable power level.

7. The command control method for shared charging equipment based on multi-modal signal fusion according to claim 6, characterized in that, Step S3 includes the following: The rollback command parses the current voltage step value to generate the voltage value after cancellation, and extracts the value of the previous stable power level from the historical power record to generate the power value after switching.

8. The command control method for shared charging equipment based on multi-modal signal fusion according to claim 7, characterized in that, Step S3 also Includes the following: The system queries the brand-channel occupancy table to extract terminal brand and channel occupancy rate to generate a rematched output power. It selects a compatible protocol based on the terminal brand and updates the output power to the rematched output power. It extracts the residual sequence and protocol verification results to generate anomaly feature vectors to generate an updated threshold bin.

9. The command control method for shared charging equipment based on multimodal signal fusion according to claim 8, characterized in that, Step S4 includes the following: During the power stabilization phase, the target power is compared with the measured power to generate a power difference. The ratio of the power difference to the target power is adjusted by the update rate to generate a threshold update amount. The threshold update amount is then combined with the abnormal feature vector and added to the current threshold bin content to generate an updated threshold bin. The updated threshold bin is then incorporated into the loss function of the self-supervised reconstruction model. The final loss function value is generated by summing the reconstruction error and the threshold bin adjusted residual sequence.

10. A command control system for shared charging equipment based on multimodal signal fusion, used to implement the command control method for shared charging equipment based on multimodal signal fusion as described in any one of claims 1-9, characterized in that, include: Data fusion module: synchronously collects message stream, bus voltage curve and port thermal image gradient and completes time alignment, extracts instantaneous surge energy index and thermal inertia index and calculates multi-source consistency coefficient. If the multi-source consistency coefficient is lower than the preset threshold, it marks the handshake as abnormal and outputs the marking result. Anomaly detection module: Performs self-supervised reconstruction on segments without handshake anomalies to generate residual sequences. Then, sends the residual sequences and protocol verification results into the change point judgment process. If the change point judgment process and the handshake anomaly marking indicate an anomaly simultaneously, a rollback instruction is generated. Power rollback module: The rollback command cancels the current voltage step and switches back to the previous stable power level. At the same time, it rematches the output power and protocol according to the brand-channel occupancy table and writes the abnormal characteristics into the threshold bin. Threshold calibration module: During the power stabilization phase, the threshold bin is iteratively updated by comparing the difference between the target power and the measured power, and then synchronously transmitted back to the self-supervised reconstruction process to accelerate the next round of judgment convergence and continuously suppress power overshoot caused by protocol heterogeneity.

Images