Power factor correction method and system for AC-DC conversion of server power supply

Abstract

The invention provides a power factor correction method and system for AC-DC conversion of a server power supply, and relates to the technical field of server power supplies, and the method comprises the steps: collecting an input AC signal and a DC bus output signal, calculating a real-time power factor value, constructing a compensation function based on the power factor value, and calculating a deviation value; and generating a compensation parameter matrix through recursive calculation, processing a direct current output signal by adopting an adaptive filtering algorithm, fusing the direct current output signal with the compensation parameter matrix, calculating to obtain a current adjusting signal and a duty ratio adjusting signal, and accurately adjusting a power factor.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of server power supply, in particular to a power factor correction method and system for AC-DC conversion of server power supply. BACKGROUND

[0002] The power factor is an important indicator to measure the utilization efficiency of the power supply system. Low power factor not only leads to energy waste, but also causes harmonic pollution of the power grid and overheating of equipment. The power factor correction technology aims to improve the utilization of electric energy, reduce the harmonic content of input current, make the current waveform closer to the sine wave, thereby reducing system loss and improving the reliability and stability of the power supply system. The current power factor correction technology of server power supply mainly adopts passive correction and active correction. Passive correction is realized by adding passive components such as LC filter. Active correction adjusts the input current waveform by controlling the conduction time of power electronic switching devices. With the increasingly complex and variable characteristics of server load, the traditional power factor correction method has poor adaptability to dynamic changes of load. When the server load changes suddenly, it is difficult to quickly respond and maintain good power factor, resulting in poor correction effect in complex working environment. The control strategy generally uses fixed parameters, lacks adaptive adjustment ability, and cannot flexibly adjust the compensation parameters according to the real-time system state, resulting in large differences in correction effect under different working conditions, lack of effective DC side fluctuation component extraction and compensation mechanism, and difficulty in realizing accurate power factor control. SUMMARY

[0003] The embodiments of the present application provide a power factor correction method and system for AC-DC conversion of server power supply, which can at least solve some problems in the prior art.

[0004] In a first aspect, the present application provides a power factor correction method for AC-DC conversion of server power supply, comprising: Collecting the input AC signal and DC bus output signal of the server power supply, and calculating the real-time power factor value according to the input AC signal; If the real-time power factor value is less than the preset power factor threshold, a compensation function is constructed based on the real-time power factor value and the real-time deviation is calculated. The real-time deviation and the historical compensation value are weighted, accumulated and normalized to obtain an initial compensation factor. The initial compensation factor is error corrected by recursive calculation to generate a stable value. The iteration step is adjusted according to the change rate of the stable value to generate a fast compensation amount and a gradual compensation amount. The compensation parameter matrix is obtained by combination. The adaptive filtering algorithm is used to process the DC bus output signal and extract the effective fluctuation component, the effective fluctuation component is subjected to multi-dimensional fusion operation with a compensation parameter matrix to generate a comprehensive adjustment matrix, parameter decomposition is performed based on the comprehensive adjustment matrix, and a current adjustment signal and a duty cycle adjustment signal are respectively calculated, the input current waveform is adjusted according to the current adjustment signal, and the on-time of the power switch tube is adjusted according to the duty cycle adjustment signal. The adjusted power factor value and the DC output voltage value are obtained, and it is judged whether the preset target range is met, if yes, power correction is performed through the power factor correction circuit.

[0005] In an optional implementation, The input AC signal and the DC bus output signal of the server power supply are collected, the real-time power factor value is calculated according to the input AC signal, and the calculation includes: The input AC signal and the DC bus output signal of the server power supply are collected to obtain a voltage sampling sequence and a current sampling sequence, zero-crossing point information in the voltage sampling sequence and the current sampling sequence is extracted, and a phase feature vector is generated; Discrete Fourier transform is performed on the voltage sampling sequence and the current sampling sequence respectively, the amplitude and phase information of the fundamental frequency component are extracted, and the phase difference value of the fundamental frequency component is calculated; The phase difference value is substituted into a cosine function for operation, and weighted correction is performed in combination with the phase feature vector to obtain the real-time power factor value.

[0006] In an optional implementation, If the real-time power factor value is less than the preset power factor threshold, a compensation function is constructed based on the real-time power factor value, a real-time deviation amount is calculated, the real-time deviation amount is subjected to weighted accumulation and normalization processing with a historical compensation value to obtain an initial compensation factor, and the calculation includes: The real-time power factor value is obtained, it is judged whether the real-time power factor value is less than the preset power factor threshold, if yes, a static deviation term is obtained by performing difference square operation on the real-time power factor value and a target power factor value, a dynamic change rate term is obtained by calculating the absolute value of the change rate of the real-time power factor value, and a compensation function is constructed by weighted combination of the static deviation term and the dynamic change rate term; The compensation function is subjected to integral operation within a preset integral time window to obtain a real-time deviation amount, and the real-time deviation amount is added to a historical compensation value at the previous moment to obtain a cumulative compensation value at the current moment; The rate of change of the real-time power factor value is calculated, a reference gain coefficient is taken as a base, and an inverse of the rate of change is taken as an index to obtain an adaptive gain factor by power operation. The maximum value and the minimum value of the cumulative compensation value are calculated. The cumulative compensation value is divided by the difference between the maximum value and the minimum value after the minimum value is subtracted from the cumulative compensation value. The normalized result is multiplied by the adaptive gain factor to obtain an initial compensation factor.

[0007] In an optional embodiment, The initial compensation factor is error-corrected by recursive calculation to generate a stable value. The iteration step is adjusted according to the rate of change of the stable value to generate a fast compensation amount and a gradual compensation amount. The compensation parameter matrix is obtained by combination. The initial compensation factor is taken as an initial value of recursive operation. The real-time deviation amount is multiplied by a preset weight coefficient and added to the initial value to obtain a first iteration value. Recursive operation is performed based on the first iteration value to obtain a recursive compensation result. The result of each recursive operation is equal to the sum of the result of the last recursive operation and the product of the real-time deviation amount and the preset weight coefficient. The real-time deviation amount, the rate of change of the real-time deviation amount, and the real-time power factor value are mapped to a three-dimensional fuzzy space and a fuzzy rule base is established. The fuzzy membership function of the fuzzy rule base is dynamically adjusted to obtain an optimized fuzzy rule set. The recursive compensation result and the stable value at the last moment are weighted and smoothed according to the optimized fuzzy rule set to generate a stable value at the current moment. The real-time deviation amount and the rate of change of the stable value at the current moment are substituted into a sliding mode calculation to obtain a sliding mode operation result. The rate of change of the stable value at the current moment is calculated based on the sliding mode operation result to obtain a fast compensation amount. The real-time deviation amount is integrated to obtain a gradual compensation amount. An inner loop compensation matrix is constructed based on the fast compensation amount and its change amount. An outer loop compensation matrix is constructed based on the gradual compensation amount and its change amount. A weighting coefficient is determined based on the optimized fuzzy rule set. The inner loop compensation matrix and the outer loop compensation matrix are adaptively weighted and combined to obtain a compensation parameter matrix.

[0008] In an optional embodiment, The DC bus output signal is processed by an adaptive filtering algorithm to extract an effective fluctuation component. The effective fluctuation component and the compensation parameter matrix are subjected to multi-dimensional fusion operation to generate a comprehensive adjustment matrix. The DC bus output signal is subjected to recursive calculation based on an adaptive filtering algorithm. A signal vector group is obtained by simultaneous solution of the results of the recursive calculation. A state estimation error covariance matrix is constructed based on the signal vector group. The state estimation error covariance matrix is subjected to matrix operation with an observation matrix and a measurement noise covariance matrix in sequence to obtain an adaptive gain matrix. The signal vector group is weighted by the adaptive gain matrix to obtain a filtered signal, and a wavelet transform is performed on the filtered signal to obtain a decomposition result, the decomposition result is combined with a preset wavelet base function and a scale factor to obtain a fluctuation characteristic of the filtered signal, and the fluctuation characteristics are combined to form an effective fluctuation component; A dynamic weight is obtained by weight mapping on the effective fluctuation component, an adaptive weight matrix is constructed using the dynamic weight, and a tensor product operation is performed on the adaptive weight matrix and the effective fluctuation component to obtain a fusion operation result; A deviation value is obtained by subtracting a preset expected output from the fusion operation result, a gradient of a weight coefficient is obtained by substituting the deviation value into a target function for calculation, the dynamic weight is iteratively optimized based on the gradient of the weight coefficient to obtain an optimized weight coefficient, a compensation correction amount is obtained by multiplying the optimized weight coefficient and the fusion operation result, and a comprehensive adjustment matrix is obtained by adding the compensation correction amount and the fusion operation result.

[0009] In an optional implementation, A parameter decomposition is performed based on the comprehensive adjustment matrix to respectively calculate a current adjustment signal and a duty cycle adjustment signal, an input current waveform is adjusted according to the current adjustment signal, and a conduction time of a power switch tube is adjusted according to the duty cycle adjustment signal, including: Eigenvalue decomposition is performed on the comprehensive adjustment matrix to obtain an eigenvalue matrix and an eigenvector matrix, and initial adjustment parameters are obtained by performing parameter mapping operation on the eigenvalue matrix and a preset current adjustment coefficient and a preset duty cycle adjustment coefficient; A current adjustment signal is obtained by performing a sine function operation on the initial adjustment parameters and eigenvalues in the eigenvalue matrix and multiplying the result by a preset current gain coefficient, a dynamic compensation signal is obtained by substituting the current adjustment signal into differential operation and integral operation, and a compensated current adjustment signal is obtained by combining the dynamic compensation signal and the current adjustment signal; A duty cycle adjustment signal is obtained by performing a cosine function operation on the initial adjustment parameters and eigenvalues in the eigenvalue matrix and adding a preset initial duty cycle value, a compensated duty cycle adjustment signal is obtained by substituting the duty cycle adjustment signal into a hyperbolic tangent function for dead zone compensation operation; A feedback control signal is calculated based on the compensated current adjustment signal and the compensated duty cycle adjustment signal, an error of the feedback control signal is substituted into integral operation to obtain an adaptive gain signal, a real-time control gain coefficient is obtained by adding the adaptive gain signal and a preset initial gain value, and input current waveforms of the server and conduction times of the power switch tube are adjusted based on the real-time control gain coefficient.

[0010] In an alternative embodiment, The adjusted power factor value and the DC output voltage value are obtained, and it is determined whether a preset target range is met, and if so, power correction is performed by the power factor correction circuit, including: The adjusted power factor value and the DC output voltage value are obtained. The power factor value and the DC output voltage value are compared with a preset power factor target range and a preset voltage target range respectively, and it is determined whether the power factor value and the DC output voltage value meet the preset power factor target range and the preset voltage target range at the same time. When the power factor value and the DC output voltage value meet the preset power factor target range and the preset voltage target range at the same time, real-time control gain coefficients are obtained in advance, and power correction is performed by the power factor correction circuit.

[0011] In a second aspect of the embodiment, a power factor correction system for AC-DC conversion of a server power supply is provided, including: A first unit is configured to collect an input AC signal and a DC bus output signal of the server power supply, and calculate a real-time power factor value based on the input AC signal. A second unit is configured to, if the real-time power factor value is less than a preset power factor threshold, construct a compensation function based on the real-time power factor value and calculate a real-time deviation, perform weighted accumulation and normalization processing on the real-time deviation and a historical compensation value to obtain an initial compensation factor, perform error correction on the initial compensation factor by recursive calculation to generate a stable value, adjust an iteration step length according to a change rate of the stable value, generate a fast compensation amount and a gradual compensation amount, and combine to obtain a compensation parameter matrix. A third unit is configured to process the DC bus output signal by using an adaptive filtering algorithm and extract an effective fluctuation component, perform multi-dimensional fusion operation on the effective fluctuation component and the compensation parameter matrix to generate a comprehensive adjustment matrix, perform parameter decomposition based on the comprehensive adjustment matrix to calculate a current adjustment signal and a duty cycle adjustment signal respectively, adjust an input current waveform according to the current adjustment signal, and adjust a conduction time of a power switch tube according to the duty cycle adjustment signal. A fourth unit is configured to obtain an adjusted power factor value and a DC output voltage value, and determine whether a preset target range is met, and if so, perform power correction by a power factor correction circuit.

[0012] In a third aspect of the embodiment, an electronic device is provided, including: A processor and a memory for storing processor-executable instructions, wherein the processor is configured to invoke the instructions stored in the memory to execute the method described above.

[0013] In a fourth aspect, the present application provides a computer readable storage medium having computer program instructions stored thereon, which, when executed by a processor, implement the method described above.

[0014] In the present application, by monitoring and calculating the power factor value in real time, and starting the correction process when the power factor is lower than the threshold value, the weighted cumulative and normalization processing method is used to optimize the compensation factor, and then the error is corrected through recursive calculation, which effectively improves the accuracy and stability of the power factor correction. By analyzing the change rate of the stable value, the iteration step is dynamically adjusted, the fast compensation amount and the gradual compensation amount are generated and combined to form the compensation parameter matrix, which greatly improves the response speed and adaptability of the system to load changes. The adaptive filtering algorithm is used to process the DC bus output signal, and after extracting the effective fluctuation component, multi-dimensional fusion operation is performed with the compensation parameter matrix to realize accurate adjustment of the current waveform and the switching tube duty cycle. The input power factor and output stability of the server power supply are significantly improved, and the system energy consumption and harmonic interference are reduced. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 A flowchart of the power factor correction method for the AC-DC conversion of the server power supply of the present application embodiment is shown in Figure 2 A server power control flowchart of the power factor correction method for the AC-DC conversion of the server power supply of the present application embodiment is shown in DETAILED DESCRIPTION

[0016] In order to make the purpose, technical scheme and advantages of the present application embodiment clearer, the technical scheme of the present application embodiment will be described clearly and completely below in combination with the drawings of the present application embodiment. Obviously, the described embodiments are only a part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of the present application.

[0017] The technical scheme of the present application will be described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in some embodiments.

[0018] Figure 1 A flowchart of the power factor correction method for the AC-DC conversion of the server power supply of the present application embodiment is shown in Figure 1 As shown, the method comprises: Collecting the input AC signal and DC bus output signal of the server power supply, calculating the real-time power factor value according to the input AC signal; If the real-time power factor value is less than a preset power factor threshold, a compensation function is constructed based on the real-time power factor value and a real-time deviation is calculated, the real-time deviation is weighted, accumulated and normalized with a historical compensation value to obtain an initial compensation factor, the initial compensation factor is error-corrected through recursive calculation to generate a stable value, an iteration step is adjusted according to a change rate of the stable value, a fast compensation amount and a gradual compensation amount are generated, and a compensation parameter matrix is obtained by combination; The DC bus output signal is processed by using an adaptive filtering algorithm to extract an effective fluctuation component, the effective fluctuation component is subjected to multi-dimensional fusion operation with the compensation parameter matrix to generate a comprehensive adjustment matrix, parameter decomposition is performed based on the comprehensive adjustment matrix to calculate a current adjustment signal and a duty cycle adjustment signal respectively, the input current waveform is adjusted according to the current adjustment signal, and the on-time of the power switch tube is adjusted according to the duty cycle adjustment signal. The adjusted power factor value and the DC output voltage value are obtained and it is determined whether the preset target range is met, and if so, power correction is performed by a power factor correction circuit.

[0019] In an alternative embodiment, The input AC signal and the DC bus output signal of the server power supply are collected, and the real-time power factor value is calculated based on the input AC signal, including: The input AC signal and the DC bus output signal of the server power supply are collected to obtain a voltage sampling sequence and a current sampling sequence, and zero-crossing point information in the voltage sampling sequence and the current sampling sequence is extracted to generate a phase feature vector. Discrete Fourier transform is performed on the voltage sampling sequence and the current sampling sequence respectively, the amplitude and phase information of the fundamental frequency component are extracted, and the phase difference value of the fundamental frequency component is calculated. The phase difference value is substituted into a cosine function for operation, and weighted correction is performed in combination with the phase feature vector to obtain the real-time power factor value.

[0020] The input AC signal and the DC bus output signal of the server power supply are collected by installing voltage sensors and current sensors at the AC input end and the DC bus output end of the server power supply. For voltage sampling, a voltage sampling circuit composed of a resistance voltage dividing network combined with an operational amplifier is used, and the sampling frequency is set to 20 kHz to ensure that the change characteristics of the 50 Hz power grid signal can be accurately captured. For current sampling, a current sampling circuit composed of a Hall current sensor or a precision sampling resistor combined with a differential amplifier is used, and a sampling frequency of 20 kHz is adopted. During sampling, appropriate anti-interference measures such as the combination of analog filtering and digital filtering should be taken to reduce interference and improve signal quality. Through the sampling circuit, the voltage sampling sequence v(n) and the current sampling sequence i(n) can be obtained, where n is the sequence number of the sampling point.

[0021] After obtaining the voltage sample sequence and the current sample sequence, the zero-crossing time information in the sequence is extracted. The zero-crossing detection is realized by judging the sign change of adjacent two sample points. For the voltage sequence v(n), when v(n-1) < 0 and v(n) ≥ 0, it can be determined that there is a zero-crossing from negative to positive between n-1 and n; when v(n-1) ≥ 0 and v(n) < 0, it can be determined that there is a zero-crossing from positive to negative between n-1 and n.

[0022] The discrete Fourier transform is performed on the voltage sample sequence v(n) and the current sample sequence i(n) respectively, and the amplitude and phase information of the fundamental frequency component are extracted. In practical applications, a complete period of data (for example, for a 50Hz signal, 400 sample points at a sampling rate of 20kHz) is selected for discrete Fourier transform. Through Fourier transform, the amplitude and phase information of the voltage and current signals at different frequencies can be obtained. For the calculation of power factor, the component at the fundamental frequency (i.e. 50Hz) is mainly concerned. The amplitude Av and phase φv of the voltage fundamental wave, and the amplitude Ai and phase φi of the current fundamental wave are extracted from the transformation result. The phase difference value φd of the fundamental frequency component is calculated as φv- φi.

[0023] After obtaining the phase difference value φd of the fundamental frequency component, it is substituted into the cosine function for operation, that is, cos(φd) is calculated. In ideal cases, this value is the power factor. However, considering various nonlinear factors and harmonic interference existing in actual systems, it is necessary to combine the previously obtained phase feature vector for weighted correction. The process of weighted correction can provide multiple cycle phase difference estimation values φz1, φz2,..., φzn based on the phase feature vector generated by the zero-crossing time information, and calculate the cosine values cos(φz1), cos(φz2),..., cos(φzn) of the estimation values. The cosine values of the estimation values are weighted and averaged with cos(φd) obtained by Fourier transform, and the weight can be determined according to the signal quality and system characteristics. For example, if the reliability of the Fourier transform result is high, a larger weight (such as 0.7) can be given to cos(φd), and the remaining weight (0.3) is evenly distributed to the zero-crossing based estimation values.

[0024] Exemplarily, for testing a certain server power supply, the peak value of the collected voltage sequence in a cycle is 325V, and the peak value of the current sequence is 5.2A. The voltage leads the current by 25 degrees through zero-crossing point detection. Discrete Fourier transform analysis shows that the voltage fundamental wave amplitude is 311V, and the phase is 3 degrees; the current fundamental wave amplitude is 4.9A, and the phase is -22 degrees, with a phase difference of 25 degrees. Substituting into the cosine function to calculate the power factor based on Fourier analysis: cos(25°) = 0.906. The power factor estimated value obtained by zero-crossing point analysis is 0.902, 0.909, and 0.904 for three consecutive cycles. The final power factor is calculated using weight distribution (0.7:0.1:0.1:0.1): 0.7 x 0.906 + 0.1 x 0.902 + 0.1 x 0.909 + 0.1 x 0.904 = 0.906, i.e. the real-time power factor value is 0.906.

[0025] In this embodiment, by simultaneously collecting the electrical signals of the input end and the output end, more complete system characteristic information is obtained, laying a reliable data foundation for subsequent analysis. By extracting the fundamental frequency component through discrete Fourier transform, high-order harmonic interference can be effectively suppressed, and the anti-interference ability of the measurement is improved. The innovative method of using phase feature vectors for weighted correction can compensate for various errors in the measurement process, further improving the accuracy of power factor calculation. Continuous acquisition and processing of voltage and current signals, real-time calculation of power factor through efficient digital signal processing algorithms, timely reflection of dynamic changes in load characteristics, and accurate basis for monitoring and adjusting the power supply system.

[0026] In an alternative embodiment, If the real-time power factor value is less than the preset power factor threshold, a compensation function is constructed based on the real-time power factor value, and a real-time deviation amount is calculated. The real-time deviation amount and the historical compensation value are weighted, accumulated, and normalized to obtain an initial compensation factor, including: The real-time power factor value is obtained, and it is determined whether the real-time power factor value is less than the preset power factor threshold. If it is less than the preset power factor threshold, a static deviation term is obtained by performing a difference square operation on the real-time power factor value and a target power factor value. A dynamic change rate term is obtained by calculating the absolute value of the change rate of the real-time power factor value. The static deviation term and the dynamic change rate term are combined to construct a compensation function. The compensation function is integrated within a preset integration time window to obtain a real-time deviation amount. The real-time deviation amount and the historical compensation value at the previous moment are added to obtain the cumulative compensation value at the current moment. The rate of change of the real-time power factor value is calculated, a reference gain coefficient is taken as the base, and the reciprocal of the rate of change is taken as the exponent to obtain an adaptive gain factor. The maximum value and the minimum value of the cumulative compensation value are calculated, the cumulative compensation value is divided by the difference between the maximum value and the minimum value after the minimum value is subtracted, the normalized result is multiplied by the adaptive gain factor to obtain an initial compensation factor.

[0027] A real-time power factor value is obtained, and the real-time power factor value is compared with a preset power factor threshold value. The preset power factor threshold value is generally set to 0.95, which is a relatively ideal power factor level recommended by the power system. When the real-time power factor value is less than 0.95, a power factor correction process is started.

[0028] In the correction process, a static deviation term is calculated, and the real-time power factor value is subtracted from a target power factor value to obtain a difference value. The difference value is squared. For example, when the real-time power factor is 0.85 and the target power factor is 0.98, the difference value is 0.13, and the square of the difference value is 0.0169. The static deviation term reflects the deviation of the current power factor from the target value. The greater the deviation, the greater the correction effort required.

[0029] A dynamic change rate term is calculated. The dynamic change rate term is obtained by calculating the absolute value of the change rate of the real-time power factor value per unit time. The difference between the power factor values of the current sampling point and the previous sampling point is divided by the sampling time interval to obtain the absolute value. For example, if the current sampling point power factor is 0.85 and the previous sampling point is 0.83, and the sampling time interval is 0.1 milliseconds, then the change rate is 20 / second, and the absolute value is the dynamic change rate term.

[0030] The static deviation term and the dynamic change rate term are combined by weighting to construct a compensation function. The weighting coefficients can be adjusted according to the system characteristics. In general, the weight of the static deviation term is 0.7, and the weight of the dynamic change rate term is 0.3. Using the above data as an example, assuming that the dynamic change rate term calculation result is 10, then the compensation function value is 0.0169x0.7+10x0.3=3.0118.

[0031] The compensation function is integrated within a preset integration time window to obtain a real-time deviation. The integration time window is generally set to 20 milliseconds, and the trapezoidal rule is used for integration. The values of the compensation function within the time window are accumulated and multiplied by the sampling time interval. Assuming that the average value of the compensation function within the integration window is 3.0, and the integration time is 0.02 seconds, then the real-time deviation is 3.0x0.02=0.06.

[0032] The real-time deviation is added to the historical compensation value at the last time to obtain the cumulative compensation value at the current time. Assuming that the historical compensation value at the last time is 0.25, the current cumulative compensation value is 0.06+0.25=0.31. The cumulative compensation value reflects the total amount of adjustment required by the system and has memory, which can effectively handle persistent deviations.

[0033] The rate of change of the real-time power factor value is calculated, which is used to adjust the gain coefficient subsequently. The rate of change is calculated in a similar way to the dynamic change rate described above, but without taking the absolute value, and the sign is retained to reflect the change direction. For example, if the current power factor is increasing at a rate of 0.02 per second, the rate of change is +0.02.

[0034] The adaptive gain factor is obtained by performing an exponential operation with the reference gain coefficient as the base and the inverse of the power factor change rate as the exponent. The reference gain coefficient is usually set to 2.5, which has good response characteristics. If the change rate is +0.02, the exponent is -0.02, and the adaptive gain factor is 2.5^(-0.02)≈2.44. When the power factor is increasing, the gain factor is slightly reduced to avoid overcorrection; when the power factor decreases, the gain factor increases to strengthen the correction strength.

[0035] The maximum and minimum values of the cumulative compensation value are calculated for normalization processing. Through historical data analysis, it is found that the reasonable range of the cumulative compensation value is [0.1, 0.5], so the minimum value is set to 0.1 and the maximum value is set to 0.5. The normalized result is obtained by subtracting the minimum value 0.1 from the current cumulative compensation value 0.31 and dividing by the difference between the maximum value and the minimum value 0.4, i.e., (0.31-0.1) / 0.4=0.525.

[0036] The normalized result is multiplied by the adaptive gain factor to obtain the initial compensation factor. Taking the above calculation result as an example, the initial compensation factor is 0.525×2.44≈1.28. This compensation factor will be used to adjust the duty cycle or switching frequency of the PWM wave to achieve power factor correction.

[0037] In this embodiment, the square of the difference between the real-time power factor value and the target value is taken as the static deviation term, and the absolute value of the power factor change rate is introduced as the dynamic change rate term, to construct a compensation function that comprehensively reflects the compensation demand of the system, considering both the compensation demand at the current time and the dynamic change characteristics of the system, making the compensation process more accurate and stable. The real-time deviation is obtained by integrating the compensation function, and the cumulative compensation value is obtained by superimposing the historical compensation value, effectively avoiding the violent fluctuations in the compensation process and enhancing the stability of the system.

[0038] In an alternative embodiment, The initial compensation factor is error-corrected by recursive calculation to generate a stable value and adjust the iteration step according to the change rate of the stable value, and the fast compensation amount and the gradual compensation amount are generated, and the compensation parameter matrix is obtained by combination, including: The initial compensation factor is taken as the initial value of recursive operation, the real-time deviation is multiplied by a preset weight coefficient and then added to the initial value to obtain a first iteration value, and recursive operation is performed based on the first iteration value to obtain a recursive compensation result, and the result of each recursive operation is equal to the sum of the result of the last recursive operation and the product of the real-time deviation and the preset weight coefficient. The real-time deviation, the change rate of the real-time deviation, and the real-time power factor value are mapped to a three-dimensional fuzzy space and a fuzzy rule base is established, and the fuzzy membership function of the fuzzy rule base is dynamically adjusted to obtain an optimized fuzzy rule set. The recursive compensation result and the stable value at the last time are weighted and smoothed according to the optimized fuzzy rule set to generate a stable value at the current time, and the real-time deviation and the change rate of the stable value at the current time are substituted into a sliding mode calculation to obtain a sliding mode operation result. The change rate of the stable value at the current time is calculated based on the sliding mode operation result to obtain a fast compensation amount, and the real-time deviation is integrated to obtain a gradual compensation amount, an inner loop compensation matrix is constructed by the fast compensation amount and its change amount, an outer loop compensation matrix is constructed by the gradual compensation amount and its change amount, a weighting coefficient is determined based on the optimized fuzzy rule set, and the inner loop compensation matrix and the outer loop compensation matrix are adaptively weighted and combined to obtain a compensation parameter matrix.

[0039] The initial compensation factor is taken as the initial value of recursive operation, and multi-level optimization compensation is performed. The initial compensation factor is taken as the starting point of the recursive algorithm to ensure the stability and convergence of the compensation process. The recursive operation continuously optimizes the compensation result through iteration, so that the power factor gradually approaches the target value. In specific implementation, the real-time deviation is multiplied by a preset weight coefficient and then added to the initial value to obtain a first iteration value. The preset weight coefficient is usually set to 0.35, and the weight coefficient value can achieve a good balance between system stability and response speed through experimental verification. For example, when the initial compensation factor is 1.28 and the real-time deviation is 0.06, the first iteration value is 1.28+(0.06×0.35)=1.301.

[0040] The recursive compensation result is obtained based on the first iteration value. The maximum number of iterations is set to 8 in the recursive process, and the result of each recursive operation is equal to the sum of the result of the last recursive operation and the product of the real-time deviation and the preset weight coefficient. Taking the foregoing data as an example, the second iteration value is 1.301 + (0.06 x 0.35) = 1.322, the third iteration value is 1.322 + (0.06 x 0.35) = 1.343, and so on. When the iteration result changes less than 0.001 or reaches the maximum number of iterations, the recursive process is terminated, and the calculation result at this time is the recursive compensation result, for example, the final value is 1.371.

[0041] The real-time deviation, the change rate of the real-time deviation, and the real-time power factor value are mapped to a three-dimensional fuzzy space and a fuzzy rule base is established. In the fuzzy space, the domain range of the real-time deviation is [-0.2, 0.2], which is divided into five fuzzy sets: large negative deviation, small negative deviation, zero deviation, small positive deviation, and large positive deviation; the domain range of the change rate of the real-time deviation is [-30, 30], which is divided into five fuzzy sets: rapid decrease, slow decrease, stable, slow increase, and rapid increase; and the domain range of the real-time power factor value is [0.8, 1.0], which is divided into three fuzzy sets: insufficient, general, and ideal. The fuzzy rule base includes 75 rules for describing the control strategy corresponding to the combination of each input variable. For example, when the real-time deviation is "small positive deviation", the change rate is "slow decrease", and the power factor is "general", the corresponding rule is "moderate enhancement compensation".

[0042] The fuzzy membership function of the fuzzy rule base is dynamically adjusted to obtain an optimized fuzzy rule set. The adjustment process is based on the change of the server load, and when the load changes rapidly, the shape parameter of the membership function is adjusted to increase its sensitivity. For example, in the case of a sudden load surge, the width parameter of the membership function of "large positive deviation" is adjusted from 0.08 to 0.06, so that the system can respond quickly to smaller deviations.

[0043] The current time stable value is generated by weighted smoothing operation of the recursive compensation result and the last time stable value according to the optimized fuzzy rule set. The weighted smoothing operation adopts the exponential smoothing method, and the smoothing factor is dynamically adjusted according to the fuzzy rule reasoning result, ranging from 0.1 to 0.4. Assuming that the recursive compensation result is 1.371, the last time stable value is 1.25, and the current smoothing factor is 0.25, then the current time stable value is 1.25 x (1-0.25) + 1.371 x 0.25 = 1.28.

[0044] Substituting the real-time deviation and the rate of change of the current stable value into the sliding mode calculation yields the sliding mode result. The sliding mode calculation uses a convergence rate function with a sliding mode coefficient of 0.85, achieving both rapid convergence and stable retention. Assuming the real-time deviation is 0.06 and the rate of change of the current stable value is 0.03, the sliding mode calculation result is 0.081.

[0045] The rapid compensation amount is calculated based on the sliding mode operation result, calculating the rate of change of the current stable value. This is calculated by multiplying the sliding mode operation result by the current stable value and then multiplying it by the adjustment coefficient of 0.75. Based on the above data, the rapid compensation amount is 0.081 × 1.28 × 0.75 = 0.0778.

[0046] The real-time deviation is integrated to obtain the progressive compensation. The integration uses the trapezoidal rule, with an integration window of 50 milliseconds and an integral gain of 0.6. Assuming the average real-time deviation within the integration window is 0.055, the progressive compensation is 0.055 × 0.05 × 0.6 = 0.00165.

[0047] The fast compensation value and its variation are used to construct the inner-loop compensation matrix. The inner-loop compensation matrix is ​​a 2×2 matrix containing the fast compensation value and its first-order and second-order variations. Assuming the fast compensation value is 0.0778 and its first-order variation is 0.0025, the elements of the inner-loop compensation matrix are [0.0778, 0.0025; 0.0025, 0.0001], respectively.

[0048] The outer-loop compensation matrix is ​​constructed by combining the progressive compensation amount and its variation. This outer-loop compensation matrix is ​​also a 2×2 matrix, containing the progressive compensation amount and its variation trend. Assuming the progressive compensation amount is 0.00165 and its variation trend is 0.0003, the elements of the outer-loop compensation matrix are [0.00165, 0.0003; 0.0003, 0.00006].

[0049] The weighting coefficients are determined based on an optimized fuzzy rule set, and the inner and outer loop compensation matrices are adaptively weighted together to produce the compensation parameter matrix. The inner loop weight, determined through fuzzy reasoning, is 0.65, and the outer loop weight is 0.35. The weighted combination of these two matrices yields the compensation parameter matrix [0.0515, 0.00175; 0.00175, 0.000074].

[0050] In this embodiment, the initial compensation factor is taken as the starting point, and the compensation result is continuously optimized through iterative calculation, which can make full use of historical information, make the compensation process continuous and smooth, effectively avoid the mutation of compensation amount, and realize the flexible adjustment of recursive calculation intensity by introducing the preset weight coefficient, enhance the adaptability of the algorithm, take the real-time deviation, the deviation change rate and the power factor value as the input variables, realize the adaptive adjustment of the control rule by dynamically optimizing the membership function of the fuzzy rule base, the fuzzy logic based control mechanism can better handle the uncertainty and nonlinearity of the system, and improve the robustness of the control, the real-time deviation and the stable value change rate are substituted into the sliding mode operation to obtain a more accurate compensation amount, the sliding mode control has good anti-interference ability and fast convergence characteristics, and can effectively cope with system parameter fluctuations and external disturbances.

[0051] In an alternative embodiment, The adaptive filtering algorithm is used to process the DC bus output signal and extract the effective fluctuation component, and the effective fluctuation component is subjected to multi-dimensional fusion operation with the compensation parameter matrix to generate a comprehensive adjustment matrix, which comprises: The adaptive filtering algorithm is used to process the DC bus output signal and extract the effective fluctuation component, and the effective fluctuation component is subjected to multi-dimensional fusion operation with the compensation parameter matrix to generate a comprehensive adjustment matrix, which comprises: The adaptive filtering algorithm is used to process the DC bus output signal and extract the effective fluctuation component, and the effective fluctuation component is subjected to multi-dimensional fusion operation with the compensation parameter matrix to generate a comprehensive adjustment matrix, which comprises: The adaptive filtering algorithm is used to process the DC bus output signal and extract the effective fluctuation component, and the effective fluctuation component is subjected to multi-dimensional fusion operation with the compensation parameter matrix to generate a comprehensive adjustment matrix, which comprises: The adaptive filtering algorithm is used to process the DC bus output signal and extract the effective fluctuation component, and the effective fluctuation component is subjected to multi-dimensional fusion operation with the compensation parameter matrix to generate a comprehensive adjustment matrix, which comprises:

[0052] An adaptive filtering algorithm is used to process the DC bus output signal. Based on the Kalman filter principle, a recursive calculation of the bus voltage signal is performed to eliminate the adverse effects of voltage fluctuations on power factor correction. During implementation, the DC bus output signal is acquired via an analog-to-digital converter (ADC) with a sampling frequency of 20kHz to ensure real-time and accurate signal acquisition. After preprocessing, the acquired signal enters the recursive calculation module, which uses state equations to describe the system's dynamic characteristics and observation equations to describe the measurement process.

[0053] The recursive calculation results are solved simultaneously to obtain a signal vector set. This simultaneous solution uses the Gauss-Seidel iterative method, with five iterations and a convergence threshold of 0.001. For example, when the DC bus voltage is 380 V with a ±5 V fluctuation range, the recursive calculation and simultaneous solution yield the signal vector set [380.2, 0.5, -0.3, 0.1], representing the estimated voltage value and its first-order, second-order, and third-order rates of change, respectively. This signal vector set is calculated using a digital signal processor with a computational cycle of 50 microseconds.

[0054] The state estimation error covariance matrix is ​​constructed based on the signal vector group. This matrix is ​​a 4×4 symmetric matrix, with the diagonal elements representing the variance of each state estimation error and the off-diagonal elements representing the covariance between state estimation errors. The initial covariance matrix is ​​set to a diagonal matrix with diagonal elements in the range [0.5, 0.1, 0.05, 0.02], representing the uncertainty of the initial state estimate. The covariance matrix is ​​updated recursively, calculated once per control cycle.

[0055] The adaptive gain matrix is ​​obtained by performing matrix operations on the state estimation error covariance matrix, the observation matrix, and the measurement noise covariance matrix. The observation matrix is ​​set to [1, 0, 0, 0], indicating that only the voltage value is directly measured. The measurement noise covariance matrix is ​​a single-element matrix with a value of 0.25, representing the variance of the measurement noise. Matrix operations include matrix multiplication, matrix addition, and matrix inversion, implemented using the floating-point unit. The resulting adaptive gain matrix is ​​a 4×1 column vector, for example, [0.85, 0.32, 0.15, 0.06], which indicates how each state estimate should be corrected based on the measurement residuals.

[0056] The adaptive gain matrix is used to weight the signal vector group to obtain the filtered signal. The weighting operation process is to calculate the difference between the measured value and the predicted value (measurement residual), multiply the residual by the adaptive gain matrix to obtain the state correction, and add the correction to the predicted state to obtain the filtered state estimation. Taking the above data as an example, if the measured value is 381V and the predicted value is 380.2V, the residual is 0.8V, and the state correction [0.68, 0.256, 0.12, 0.048] is obtained after weighting by the adaptive gain, and the updated filtered signal is [380.88, 0.756, -0.18, 0.148].

[0057] The filtered signal is decomposed by wavelet transform to obtain the decomposition result. The wavelet transform uses a multi-scale analysis method to decompose the signal into sub-signals of different frequency bands. The wavelet transform uses Daubechies wavelet with 3 layers of decomposition. Taking the filtered voltage signal as an example, 3 detail coefficient sequences and 1 approximation coefficient sequence are obtained after wavelet transform decomposition, and the detail coefficients reflect the high frequency, medium frequency and low frequency fluctuation characteristics of the signal, and the approximation coefficient reflects the trend characteristics of the signal.

[0058] The decomposition result is combined with the preset wavelet basis function and scale factor to obtain the fluctuation characteristics of the filtered signal. The wavelet basis function is selected as db4 function, and the scale factor is set as 1, 2 and 4 respectively, corresponding to the signal characteristics of different frequency bands. The combination operation adopts the form of inner product, and the matching degree of each detail coefficient and the basis function under the corresponding scale is calculated. The calculated fluctuation characteristics are represented as [0.28, 0.15, 0.08], respectively corresponding to the intensity of high frequency, medium frequency and low frequency fluctuation.

[0059] The fluctuation characteristics are combined to form effective fluctuation components. The combination process adopts a weighted sum method, and the weight coefficients are set by experiments as [0.5, 0.3, 0.2], representing the importance of the fluctuations of each frequency band on the power factor. Taking the above fluctuation characteristics as an example, the effective fluctuation component is calculated as 0.28*0.5+0.15*0.3+0.08*0.2=0.199, representing the comprehensive influence degree of the current voltage fluctuation on the power factor.

[0060] The effective fluctuation component is mapped to a dynamic weight through a weight mapping. The weight mapping uses a S-shaped function to map the fluctuation component to the interval [0.1, 0.9], and the mapping parameters are set according to the system characteristics. Taking the effective fluctuation component 0.199 as an example, the dynamic weight obtained after mapping is 0.35, representing the adjustment intensity of the correction parameter under the current fluctuation condition.

[0061] The dynamic weight is used to construct an adaptive weight matrix. The adaptive weight matrix is a 2x2 symmetric matrix, the diagonal elements are set to the dynamic weight and its complementary value, and the non-diagonal elements are set to the geometric mean of the two. Taking the dynamic weight 0.35 as an example, the adaptive weight matrix is constructed as [[0.35, 0.3], [0.3, 0.65]], which represents the adjustment intensity of different correction parameters.

[0062] The adaptive weight matrix is subjected to tensor product operation with the effective fluctuation component to obtain a fusion operation result. The tensor product operation expands the matrix multiplication, and the multiplication of a scalar and a matrix obtains a weighted matrix. Taking the effective fluctuation component 0.199 and the aforementioned adaptive weight matrix as an example, the fusion operation result is [[0.0697, 0.0597], [0.0597, 0.1294]], which represents the compensation parameter adjustment amount after considering the influence of voltage fluctuation.

[0063] The fusion operation result is subtracted from the pre-set expected output to obtain a deviation value. The expected output matrix is set to [[0.05, 0.02], [0.02, 0.1]], which represents the compensation parameters under ideal working conditions, and the deviation value is calculated as the difference between the fusion operation result and the expected output, i.e. [[0.0197, 0.0397], [0.0397, 0.0294]], which represents the gap between the current parameters and the ideal parameters.

[0064] The deviation value is substituted into the objective function to calculate the gradient of the weight coefficient. The objective function adopts the form of mean square error, which represents the deviation degree of the current compensation effect from the ideal effect. The gradient calculation adopts analytical derivative to obtain the weight adjustment direction. Taking the aforementioned deviation value as an example, the calculated gradient value is [0.023, -0.015], which represents that the first weight should be increased and the second weight should be decreased.

[0065] The dynamic weight is iteratively optimized based on the gradient of the weight coefficient to obtain an optimized weight coefficient. The iterative optimization adopts the gradient descent method, the learning rate is set to 0.2, and the iteration number is 3 times. After iterative optimization, the dynamic weight is adjusted from 0.35 to 0.3846, and the optimized weight coefficient is [0.3846, 0.6154].

[0066] The optimized weight coefficient is multiplied by the fusion operation result to obtain a compensation correction amount. The multiplication operation is the multiplication of a matrix and a vector, and the calculation result is [0.0427, 0.0853], which represents the adjustment amount of the original compensation parameters.

[0067] The compensation correction amount and the fusion operation result are added to obtain a comprehensive adjustment matrix, the addition operation is matrix addition, and the obtained comprehensive adjustment matrix is [[0.1124, 0.1024], [0.1024, 0.2147]]. The matrix is used for adjusting the control parameters of the power factor correction circuit, including the PWM duty cycle, the switching frequency and the like, so that accurate power factor correction under the condition of DC bus voltage fluctuation is realized.

[0068] In the embodiment, by constructing a state estimation error covariance matrix and combining the observation matrix and the measurement noise covariance matrix to perform operation, an adaptive gain matrix is obtained, which can dynamically adjust the filtering parameters according to the signal characteristics, effectively suppress the influence of measurement noise, and improve the accuracy of signal processing. Through wavelet decomposition of the filtered signal and combined operation with the preset wavelet basis function and scale factor, multi-scale analysis of the signal fluctuation characteristics is realized, which can effectively capture the local characteristics and mutation information of the signal, and provide a reliable basis for subsequent compensation control. By substituting the deviation between the fusion operation result and the expected output into the objective function, the gradient of the weight coefficient is calculated, and iterative optimization is performed based on this, so as to realize dynamic adjustment of the weight coefficient, ensure the accuracy and convergence of the weight adjustment, and improve the adaptive ability of the system.

[0069] In an alternative embodiment, Based on the comprehensive adjustment matrix, a current adjustment signal and a duty cycle adjustment signal are calculated respectively, the input current waveform is adjusted according to the current adjustment signal, and the on-time of the power switch tube is adjusted according to the duty cycle adjustment signal, including: Eigenvalue decomposition is performed on the comprehensive adjustment matrix to obtain an eigenvalue matrix and an eigenvector matrix, and parameter mapping operation is performed on the eigenvalue matrix, a preset current adjustment coefficient and a preset duty cycle adjustment coefficient to obtain initial adjustment parameters; The initial adjustment parameters and the eigenvalues in the eigenvalue matrix are subjected to sine function operation and multiplied by a preset current gain coefficient to obtain a current adjustment signal, the current adjustment signal is substituted into differential operation and integral operation to obtain a dynamic compensation signal, and the dynamic compensation signal and the current adjustment signal are combined to obtain a compensated current adjustment signal; The initial adjustment parameters and the eigenvalues in the eigenvalue matrix are subjected to cosine function operation and superimposed with a preset initial duty cycle value to obtain a duty cycle adjustment signal, and the duty cycle adjustment signal is subjected to dead zone compensation operation by substituting it into a hyperbolic tangent function to obtain a compensated duty cycle adjustment signal; The feedback control signal is calculated based on the compensated current regulation signal and the compensated duty cycle regulation signal, the error of the feedback control signal is substituted into integral operation to obtain an adaptive gain signal, the adaptive gain signal is added with a preset initial gain value to obtain a real-time control gain coefficient, and the input current waveform of the server and the on-time of the power switch tube are adjusted based on the real-time control gain coefficient.

[0070] The eigenvalue decomposition is realized by the Jacobi iteration method, the iteration accuracy is set to 0.0001, and the maximum iteration number is 20. Taking the aforementioned comprehensive adjustment matrix [[0.1124, 0.1024], [0.1024, 0.2147]] as an example, the eigenvalue matrix obtained by eigenvalue decomposition is a diagonal matrix [[0.0387, 0], [0, 0.2884]], and the corresponding eigenvector matrix is [[0.7352, -0.6778], [0.6778, 0.7352]]. The eigenvalue reflects the response strength of the system in different directions, and the eigenvector represents the main direction of the response.

[0071] The initial adjustment parameter is obtained by parameter mapping operation of the eigenvalue matrix, the preset current regulation coefficient and the preset duty cycle regulation coefficient. The preset current regulation coefficient is 1.5, the preset duty cycle regulation coefficient is 0.8, the parameter mapping adopts linear transformation mode, and the eigenvalue is scaled according to the preset coefficient. In specific calculation, the first eigenvalue of the eigenvalue matrix is multiplied by the current regulation coefficient, and the second eigenvalue is multiplied by the duty cycle regulation coefficient to obtain the initial adjustment parameter [0.0581, 0.2307], which is used in the subsequent current control loop and duty cycle control loop.

[0072] The initial adjustment parameter and the eigenvalue in the eigenvalue matrix are subjected to sine function operation and multiplied by a preset current gain coefficient to obtain a current regulation signal. The sine function operation is used to introduce periodic regulation characteristics, so that the control signal has a smooth change process. The current gain coefficient is set to 2.5, which is used to adjust the amplitude of the control signal. Taking the initial adjustment parameter as an example, the first parameter 0.0581 and the first eigenvalue 0.0387 are substituted into the sine function to obtain the sine value 0.0581, and then multiplied by the current gain coefficient 2.5 to obtain the current regulation signal 0.1453. The current regulation signal is used to adjust the gain of the current control loop and affects the waveform of the input current.

[0073] The current regulation signal is substituted into the differential operation and integral operation to obtain a dynamic compensation signal. The differential operation adopts a second-order central difference method, and the time step is 0.1 millisecond, so as to improve the numerical precision and stability. The integral operation adopts a trapezoidal rule, and the integral time window is 20 milliseconds. Taking the current regulation signal 0.1453 as an example, assuming that the signal at the previous time is 0.1425, the differential calculation obtains a change rate of 0.28 / second, and the integral calculation obtains an accumulated value of 0.0029. The dynamic compensation signal is obtained by weighted combination of the differential result and the integral result, and the weights are 0.4 and 0.6 respectively, and the calculation result is 0.28*0.4+0.0029*0.6=0.1137. The dynamic compensation signal is used to process the dynamic change of the current waveform, and improve the response speed and stability of the system.

[0074] The dynamic compensation signal and the current regulation signal are combined to obtain a compensated current regulation signal. The combination mode is weighted summation, the weight of the dynamic compensation signal is 0.3, and the weight of the current regulation signal is 0.7, and the calculation obtains a compensated current regulation signal of 0.1137*0.3+0.1453*0.7=0.1358. The compensated current regulation signal introduces a dynamic response characteristic on the basis of the original current regulation signal, and can better cope with load changes and input voltage fluctuations.

[0075] The initial regulation parameter and the eigenvalue in the eigenvalue matrix are subjected to cosine function operation and superimposed with a pre-set initial duty ratio value to obtain a duty ratio regulation signal. The cosine function operation is used to introduce a periodic regulation characteristic, but has different phase characteristics compared with the sine function. The initial duty ratio value is set to 0.45, indicating the proportion of the power switch tube conduction time to the period. Taking the second initial regulation parameter 0.2307 and the second eigenvalue 0.2884 as an example, the two are substituted into the cosine function to obtain a cosine value of 0.9656, and then the initial duty ratio value 0.45 is superimposed to obtain a duty ratio regulation signal of 0.4656. The signal is used to adjust the duty ratio of the PWM controller and affect the conduction time of the power switch tube.

[0076] The duty ratio regulation signal is substituted into the hyperbolic tangent function for dead zone compensation operation to obtain a compensated duty ratio regulation signal, and the scaling factor of the hyperbolic tangent function is set to 2 and the offset is set to 0. Taking the duty ratio regulation signal 0.4656 as an example, the calculation by substituting into the hyperbolic tangent function obtains 0.4342 as the compensated duty ratio regulation signal. After the dead zone compensation, the signal can more accurately control the conduction time of the power switch tube, reduce the switching loss, and improve the energy efficiency.

[0077] The feedback control signal is calculated based on the compensated current regulation signal and the compensated duty cycle regulation signal. The calculation method is weighted average, and the weight coefficient is dynamically adjusted through the system state. When the input voltage is stable, the duty cycle regulation signal weight is higher; when the input voltage fluctuates, the current regulation signal weight is higher. Taking the current state current regulation signal weight as 0.45 and the duty cycle regulation signal weight as 0.55 as an example, the feedback control signal is calculated as 0.1358x0.45+0.4342x0.55=0.3001. The feedback control signal considers the control requirements of current waveform and switching timing, and is used to adjust the operating state of the power factor correction circuit.

[0078] The error of the feedback control signal is substituted into the integral operation to obtain the adaptive gain signal. The error is defined as the difference between the feedback control signal and the expected control signal, and the expected control signal is set to 0.32. The error integration uses discrete accumulation method, and the integral gain is 0.15 and the integral limit is ±0.2. Taking the feedback control signal 0.3001 as an example, the error is 0.3001-0.32=-0.0199, and the gain adjustment amount is-0.0030 after one cycle of integration. The adaptive gain signal is continuously accumulated through error integration, which can automatically adjust the control parameters and adapt to different working conditions.

[0079] The adaptive gain signal is added to the pre-set initial gain value to obtain the real-time control gain coefficient. The initial gain value is set to 1.8, representing the basic response strength of the system. Taking the adaptive gain signal-0.0030 as an example, the real-time control gain coefficient is calculated as 1.8+(-0.0030)=1.797.

[0080] The input current waveform of the server and the conduction time of the power switch tube are adjusted based on the real-time control gain coefficient. The adjustment process is realized through a digital controller, and the output PWM signal controls the switching state of the power switch tube. The adjustment of the input current waveform is realized by changing the gain of the current loop, and the adjustment of the conduction time of the power switch tube is realized by changing the duty cycle of the PWM signal.

[0081] In the embodiment, the initial adjustment parameters are established by parameter mapping of the eigenvalue matrix, the preset current adjustment coefficient and the duty ratio adjustment coefficient, the essential characteristics of the system can be fully extracted, and the foundation for subsequent accurate control is laid, the generation of the current adjustment signal is realized by the sine function operation of the initial adjustment parameters and the eigenvalue and in combination with the current gain coefficient, the duty ratio adjustment signal is generated in the manner of superposition of the cosine function operation and the initial duty ratio, and the dead zone compensation is realized through the hyperbolic tangent function, which not only ensures the continuity and smoothness of the duty ratio adjustment, but also effectively solves the dead zone problem in the switching tube control, improves the accuracy of the switching control, and through the generation of the feedback control signal by the comprehensive current adjustment signal and the duty ratio adjustment signal and the conversion of the error into the adaptive gain signal through the integral operation, the control strength can be dynamically adjusted according to the system state, and the adaptability and robustness of the system are improved.

[0082] Figure 2 The server power control flowchart of the power factor correction method for the server power AC / DC conversion of the embodiment.

[0083] In an optional embodiment, The adjusted power factor value and the DC output voltage value are acquired, and it is judged whether the preset target range is met, if yes, the power correction is executed through the power factor correction circuit, including: The adjusted power factor value and the DC output voltage value are acquired; The power factor value and the DC output voltage value are compared with the preset power factor target range and the preset voltage target range respectively, and it is judged whether the power factor value and the DC output voltage value meet the preset power factor target range and the preset voltage target range at the same time; When the power factor value and the DC output voltage value meet the preset power factor target range and the preset voltage target range at the same time, the real-time control gain coefficient and the power factor correction circuit are executed according to the power correction.

[0084] The adjusted power factor value and the DC output voltage value are obtained, and the power factor value is obtained by real-time measurement through a special detection circuit. The circuit is composed of a high-precision voltage sensor and a current sensor, and the sampling frequency is 50 kHz, which ensures that the subtle changes of the voltage and current waveforms can be captured. The voltage sensor adopts a resistance voltage division principle, and the voltage division ratio is 100:1. The measurement range is 0-400V AC. The current sensor adopts a Hall effect principle, and the sensitivity is 100mV / A. The measurement range is 0-20A. The sensor output signal is sent to a 16-bit analog-to-digital converter after being processed by a signal conditioning circuit. The conversion result is read by a digital signal processor and the power factor is calculated. The DC output voltage value is also obtained by a voltage sensor, which adopts a resistance voltage division principle with a voltage division ratio of 50:1. The measurement range is 0-800V DC, and the sampling frequency is synchronized with the power factor detection.

[0085] The power factor value and the DC output voltage value are compared with the preset power factor target range and the preset voltage target range, respectively, to determine whether they meet the preset target at the same time. The power factor target range is set to 0.95-1.0, which meets the requirements of high-efficiency server power supplies and can effectively reduce grid harmonic pollution and reactive power loss. The voltage target range is set according to the server type and specifications. For a typical 12V output server power supply, the DC output voltage target range is set to 11.8V-12.2V, allowing a ±1.67% fluctuation. The comparison process is performed by a digital controller using an interval judgment method, i.e., checking whether the measured value falls within the target interval. Taking an actual application as an example, when the measured power factor value is 0.974 and the DC output voltage is 12.05V, both fall within the target range, and the judgment result is that the requirements are met.

[0086] In the comparison and judgment process, to improve the stability and anti-interference ability of the system, the measured value is processed by a sliding average filter with a window length of 10 sampling points. To prevent the system from frequently switching between states under critical conditions, a hysteresis comparison mechanism is introduced. When the measured value changes from not meeting the requirements to meeting the requirements, it needs to meet the requirements for 20 consecutive control periods to trigger state conversion. When the measured value changes from meeting the requirements to not meeting the requirements, it needs to not meet the requirements for 5 consecutive control periods to trigger state conversion. This hysteresis comparison mechanism effectively avoids the oscillation phenomenon of the system under critical conditions.

[0087] When the power factor value and the DC output voltage value meet the preset power factor target range and the preset voltage target range at the same time, the power correction is performed according to the real-time control gain coefficient obtained in advance and the power factor correction circuit. The real-time control gain coefficient is obtained by the foregoing calculation process, for example, 1.797, which is used to adjust the response strength of the control loop. The power factor correction circuit adopts a digital control boost topology, including power switch tube, diode, inductor and filter capacitor and other elements. The power switch tube adopts MOSFET device, the breakdown voltage is 650V, and the on-resistance is 55mΩ; the inductor selects a ferrite core, the inductance value is 500μH, and the saturation current is 15A; the filter capacitor adopts low equivalent series resistance design, the capacitance value is 470μF, and the withstand voltage is 450V.

[0088] During the power correction execution process, the controller adjusts the generation parameters of the PWM signal according to the real-time control gain coefficient. The PWM signal frequency is set to 65kHz, and the duty cycle range is 0.1-0.9, which is generated by a digital pulse width modulator. The control algorithm is based on the current mode control principle, which combines the advantages of average current control and edge current control, and realizes accurate control of the input current waveform. When the zero crossing point of the input voltage is detected, the controller resets the internal parameters and starts a new control process for each switching period. For each switching period, the controller calculates the ideal switching on time according to the input voltage, output voltage and load current, and then adjusts it in combination with the real-time control gain coefficient to generate the final PWM control signal.

[0089] Exemplarily, when the server runs under 50% load condition, the measured power factor is 0.974, and the DC output voltage is 12.05V, both of which meet the preset target range. At this time, the real-time control gain coefficient is 1.797, the PWM signal duty cycle generated by the controller is 0.432, the power switch tube opening time is 6.65μs, and the closing time is 8.73μs. The input current waveform of the power factor correction circuit is approximately in phase with the input voltage waveform, the current total harmonic distortion rate is 3.6%, and the system efficiency reaches 94.2%. When the server load increases to 80%, the controller automatically adjusts the real-time control gain coefficient to 1.825, the PWM signal duty cycle increases to 0.563, the power switch tube opening time is extended to 8.66μs, the power factor is maintained at 0.985, the DC output voltage is stabilized at 11.96V, and the system efficiency is improved to 95.7%.

[0090] For smooth transition of different working states, the controller adopts gradual adjustment strategy to avoid transient impact caused by parameter mutation. When detecting load change, the control gain coefficient is adjusted at a rate of not more than 0.5% per control cycle, ensuring system stability. At the same time, the controller also has abnormal processing capability, which can quickly respond and take protective measures when detecting input voltage abnormality, output overload or short circuit fault state, such as reducing PWM frequency, limiting duty cycle maximum value or completely shutting down power switch tube, to prevent component damage.

[0091] To adapt to different server load characteristics, the controller has multiple working modes, including high efficiency mode, low noise mode and balance mode. In high efficiency mode, the controller gives priority to system efficiency, and the power factor target range is appropriately relaxed to 0.92-1.0; in low noise mode, the PWM frequency is increased to 100kHz, which is beyond the audible range of human ear, but the power factor and efficiency will be reduced; the balance mode balances between efficiency and noise, which is the default working mode of the system. Users can select the appropriate working mode according to the data center environment and server purpose.

[0092] The whole power factor correction method is realized by a digital controller, and the control code is stored in the program memory, which occupies about 35KB of storage space when running, and the main loop execution time is less than 15 microseconds, meeting the real-time control requirements. The controller also provides a communication interface to support interaction with the server management system, report running parameters and state information, receive configuration commands, and realize remote monitoring and management.

[0093] In this embodiment, by comparing the actual measured power factor value and DC output voltage value with the preset target range in real time, the system running state can be more comprehensively reflected, ensuring that the power supply system achieves optimal operation on multiple key parameters. A condition triggered control mechanism is adopted, and only when the power factor value and DC output voltage value meet the preset target range at the same time, the power correction operation is performed, avoiding one-sided optimization caused by single index reaching the target, and ensuring the improvement of the overall performance of the system.

[0094] In a second aspect of the embodiment of the present application, a power factor correction system for AC-DC conversion of a server power supply is provided, comprising: The first unit is used for collecting the input AC signal and DC bus output signal of the server power supply, and calculating the real-time power factor value according to the input AC signal; The second unit is configured to, if the real-time power factor value is less than a preset power factor threshold, construct a compensation function based on the real-time power factor value and calculate a real-time deviation amount, weight, accumulate and normalize the real-time deviation amount and a historical compensation value to obtain an initial compensation factor, correct the initial compensation factor by recursive calculation, generate a stable value and adjust an iteration step length according to a change rate of the stable value, generate a fast compensation amount and a gradual compensation amount, and combine the fast compensation amount and the gradual compensation amount to obtain a compensation parameter matrix. The third unit is configured to process the DC bus output signal by using an adaptive filtering algorithm and extract an effective fluctuation component, perform multi-dimensional fusion operation on the effective fluctuation component and the compensation parameter matrix to generate a comprehensive adjustment matrix, perform parameter decomposition based on the comprehensive adjustment matrix to separately calculate a current adjustment signal and a duty cycle adjustment signal, adjust an input current waveform according to the current adjustment signal, and adjust a conduction time of a power switch tube according to the duty cycle adjustment signal. The fourth unit is configured to acquire an adjusted power factor value and a DC output voltage value and determine whether the adjusted power factor value and the DC output voltage value meet a preset target range, and if so, perform power correction by using a power factor correction circuit.

[0095] In a third aspect, an electronic device is provided, including: A processor and a memory for storing processor-executable instructions, wherein the processor is configured to invoke the instructions stored in the memory to execute the method described above.

[0096] In a fourth aspect, a computer-readable storage medium is provided, which stores computer program instructions, and the computer program instructions are executed by a processor to implement the method described above.

[0097] The present application can be a method, device, system and / or computer program product. The computer program product can include a computer readable storage medium having computer readable program instructions stored therein, which are used to perform various aspects of the present application.

[0098] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A power factor correction method for AC / DC conversion of a server power supply, characterized in that: include: Collect the input AC signal and DC bus output signal of the server power supply, and calculate the real-time power factor value based on the input AC signal; If the real-time power factor value is less than a preset power factor threshold, a compensation function is constructed based on the real-time power factor value and a real-time deviation is calculated. The real-time deviation is weightedly accumulated and normalized with a historical compensation value to obtain an initial compensation factor. The initial compensation factor is error-corrected through recursive calculation to generate a stable value and adjust the iteration step size according to the rate of change of the stable value to generate a rapid compensation amount and a gradual compensation amount, which are combined to obtain a compensation parameter matrix; Adopting an adaptive filtering algorithm to process the DC bus output signal and extract an effective fluctuation component, performing a multi-dimensional fusion operation on the effective fluctuation component and a compensation parameter matrix to generate a comprehensive adjustment matrix, performing parameter decomposition based on the comprehensive adjustment matrix, and respectively calculating and obtaining a current adjustment signal and a duty cycle adjustment signal, adjusting the input current waveform according to the current adjustment signal, and adjusting the on-time of the power switch tube according to the duty cycle adjustment signal; The adjusted power factor value and DC output voltage value are obtained and judged whether they meet the preset target range. If so, power correction is performed through the power factor correction circuit.

2. The method according to claim 1, characterized in that Collect the input AC signal and DC bus output signal of the server power supply, and calculate the real-time power factor value based on the input AC signal, including: Collecting the input AC signal and DC bus output signal of the server power supply to obtain a voltage sampling sequence and a current sampling sequence, extracting zero-crossing time information in the voltage sampling sequence and the current sampling sequence, and generating a phase feature vector; Performing discrete Fourier transform on the voltage sampling sequence and the current sampling sequence respectively, extracting the amplitude and phase information of the fundamental frequency component and calculating the phase difference of the fundamental frequency component; The phase difference value is substituted into the cosine function for operation, and weighted correction is performed in combination with the phase characteristic vector to obtain a real-time power factor value.

3. The method according to claim 1, characterized in that If the real-time power factor value is less than a preset power factor threshold, a compensation function is constructed based on the real-time power factor value and a real-time deviation is calculated. The real-time deviation is weightedly accumulated and normalized with a historical compensation value to obtain an initial compensation factor, which includes: Obtaining a real-time power factor value, determining whether the real-time power factor value is less than a preset power factor threshold, and if so, performing a square operation on the difference between the real-time power factor value and a target power factor value to obtain a static deviation term, calculating an absolute value of a rate of change of the real-time power factor value to obtain a dynamic change rate term, and performing a weighted combination of the static deviation term and the dynamic change rate term to construct a compensation function; Integrating the compensation function within a preset integration time window to obtain a real-time deviation, and adding the real-time deviation to the historical compensation value at the previous moment to obtain a cumulative compensation value at the current moment; Calculate the rate of change of the real-time power factor value, use a reference gain coefficient as a base, and perform a power operation with the inverse of the rate of change as an exponent to obtain an adaptive gain factor, calculate the maximum and minimum values ​​of the cumulative compensation value, subtract the minimum value from the cumulative compensation value and divide the result by the difference between the maximum and the minimum values, and multiply the obtained normalized result by the adaptive gain factor to obtain an initial compensation factor.

4. The method according to claim 1, wherein The initial compensation factor is corrected by recursive calculation to generate a stable value. The iteration step size is adjusted according to the rate of change of the stable value to generate a rapid compensation amount and a gradual compensation amount. The compensation parameter matrix obtained by combining them includes: Using the initial compensation factor as an initial value of a recursive operation, multiplying the real-time deviation by a preset weight coefficient and then adding the result to the initial value to obtain a first iteration value, performing a recursive operation based on the first iteration value to obtain a recursive compensation result, wherein the result of each recursive operation is equal to the sum of the result of the previous recursive operation and the product of the real-time deviation and the preset weight coefficient; Mapping the real-time deviation, the rate of change of the real-time deviation, and the real-time power factor value to a three-dimensional fuzzy space and establishing a fuzzy rule base, and dynamically adjusting the fuzzy membership function of the fuzzy rule base to obtain an optimized fuzzy rule set; Performing a weighted smoothing operation on the recursive compensation result and the stable value at the previous moment according to the optimized fuzzy rule set to generate a stable value at the current moment, and substituting the real-time deviation and the rate of change of the stable value at the current moment into a sliding mode calculation to obtain a sliding mode calculation result; The rate of change of the stable value at the current moment is calculated based on the sliding mode operation result to obtain a rapid compensation amount, the real-time deviation amount is integrated to obtain a progressive compensation amount, the rapid compensation amount and its variation amount are used to construct an inner-loop compensation matrix, the progressive compensation amount and its variation amount are used to construct an outer-loop compensation matrix, the weighting coefficient is determined based on the optimized fuzzy rule set, and the inner-loop compensation matrix and the outer-loop compensation matrix are adaptively weighted and combined to obtain a compensation parameter matrix.

5. The method according to claim 1, wherein Adopting an adaptive filtering algorithm to process the DC bus output signal and extract the effective fluctuation component, and performing a multi-dimensional fusion operation on the effective fluctuation component and the compensation parameter matrix to generate a comprehensive adjustment matrix includes: Recursively calculating the DC bus output signal based on an adaptive filtering algorithm, and simultaneously solving the results of the recursive calculation to obtain a signal vector group, constructing a state estimation error covariance matrix based on the signal vector group, and sequentially performing matrix operations on the state estimation error covariance matrix with an observation matrix and a measurement noise covariance matrix to obtain an adaptive gain matrix; performing a weighted operation on the signal vector group using the adaptive gain matrix to obtain a filtered signal, performing a wavelet transform on the filtered signal to obtain a decomposition result, combining the decomposition result with a preset wavelet basis function and a scale factor to obtain a fluctuation feature of the filtered signal, and combining the fluctuation features to form an effective fluctuation component; Performing weight mapping on the effective fluctuation component to obtain a dynamic weight, constructing an adaptive weight matrix using the dynamic weight, and performing a tensor product operation on the adaptive weight matrix and the effective fluctuation component to obtain a fusion operation result; Subtract the preset expected output from the fusion operation result to obtain a deviation value, substitute the deviation value into the objective function for calculation to obtain the gradient of the weight coefficient, iteratively optimize the dynamic weight based on the gradient of the weight coefficient to obtain the optimized weight coefficient, multiply the optimized weight coefficient by the fusion operation result to obtain a compensation correction amount, and add the compensation correction amount to the fusion operation result to obtain a comprehensive adjustment matrix.

6. The method according to claim 1, characterized in that Performing parameter decomposition based on the comprehensive regulation matrix, respectively calculating and obtaining a current regulation signal and a duty cycle regulation signal, adjusting the input current waveform according to the current regulation signal, and adjusting the on-time of the power switch tube according to the duty cycle regulation signal includes: Performing eigenvalue decomposition on the comprehensive adjustment matrix to obtain an eigenvalue matrix and an eigenvector matrix, and performing parameter mapping operation on the eigenvalue matrix, a preset current adjustment coefficient, and a preset duty cycle adjustment coefficient to obtain initial adjustment parameters; performing a sine function operation on the initial adjustment parameter and the eigenvalues ​​in the eigenvalue matrix and multiplying the resultant result by a preset current gain coefficient to obtain a current adjustment signal, substituting the current adjustment signal into a differential operation and an integral operation to obtain a dynamic compensation signal, and combining the dynamic compensation signal with the current adjustment signal to obtain a compensated current adjustment signal; Performing a cosine function operation on the initial adjustment parameter and the eigenvalue in the eigenvalue matrix and superimposing a preset initial duty cycle value to obtain a duty cycle adjustment signal, and substituting the duty cycle adjustment signal into a hyperbolic tangent function to perform a dead zone compensation operation to obtain a compensated duty cycle adjustment signal; A feedback control signal is calculated based on the compensated current regulation signal and the compensated duty cycle regulation signal, the error of the feedback control signal is substituted into the integral operation to obtain an adaptive gain signal, the adaptive gain signal is added to the preset initial gain value to obtain a real-time control gain coefficient, and the input current waveform of the server and the conduction time of the power switch tube are adjusted based on the real-time control gain coefficient.

7. The method according to claim 1, characterized in that Obtaining the adjusted power factor value and DC output voltage value and determining whether they meet a preset target range, and if so, performing power correction through a power factor correction circuit includes: Obtain the adjusted power factor value and DC output voltage value; Comparing the power factor value and the DC output voltage value with a preset power factor target range and a preset voltage target range, respectively, to determine whether the power factor value and the DC output voltage value simultaneously satisfy the preset power factor target range and the preset voltage target range; When the power factor value and the DC output voltage value simultaneously satisfy the preset power factor target range and the preset voltage target range, power correction is performed according to the pre-acquired real-time control gain coefficient and the power factor correction circuit.

8. A power factor correction system for AC / DC conversion of a server power supply, used to implement the method according to any one of claims 1 to 7, characterized in that: include: The first unit is used to collect the input AC signal and DC bus output signal of the server power supply and calculate the real-time power factor value based on the input AC signal; a second unit configured to, if the real-time power factor value is less than a preset power factor threshold, construct a compensation function based on the real-time power factor value and calculate a real-time deviation, perform weighted accumulation and normalization processing on the real-time deviation and a historical compensation value to obtain an initial compensation factor, perform error correction on the initial compensation factor through recursive calculation, generate a stable value, adjust an iterative step size according to a rate of change of the stable value, generate a rapid compensation amount and a gradual compensation amount, and combine them to obtain a compensation parameter matrix; a third unit, configured to process the DC bus output signal using an adaptive filtering algorithm and extract an effective fluctuation component, perform a multi-dimensional fusion operation on the effective fluctuation component and a compensation parameter matrix to generate a comprehensive adjustment matrix, perform parameter decomposition based on the comprehensive adjustment matrix, and respectively calculate a current adjustment signal and a duty cycle adjustment signal, adjust the input current waveform according to the current adjustment signal, and adjust the conduction time of the power switch tube according to the duty cycle adjustment signal; The fourth unit is used to obtain the adjusted power factor value and DC output voltage value and determine whether they meet the preset target range. If so, power correction is performed through the power factor correction circuit.

9. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images