Power allocation method and system of liquid cooling charging pile

Abstract

The invention relates to the technical field of power allocation, and discloses a power allocation method and system for a liquid cooling charging pile, and the method comprises the steps: collecting the charging power data, liquid cooling temperature data, charging voltage data, charging current data and liquid cooling flow data of the liquid cooling charging pile, and carrying out the signal processing, obtaining a charging operation standard data set; performing non-uniform time delay characteristic analysis on the charging operation standard data set, and calculating time sequence correlation through a joint mutual information criterion to obtain charging power time characteristic data; inputting the charging power time characteristic data into a charging power prediction model for charging power prediction to obtain charging power prediction data; and performing proportional robust diffusion recursion operation on the charging power prediction data, and performing parameter optimization in combination with a hyperbolic cosine function to obtain a dynamic power control parameter. According to the invention, the robustness of power allocation is improved, and the power allocation of the liquid cooling charging pile is more accurate and efficient.

Description

Technical Field

[0001] The present invention relates to the technical field of power allocation, and in particular, to a power allocation method and system for a liquid-cooled charging pile. Background Art

[0002] In the actual operation process, the power allocation of liquid-cooled charging piles faces many challenges. Traditional power allocation methods mainly rely on fixed control strategies and cannot effectively cope with the dynamic changes during the charging process, resulting in low charging efficiency and unable to ensure charging safety.

[0003] Currently, there are generally problems such as insufficient signal acquisition accuracy, lag in data processing, and inaccurate feature extraction in the power allocation of liquid-cooled charging piles. Especially in high-power charging scenarios, due to drastic temperature changes and frequent current fluctuations, traditional power allocation methods are difficult to accurately capture the dynamic characteristics of the system, easily causing uneven power distribution and affecting charging efficiency and the service life of the equipment. Summary of the Invention

[0004] The present invention provides a power allocation method and system for a liquid-cooled charging pile, which improves the robustness of power allocation and makes the power allocation of the liquid-cooled charging pile more accurate and efficient.

[0005] In the first aspect, the present invention provides a power allocation method for a liquid-cooled charging pile, and the power allocation method for the liquid-cooled charging pile includes: Collect the charging power data, liquid-cooled temperature data, charging voltage data, charging current data, and liquid-cooled flow rate data of the liquid-cooled charging pile, and perform signal processing through a multiple-delay signal cancellation pre-filter to obtain a charging operation standard data set; Perform non-uniform time-delay feature analysis on the charging operation standard data set, and calculate the time series correlation through the joint mutual information criterion to obtain charging power time feature data; Input the charging power time feature data into a charging power prediction model for charging power prediction to obtain charging power prediction data; Perform a proportional robust diffusion recursive operation on the charging power prediction data, and optimize the parameters in combination with the hyperbolic cosine function to obtain dynamic power control parameters.

[0006] In the second aspect, the present invention provides a power allocation system for a liquid-cooled charging pile, and the power allocation system for the liquid-cooled charging pile includes: A collection module, configured to collect the charging power data, liquid-cooled temperature data, charging voltage data, charging current data, and liquid-cooled flow rate data of the liquid-cooled charging pile, and perform signal processing through a multiple-delay signal cancellation pre-filter to obtain a charging operation standard data set; An analysis module for performing non-uniform time-delay feature analysis on the charging operation standard data set, calculating the time series correlation degree through the joint mutual information criterion, and obtaining the charging power time feature data; A prediction module for inputting the charging power time feature data into a charging power prediction model to predict the charging power and obtaining the charging power prediction data; An optimization module for performing proportional robust diffusion recursive operation on the charging power prediction data and optimizing the parameters in combination with the hyperbolic cosine function to obtain the dynamic power control parameters.

[0007] The third aspect of the present invention provides a power allocation device for a liquid-cooled charging pile, including: a memory and at least one processor, wherein instructions are stored in the memory; the at least one processor calls the instructions in the memory to enable the power allocation device of the liquid-cooled charging pile to execute the above-mentioned power allocation method for the liquid-cooled charging pile.

[0008] The fourth aspect of the present invention provides a computer-readable storage medium, in which instructions are stored, and when it runs on a computer, it enables the computer to execute the above-mentioned power allocation method for the liquid-cooled charging pile.

[0009] In the technical solution provided by the present invention, by adopting a multiple-delay signal cancellation pre-filter for signal processing and combining with the frequency-locked loop phase correction technology, the accuracy of data acquisition is effectively improved, and the harmonic distortion and DC offset in the signal are reduced. By using the non-uniform time-delay feature analysis method and combining with the joint mutual information criterion, the accurate extraction of the charging power time feature is realized, and the dynamic change characteristics during the charging process are accurately captured; by combining the extreme gradient boosting algorithm and the support vector regression, a multi-level charging power prediction model is constructed, significantly improving the accuracy of power prediction; through the proportional robust diffusion recursive operation and the hyperbolic cosine function parameter optimization, the adaptive adjustment of the dynamic power control parameters is realized, ensuring the stability of the charging process; by adopting a multi-layer decision tree structure for feature reconstruction and through the adaptive weighted integration method, the robustness of power allocation is improved; by designing a complete power allocation control process, the full-process intelligent control from data acquisition, feature extraction to power prediction and parameter optimization is realized, making the power allocation of the liquid-cooled charging pile more accurate and efficient.

[0010] Other features and advantages of the present invention will be described in the following specification, and, in part, will be obvious from the specification, or will be understood by implementing the present invention. The objectives and other advantages of the present invention are achieved and obtained by the structures specifically pointed out in the specification, claims and drawings.

[0011] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following provides preferred embodiments in conjunction with the accompanying drawings and detailed descriptions are as follows. Description of the Drawings

[0012] Figure 1 It is a schematic diagram of an embodiment of the power allocation method for a liquid-cooled charging pile in an embodiment of the present invention; Figure 2 It is a schematic diagram of an embodiment of the power allocation system for a liquid-cooled charging pile in an embodiment of the present invention; Figure 3 It is a schematic diagram of an embodiment of the power allocation device for a liquid-cooled charging pile in an embodiment of the present invention. Detailed Embodiments

[0013] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are some but not all of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of the present invention.

[0014] The terms "including" and "having" and any variations thereof mentioned in the embodiments of the present invention are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the listed steps or units, but optionally further includes other unlisted steps or units, or optionally further includes other steps or units inherent to these processes, methods, products, or devices.

[0015] For ease of understanding of this embodiment, first, a detailed introduction is given to a power allocation method for a liquid-cooled charging pile disclosed in the embodiments of the present invention. As Figure 1 shown, the method includes the following steps: 101. Collect the charging power data, liquid-cooling temperature data, charging voltage data, charging current data, and liquid-cooling flow data of the liquid-cooled charging pile, and perform signal processing through a multiple-delay signal cancellation pre-filter to obtain a charging operation standard data set; It can be understood that the execution subject of the present invention can be a power allocation system for a liquid-cooled charging pile, or a terminal or a server, and specifically, it is not limited here. In the embodiments of the present invention, the server is taken as an example of the execution subject for illustration.

[0016] Specifically, various operating parameters in the liquid-cooled charging pile are collected and classified in real time, including charging power data, liquid-cooled temperature data, charging voltage data, charging current data, and liquid-cooled flow data. Signal conditioning and normalization processing are performed on the charging power data, liquid-cooled temperature data, charging voltage data, charging current data, and liquid-cooled flow data. The purpose of signal conditioning is to remove the noise, error, and interference in the acquisition process so that the data is closer to the actual operating state, while the normalization processing ensures the comparability and applicability of various types of data in subsequent calculations by unifying the dimensions and numerical ranges of different data types. After conditioning and normalization processing, a standardized signal data set is obtained. The standardized signal data set is input into a multi-delay signal elimination prefilter to filter out the harmonic distortion in the signal. The multi-delay signal elimination prefilter can efficiently identify and eliminate the harmonic components in the power system and improve the signal quality and accuracy. A frequency-locked loop phase correction model is established based on the filtered signal data. Through this model, the phase deviation of the signal is compensated and calculated to obtain accurately corrected phase correction data. The DC offset of the phase correction data is analyzed to identify and quantify the DC offset in the data, and the DC offset compensation value is calculated. The DC offset compensation value is substituted into the compensation operation unit to generate offset compensation data to eliminate the influence of DC deviation in the data. After the offset compensation is completed, the offset compensation data is sorted according to the time series, and a data correspondence matrix is established based on these time series data. Through this matrix structure, the correlation of various types of data in the time dimension is described to form a complete data set with time series correlation. In order to improve the reliability and accuracy of the data, the outliers in the time series correlation data are identified and processed. According to the preset data valid range, the abnormal data that exceeds the reasonable range is removed to ensure the validity of the data set. The valid data set after the outliers are removed is integrated according to the corresponding relationship of charging power, liquid cooling temperature, charging voltage, charging current, and liquid cooling flow. In this process, the logical association and matching relationship of different data types are ensured to remain consistent. The integrated data set can reflect the operating status of the liquid cooling charging pile and provide an accurate and reliable charging operation standard data set.

[0017] 102. Perform non-uniform time delay feature analysis on the charging operation standard data set, and calculate the time series correlation by joint mutual information criterion to obtain charging power time feature data; Specifically, the charging power change rate is extracted from the charging operation standard data set, and the power change trend data is obtained by calculating the change relationship between the power increment and the basic power value at each time point, reflecting the dynamic change characteristics of power over time during the charging process. Different time delay compensation values are set for the power change trend data to generate multiple sets of time delay compensation data to cope with the characteristic hysteresis and non-uniformity of power changes in different time ranges. Multiple sets of time delay compensation data are input into the non-uniform time delay calculation model, and the charging power delay feature data is extracted by analyzing the dynamic response of the data. The non-uniform time delay calculation model is designed to deal with the nonlinear delay relationship existing in the time series. Through modeling and calculation, the delay feature data can more accurately reflect the power change law in different time dimensions. At the same time, the charging power delay feature data and the liquid cooling temperature data are jointly calculated by mutual information, the correlation between power characteristics and temperature characteristics is quantified, and the power-temperature correlation feature is generated. Through joint mutual information calculation, the influence mechanism of liquid cooling temperature on charging power fluctuation is explored. The mutual information entropy value is calculated based on the power-temperature correlation feature to measure the correlation strength of each feature variable. According to the preset mutual information entropy value threshold, the features with low correlation are screened out, and only the target influencing features are retained to ensure the simplicity and effectiveness of data analysis. Time series analysis is performed on the target influencing features to extract the periodic characteristics of charging power fluctuations and obtain power fluctuation periodic data. The power fluctuation periodic data is combined with the target influencing features to generate a feature mapping matrix. The feature mapping matrix comprehensively considers the multidimensional correlation between the time dimension and the physical characteristics, and the complex dynamic behavior of the charging power can be more intuitively reflected through matrix operations. In order to reduce the computational complexity and improve the efficiency of the model, the matrix is subjected to dimensionality reduction processing. Principal component analysis or other optimization algorithms are used to remove redundant features while retaining the main information to obtain the charging power time characteristic data.

[0018] Sort the multi-group time delay compensation data, set different time windows, and generate a multi-scale delay data matrix. The setting of the time window can capture the dynamic change characteristics in different time ranges during the charging process, forming a data matrix with multi-scale characteristics. Input the multi-scale delay data matrix into the non-uniform time delay calculation model, analyze the data dynamic characteristics within each time window, generate a time delay feature sequence through delay correlation calculation of the data, reflecting the lag characteristics and regularity of power changes in different time ranges. Match the time delay feature sequence with the charging operation standard data set, and generate delay compensation feature data through data alignment and feature extraction. Conduct autocorrelation analysis on the delay compensation feature data, identify potential periodic or repetitive patterns in the data, extract the charging power delay feature data by analyzing the similarity between different time points, reflecting the dynamic characteristics of power changes during the charging process. Align the charging power delay feature data and the liquid cooling temperature data in time to generate a synchronous feature matrix, ensuring the consistency of different data sources on the same time scale, thus avoiding data association errors caused by time misalignment. In the synchronous feature matrix, calculate the mutual information value for each pair of variables. The mutual information value is an important indicator to measure the information correlation between variables, and can quantify the correlation degree between the power feature and the temperature feature. Generate a mutual information sequence through the calculation of the mutual information values for all variable pairs. Based on the mutual information sequence, establish a feature correlation coefficient matrix to obtain the variable correlation degree data. To ensure the comparability and consistency of the correlation data, normalize the variable correlation degree data. Through standardizing the data transformation, map the variable correlation degree data to a unified numerical range, thereby eliminating the influence of different data scales and obtaining the power-temperature correlation feature.

[0019] 103. Input the charging power time feature data into the charging power prediction model for charging power prediction to obtain the charging power prediction data; Specifically, the charging power time feature data is input into the feature extraction layer of the charging power prediction model, which is composed of an extreme gradient boosting unit and a multi-layer perceptron unit. Among them, the charging power time feature data is input into the extreme gradient boosting unit for feature reconstruction. The extreme gradient boosting unit performs hierarchical processing on the input data through its built-in 3 decision tree modules, and uses the step-by-step optimization ability of the decision tree modules to extract key patterns and non-linear features in the data, generating initial feature mapping data. The initial feature mapping data is input into the multi-layer perceptron unit, which contains 4 fully connected layers. Each fully connected layer uses the ReLU activation function to introduce non-linear features and uses the Dropout regularization method to prevent overfitting. Through the deep learning structure of the multi-layer neural network, the initial feature mapping data is further enhanced in features, generating feature enhanced data with stronger expression ability. The feature enhanced data is input into the support vector regression layer of the charging power prediction model. The support vector regression layer uses its regression ability to map the input data to the charging power prediction result space through an optimized hyperplane, obtaining a regression prediction result, which reflects the prediction trend of the charging power changing with time. In order to eliminate short-term fluctuations and noise in the prediction data, hyperbolic tangent function operation and moving average filtering are performed on the regression prediction result. The hyperbolic tangent function operation can smooth the non-linear change characteristics of the data, while the moving average filtering further reduces the high-frequency noise in the data, generating smoothed data. Anti-normalization calculation is performed on the smoothed data. The anti-normalization calculation is implemented through a linear mapping function, using the rated power range of the charging pile as the upper and lower limits of the mapping, and converting the processed data into power mapping data within the actual power value range. The power mapping data is input into the threshold clipping unit for boundary constraint. The threshold clipping unit sets the maximum power value and the minimum power value of the charging pile as the upper and lower limits to ensure that the output charging power prediction data does not exceed the working range of the charging pile. Through the constraint operation, the generated charging power prediction data not only conforms to the physical boundary conditions of actual operation, but also has the dynamic characteristics calculated by the prediction model.

[0020] The charging power time feature data is divided into intervals according to the charging power value. By setting the power intervals, the data with a charging power greater than 80% of the rated power is divided into the high-power data segment, the data with a charging power between 30% and 80% of the rated power is divided into the medium-power data segment, and the data with a charging power less than 30% of the rated power is divided into the low-power data segment, generating stratified feature data. The stratified feature data is input into the first decision tree module in the extreme gradient boosting unit for feature space partitioning and regression tree splitting operations. The maximum depth of the first decision tree module is set to 6 layers, the minimum number of samples per node in each layer is 50, and the number of leaf nodes is limited to 32. These parameter settings ensure that the decision tree can effectively prevent overfitting while guaranteeing the learning ability, generating the first-layer power feature tree. The first-layer power feature tree performs non-linear feature extraction on the data through splitting operations, initially revealing the key distribution laws in the power time feature data. The difference calculation is performed between the output data of the first-layer power feature tree and the stratified feature data to generate the first-layer residual data, which reflects the unexplained error information in the current feature reconstruction process and forms the basis for inputting into the second decision tree module of the extreme gradient boosting unit. The splitting criterion of the second decision tree module adopts the mean square error minimization criterion, and the learning rate is set to 0.1. This configuration can effectively optimize the data splitting process, extract the deep features in the residual data, and generate the second-layer power feature tree. On this basis, the output data of the first-layer power feature tree and the output data of the second-layer power feature tree are subjected to weighted combination operations according to the weight coefficients, where the combination weights are set to 0.6 and 0.4 respectively, and the weight ratio reflects the contribution difference between the first-layer feature tree and the second-layer feature tree in terms of feature expression ability. The difference calculation is performed again between the combined feature data and the stratified feature data to generate the second-layer residual data. The second-layer residual data is input into the third decision tree module of the extreme gradient boosting unit, which adopts L2 regularization constraints, and the regularization coefficient is set to 0.01. L2 regularization can prevent the degradation of generalization ability caused by overfitting of the data while constraining the model complexity. Through the feature compensation calculation of the second-layer residual data, the third-layer power feature tree is generated. The output data of the first-layer power feature tree, the second-layer power feature tree, and the third-layer power feature tree are subjected to adaptive weighted integration. The weight coefficients are determined by minimizing the error of the validation set, thereby ensuring the optimal global performance of the integrated feature data. The generated integrated feature data is subjected to interval mapping in combination with the power range of the charging pile, where the upper limit of the mapping is set to the maximum allowable charging power value of the charging pile, and the lower limit is set to the minimum allowable charging power value of the charging pile. The feature values are reconstructed through a piecewise linear mapping function, and finally the initial feature mapping data is obtained.

[0021] 104. Perform proportional robust diffusion recursive operations on the charging power prediction data and optimize the parameters in combination with the hyperbolic cosine function to obtain dynamic power control parameters.

[0022] Specifically, time - series segmentation processing is performed on the charging power prediction data. The continuous charging power data is divided into multiple power analysis units at fixed time intervals to generate power time - series segmented data. Through time - series segmentation, the variation characteristics of the charging power at different time periods can be captured at a finer granularity. The power time - series segmented data is input into the proportional robust diffusion calculation unit, which calculates the power change characteristic data of each power analysis unit by the least - squares method. The least - squares method can efficiently fit the change trend in the data and extract the power change law in each time period. Recursive iterative calculation is performed on the power change characteristic data. The recursive process can gradually optimize the accuracy of the characteristic data and disperse the noise impact through the diffusion mechanism, making the calculation result more robust and generating recursively optimized data. After completing the recursive optimization, the recursively optimized data is input into the hyperbolic cosine function calculation unit. The hyperbolic cosine function plays a role in parameter optimization in this process. Its period parameter is set to the standard period of the charging process, and the amplitude parameter is set to the fluctuation range of the charging power. Through these parameter configurations, the hyperbolic cosine function can effectively capture the periodic characteristics of the charging power and generate period characteristic data. Parameter sensitivity analysis is performed on the period characteristic data. By analyzing the influence degree of the power control parameters on the charging process, the contribution of different parameters to the system performance is evaluated, and a parameter weight matrix is established to reflect the importance ranking of each control parameter. After inputting the parameter weight data into the feedback control unit, constraint optimization is performed in combination with the power limit conditions of the charging pile. During the constraint optimization process, the upper power limit of the charging pile is set to the maximum allowable power, and the lower limit is set to the minimum power to ensure that the optimization result operates within the physical range. Through constraint optimization, dynamic management of the charging power is achieved while meeting the hardware limitations, generating constraint - optimized data. Stability analysis is performed on the constraint - optimized data. By evaluating the dynamic response characteristics of the system under different power control conditions, potential unstable factors are identified and adjusted. The stability control data generated through stability analysis can provide important safety guarantees for the system operation. The stability control data is input into the dynamic parameter tuning unit to online adjust the control parameters according to the dynamic characteristics of the charging process. The process of dynamic parameter tuning considers the non - linear behavior and real - time change characteristics during the charging process, enabling the control parameters to be continuously optimized during operation and generating dynamic power control parameters.

[0023] In the embodiment of the present invention, by adopting multiple delayed signal elimination pre-filters for signal processing and combining with the frequency-locked loop phase correction technology, the accuracy of data acquisition is effectively improved, the harmonic distortion and DC offset in the signal are reduced, and the non-uniform time delay feature analysis method is adopted in combination with the joint mutual information criterion to achieve accurate extraction of charging power time characteristics and accurately capture the dynamic change characteristics in the charging process; the extreme gradient boosting algorithm is combined with support vector regression to construct a multi-level charging power prediction model, which significantly improves the accuracy of power prediction; through proportional robust diffusion recursive operation and hyperbolic cosine function parameter optimization, the adaptive adjustment of dynamic power control parameters is achieved to ensure the stability of the charging process; a multi-layer decision tree structure is used for feature reconstruction, and the robustness of power allocation is improved through adaptive weighted integration; by designing a complete power allocation control process, the whole process intelligent control from data acquisition, feature extraction to power prediction and parameter optimization is achieved, making the power allocation of liquid-cooled charging piles more accurate and efficient.

[0024] In a specific embodiment, the process of executing step 101 may specifically include the following steps: Collect and sort various operating parameters of the liquid-cooled charging pile to obtain charging power data, liquid-cooling temperature data, charging voltage data, charging current data, and liquid-cooling flow data; Perform signal conditioning and normalization processing on charging power data, liquid cooling temperature data, charging voltage data, charging current data and liquid cooling flow data to obtain a standardized signal data set; The standardized signal data set is input into a multi-delay signal elimination pre-filter to filter out the harmonic distortion in the signal to obtain filtered signal data, and a frequency-locked loop phase correction model is established based on the filtered signal data to compensate for the phase deviation of the signal to obtain phase correction data; Performing DC offset analysis on the phase correction data, and substituting the calculated DC offset compensation value into the compensation operation unit to obtain offset compensation data; The offset compensation data is sorted according to the time series, and a data correspondence matrix is established to obtain time series correlation data, outliers in the time series correlation data are identified, and outliers are removed according to the preset data valid range to obtain a valid data set; The effective data set is integrated according to the corresponding relationship among charging power, liquid cooling temperature, charging voltage, charging current and liquid cooling flow rate to obtain the charging operation standard data set.

[0025] Specifically, various operating parameters in the liquid-cooled charging pile are collected in real time, including charging power data, liquid-cooled temperature data, charging voltage data, charging current data, and liquid-cooled flow rate data. Through the high-precision sensor module and data acquisition device, parameter data is obtained at fixed time intervals to form an original data set containing multi-dimensional features. The collected data is subjected to signal conditioning and normalization processing. Signal conditioning improves the data quality through methods such as filtering and denoising. For example, a low-pass filter is used to remove high-frequency noise, or mean smoothing processing is used to reduce the impact of sudden fluctuations. Normalization processing is to convert data with different dimensions to a unified range through mathematical mapping, and the linear normalization formula is selected: ; where, is the original data value, and are the minimum and maximum values of the data respectively, is the normalized data value.

[0026] For example, for the charging power data, the actual sampling data in a certain test interval is taken: P = {150kW, 165kW, 142kW, 158kW, 170kW}; Applying the normalization formula: x' = (x - x min ) / (x max - x min ) x min = 142kW, x max = 170kW; The normalized data is obtained: P' = {0.286, 0.821, 0, 0.571, 1.0}.

[0027] Through this processing, a standardized signal data set is generated, making various data features comparable and having a unified scale. The standardized signal data set is input into a multiple-delay signal cancellation pre-filter to filter out harmonic distortion in the signal. Harmonic distortion of the signal is usually caused by high-frequency interference or unstable device operation and is eliminated through frequency domain analysis and filtering algorithms. Assume the Fourier decomposition form of the signal is: ; where, is the signal value at time , expressed as the superposition of harmonics; is the DC component, that is, the average value of the signal; , are the harmonic amplitude coefficients, reflecting the weights of the signal on different frequency components; is the harmonic order, representing the multiple of the fundamental frequency (e.g., 2 represents the second harmonic); is the fundamental frequency, the basic oscillation frequency of the signal.

[0028] Perform Fourier decomposition on the power signal and take the first 3 harmonics: S(t) = 160 + 12cos(ωt) + 8sin(ωt) + 5cos(2ωt) + 3sin(2ωt) + 2cos(3ωt) + sin(3ωt); where ω = 2π / T, and T is the sampling period (e.g., 10 s); Set the cut-off frequency of the filter to 1.5ω to eliminate high-order harmonics and obtain: Sf(t) = 160 + 12cos(ωt) + 8sin(ωt).

[0029] The filter removes high-order harmonic terms by setting an appropriate cut-off frequency to generate filtered signal data . Based on the filtered signal data, a phase-locked loop phase correction model is established to correct the phase deviation of the signal. The phase deviation can be calculated by the following formula: where, is the phase deviation, representing the difference between the actual phase and the reference phase of the signal; Im is the imaginary part of the filtered signal, representing the amplitude of the sine component; is the real part of the filtered signal, representing the amplitude of the cosine component.

[0030] Calculate the phase deviation for the filtered signal: Δφ = arctan(Im(Sf) / Re(Sf)) = arctan(8 / 12) = 33.69° After phase correction: Sc(t) = 160 + 14.42cos(ωt - 33.69°).

[0031] The corrected signal data generates a corrected signal by compensating for the phase error . Perform DC offset analysis on the corrected signal data. The DC offset is calculated by the formula: ; where, is the DC offset, representing the average offset value of the signal; is the number of time points, i.e., the total number of sampling points; is the time point and is the value of the corrected signal at the time point.

[0032] Take N = 100 sampling points and calculate the average offset: Dc = (1 / 100)∑Sc(ti) = 158.5kW; After offset compensation: Sb(t) = Sc(t) - 158.5.

[0033] By calculation And perform compensation to obtain the offset compensation data: ; Among them, is the signal value after compensation; is the calibration signal value; is the DC offset. Sort the compensated data according to the time series and construct the corresponding relationship matrix : ; The rows of the matrix represent time points, and the columns represent different operating parameters. On the basis of generating the matrix, perform outlier identification and elimination operations. Detect and eliminate the outlier data beyond the range through the preset data valid range. For example, use the upper and lower limit constraints to eliminate the data points that do not conform to the physical reality and generate the effective data set. Integrate the effective data set according to the corresponding relationship between the parameters. The corresponding relationships of charging power, liquid cooling temperature, charging voltage, charging current, and liquid cooling flow rate are used to form the standardized charging operation data record. The integration process ensures the logical association consistency among the parameters and finally generates the charging operation standard data set.

[0034] In a specific embodiment, the process of executing step 102 may specifically include the following steps: Calculate the charging power change rate in the charging operation standard data set to obtain the power change trend data, and set different time delay compensation values for the power change trend data to obtain multiple groups of time delay compensation data; Substitute the multiple groups of time delay compensation data into the non-uniform time delay calculation model to obtain the charging power delay characteristic data, and perform joint mutual information calculation on the charging power delay characteristic data and the liquid cooling temperature data to obtain the power-temperature correlation characteristic; Calculate the mutual information entropy value based on the power-temperature correlation characteristic, perform feature screening according to the preset threshold to obtain the target influence feature, and perform time series analysis on the target influence feature to obtain the power fluctuation period data; Perform combined operation on the power fluctuation period data and the target influence feature to obtain the feature mapping matrix, and perform dimensionality reduction processing on the feature mapping matrix to obtain the charging power time characteristic data.

[0035] Specifically, calculate the change rate of the charging power in the charging operation standard dataset to capture the dynamic change characteristics of the charging power. The change rate of the charging power is calculated by the following formula: ; Wherein, represents the change rate of the charging power at time , is the charging power at time , is the time interval. By calculating the power change rate at each time point, a complete power change trend data sequence is generated.

[0036] Take the power data of 5 consecutive time points (unit: kW): P(t1) = 160, P(t2) = 168, P(t3) = 172, P(t4) = 165, P(t5) = 170; The time interval Δt = 1s; Calculate the change rate: Rp(t1) = (P(t2)-P(t1)) / Δt = (168-160) / 1 = 8 kW / s; Rp(t2) = (P(t3)-P(t2)) / Δt = (172-168) / 1 = 4 kW / s; Rp(t3) = (P(t4)-P(t3)) / Δt = (165-172) / 1 = -7 kW / s; Rp(t4) = (P(t5)-P(t4)) / Δt = (170-165) / 1 = 5 kW / s; Obtain the power change trend data sequence: Rp = {8, 4, -7, 5} kW / s.

[0037] Set different time delay compensation values for the power change trend data. The core of time delay compensation is to introduce the time lag effect to analyze the characteristics of power change under different delays. Let the delay compensation value be , then the calculation formula for the data after time delay compensation is: ; By selecting multiple different values (for example, ), multiple groups of time delay compensation data are generated.

[0038] Set 3 different delay values: τ1 = 1s, τ2 = 2s, τ3 = 3s; Perform delay compensation on the power change rate at the moment of t = 5s: Rpτ1(5) = Rp(4) = 5 kW / s; Rpτ2(5) = Rp(3) = -7 kW / s; Rpτ3(5) = Rp(2) = 4 kW / s.

[0039] Calculate the delay factor (take the weight w(t)=1): Dτ1 = |5 - 5| = 0; Dτ2 = |(-7) - 5| = 12; Dτ3 = |4 - 5| = 1.

[0040] These delay compensation data can reflect the characteristics of the power change trend at different time lags. Substitute the time delay compensation data into the non-uniform time delay calculation model, which can handle the non-uniform delay characteristics in the time series. The non-uniform time delay calculation model generates the charging power delay characteristic data by optimizing the calculation of the delay factor. The calculation of the delay factor is expressed as: ; where, represents the delay factor, is the time weight, which is set to a normal distribution or a uniform distribution according to the actual application, is the number of time points. Combine the charging power delay characteristic data with the liquid cooling temperature data to perform joint mutual information calculation to analyze the correlation between the power change characteristics and the temperature. The calculation formula of the joint mutual information is: ; where, represents the mutual information between the charging power delay characteristic data and the liquid cooling temperature data, is the joint probability distribution, and are the marginal probability distributions respectively. By calculating the mutual information, the power-temperature correlation characteristics are obtained. Based on the power-temperature correlation characteristics, calculate the mutual information entropy value to measure the overall importance of the characteristics. The formula of the mutual information entropy value is: ; where, is the mutual information entropy value, is the probability of the th feature. By setting the preset threshold , filter out the target influence characteristics higher than the threshold.

[0041] Take power and temperature data samples: Power (kW): P = {160, 168, 172, 165, 170}; Temperature (°C): T = {35, 37, 38, 36, 37}; Discretize the data into 3 intervals and calculate the joint probability: p(P low, T low) = 0.2 p(P low, T medium) = 0.1 p(P low, T high) = 0; p(P medium, T low) = 0.1 p(P medium, T medium) = 0.2 p(P medium, T high) = 0.1; p(P high, T low) = 0 p(P high, T medium) = 0.1 p(P high, T high) = 0.2; Calculate the mutual information: I(P;T) = Σp(p,t)log(p(p,t) / (p(p)p(t))) = 0.2log(0.2 / 0.3×0.3) +... + 0.2log(0.2 / 0.3×0.3) = 0.322 bits.

[0042] Perform time series analysis on the target impact features to extract power fluctuation period data. Time series analysis uses the autocorrelation function or the fast Fourier transform for periodic feature extraction, and the power fluctuation period is determined by the reciprocal of the frequency: ; where, represents the power fluctuation period, is the main frequency. Combine the power fluctuation period data with the target impact features to generate a feature mapping matrix. The form of the feature mapping matrix is: ; where, represents the periodic feature at the th time point, represents the target impact feature. To simplify the feature dimension, perform dimensionality reduction on the feature mapping matrix using the principal component analysis method. The dimensionality reduction formula for principal component analysis is: ; where, is the feature matrix after dimensionality reduction, is the original feature mapping matrix, is the principal component matrix. Obtain the charging power time feature data through dimensionality reduction processing.

[0043] In a specific embodiment, the process of performing the steps of substituting multiple sets of time delay compensation data into a non-uniform time delay calculation model to obtain charging power delay characteristic data, and calculating the joint mutual information of the charging power delay characteristic data and the liquid cooling temperature data to obtain power-temperature correlation characteristics may specifically include the following steps: Sort the multiple sets of time delay compensation data and set different time windows to obtain a multi-scale delay data matrix; Input the multi-scale delay data matrix into the non-uniform time delay calculation model, calculate the delay factor for each time window, and obtain a time delay characteristic sequence; Match the time delay characteristic sequence with the charging operation standard data set to obtain delay compensation characteristic data, and perform autocorrelation analysis on the delay compensation characteristic data to obtain charging power delay characteristic data; Align the charging power delay characteristic data and the liquid cooling temperature data according to time to obtain a synchronous characteristic matrix, and calculate the mutual information value for each pair of variables in the synchronous characteristic matrix to obtain a mutual information sequence; Establish a characteristic correlation coefficient matrix based on the mutual information sequence to obtain variable correlation degree data, and perform normalization processing on the variable correlation degree data to obtain power-temperature correlation characteristics.

[0044] Specifically, sort the multiple sets of time delay compensation data to ensure the consistency of the time sequence. Assume that the time delay compensation data is , where represents different delay compensation values, and represents the time point. During the sorting process, all delay data are sorted in ascending order according to the time point to generate an ordered time sequence data set: Set different time windows to generate a multi-scale delay data matrix. Assume that the length of the time window is , then summarize the data within each window to form a multi-scale characteristic matrix . Its element represents the average value of the th delay compensation data in the th time window: ; Input the multi-scale delay data matrix into the non-uniform time delay calculation model, and the model calculates the delay factor for each time window. The calculation of the delay factor is based on the change trend of the delay data, and its formula is: ; where is the delay compensation weight, which is used to adjust the contribution of different delay data. The delay factor Indicate the intensity of the characteristic change of the delay compensation data within the time window to obtain the time delay feature sequence . Match the time delay feature sequence with the charging operation standard data set to generate delay compensation feature data. The matching process aligns the operations according to time points, for the time delay feature sequence and the charging power , liquid cooling temperature , voltage , current , flow rate in the standard data set for point-by-point matching to form a new data set: ; Perform autocorrelation analysis on the delay compensation feature data to extract the potential periodicity or correlation in the time series. The calculation formula of the autocorrelation function is: ; where is the autocorrelation coefficient, is the time lag, is the mean value of the delay feature data. Align the obtained charging power delay feature data with the liquid cooling temperature data in terms of time to generate a synchronous feature matrix . Each row in the matrix represents the charging power delay feature and the liquid cooling temperature data at the same time point. For each pair of variables in the synchronous feature matrix, calculate their mutual information value to quantify the correlation. The calculation formula of the mutual information is: ; where is the mutual information value, is the joint probability distribution, and are the marginal probability distributions. The obtained mutual information sequence represents the degree of association between the delay feature and the liquid cooling temperature. Based on the mutual information sequence, establish a feature correlation coefficient matrix , whose element represents the correlation between the th variable and the th variable. To ensure consistency, normalize the correlation data, and the normalization formula is: ; where and are the minimum and maximum values in the correlation matrix respectively, is the value after normalization. The normalized correlation matrix provides the association characteristics between power and temperature.

[0045] In a specific embodiment, the process of executing step 103 may specifically include the following steps: Input the charging power time feature data into the feature extraction layer of the charging power prediction model. The feature extraction layer includes an extreme gradient boosting unit and a multi-layer perceptron unit; Input the charging power time feature data into the extreme gradient boosting unit for feature reconstruction. The extreme gradient boosting unit includes 3 decision tree modules to obtain initial feature mapping data; Input the initial feature mapping data into the multi-layer perceptron unit. The multi-layer perceptron unit includes 4 fully connected layers, and each fully connected layer uses the ReLU activation function and Dropout regularization to obtain feature enhanced data; Input the feature enhanced data into the support vector regression layer of the charging power prediction model for regression prediction to obtain a regression prediction result, and perform hyperbolic tangent function operation and moving average filtering on the regression prediction result to obtain smoothed data; Perform denormalization calculation on the smoothed data. The denormalization calculation uses a linear mapping function, and the mapping range is the rated power interval of the charging pile to obtain power mapping data; Input the power mapping data into the threshold clipping unit for boundary constraint. The upper limit of the threshold clipping unit is the maximum power value of the charging pile, and the lower limit is the minimum power value of the charging pile to obtain the charging power prediction data.

[0046] Specifically, input the charging power time feature data into the feature extraction layer of the charging power prediction model. The feature extraction layer is composed of an extreme gradient boosting unit and a multi-layer perceptron unit. The charging power time feature data is represented as a matrix , where represents the feature vector at the -th time point, including multi-dimensional features such as the historical change of charging power, liquid cooling temperature, voltage, and current. Input the feature data into the extreme gradient boosting unit for feature reconstruction. The extreme gradient boosting unit is composed of 3 decision tree modules, and each decision tree realizes non-linear segmentation and feature extraction of data through recursive splitting. Let the depth of the decision tree be , and the data splitting criterion is to minimize the mean squared error (MSE). Its splitting formula is: ; where, is the number of samples, is the actual value, is the predicted value. The output of the decision tree module is the set of segmentation features on the leaf nodes. Through the weighted combination of the outputs of all decision trees, the initial feature mapping data is generated, where represents the A mapping feature. The initial feature mapping data is input into a multi-layer perceptron unit for feature enhancement processing. The multi-layer perceptron contains 4 fully connected layers, and each layer uses the ReLU activation function and Dropout regularization. The formula of the ReLU activation function is: ; where is the input feature value. The role of ReLU is to introduce non-linearity and enhance the network's ability to express complex features. Dropout regularization prevents overfitting by randomly masking some neurons, and its operation is expressed as: ; where is the activation value of the neuron, and mask is a randomly generated binary vector. After being processed by the multi-layer perceptron, the initial feature mapping data is mapped to a high-dimensional space to generate feature enhancement data . The feature enhancement data is input into the support vector regression layer for regression prediction. The support vector regression model realizes prediction by optimizing the following objective function: ; where are the model parameters, is the regularization coefficient, is the loss tolerance interval. The output of the support vector regression is the regression prediction result To smooth the regression result, the prediction result is subjected to a hyperbolic tangent function operation and a moving average filter. The formula of the hyperbolic tangent function is: ; The moving average filter reduces the high-frequency fluctuations in the data by taking the neighborhood mean, and the formula is: where is the moving window size. After the smoothing process, the smoothed data is obtained. The smoothed data is subjected to an inverse normalization calculation to map the dimensionless normalized data back to the actual power range. The formula of the inverse normalization is: ; where is the normalized data, and are the rated maximum power and minimum power of the charging pile respectively. After the inverse normalization process, the power mapping data is obtained. The power mapping data is input into the threshold clipping unit for boundary constraint. The clipping unit ensures that the predicted power is within the physical range through the following rules: ; where is the final charging power prediction data.

[0047] In a specific embodiment, the step of inputting the charging power time feature data into the extreme gradient boosting unit for feature reconstruction is performed. The extreme gradient boosting unit includes 3 decision tree modules. The process of obtaining the initial feature mapping data may specifically include the following steps: The charging power time feature data is divided into power intervals according to the charging power value. The data with a charging power greater than 80% of the rated power is divided into a high-power data segment, the data with a charging power between 30% and 80% of the rated power is divided into a medium-power data segment, and the data with a charging power less than 30% of the rated power is divided into a low-power data segment, obtaining hierarchical feature data; The hierarchical feature data is input into the first decision tree module of the extreme gradient boosting unit. The maximum depth of the first decision tree module is set to 6 layers, the minimum number of samples per layer of nodes is set to 50, and the number of leaf nodes is set to 32. Feature space partitioning and regression tree splitting operations are performed on the hierarchical feature data to obtain the first-layer power feature tree; The output data of the first-layer power feature tree and the hierarchical feature data are subjected to a difference calculation to obtain the first-layer residual data, and the first-layer residual data is input into the second decision tree module of the extreme gradient boosting unit. The splitting criterion of the second decision tree module adopts the mean squared error minimization criterion, and the learning rate is set to 0.1, obtaining the second-layer power feature tree; Weighted combination operations are performed on the output data of the first-layer power feature tree and the output data of the second-layer power feature tree. The combination weights are respectively set to 0.6 and 0.4, obtaining combined feature data, and the combined feature data and the hierarchical feature data are subjected to a difference calculation to obtain the second-layer residual data; The second-layer residual data is input into the third decision tree module of the extreme gradient boosting unit. The third decision tree module adopts L2 regularization constraints, and the regularization coefficient is set to 0.01. Feature compensation calculations are performed on the second-layer residual data to obtain the third-layer power feature tree; The output data of the first-layer power feature tree, the second-layer power feature tree, and the third-layer power feature tree are adaptively weighted and integrated. The weight coefficients are determined by minimizing the validation set error, obtaining integrated feature data; The integrated feature data is subjected to interval mapping according to the power range of the charging pile. The upper limit of the mapping interval is set to the maximum allowable charging power value of the charging pile, and the lower limit is set to the minimum allowable charging power value of the charging pile. A piecewise linear mapping function is used for eigenvalue reconstruction to obtain the initial feature mapping data.

[0048] Specifically, for the charging power time feature data is divided into intervals according to the power value. Assuming that the power value at each time point is and the rated power is Then, the data is divided into high-power data segments, medium-power data segments, and low-power data segments. The division rule is as follows: Through this division, the original data is stratified into three subsets, forming stratified feature data , corresponding to high, medium, and low power intervals respectively. The stratified feature data is input into the first decision tree module of the Extreme Gradient Boosting (XGBoost) unit for feature reconstruction. The maximum depth of the first decision tree is set to 6 layers, and the minimum number of samples per node is 50, and the number of leaf nodes is 32. The decision tree realizes the division of the feature space through recursive splitting, and the splitting criterion is based on the minimization of the mean squared error. The formula is: ; where, is the number of samples, is the actual value, is the predicted value. The first decision tree module generates the first-layer power feature tree by optimizing the objective function, and its output data is expressed as . The output data of the first-layer power feature tree and the stratified feature data are used for difference calculation to generate the first-layer residual data : ; The residual data represents the unexplained error information in the first-layer feature reconstruction. is input into the second decision tree module. The splitting criterion of this module is also based on the minimization of the mean squared error, and the learning rate is set to 0.1. By optimizing the residual data, the second-layer power feature tree is generated. On this basis, the output data of the first-layer power feature tree and the second-layer power feature tree are subjected to weighted combination operation, and the weights are set to 0.6 and 0.4 respectively. The formula is: ; The combined feature data represents the comprehensive characteristics of the two feature levels. and the stratified feature data are used for difference calculation to generate the second-layer residual data : ; The second-layer residual data is input into the third decision tree module. This module adopts L2 regularization constraint, and its objective function is: ; Among them, is the regularization coefficient, set to 0.01, represents the output data of the third-layer power feature tree. Regularization constraints can suppress model complexity and prevent overfitting. The first-layer power feature tree , the second-layer power feature tree and the third-layer power feature tree output data are adaptively weighted and integrated, and the weight coefficients are determined by minimizing the validation set error. The integrated feature data is expressed as: ; Among them, the weight satisfies .

[0049] Use 3 decision trees for feature reconstruction: The first-layer decision tree (depth = 6): Number of input samples = 1000; Minimum number of samples per layer = 50; Number of leaf nodes = 32; Training MSE = 0.015; The second-layer decision tree: The input is the first-layer residual; Learning rate = 0.1; Training MSE = 0.008; The third-layer decision tree: L2 regularization coefficient = 0.01; Training MSE = 0.005; Weight coefficients of the three-layer tree: w1 = 0.6, w2 = 0.4, w3 = 0.3; The MSE of the integrated validation set = 0.003; For the integrated feature data perform interval mapping according to the power range of the charging pile. The upper limit of the mapping interval is set to the maximum allowable charging power of the charging pile , and the lower limit is the minimum power . The formula of the piecewise linear mapping function is: ; Through this mapping, the eigenvalue is reconstructed into a power range with physical meaning, and finally the initial feature mapping data is generated.

[0050] In a specific embodiment, the process of executing step 104 may specifically include the following steps: Perform time series segmentation on the charging power prediction data, divide the continuous charging power data into multiple power analysis units at fixed time intervals, and obtain the power time series segmented data; Input the power time series segmented data into the proportional robust diffusion calculation unit, calculate the power change characteristic data of each power analysis unit using the least squares method, and perform recursive iterative calculation on the power change characteristic data to obtain the recursively optimized data; Input the recursively optimized data into the hyperbolic cosine function calculation unit, set the period parameter of the hyperbolic cosine function to the standard period of the charging process, and set the amplitude parameter to the fluctuation range of the charging power to obtain the period characteristic data; Perform parameter sensitivity analysis on the period characteristic data, calculate the influence degree of the power control parameter on the charging process, and establish a parameter weight matrix to obtain the parameter weight data; Input the parameter weight data into the feedback control unit, perform constraint optimization in combination with the power limit conditions of the charging pile, set the power upper limit to the maximum power of the charging pile, and the power lower limit to the minimum power of the charging pile to obtain the constraint optimization data; Perform stability analysis on the constraint optimization data to obtain the stability control data, and process the stability control data through the dynamic parameter tuning unit to perform online adjustment of the control parameter according to the dynamic characteristics of the charging process to obtain the dynamic power control parameter.

[0051] Specifically, perform time series segmentation on the charging power prediction data, and divide the continuous power data at fixed time intervals into multiple power analysis units. Assume that the charging power prediction data is , where represents the time point, then each power analysis unit contains data points. After segmentation, a power time series segmented data set is generated, where represents the th analysis unit.

[0052] Input the power time series segmented data into the proportional robust diffusion calculation unit, and calculate the power change characteristic data of each power analysis unit using the least squares method. Let of the data points be , corresponding to the time point , and the linear model of the least squares fit is: ; where, is the power change rate, is the intercept. Calculate and by minimizing the sum of squared errors: ; Solve to get: ; Power change rate is an important feature of each power analysis unit, and the finally generated set of power change feature data is . Perform recursive iterative calculation on the power change feature data, and adopt a diffusion mechanism to enhance the robustness of the data. The recursive formula is: ; where is the eigenvalue of the th unit in the th iteration, is the weight coefficient, which controls the balance between the historical value and the neighborhood value.

[0053] After the recursive iteration converges, the recursively optimized data is obtained. Input the recursively optimized data into the hyperbolic cosine function calculation unit, and use the hyperbolic cosine function to generate periodic feature data. The formula of the hyperbolic cosine function is: ; where is the amplitude parameter, indicating the power fluctuation range, is the standard period.

[0054] Segment the predicted power sequence: Time window T = 10s; Number of samples per segment = 100; Least squares fitting: p(t) = at + b; a = (100×Σtipi - ΣtiΣpi) / (100×Σti² - (Σti)²); b = (Σpi - aΣti) / 100; Get the power change rate a = 2.5 kW / s; Intercept b = 158 kW; Hyperbolic cosine function parameters: Period Tstd = 60s; Amplitude A = 15kW; Sensitivity analysis: ∂C / ∂A = 0.08; ∂C / ∂Tstd = -0.03; Parameter update (learning rate η = 0.01): A(k + 1)=15 - 0.01×0.08 = 14.992; Tstd(k + 1)=60 - 0.01×(-0.03)=60.0003。

[0055] Recursively optimize the data Map to , generating a set of periodic feature data . Conduct a parameter sensitivity analysis on the periodic feature data to quantify the influence degree of the power control parameter on the charging process. The sensitivity analysis formula is: ; where represents the th control parameter, represents its sensitivity. By calculating the sensitivities of all parameters, a parameter weight matrix is generated, and its element represents the weight of the th periodic feature to the th parameter. Input the parameter weight matrix into the feedback control unit and perform constraint optimization in combination with the charging pile power limit condition. The optimization goal is to make the output power meet the upper and lower limit constraints: ; Through the optimization function: ; Generate constraint optimization data . Conduct a stability analysis on the constraint optimization data and evaluate the stability of the system under perturbations through the eigenvalue method. Let the stability matrix be , and the stability judgment formula is: ; where is the maximum eigenvalue of the matrix .

[0056] Construct the stability matrix M: M = [-0.05 0.02 -0.01 -0.03]; Calculate the eigenvalues: λ1 = -0.06; λ2 = -0.02; Since λmax = -0.02 < 0; The system stability meets the requirements.

[0057] After the stability analysis is qualified, the data is input into the dynamic parameter tuning unit, and online adjustment is performed in combination with the dynamic characteristics of the charging process to adjust the control parameters The update formula of ; Among them, is the learning rate, and the final dynamic power control parameters are generated.

[0058] The power allocation method of the liquid-cooled charging pile in the embodiment of the present invention is described above. Next, the power allocation system of the liquid-cooled charging pile in the embodiment of the present invention will be described. Please refer to Figure 2 , an embodiment of the power allocation system of the liquid-cooled charging pile in the embodiment of the present invention includes: The acquisition module 201 is used to acquire the charging power data, liquid-cooled temperature data, charging voltage data, charging current data and liquid-cooled flow data of the liquid-cooled charging pile, and perform signal processing through a multi-delay signal cancellation pre-filter to obtain a charging operation standard data set; The analysis module 202 is used to perform non-uniform time-delay feature analysis on the charging operation standard data set, and calculate the time series correlation degree through the joint mutual information criterion to obtain the charging power time feature data; The prediction module 203 is used to input the charging power time feature data into the charging power prediction model for charging power prediction to obtain the charging power prediction data; The optimization module 204 is used to perform proportional robust diffusion recursive operation on the charging power prediction data, and perform parameter optimization in combination with the hyperbolic cosine function to obtain the dynamic power control parameters.

[0059] Through the collaborative cooperation of the above-mentioned various components, by using a multi-delay signal cancellation pre-filter for signal processing and combining the frequency-locked loop phase correction technology, the accuracy of data acquisition is effectively improved, the harmonic distortion and DC offset in the signal are reduced, and the non-uniform time-delay feature analysis method is used. Combining the joint mutual information criterion, the accurate extraction of the charging power time characteristics is realized, and the dynamic change characteristics during the charging process are accurately captured; the extreme gradient boosting algorithm and support vector regression are combined to construct a multi-level charging power prediction model, which significantly improves the accuracy of power prediction; through proportional robust diffusion recursive operation and hyperbolic cosine function parameter optimization, the adaptive adjustment of the dynamic power control parameters is realized, ensuring the stability of the charging process; the multi-layer decision tree structure is used for feature reconstruction, and through the adaptive weighted integration method, the robustness of power allocation is improved; by designing a complete power allocation control process, the full-process intelligent control from data acquisition, feature extraction to power prediction and parameter optimization is realized, making the power allocation of the liquid-cooled charging pile more accurate and efficient.

[0060] AboveFigure 2 The power distribution system of the liquid-cooled charging pile in the embodiment of the present invention will be described in detail from the perspective of modular functional entities. Next, the power distribution device of the liquid-cooled charging pile in the embodiment of the present invention will be described in detail from the perspective of hardware processing.

[0061] Figure 3 FIG. is a schematic structural diagram of a power distribution device of a liquid-cooled charging pile provided by an embodiment of the present invention. The power distribution device 300 of the liquid-cooled charging pile may vary greatly due to different configurations or performances, and may include one or more processors (central processing units, CPU) 310 (for example, one or more processors) and a memory 320, and one or more storage media 330 for storing application programs 333 or data 332 (for example, one or more mass storage device terminals). Among them, the memory 320 and the storage media 330 may be transient storage or persistent storage. The program stored in the storage media 330 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations on the power distribution device 300 of the liquid-cooled charging pile. Further, the processor 310 may be configured to communicate with the storage media 330 and execute a series of instruction operations in the storage media 330 on the power distribution device 300 of the liquid-cooled charging pile to implement the steps of the above-mentioned power distribution method of the liquid-cooled charging pile.

[0062] The power distribution device 300 of the liquid-cooled charging pile may further include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input / output interfaces 360, and / or one or more operating systems 331, such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, and so on. Those skilled in the art can understand that Figure 3 The shown structural diagram of the power distribution device of the liquid-cooled charging pile does not limit the power distribution device of the liquid-cooled charging pile provided by the present invention, and may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0063] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions run on a computer, the computer is caused to execute the steps of the power distribution method of the liquid-cooled charging pile.

[0064] Those skilled in the art can clearly understand that for the convenience and conciseness of description, the specific working processes of the systems, systems, and units described above can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0065] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The foregoing storage medium includes: various media that can store program codes such as USB flash drives, mobile hard disks, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical discs.

[0066] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments or perform equivalent replacements for some of the technical features; and 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.

Claims

1. A power allocation method for a liquid-cooled charging pile, characterized in that, The method comprises: The charging power data, liquid cooling temperature data, charging voltage data, charging current data and liquid cooling flow data of the liquid cooling charging pile are collected, and the signal is processed through a multi-delay signal elimination pre-filter to obtain a charging operation standard data set; Performing non-uniform time delay feature analysis on the charging operation standard data set, and calculating time series correlation by joint mutual information criterion to obtain charging power time feature data; Inputting the charging power time characteristic data into a charging power prediction model to perform charging power prediction to obtain charging power prediction data; A proportional robust diffusion recursive operation is performed on the charging power prediction data, and a hyperbolic cosine function is combined to perform parameter optimization to obtain dynamic power control parameters.

2. The power allocation method of the liquid-cooled charging pile according to claim 1, wherein, The charging power data, liquid cooling temperature data, charging voltage data, charging current data and liquid cooling flow data of the liquid cooling charging pile are collected, and signal processing is performed through a multiple delay signal elimination pre-filter to obtain a charging operation standard data set, including: Collect and sort various operating parameters of the liquid-cooled charging pile to obtain charging power data, liquid-cooling temperature data, charging voltage data, charging current data, and liquid-cooling flow data; Performing signal conditioning and normalization processing on the charging power data, the liquid cooling temperature data, the charging voltage data, the charging current data, and the liquid cooling flow rate data to obtain a standardized signal data set; Inputting the standardized signal data set into a multiple delay signal elimination pre-filter, filtering out the harmonic distortion in the signal to obtain filtered signal data, and establishing a frequency-locked loop phase correction model based on the filtered signal data, performing compensation calculation on the phase deviation of the signal to obtain phase correction data; Performing a DC offset analysis on the phase correction data, and substituting the calculated DC offset compensation value into a compensation operation unit to obtain offset compensation data; The offset compensation data is sorted according to the time series, and a data correspondence matrix is established to obtain time series correlation data, abnormal values in the time series correlation data are identified, and abnormal data are removed according to a preset data valid range to obtain a valid data set; The effective data set is integrated according to the corresponding relationship among charging power, liquid cooling temperature, charging voltage, charging current and liquid cooling flow rate to obtain a charging operation standard data set.

3. The power allocation method of the liquid-cooled charging pile according to claim 2, wherein The non-uniform time delay feature analysis is performed on the charging operation standard data set, and the time series correlation is calculated by the joint mutual information criterion to obtain the charging power time feature data, including: Calculating the charging power change rate in the charging operation standard data set to obtain power change trend data, and setting different time delay compensation values for the power change trend data to obtain multiple groups of time delay compensation data; Substituting the multiple groups of time delay compensation data into a non-uniform time delay calculation model to obtain charging power delay feature data, and performing joint mutual information calculation on the charging power delay feature data and liquid cooling temperature data to obtain power-temperature correlation features; Calculate the mutual information entropy value based on the power-temperature correlation feature, perform feature screening according to a preset threshold to obtain the target influencing feature, and perform time series analysis on the target influencing feature to obtain power fluctuation period data; Perform a combined operation on the power fluctuation period data and the target influencing feature to obtain a feature mapping matrix, and perform dimensionality reduction processing on the feature mapping matrix to obtain charging power time feature data.

4. The power allocation method of the liquid-cooled charging pile according to claim 3, wherein The step of substituting the multiple groups of time delay compensation data into the non-uniform time delay calculation model to obtain charging power delay feature data, and performing joint mutual information calculation on the charging power delay feature data and the liquid cooling temperature data to obtain the power-temperature correlation feature includes: Sort the multiple groups of time delay compensation data and set different time windows to obtain a multi-scale delay data matrix; Input the multi-scale delay data matrix into the non-uniform time delay calculation model, calculate the delay factor of each time window to obtain a time delay feature sequence; Match the time delay feature sequence with the charging operation standard data set to obtain delay compensation feature data, and perform autocorrelation analysis on the delay compensation feature data to obtain charging power delay feature data; Align the charging power delay feature data and the liquid cooling temperature data in time to obtain a synchronous feature matrix, and calculate the mutual information value for each pair of variables in the synchronous feature matrix to obtain a mutual information sequence; Establish a feature correlation coefficient matrix based on the mutual information sequence to obtain variable correlation degree data, and perform normalization processing on the variable correlation degree data to obtain the power-temperature correlation feature.

5. The power allocation method of the liquid-cooled charging pile according to claim 4, characterized in that, The step of inputting the charging power time feature data into the charging power prediction model to predict the charging power to obtain the charging power prediction data includes: Input the charging power time feature data into the feature extraction layer of the charging power prediction model, and the feature extraction layer includes an extreme gradient boosting unit and a multi-layer perceptron unit; Input the charging power time feature data into the extreme gradient boosting unit for feature reconstruction, and the extreme gradient boosting unit includes 3 decision tree modules to obtain initial feature mapping data; Input the initial feature mapping data into the multi-layer perceptron unit, and the multi-layer perceptron unit includes 4 fully connected layers, and each fully connected layer uses the ReLU activation function and Dropout regularization to obtain feature enhanced data; Input the feature enhanced data into the support vector regression layer of the charging power prediction model for regression prediction to obtain a regression prediction result, and perform hyperbolic tangent function operation and moving average filtering on the regression prediction result to obtain smoothed data; Perform anti-normalization calculation on the smoothed data, and the anti-normalization calculation uses a linear mapping function with a mapping range of the rated power interval of the charging pile to obtain power mapping data; Input the power mapping data into the threshold limiting unit for boundary constraint, and the upper limit of the threshold limiting unit is the maximum power value of the charging pile, and the lower limit is the minimum power value of the charging pile to obtain the charging power prediction data.

6. The power allocation method of the liquid-cooled charging pile according to claim 5, wherein Inputting the charging power time feature data into the extreme gradient boosting unit for feature reconstruction, the extreme gradient boosting unit includes 3 decision tree modules, and obtaining initial feature mapping data, including: Dividing the charging power time feature data according to the charging power value, dividing the data with a charging power greater than 80% of the rated power into a high-power data segment, dividing the data with a charging power between 30% and 80% of the rated power into a medium-power data segment, and dividing the data with a charging power less than 30% of the rated power into a low-power data segment, to obtain hierarchical feature data; Inputting the hierarchical feature data into the first decision tree module of the extreme gradient boosting unit, setting the maximum depth of the first decision tree module to 6 layers, setting the minimum number of samples per layer of nodes to 50, and setting the number of leaf nodes to 32, performing feature space partitioning and regression tree splitting operations on the hierarchical feature data, to obtain the first-layer power feature tree; Calculating the difference between the output data of the first-layer power feature tree and the hierarchical feature data, to obtain the first-layer residual data, and inputting the first-layer residual data into the second decision tree module of the extreme gradient boosting unit, using the mean squared error minimization criterion as the splitting criterion for the second decision tree module, and setting the learning rate to 0.1, to obtain the second-layer power feature tree; Performing a weighted combination operation on the output data of the first-layer power feature tree and the output data of the second-layer power feature tree, setting the combination weights to 0.6 and 0.4 respectively, to obtain combined feature data, and calculating the difference between the combined feature data and the hierarchical feature data, to obtain the second-layer residual data; Inputting the second-layer residual data into the third decision tree module of the extreme gradient boosting unit, using L2 regularization constraint for the third decision tree module, and setting the regularization coefficient to 0.01, performing feature compensation calculation on the second-layer residual data, to obtain the third-layer power feature tree; Performing adaptive weighted integration on the output data of the first-layer power feature tree, the second-layer power feature tree, and the third-layer power feature tree, determining the weight coefficients by minimizing the validation set error, to obtain integrated feature data; Performing interval mapping on the integrated feature data according to the power range of the charging pile, setting the upper limit of the mapping interval to the maximum allowable charging power value of the charging pile, setting the lower limit to the minimum allowable charging power value of the charging pile, and using a piecewise linear mapping function for feature value reconstruction, to obtain the initial feature mapping data.

7. The power allocation method of the liquid-cooled charging pile according to claim 6, characterized in that, Performing a proportional robust diffusion recursive operation on the charging power prediction data and optimizing the parameters in combination with the hyperbolic cosine function, to obtain dynamic power control parameters, including: Performing time series segmentation on the charging power prediction data, dividing the continuous charging power data into multiple power analysis units at a fixed time interval, to obtain power time series segmentation data; Inputting the power time series segmentation data into the proportional robust diffusion calculation unit, calculating the power change feature data of each power analysis unit using the least squares method, and performing recursive iterative calculation on the power change feature data, to obtain recursively optimized data; Input the recursively optimized data into the hyperbolic cosine function calculation unit, set the period parameter of the hyperbolic cosine function to the standard period of the charging process, and set the amplitude parameter to the fluctuation range of the charging power to obtain period feature data; Conduct parameter sensitivity analysis on the period feature data, calculate the influence degree of the power control parameter on the charging process, and establish a parameter weight matrix to obtain parameter weight data; Input the parameter weight data into the feedback control unit, perform constraint optimization in combination with the power limit condition of the charging pile, set the power upper limit to the maximum power of the charging pile and the power lower limit to the minimum power of the charging pile to obtain constraint optimization data; Conduct stability analysis on the constraint optimization data to obtain stability control data, and process the stability control data through a dynamic parameter tuning unit to perform online adjustment of the control parameter according to the dynamic characteristics of the charging process to obtain dynamic power control parameters.

8. A power allocation system for a liquid-cooled charging pile, characterized in that, For executing the power allocation method of the liquid-cooled charging pile according to any one of claims 1-7, the system includes: A collection module, configured to collect the charging power data, liquid-cooled temperature data, charging voltage data, charging current data, and liquid-cooled flow data of the liquid-cooled charging pile, and perform signal processing through a multi-delay signal cancellation pre-filter to obtain a charging operation standard data set; An analysis module, configured to perform non-uniform time delay feature analysis on the charging operation standard data set, and calculate the time series correlation degree through the joint mutual information criterion to obtain charging power time feature data; A prediction module, configured to input the charging power time feature data into a charging power prediction model to predict the charging power and obtain charging power prediction data; An optimization module, configured to perform a proportional robust diffusion recursive operation on the charging power prediction data, and perform parameter optimization in combination with the hyperbolic cosine function to obtain dynamic power control parameters.

9. A power allocation device for a liquid-cooled charging pile, characterized in that, The power allocation device of the liquid-cooled charging pile includes: a memory and at least one processor, and instructions are stored in the memory; The at least one processor invokes the instructions in the memory so that the power allocation device of the liquid-cooled charging pile executes the power allocation method of the liquid-cooled charging pile according to any one of claims 1-7.

10. A computer-readable storage medium having instructions stored thereon, characterized in that, When the instructions are executed by the processor, the power allocation method of the liquid-cooled charging pile according to any one of claims 1-7 is implemented.

Images