A charging host security early warning method and system fusing multi-source sensing data

Abstract

The application belongs to the technical field of alarm systems, and relates to a charging host safety early warning method and system fusing multi-source sensing data, comprising the following steps: obtaining an encoded micro-disturbance sequence; obtaining a composite modulation driving signal; obtaining a multi-dimensional original response data stream; performing a solving and purification operation to obtain a cross-domain response signal; obtaining an electric-thermal transfer function characteristic signature; obtaining a standard state transfer function baseline matrix; acquiring the electric-thermal transfer function characteristic signature and the standard state transfer function baseline matrix, performing a multi-space distance fusion comparison operation to obtain a deviation rating quantitative score; acquiring the deviation rating quantitative score, performing a warning instruction reorganization operation to obtain a hierarchical safety early warning instruction. The application solves the problem that early latent physical degradation in a charging host is difficult to detect and locate with high signal-to-noise ratio in the prior art.

Description

Technical Field

[0001] This invention belongs to the technical field of alarm systems and relates to a method and system for safety early warning of charging host that integrates multi-source sensor data. Background Technology

[0002] Under long-term high-power-density operation, the core components of high-power DC charging systems are prone to gradual physical degradation, such as insulation aging or decreased thermal conductivity. The macroscopic electrical parameter anomalies caused by such early latent faults are often not significant, and conventional fixed-threshold monitoring mechanisms are difficult to effectively address.

[0003] Chinese patent application publication number CN120853314A discloses a method and system for early warning of fires in distribution boxes based on multi-source information fusion. This method simultaneously collects multi-source signals such as temperature and current to generate a dynamic baseline, extracts features and clusters them to calibrate operating modes, retrieves and matches baseline signatures to calculate multimodal standardized deviation, obtains coupling risk coefficients through a fusion model, and finally performs multi-scale change analysis and uses dynamic thresholds for graded early warning.

[0004] The aforementioned existing technologies mainly rely on passive acquisition and fusion of multi-source signals. In environments with strong electromagnetic interference, the weak degradation characteristics passively acquired are easily affected by environmental fluctuations and switching harmonics, limiting the detection signal-to-noise ratio. Furthermore, given the complex conduction paths of deep electrothermal interaction within the equipment, this approach struggles to establish a clear mapping between external monitoring data and internal microscopic physical degradation mechanisms, increasing the difficulty of decoupling and locating latent faults.

[0005] Based on the above problems, the present invention aims to solve the problem of difficulty in detecting and locating early latent physical degradation inside the charging host with high signal-to-noise ratio in the prior art. Summary of the Invention

[0006] In a first aspect, the present invention provides a charging host safety early warning method that integrates multi-source sensor data, comprising the following steps:

[0007] S1. Obtain the steady-state operating parameters of the control loop of the charging host, perform modulation and coding operation, and obtain the coded micro-perturbation sequence;

[0008] S2. Obtain the coded micro-perturbation sequence, perform signal aliasing injection operation, and obtain the composite modulation drive signal;

[0009] S3. Acquire sensor network sampling data under the excitation of composite modulation driving signal, perform synchronous collection operation, and obtain multi-dimensional raw response data stream;

[0010] S4. Obtain the multi-dimensional raw response data stream and encoded micro-perturbation sequence, perform the solution and purification operation to obtain the cross-domain response signal;

[0011] S5. Obtain the cross-domain response signal, perform feature reconstruction operation, and obtain the feature signature of the electro-thermal transfer function;

[0012] S6. Obtain the real-time ambient temperature parameters and real-time power load ratio, perform baseline matching operation, and obtain the standard state transfer function baseline matrix.

[0013] S7. Obtain the feature signature of the electro-thermal transfer function and the baseline matrix of the standard state transfer function, perform multiple spatial distance fusion comparison operations, and obtain the deviation rating quantitative score;

[0014] S8. Obtain the deviation rating quantification score, execute the early warning instruction reorganization operation, and obtain the graded safety early warning instruction.

[0015] A further aspect of the present invention, step S1, includes the following steps:

[0016] Obtain the bus reference operating frequency from the steady-state operating parameters of the control loop, and generate the test signal period based on the bus reference operating frequency and the preset test band modulation coefficient;

[0017] Based on the test signal period and the preset Barker code basic waveform, combined with the preset amplitude gain weight, an initial disturbance signal stream is generated, a global clock synchronization broadcast pulse frame is obtained, and the initial disturbance signal stream is phase aligned using the global clock synchronization broadcast pulse frame to obtain the coded micro-disturbance sequence.

[0018] A further aspect of the present invention, step S2, includes the following steps:

[0019] Obtain the steady-state duty cycle adjustment value output by the microprocessor, and add the encoded micro-perturbation sequence to the steady-state duty cycle adjustment value to obtain the temporary duty cycle value;

[0020] Obtain the preset maximum safe duty cycle limit and minimum safe duty cycle limit, take the maximum value between the temporary duty cycle value and the minimum safe duty cycle limit, and then take the minimum value between the obtained maximum value and the maximum safe duty cycle limit for amplitude limiting to obtain the composite duty cycle adjustment value.

[0021] The composite duty cycle adjustment value is converted into a square wave sequence to obtain the composite modulation drive signal.

[0022] A further aspect of the present invention, step S3, includes the following steps:

[0023] Obtain the time-stamped counter within the global clock synchronization broadcast pulse frame, and according to the unified beat time defined by the time-stamped counter, obtain the current ripple data and voltage fluctuation data generated by the main circuit electrical circuit excited by the composite modulation drive signal;

[0024] According to the unified beat time, the heat transfer temperature gradient sequence of the heating device and the fluid dynamic pressure change data of the pipeline are acquired. The current ripple data, voltage fluctuation data, heat transfer temperature gradient sequence and fluid dynamic pressure change data are merged and packaged to obtain a multi-dimensional raw response data stream.

[0025] A further aspect of the present invention, step S4, includes the following steps:

[0026] Notch filtering is applied to the multi-dimensional raw response data stream to obtain the noise-reduced input data sequence. The reference frequency of the encoded micro-perturbation sequence is obtained, and in-phase and quadrature component waveforms are generated based on the reference frequency.

[0027] The input data sequence is multiplied and accumulated with the in-phase component waveform and the quadrature component waveform respectively to obtain the in-phase component and the quadrature component. Based on the in-phase component and the quadrature component, the sum of squares and the square root operation are performed to obtain the test frequency amplitude variable.

[0028] The characteristic phase delay parameter is obtained by performing arctangent operation on the in-phase and quadrature components. The cross-domain response signal is obtained by combining the test frequency amplitude variable and the characteristic phase delay parameter.

[0029] A further aspect of the present invention, step S5, includes the following steps:

[0030] Obtain the pre-saved transmission origin reference strength scalar, divide the test frequency amplitude variable by the transmission origin reference strength scalar, and obtain the effective node transmission cross-domain gain rate.

[0031] Obtain the preset spatial physical hindrance constant, convert the spatial physical hindrance constant into a reference phase delay based on the reference frequency, and subtract the reference phase delay from the characteristic phase delay parameter to obtain the offset phase shift angle;

[0032] By using the effective node transmission cross-domain gain rate as the diagonal element and the bias phase shift angle as the off-diagonal element, a matrix is ​​constructed to obtain the characteristic signature of the electro-thermal transfer function.

[0033] A further aspect of the present invention, step S6, includes the following steps:

[0034] Obtain real-time ambient temperature parameters and real-time power load ratio;

[0035] The real-time ambient temperature parameters and real-time power load ratio are mapped to discrete interval codes and concatenated to obtain a multi-way tree hash index code.

[0036] Use multi-way tree hash index encoding to locate a subset of candidate baseline files in a pre-defined read-only parameter template clustering library;

[0037] Calculate the weighted Euclidean distance between the real-time ambient temperature parameters and the real-time power load ratio and the reference operating condition parameters within the subset of candidate baseline files, and select the candidate baseline file with the smallest distance;

[0038] Decode the selected candidate baseline file to obtain the standard state transfer function baseline matrix.

[0039] A further aspect of the present invention, step S7, includes the following steps:

[0040] Convert the electro-thermal transfer function feature signature and the standard state transfer function baseline matrix into a complex tensor set;

[0041] Obtain the weight coefficients from the pre-set knowledge base of risk weights for aging and runaway components, and calculate the weighted Manhattan distance and weighted cosine distance between complex tensor groups based on the weight coefficients;

[0042] The weighted Manhattan distance and the weighted cosine distance are weighted and fused to obtain a multi-spatial distance. The multi-spatial distance is then input into a preset saturation curve mapping equation based on the arctangent function for nonlinear mapping to obtain a deviation rating quantification score.

[0043] A further aspect of the present invention, step S8, includes the following steps:

[0044] Obtain the threshold limit of the preset distribution of digital red line safety zone, compare the deviation rating quantification score with the threshold limit, and obtain the early warning response level;

[0045] Subtracting the standard state transfer function baseline matrix from the characteristic signature of the electro-thermal transfer function yields the difference matrix;

[0046] The gradient direction of the difference matrix is ​​extracted to obtain the multidimensional partial derivative orientation trajectory. The multidimensional partial derivative orientation trajectory is then matched with the pre-stored failure feature template to obtain the precise location data of the precursor physical degradation components.

[0047] By structuring and assembling the early warning response level with the precise location data of precursory physical degradation components, a graded safety early warning instruction is obtained.

[0048] Secondly, the present invention provides a charging host safety early warning system that integrates multi-source sensor data, comprising the following modules:

[0049] The micro-perturbation sequence generation module obtains the steady-state operating parameters of the control loop of the charging host, performs modulation and coding operations, and obtains the coded micro-perturbation sequence.

[0050] The composite signal injection module acquires the coded micro-perturbation sequence, performs signal aliasing injection operation, and obtains the composite modulation drive signal;

[0051] The synchronous response acquisition module acquires the sensor network sampling data under the excitation of the composite modulation driving signal, performs synchronous collection operation, and obtains a multi-dimensional raw response data stream.

[0052] The response signal purification module acquires multi-dimensional raw response data streams and encoded micro-perturbation sequences, performs decomposition and purification operations, and obtains cross-domain response signals.

[0053] The feature signature generation module acquires the cross-domain response signal, performs feature reconstruction operation, and obtains the feature signature of the electro-thermal transfer function;

[0054] The baseline matrix retrieval module obtains real-time ambient temperature parameters and real-time power load ratio, performs baseline matching operation, and obtains the standard state transfer function baseline matrix.

[0055] The deviation rating calculation module obtains the feature signature of the electro-thermal transfer function and the baseline matrix of the standard state transfer function, performs multiple spatial distance fusion comparison operations, and obtains the deviation rating quantitative score.

[0056] The safety warning instruction generation module obtains the deviation rating quantification score, performs a warning instruction reorganization operation, and obtains graded safety warning instructions.

[0057] In summary, the present invention has the following beneficial technical effects:

[0058] 1. By actively injecting coded micro-perturbation sequences into the digital control loop and combining this with software lock-in amplification technology, the multi-dimensional raw response data stream is processed and purified. This effectively suppresses broadband harmonics and environmental interference, separating the cross-domain response signal. This active detection method improves upon the limited signal-to-noise ratio of passive monitoring and enhances the detection sensitivity for early latent physical degradation characteristics.

[0059] 2. By analyzing the gain and bias phase shift characteristics in the energy transfer path, different degradation mechanisms such as insulation degradation and thermal conductivity decline can be effectively distinguished, thereby providing data support for the accurate location of precursory physical degradation components and improving the accuracy of fault location.

[0060] 3. By comparing the characteristic signature of the electro-thermal transfer function with the baseline matrix using spatial distance differentiation, normal operating condition drift is separated from irreversible physical degradation. This adaptive matching mechanism reduces the probability of false alarms caused by environmental fluctuations, ensures that the deviation rating quantification score can objectively reflect the inherent health status of the equipment, and improves the reliability and adaptability of the safety early warning method. Attached Figure Description

[0061] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. The drawings are used to provide a further understanding of the present invention.

[0062] Figure 1 A flowchart illustrating an embodiment of this application is disclosed.

[0063] Figure 2 A waveform diagram showing the change of duty cycle adjustment value over time in an embodiment of this application is disclosed.

[0064] Figure 3 A schematic diagram of the temperature dimension weighted Euclidean distance spatial distribution tolerance line in an embodiment of this application is disclosed.

[0065] Figure 4 Structural schematic diagrams of embodiments of this application are disclosed. Detailed Implementation

[0066] See attached document Figure 1 This invention proposes a safety early warning method for charging host that integrates multi-source sensor data, comprising the following steps:

[0067] S1. Obtain the steady-state operating parameters of the control loop of the charging host, perform modulation and coding operation, and obtain the coded micro-perturbation sequence;

[0068] S2. Obtain the coded micro-perturbation sequence, perform signal aliasing injection operation, and obtain the composite modulation drive signal;

[0069] S3. Acquire sensor network sampling data under the excitation of composite modulation driving signal, perform synchronous collection operation, and obtain multi-dimensional raw response data stream;

[0070] S4. Obtain the multi-dimensional raw response data stream and encoded micro-perturbation sequence, perform the solution and purification operation to obtain the cross-domain response signal;

[0071] S5. Obtain the cross-domain response signal, perform feature reconstruction operation, and obtain the feature signature of the electro-thermal transfer function;

[0072] S6. Obtain the real-time ambient temperature parameters and real-time power load ratio, perform baseline matching operation, and obtain the standard state transfer function baseline matrix.

[0073] S7. Obtain the feature signature of the electro-thermal transfer function and the baseline matrix of the standard state transfer function, perform multiple spatial distance fusion comparison operations, and obtain the deviation rating quantitative score;

[0074] S8. Obtain the deviation rating quantification score, execute the early warning instruction reorganization operation, and obtain the graded safety early warning instruction.

[0075] The underlying control loop operating parameters of the charging host are obtained, and combined with the underlying pulse width modulation control reference rules, the original digital pulse waveform carrying the amplitude and frequency information of the preset test frequency band is generated, and the coded micro-perturbation sequence in a strict cross-domain time synchronization state is output.

[0076] The micro-disturbance injection module inside the system continuously monitors and reads the operating data frames of the main control module of the charging host through the local area network bus, and parses the steady-state operating parameters of the control loop from them. These steady-state operating parameters are the basis for determining whether the charging host is in a suitable operating condition for active detection. Specifically, these parameters include the DC bus voltage being stable within ±1% of the rated value, the output current fluctuation being less than 0.5% of the rated value, and the temperature gradient change rate of all power devices being less than 0.1℃ per second, which are used to determine whether the system is in a stable power output state.

[0077] Once the steady-state conditions are confirmed to be met, the injection module will lock the bus reference operating frequency used by the current power control node. The frequency range is typically set between 30kHz and 80kHz. This range is chosen because it balances the switching losses of wide-bandgap power devices in megawatt-level charging piles with the size and weight requirements of magnetic components. If the frequency is below 30kHz, the internal transformers and inductors will become too large and expensive; if the frequency is above 80kHz, it will increase switching losses and the heat dissipation pressure on the thermal management system. The micro-perturbation injection module sets the modulation coefficient for the test frequency band. It is a dimensionless positive integer, and its setting must ensure the testing frequency. Operating frequency with bus reference The low-order harmonics have a sufficiently wide protection interval, typically set between 15 and 50. This setting is based on two principles: firstly, it ensures that the generated test fundamental frequency is much higher than the main electromagnetic interference frequency band on the main power control loop line, usually concentrated within the 15th harmonic, thus avoiding low-frequency grid-side harmonic interference; secondly, limited by the highest clock frequency of the controller's on-chip peripherals and the sampling processing power of the digital signal processor (DSP), the modulation coefficient should not exceed 50 to prevent the microprocessor from experiencing task overload or crashing due to excessively high interrupt frequencies. The test signal period is thus determined. Operating frequency based on bus reference Modulation coefficients of the test band The decision was made jointly, and the calculation formula is as follows:

[0078]

[0079] The micro-perturbation injection module calls the digital signal generator and uses coefficients. Generates a basic waveform of Barker code with preset encoding and decoding rules. The waveform is selected from binary sequences with sharp autocorrelation peaks to facilitate pulse compression and recognition in subsequent signal processing. This embodiment uses a 13-bit Barker code. In the formula... The Barker code is an index for an integer sequence that is counted starting from 0. It is preset to contain only a finite-length sequence of numbers with a value of +1 or -1.

[0080] To ensure that the injected disturbance does not affect the power transmission of the charging host, the injection module applies a preset amplitude gain weight to the Barker code fundamental waveform. , It is a dimensionless constant and a key parameter determining the intensity of the disturbance. Its value is determined by the need to ensure that the energy of the injected signal is sufficient to be detected in the subsequent phase-locked amplification stage, while also ensuring that its disturbance to the main circuit is small enough not to affect charging safety and efficiency. For example, assuming the duty cycle adjustment resolution is 16 bits, then... The typical value range is 0.0005 to 0.002. The amplitude of the Barker code underlying waveform is scaled to a minimum level through scalar multiplication, thereby generating an initial perturbation signal stream with an unaligned digital timestamp.

[0081] To achieve system-wide time synchronization, the injection module, acting as a slave node, listens for and captures the global clock synchronization broadcast pulse frames periodically sent by the master control module to all subordinate functional nodes via the network bus. The micro-perturbation injection module uses the rising edge of the global clock synchronization broadcast pulse frame as a trigger reference to phase-align the starting point of the initial perturbation signal stream with its misaligned digital timestamps, and assigns a timestamp anchor point to each data point in the sequence. The unit is Output encoded micro-perturbation sequence At the synchronization moment when the global clock synchronization broadcast pulse frame is captured. After that, the first in the sequence The logic for generating each data point can be defined by the formula: .

[0082] In a specific embodiment, assuming the current charging host is operating stably, the bus reference operating frequency reported by its main control module... The frequency is 50kHz. The micro-perturbation injection module sets the modulation coefficient of the test frequency band according to its internal rules. The value is 20. Therefore, the calculated reference frequency of the disturbance signal is... for The corresponding sampling period The duration is 1 μs. The micro-perturbation injection module selects a 13-bit Barker code as the basic waveform of the Barker code. The sequence is {+1,+1,+1,+1,+1,-1,-1,+1,+1,-1,+1,-1,-1,+1}.

[0083] The module sets the amplitude gain weight according to safety regulations. The value is 0.001. The generated initial perturbation signal stream with unaligned timestamps is a sequence of 13 values: {+0.001,+0.001,+0.001,+0.001,+0.001,-0.001,-0.001,+0.001,+0.001,-0.001,+0.001,-0.001,+0.001,-0.001,+0.001}.

[0084] At a certain point in the system's operation, for example The micro-perturbation injection module captures the global clock synchronization broadcast pulse frame and, using this time as a starting point, binds the aforementioned numerical sequence with a timestamp. The first data point of the output encoded micro-perturbation sequence is... The second data point is And so on, the thirteenth data point is This time-anchored sequence will serve as the input source for subsequent injection stages.

[0085] To securely inject the coded micro-perturbation sequence into the digital control system, it is connected to the microcontroller logic link within the target power control node. The injection control module performing this step establishes an adder function entry point in the microprocessor firmware code storage area at the core of the microcontroller logic link, using a pre-defined hook function or a direct memory access channel. This entry point constitutes a secure channel. The data stream of the coded micro-perturbation sequence is guided through this secure channel and loaded into the logic layer memory buffer of the digital isolation barrier, awaiting aliasing with the main control signal.

[0086] In each pulse width modulation control cycle, the microprocessor, based on its internal closed-loop control algorithm (such as proportional-integral-derivative control), periodically calculates and outputs a steady-state duty cycle adjustment value according to the energy transfer requirements of the DC reactance stage. This value is the value of the microprocessor's main control algorithm at time [time value missing]. The output reference control quantity is a dimensionless parameter, which is dynamically calculated by the microprocessor based on the real-time feedback voltage and current values. Its value directly determines the proportion of the conduction time of the power switching device in one switching cycle.

[0087] Before sending this value to the pulse width modulation generator, a forced saturation limiting logic is set based on the anti-pulse truncation protection rule. This is to prevent the duty cycle value after superimposed disturbance from exceeding the hardware's allowable range, such as between 5% and 95%. The logic calls the adder function located in the digital isolation barrier logic layer to set the current value. The dimensionless scalar sampling amplitude of the corresponding encoded micro-perturbation sequence With steady-state duty cycle adjustment value Perform pointwise linear algebraic aliasing in the time domain.

[0088] The combined square wave sequence generated after addition is amplitude-limited before being sent out to ensure that it does not exceed the minimum safe duty cycle limit set by the pulse width modulation generator. With the maximum safe duty cycle limit Both physical limits and their values ​​are dimensionless constants, set based on the physical characteristics of power semiconductor devices and system safety specifications. For example, they are typically set as follows: It is 0.95. The value is 0.05. The mathematical implementation process of this linear algebraic aliasing and anti-pulse truncation protection rule generates the composite duty cycle adjustment value corresponding to the integrated square wave sequence. It can be expressed by the following formula:

[0089]

[0090] In the formula, and These represent the standard mathematical functions for taking the minimum and maximum values, respectively. This integrated square wave sequence, after linear algebraic aliasing and amplitude limiting, is sent to the pre-buffer stage, which then controls the driver chip. This causes the connected wide-bandgap semiconductor power devices to perform nanosecond-level switching according to the specific superposition timing defined by this integrated square wave sequence. The high-voltage square wave drive signal generated by this switching is the output composite modulation drive signal, containing a test echo probe factor.

[0091] In a specific implementation, the coded micro-perturbation sequence needs to be injected into the main control loop. Assuming the current charging host is in constant power charging mode, the microprocessor calculates the steady-state duty cycle adjustment value. It remained stable at 0.65. Maximum duty cycle. The minimum duty cycle is 0.95. It is 0.05. At time... The injection control module reads the corresponding amplitude from the encoded micro-perturbation sequence. At this point, linear algebra aliasing is performed, resulting in a temporary duty cycle value. This value is between 0.05 and 0.95, so no protection rule needs to be triggered. Therefore, the composite duty cycle adjustment value output at this moment... That is, 0.651. This value is fed into the pulse width modulation generator. For example, at time... The corresponding coded micro-perturbation sequence amplitude is The temporary duty cycle value is This value is also within the safe range, therefore the output composite duty cycle adjustment value... The value is 0.649. These tiny variations in the composite duty cycle adjustment value sequence are amplified by the driver chip to generate a composite modulation drive signal to control the switching action of the power device.

[0092] See appendix Figure 2 As shown, when the temporary duty cycle value after superimposed micro-perturbation exceeds the preset maximum safe duty cycle limit or falls below the minimum safe duty cycle limit, the system automatically cuts off the over-limit part, always limiting the black solid line within the safe range formed by the two horizontal dotted lines, which reflects the protection of the system's safety bottom line in the injection signal aliasing operation of this invention.

[0093] To capture the transient response of the charging host to the composite modulation drive signal, the system utilizes a global clock that runs throughout the entire system to synchronously broadcast pulse frames, coordinating the sampling actions of the heterogeneous multi-source sensor network deployed within the equipment chassis. This process is executed by the synchronous acquisition control module. This module broadcasts wake-up and synchronization commands to all sampling terminal hardware connected to the sensor network interface nodes. Upon receiving this command, all sampling terminal hardware detaches its autonomously running internal clock beat and instead locks onto and reads the time counter within the global clock synchronous broadcast pulse frame, using this as the sole time reference for all subsequent sampling actions.

[0094] The acquisition and control module, at each uniform tick moment defined by the time counter, is a discrete sampling time point defined by the global clock synchronous broadcast pulse frame, and its sampling frequency... The Nyquist sampling theorem must be satisfied, and the frequency must be at least the reference frequency of the encoded micro-perturbation sequence. The oversampling rate is typically set to 2 to 5 times the reference frequency. This setting is based on the fact that if only twice the frequency is used, the sampling points are prone to falling into the zero-crossing points of the AC signal, resulting in the complete loss of characteristic amplitude. Using an oversampling rate of 3 to 5 times not only ensures that enough waveform feature points are captured within one cycle for subsequent orthogonal calculations, but also fully matches the throughput limit of existing mainstream industrial-grade high-speed analog-to-digital converter chips, balancing signal fidelity and hardware cost.

[0095] On the one hand, the synchronous acquisition control module controls the analog-to-digital converters inside the current transformer module and the bus voltage isolation measurement module to perform parallel and rapid analog-to-digital conversion at a unified clock speed. This allows the acquisition of alternating current ripple data and voltage fluctuation data generated by the main circuit electrical loop under the excitation of the composite modulation drive signal, which are at the same frequency as the test signal. In other words, it is a weak alternating signal component separated from the DC component of the main circuit, whose frequency is synchronized with the reference frequency of the micro-disturbance sequence.

[0096] On the other hand, at the same unified clock cycle, the synchronous acquisition and control module controls the infrared temperature probe and the brushless impeller sensor embedded in the liquid-cooled circulation pipeline to collect their respective status information. This acquires the heat transfer temperature gradient sequence of key heat-generating components when the heat path undergoes minute changes on a millisecond timescale, as well as the fluid dynamic pressure change data caused by slight changes in pipeline resistance due to thermal expansion and contraction of the pipeline material or changes in coolant viscosity. Furthermore, the synchronous acquisition and control module merges and encapsulates data from all heterogeneous sampling channels point-by-point based on their shared timestamps, forming a time-aligned multidimensional data frame sequence—this sequence is the multidimensional raw response data stream.

[0097] In a specific embodiment, the composite modulation drive signal continuously excites the power circuit of the charging host, which contains a reference frequency of [frequency value missing]. The probe factor. The system's synchronous acquisition control module sets the sampling frequency. for That is, a uniform beat time is defined every 0.5μs. In time... At this unified timing, the synchronous acquisition control module simultaneously triggers all sensors to acquire data. The current transformer module completes analog-to-digital conversion and reports a current ripple reading of +5.1mA. Simultaneously, the bus voltage isolation measurement module reports a voltage fluctuation of +12.3mV.

[0098] At the same moment, an infrared temperature probe deployed near a power device collected an instantaneous temperature of 65.21℃. This data point will be used to calculate the heat transfer temperature gradient sequence. The brushless impeller sensor in the liquid cooling pipeline reported a rotational speed pulse frequency of 5133Hz, which, after table lookup, corresponds to a fluid dynamic pressure change of 1.05kPa. The synchronous acquisition control module encapsulates these data into a single data tuple: (10.123458000s, +5.1mA, +12.3mV, 65.21℃, 1.05kPa).

[0099] This process generates a continuous stream of multidimensional raw response data every 0.5 μs and transmits it to the subsequent data processing unit.

[0100] To separate the response signal strongly correlated with the injected excitation from the multi-dimensional raw response data stream, a software-based lock-in amplification (LLA) solution process is employed. The data processing unit controls each data channel in the multi-dimensional raw response data stream, such as the current ripple data channel, which flows through a first-stage digital deep-gain notch filter. This filter is a digital filter capable of generating an extremely deep-gain notch at the target frequency, with a quality factor of [missing information]. The Q value is typically set between 20 and 50 to ensure narrowband filtering. This range is chosen because: if the Q value is below 20, the filter's stopband is too wide, potentially attenuating useful spectral energy near the target detection signal while filtering out power frequency interference; if the Q value is above 50, although the notch is extremely narrow, it results in an excessively long transient response time in the time domain, failing to meet the low-latency requirement for millisecond-level real-time safety warnings under abnormal charging conditions. This filter incorporates multiple notch frequencies, targeting the power grid's fundamental frequency and its main odd harmonic frequencies, such as 50Hz, 150Hz, and 250Hz. It aims to forcibly remove strong periodic interference waveforms caused by transformer fundamental frequency leakage flux and power harmonic components injected from the grid side. After filtering, the first... The discrete values ​​of each sampling point, i.e., the input data sequence after noise reduction. .

[0101] The digital calculator internally allocates a virtual computing memory pool for each data channel. The system uses the coded micro-perturbation sequence generated and stored by S1 as a reference, feeding it to the external reference channel inlet of the virtual computing memory pool. Upon receiving the reference sequence, the quadrature-cosine generator inside the memory pool is triggered at the reference frequency. Two locally orthogonal contrast waveforms, strictly orthogonal in time, are generated for the core frequency, and their amplitudes are normalized to 1, i.e., in-phase component waveforms. Orthogonal component waveforms .

[0102] The data processing unit processes the noise-reduced input data sequence. Compare the waveforms with the two local orthogonal waveforms. , Perform point-by-point multiplication and summation operations. This operation is equivalent to a low-pass filter and is essentially a combination of digital mixing and low-pass filtering. It involves shifting the signal spectrum to near zero frequency and then integrating and averaging, using the total number of sampling points in the summation operation. Processing for the equivalent bandwidth factor can filter out random industrial space noise on the time coordinate set that is not correlated with the reference signal, and separate the cross-domain response signal from the strong noise background. This signal is a variable containing the amplitude of the test frequency. and characteristic phase delay parameter Complex data pairs. Demodulate the in-phase components. and orthogonal components Both units are consistent with the units of the input response signal, and their calculation formula is as follows:

[0103]

[0104]

[0105] based on , The component, and its output calculation formula is as follows: as well as .in, It is a bivariate arctangent function, whose original output is in radians, and is multiplied by... The results are correctly mapped to angular values ​​in the four quadrants.

[0106] In a specific embodiment, the data processing unit receives a sampling frequency of The multi-dimensional raw response data stream. Taking the current ripple data channel as an example, assuming its signal... It includes a 4.5mA amplitude and a 45° phase delay. The target signal is superimposed with a 50Hz power frequency interference with an amplitude of 60mA and other random noise. This data stream passes through a first-stage digital deep notch filter, whose 50Hz notch point suppresses the power frequency interference amplitude to below 0.1mA, resulting in the noise-reduced input data sequence. In the virtual computing memory pool, with Local orthogonal contrast waveforms were generated for the reference frequency. and ,in μs. Set the number of accumulated points. The value is 2 million, corresponding to an integration time of 1 second.

[0107] Will respectively with and The calculation result will converge to: Calculate the amplitude variable of the test frequency. mA. The characteristic phase delay parameter is calculated. The output response signal is a data pair: {amplitude: 4.5mA, phase: 45°}.

[0108] To extract deep characteristics representing the physical properties of the device from the cross-domain response signal, the system dissects the signal and reconstructs the internal energy transfer dynamics function describing the state transition of the component. The data processing unit starts from the... One response channel (including) Extract the amplitude component from the complex data pairs of the cross-domain response signal output from each channel, i.e., the amplitude variable at the test frequency. Retrieve the reference intensity scalar of the transmission origin corresponding to each injection channel, which was fixed and saved when the S1 injection signal was injected. That is, the dimensionless perturbation amplitude weight set in S1 This serves as a reference for subsequent channel gain calculations. By performing calculations, the effective node transmission cross-domain gain from the excitation injection point to each response acquisition point is calculated. It inherits the physical dimensions of the response channel, and the calculation formula is:

[0109]

[0110] Extracting the characteristic phase delay parameter of the cross-domain response signal To isolate the anomalous phase shift caused by equipment physical degradation, the spatial physical hindrance constant between the excitation and response points is retrieved from a pre-defined equipment hardware wiring structure parameter library. That is, the length of the signal propagation path and the actual signal transmission speed in the propagation medium The time delay in the decision The speed is determined by both the speed of light in a vacuum and the dielectric constant of the internal conductive material, and is typically the product of its wave velocity factor, approximately... m / s to Between m / s This constant is a static parameter stored in the device configuration file, reflecting the inherent time delay of the signal. It is combined with the reference frequency of the excitation signal. This time delay is converted into a fixed reference phase delay, i.e. Then, this reference portion is subtracted from the total characteristic phase delay parameter to deduce the bias phase shift angle caused by non-ideal factors such as a slight decrease in the insulation resistance of the internal isolation material or micro-oxidation at the connector interface. The calculation formula is:

[0111]

[0112] The system integrates these dynamic characteristic parameters in matrix form. The effective node transmission cross-domain gain rate calculated for each channel is set as the diagonal skeleton of the transfer tensor. The diagonal elements reflecting the cross-domain energy transfer efficiency at the test frequency form the skeleton describing the energy transfer efficiency. The bias phase shift angle is then used as a key parameter to fill in the auxiliary factor matrix parameter set, revealing the off-diagonal elements that reveal the coupling effects between different transmission paths caused by insulation degradation or parasitic capacitance changes. In this way, the system ultimately generates a signature of the electro-thermal transfer function that can describe early defects such as invisible thermal resistance and leakage current barriers. ,That The form of a square matrix can be represented as:

[0113]

[0114] diagonal elements off-diagonal elements (for ).

[0115] In a specific embodiment, consider a system with two response channels: channel 1 for current ripple and channel 2 for voltage fluctuation. Continuing from the example result in S4, the response signal for channel 1 is {amplitude: 4.5mA, phase: 45°}. Assume the response signal extracted from channel 2 is {amplitude: 11.2mV, phase: 50°}. Backtracking reveals that the amplitude of the injected signal at the source control layer is set in S1. Calculate the effective node transmission cross-domain gain rate for each channel. Gain of Channel 1. Gain of Channel 2 Next, the bias phase shift angle is calculated.

[0116] Assuming the physical path length of the signal in channel 1 is 1.0m, the signal transmission speed is set according to the material of the equipment. The inherent time delay is found to be 5ns, and the calculated reference phase delay is... .

[0117] The parasitic phase shift of channel 1 is then... Assuming the physical path of the signal in channel 2 is set to approximately 1.39m, and the transmission speed is the same... The inherent time delay is found to be 6.94 ns, and its reference phase delay is... The parasitic phase shift of channel 2 is then... Construct a 2x2 signature for the electro-thermal transfer function. The diagonal elements represent the gain, i.e. , The off-diagonal elements are the average values ​​of the parasitic phase shift angles, i.e. The final generated feature signature matrix is: This matrix encapsulates the gain and phase response characteristics of the device at the current excitation frequency.

[0118] To provide a healthy baseline for subsequent deviation comparisons, the baseline retrieval module obtains the real-time ambient temperature parameters and real-time power load ratio at the actual physical interface. The baseline retrieval module continuously reads the real-time ambient temperature parameters generated by the wide-temperature thermocouples deployed at the main air intake of the chassis, i.e., the current ambient temperature, through an analog-to-digital conversion channel. Meanwhile, this module communicates with the main charging control computer via an internal local area network bus, parsing the real-time power load percentage requested by the vehicle's battery management system after interaction with the charging protocol. This percentage, expressed as a percentage from 0% to 100%, reflects the current energy output level of the charging host and is denoted as […]. .

[0119] After acquiring these two key operating parameters, the baseline retrieval module maps continuous ambient temperature values ​​and load ratio values ​​to discrete interval codes according to preset quantization rules. The two codes are then concatenated to generate a multi-way tree hash index code that provides dual classification identification for both environment and load. This multi-way tree hash index code serves as the primary key for efficient retrieval, accessing a read-only parameter template clustering library stored in an external high-speed flash memory array. This is particularly useful before equipment shipment, when a large set of health status feature signature data is pre-collected and stored in an environmental laboratory under controlled temperature, humidity, and load conditions and active stimulation tests. This code can quickly locate the target candidate subset of the standard high-dimensional tensor family group pre-determined and validated through tens of thousands of thermal cycling aging cycles within the library.

[0120] Within this subset, the system iterates through each candidate health baseline high-concurrency matrix file record, setting the current real-time ambient temperature and load ratio, along with the reference operating condition parameters associated with each record, as follows: The reference ambient temperature for each candidate baseline file is: The reference load ratio is Weighted Euclidean distance calculations are performed to find the matrix file with the minimum spatial distribution tolerance that best fits the current operating mode. This target baseline selection formula is essentially the index for finding the minimum spatial distribution tolerance. The process can be represented by the following formula:

[0121]

[0122] in, This indicates finding the index argument that minimizes the value of the expression within the parentheses. ; and These are preset weighting coefficients with dimensions corresponding to the inverse square of the operating condition parameters (temperature or load), used to adjust the relative importance of temperature and load in distance calculations. The target matching matrix file records the target data, i.e., the reference baseline matrix corresponding to the aforementioned index, are completely read from the flash array. This is then decoded and restored to a high-dimensional matrix without degradation or distortion; this matrix is ​​the baseline matrix of the standard state transfer function. The matrix's dimensions and the physical meaning of its elements are consistent with the feature signatures generated by S5, ensuring direct element-by-element comparison. And load it into high-speed computing memory.

[0123] In a specific embodiment, assume the system currently measures a real-time ambient temperature of 28°C using a wide-temperature thermocouple, and the real-time power load ratio obtained from the charging control computer is 65%. According to quantization rules, 28°C is mapped to the 5-30°C range and encoded as 10; 65% is mapped to the 50%-75% load range and encoded as 10. These two are concatenated to generate a multi-way tree hash index encoded as 1010. The system uses this encoding 1010 to quickly locate a subset containing 5 candidate baseline files in the read-only parameter template clustering library.

[0124] Read the current working condition vector And compare them with the reference working conditions recorded in these 5 candidate files respectively. Assume that the weights are set to under the current scale. , The reference operating conditions for the five candidate documents are as follows: After calculation, The reference working condition has the smallest weighted distance from the current working condition, and its calculated value is... Therefore, the system locks in the reference operating condition. The bound health baseline file is the target baseline file. The system then reads this file from the external flash array, decodes it to obtain the standard state transfer function baseline matrix, whose content is a matrix with the same structure as the electro-thermal transfer function feature signature: This baseline matrix is ​​loaded into memory, awaiting comparison with the feature signatures generated in real time.

[0125] See appendix Figure 3 The solid line in the figure presents a V-shaped symmetrical distribution with the reference operating condition parameter of 27℃ as the minimum zero point. The punctuation marks in the figure refer to this minimum zero point, used to identify the location of the candidate baseline reference point with the smallest distance. The distance value increases linearly with the degree to which the ambient temperature deviates from 27℃. The slope of this curve fully reflects the effect of the temperature weighting coefficient set in the model, proving that the system can accurately locate and lock the nearest candidate healthy baseline file based on this distance distribution tolerance.

[0126] To quantify the degree of deviation in the health status of the charging host, the system calls the signature of the electro-thermal transfer function, which represents the current aging state. As previously stated, it is in the first... OK The elements corresponding to the column are And compare it with the baseline matrix of the standard state transfer function under healthy conditions. Let the corresponding element be A multi-spatial distance fusion and comparison logic algorithm is executed. It traverses a pre-defined knowledge base of component aging and runaway risk weights stored in local non-volatile memory. This pre-calibrated database associates each element position in the matrix with a specific physical degradation mode and assigns it a data pairing list within a quantified risk weight value. This list is then consulted to index each position in the feature signature matrix. Preset weighting coefficients .

[0127] A heavy proportional penalty function mapping coefficient with the reciprocal of the corresponding physical dimension is applied to the diagonal gain element reflecting the core insulation drying effect to eliminate the dimension and amplify the deviation effect of the critical fault mode. Simultaneously, a conventional proportional smoothing mapping coefficient with the reciprocal of the angle dimension is applied to the off-diagonal coupled phase element reflecting the mechanical damping ionization effect of the ordinary contactor. Within this comprehensive evaluation weight space coordinate environment, the algorithm deconstructs multiple spatial distances characterizing the overall deviation degree by fusing two different distance measurement methods. .

[0128] First, the weighted Manhattan distance is used to assess the sum of deviations, and the weighted Manhattan distance is calculated. At this point, operations are performed at the scalar level in the real number field, and the core calculation formula is as follows:

[0129]

[0130] Secondly, the consistency of the deviation direction is evaluated by using the cosine angle between the two endpoint vectors, and the weighted cosine distance is calculated. This parameter is dimensionless. At this point, the computational logic control module projects the feature signature and baseline matrix onto the complex plane space using Euler's formula, constructing a complex tensor set. and To ensure dimensional uniformity and preserve weight information when summing vectors across features in the complex plane, for a specific extracted element, let the value of the diagonal gain element be... The values ​​of the off-diagonal phase elements are The diagonal elements are mapped to Modulo the dimensionless value after normalization by the reciprocal of the dimensions, with an argument of 0 radians, and the off-diagonal elements are mapped as follows: Assign a constant virtual amplitude with the fundamental angular dimension. and with weights having the reciprocal of the angle. Multiplying transforms the modulus into a dimensionless pure number, and converts the phase shift in degrees to radians as the complex exponential argument. After mapping, since the modulus of all elements has been dimensionless and normalized, the core calculation formula is as follows:

[0131]

[0132] in, To take the real part of a complex number, This represents taking the imaginary part of a complex number. It combines two distance types, where... and It is a dimensionless weighting factor that integrates two distances, satisfying :

[0133]

[0134] Using a pre-defined saturation curve mapping equation, which is an S-shaped function, the aim is to map unbounded distance values ​​to a bounded, intuitively human-based percentage rating range, and to achieve passivation near 0 and 100 points to prevent drastic score jumps. This multi-dimensional spatial distance value... The data is non-linearly distributed across a normalized warning percentage scale ranging from 0 to 100, ultimately outputting an intuitive and precise deviation rating score to measure the evolution of internal media fission buds. Its formula is:

[0135]

[0136] in, The gain coefficient of the saturation curve determines the steepness of the curve at the center point; This is the distance offset, representing when... When this value is equal, the rating score is 50.

[0137] In a specific embodiment, the real-time electro-thermal transfer function feature signature is obtained. and the baseline matrix of the standard state transfer function for health In calculating the weighted Manhattan distance At that time, calculate the weighted scalar difference of each matrix element. Assume the weight library includes the reciprocal of the corresponding units. Weighted Manhattan distance In calculating the weighted cosine distance When elements are combined with weights, they are directly converted into dimensionless complex plane tensors (e.g., diagonal elements are converted to...). off-diagonal elements Convert to Then, perform complex plane dot product operations. Assume that after normalization and weighted fusion, the resulting multi-space distance... Call the saturation curve mapping equation, with parameters set as follows. , Substitute into the formula to calculate: The output deviation rating score is 70.48, indicating that the equipment status has significantly deviated from the healthy baseline and requires attention.

[0138] To transform the deviation rating quantification score into actionable operational decisions, the early warning instruction generation module reads and parses the score. This module quantifies deviation ratings and compares them with internally preset digital red-line safety zone distributions, especially the scoring intervals predefined by maintenance experts according to equipment safety operation procedures, which have threshold limits. and ,For example Set at 70 points. A score of 90 is used for comparison to determine the required early warning response level for the entire power conversion system, which is recorded as the adjudication level. This refers to the operation and maintenance response level directly corresponding to the scoring range: for example, Level 1 is the long-term hidden danger filing and attention layer; Level 2 is the forced power derating to maintain the alarm layer; and Level 3 is the disaster interruption emergency level. The logical expression for this level division is:

[0139]

[0140] Based on this ruling, the modules perform fault root cause localization in parallel. The multidimensional partial derivatives of the electro-thermal transfer function, deviating from the conventional ideal trajectory, are analyzed to determine the trajectory's characteristics. With baseline matrix The difference matrix, mathematically represented as This is identical to the feature signature structure, and the gradient direction of each element in this difference matrix as a function of time or load is analyzed. For example, if only the diagonal elements continuously decrease, it may indicate widespread aging of power devices; if a certain off-diagonal phase element shows a sharp positive drift, it may indicate degradation of the insulation medium at the corresponding connection point. The formula for calculating this deviation feature matrix is:

[0141]

[0142] This high-information-density, dimensionally normalized gradient trajectory feature fingerprint is sent to a large-scale operations and maintenance cloud platform with an established secure connection. Upon receiving this fingerprint, the cloud-based component failure diagnosis and similarity matching model initiates a pre-trained convolutional isoparametric comparison task. This task compares the input gradient trajectory fingerprint with failure feature templates of thousands of known fault modes stored internally, denoted as the i-th... Feature template of a known fault To perform high-dimensional similarity matching, similarity calculation functions such as normalized cross-correlation or cosine similarity are used, denoted as... This allows us to pinpoint the location of the specific physical degradation component causing the current state deviation, and output the fault mode identifier with the highest similarity, denoted as... For example, it could be a liquid cooling microleak or a crystallization fracture at the root of the power module lead bond. This fault identification function is defined as follows:

[0143]

[0144] in, This represents the index of the parameter that maximizes the similarity function within the parentheses. The early warning instruction generation module takes the aforementioned early warning response level and the precise location data of precursory physical degradation components, structures and assembles them, embedding them into a data package file with a fixed layout. This file is then combined to generate a maintenance guide file with a material replacement suggestion list as an attachment, constituting a graded safety early warning instruction. This is a standardized data structure, typically using Extensible Markup Language or JavaScript object representation format, containing complete maintenance information such as early warning level, device ID, fault location, diagnostic description, confidence level, recommended measures, and a list of required spare parts. This instruction is then sent to the designated troubleshooting expert's desktop via the routing protocol of the upper-layer intelligent operation network to trigger subsequent intervention and handover processes such as blocking, de-rated, or emergency shutdown.

[0145] In a specific embodiment, the early warning command generation module received a deviation rating quantification score of 70.48. The distribution status of the system's digital red line safety zone is set as follows: , Since 70≤70.48<90, the system determines the warning response level to be Level 2, i.e., forced power derating to maintain the alarm layer.

[0146] Simultaneously, the module calculates the deviation matrix. The matrix is ​​characterized by a symmetric decrease in the diagonal cross-domain gain elements and a significant symmetric positive drift in the off-diagonal phase elements. The system sends this normalized matrix as a feature fingerprint to the cloud-based component failure diagnosis and similarity matching model. The cloud system's convolutional isoparametric comparison model matches this fingerprint with a fault template library, finding that it has the highest similarity (92%) to fault mode F034, "drying thermal grease between the power module heatsink and the motherboard leading to increased thermal resistance." Based on this, the precursory physical degradation component is precisely located as power module PM-03.

[0147] The integrated information generates a fixed-format data packet file, the core content of which is: {Warning Level: 2; Level Description: "Forced Power Derating Maintenance Alarm"; Fault Location: "Power Module PM-03"; Diagnostic Conclusion: "Partial drying of thermal grease leading to increased thermal resistance"; Confidence Level: "92%"; Recommended Measures: "Reduce the output power of the charging host to 50% and arrange maintenance personnel to carry out repairs within 12 hours"; Attachment: "Maintenance Guide-PM03.pdf"; Bill of Materials: [{Materials No.: "GREASE-7783", Description: "High-performance thermal grease", Quantity: 1}]}. This complete hierarchical safety warning instruction is routed to the upper-level intelligent operation network, triggering the platform's automatic derating instruction issuance and the generation of maintenance work orders.

[0148] See appendix Figure 4 The present invention also proposes a charging host safety early warning system that integrates multi-source sensor data, including the following modules:

[0149] The micro-perturbation sequence generation module obtains the steady-state operating parameters of the control loop of the charging host, performs modulation and coding operations, and obtains the coded micro-perturbation sequence.

[0150] The composite signal injection module acquires the coded micro-perturbation sequence, performs signal aliasing injection operation, and obtains the composite modulation drive signal;

[0151] The synchronous response acquisition module acquires the sensor network sampling data under the excitation of the composite modulation driving signal, performs synchronous collection operation, and obtains a multi-dimensional raw response data stream.

[0152] The response signal purification module acquires multi-dimensional raw response data streams and encoded micro-perturbation sequences, performs decomposition and purification operations, and obtains cross-domain response signals.

[0153] The feature signature generation module acquires the cross-domain response signal, performs feature reconstruction operation, and obtains the feature signature of the electro-thermal transfer function;

[0154] The baseline matrix retrieval module obtains real-time ambient temperature parameters and real-time power load ratio, performs baseline matching operation, and obtains the standard state transfer function baseline matrix.

[0155] The deviation rating calculation module obtains the feature signature of the electro-thermal transfer function and the baseline matrix of the standard state transfer function, performs multiple spatial distance fusion comparison operations, and obtains the deviation rating quantitative score.

[0156] The safety warning instruction generation module obtains the deviation rating quantification score, performs a warning instruction reorganization operation, and obtains graded safety warning instructions.

[0157] The above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; some technical features are equivalently replaced; however, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for safety early warning of a charging host that integrates multi-source sensor data, characterized in that, Includes the following steps: S1. Obtain the steady-state operating parameters of the control loop of the charging host, perform modulation and coding operation, and obtain the coded micro-perturbation sequence; S2. Obtain the coded micro-perturbation sequence, perform signal aliasing injection operation, and obtain the composite modulation drive signal; S3. Acquire sensor network sampling data under the excitation of composite modulation driving signal, perform synchronous collection operation, and obtain multi-dimensional raw response data stream; S4. Obtain the multi-dimensional raw response data stream and encoded micro-perturbation sequence, perform the solution and purification operation to obtain the cross-domain response signal; S5. Obtain the cross-domain response signal, perform feature reconstruction operation, and obtain the feature signature of the electro-thermal transfer function; S6. Obtain the real-time ambient temperature parameters and real-time power load ratio, perform baseline matching operation, and obtain the standard state transfer function baseline matrix. S7. Obtain the feature signature of the electro-thermal transfer function and the baseline matrix of the standard state transfer function, perform multiple spatial distance fusion comparison operations, and obtain the deviation rating quantitative score; S8. Obtain the deviation rating quantification score, execute the early warning instruction reorganization operation, and obtain the graded safety early warning instruction.

2. The charging host safety early warning method according to claim 1, characterized in that, The steady-state operating parameters of the charging host's control loop are obtained, and modulation and coding operations are performed to obtain the coded micro-perturbation sequence, including the following steps: Obtain the bus reference operating frequency from the steady-state operating parameters of the control loop, and generate the test signal period based on the bus reference operating frequency and the preset test band modulation coefficient; Based on the test signal period and the preset Barker code basic waveform, combined with the preset amplitude gain weight, an initial disturbance signal stream is generated, a global clock synchronization broadcast pulse frame is obtained, and the initial disturbance signal stream is phase aligned using the global clock synchronization broadcast pulse frame to obtain the coded micro-disturbance sequence.

3. The charging host safety early warning method according to claim 1, characterized in that, The process of acquiring an encoded micro-perturbation sequence and performing a signal aliasing injection operation to obtain a composite modulation driving signal includes the following steps: Obtain the steady-state duty cycle adjustment value output by the microprocessor, and add the encoded micro-perturbation sequence to the steady-state duty cycle adjustment value to obtain the temporary duty cycle value; Obtain the preset maximum safe duty cycle limit and minimum safe duty cycle limit, take the maximum value between the temporary duty cycle value and the minimum safe duty cycle limit, and then take the minimum value between the obtained maximum value and the maximum safe duty cycle limit for amplitude limiting to obtain the composite duty cycle adjustment value. The composite duty cycle adjustment value is converted into a square wave sequence to obtain the composite modulation drive signal.

4. The charging host safety early warning method according to claim 1, characterized in that, Acquire sensor network sampling data under composite modulation driving signal excitation, perform synchronous collection operation to obtain multi-dimensional raw response data stream, including the following steps: Obtain the time-stamped counter within the global clock synchronization broadcast pulse frame, and according to the unified beat time defined by the time-stamped counter, obtain the current ripple data and voltage fluctuation data generated by the main circuit electrical circuit excited by the composite modulation drive signal; According to the unified beat time, the heat transfer temperature gradient sequence of the heating device and the fluid dynamic pressure change data of the pipeline are acquired. The current ripple data, voltage fluctuation data, heat transfer temperature gradient sequence and fluid dynamic pressure change data are merged and packaged to obtain a multi-dimensional raw response data stream.

5. A charging host safety early warning method integrating multi-source sensor data according to claim 1, characterized in that, The process of acquiring multi-dimensional raw response data streams and encoded micro-perturbation sequences, performing computational purification operations, and obtaining cross-domain response signals includes the following steps: Notch filtering is applied to the multi-dimensional raw response data stream to obtain the noise-reduced input data sequence. The reference frequency of the encoded micro-perturbation sequence is obtained, and in-phase and quadrature component waveforms are generated based on the reference frequency. The input data sequence is multiplied and accumulated with the in-phase component waveform and the quadrature component waveform respectively to obtain the in-phase component and the quadrature component. Based on the in-phase component and the quadrature component, the sum of squares and the square root operation are performed to obtain the test frequency amplitude variable. The characteristic phase delay parameter is obtained by performing arctangent operation on the in-phase and quadrature components. The cross-domain response signal is obtained by combining the test frequency amplitude variable and the characteristic phase delay parameter.

6. A charging host safety early warning method integrating multi-source sensor data according to claim 5, characterized in that, To acquire the cross-domain response signal, perform feature reconstruction operations, and obtain the feature signature of the electro-thermal transfer function, the following steps are included: Obtain the pre-saved transmission origin reference strength scalar, divide the test frequency amplitude variable by the transmission origin reference strength scalar, and obtain the effective node transmission cross-domain gain rate. Obtain the preset spatial physical hindrance constant, convert the spatial physical hindrance constant into a reference phase delay based on the reference frequency, and subtract the reference phase delay from the characteristic phase delay parameter to obtain the offset phase shift angle; By using the effective node transmission cross-domain gain rate as the diagonal element and the bias phase shift angle as the off-diagonal element, a matrix is ​​constructed to obtain the characteristic signature of the electro-thermal transfer function.

7. A charging host safety early warning method integrating multi-source sensor data according to claim 1, characterized in that, To obtain real-time ambient temperature parameters and real-time power load ratio, perform baseline matching to obtain the standard state transfer function baseline matrix, including the following steps: Obtain real-time ambient temperature parameters and real-time power load ratio; The real-time ambient temperature parameters and real-time power load ratio are mapped to discrete interval codes and concatenated to obtain a multi-way tree hash index code. Use multi-way tree hash index encoding to locate a subset of candidate baseline files in a pre-defined read-only parameter template clustering library; Calculate the weighted Euclidean distance between the real-time ambient temperature parameters and the real-time power load ratio and the reference operating condition parameters within the subset of candidate baseline files, and select the candidate baseline file with the smallest distance; Decode the selected candidate baseline file to obtain the standard state transfer function baseline matrix.

8. A charging host safety early warning method integrating multi-source sensor data according to claim 1, characterized in that, The process involves obtaining the feature signature of the electro-thermal transfer function and the baseline matrix of the standard state transfer function, performing multiple spatial distance fusion comparison operations, and obtaining a deviation rating quantification score, including the following steps: Convert the electro-thermal transfer function feature signature and the standard state transfer function baseline matrix into a complex tensor set; Obtain the weight coefficients from the pre-set knowledge base of risk weights for aging and runaway components, and calculate the weighted Manhattan distance and weighted cosine distance between complex tensor groups based on the weight coefficients; The weighted Manhattan distance and the weighted cosine distance are weighted and fused to obtain a multi-spatial distance. The multi-spatial distance is then input into a preset saturation curve mapping equation based on the arctangent function for nonlinear mapping to obtain a deviation rating quantification score.

9. A charging host safety early warning method integrating multi-source sensor data according to claim 1, characterized in that, Obtain the deviation rating quantification score, execute the early warning instruction reorganization operation to obtain the graded safety early warning instruction, including the following steps: Obtain the threshold limit of the preset distribution of digital red line safety zone, compare the deviation rating quantification score with the threshold limit, and obtain the early warning response level; Subtracting the standard state transfer function baseline matrix from the characteristic signature of the electro-thermal transfer function yields the difference matrix; The gradient direction of the difference matrix is ​​extracted to obtain the multidimensional partial derivative orientation trajectory. The multidimensional partial derivative orientation trajectory is then matched with the pre-stored failure feature template to obtain the precise location data of the precursor physical degradation components. By structuring and assembling the early warning response level with the precise location data of precursory physical degradation components, a graded safety early warning instruction is obtained.

10. A charging host safety early warning system integrating multi-source sensor data, characterized in that, Includes the following modules: The micro-perturbation sequence generation module obtains the steady-state operating parameters of the control loop of the charging host, performs modulation and coding operations, and obtains the coded micro-perturbation sequence. The composite signal injection module acquires the coded micro-perturbation sequence, performs signal aliasing injection operation, and obtains the composite modulation drive signal; The synchronous response acquisition module acquires the sensor network sampling data under the excitation of the composite modulation driving signal, performs synchronous collection operation, and obtains a multi-dimensional raw response data stream. The response signal purification module acquires multi-dimensional raw response data streams and encoded micro-perturbation sequences, performs decomposition and purification operations, and obtains cross-domain response signals. The feature signature generation module acquires the cross-domain response signal, performs feature reconstruction operation, and obtains the feature signature of the electro-thermal transfer function; The baseline matrix retrieval module obtains real-time ambient temperature parameters and real-time power load ratio, performs baseline matching operation, and obtains the standard state transfer function baseline matrix. The deviation rating calculation module obtains the feature signature of the electro-thermal transfer function and the baseline matrix of the standard state transfer function, performs multiple spatial distance fusion comparison operations, and obtains the deviation rating quantitative score. The safety warning instruction generation module obtains the deviation rating quantification score, performs a warning instruction reorganization operation, and obtains graded safety warning instructions.

Images