A method and system for overvoltage and overcurrent control of photovoltaic inverters in cold regions

Abstract

This application relates to the field of photovoltaic power generation technology, specifically to a method and system for overvoltage and overcurrent control of photovoltaic inverters in cold regions. The method includes: continuously acquiring the output current of the photovoltaic panel and the ambient light intensity, and dividing the output current into multiple sub-sequences; determining asynchronous characteristic values ​​representing interference intensity, performing local fluctuation correction on the sub-sequences, and obtaining a synchronous output sequence; extracting the differential change characteristics of the synchronous output sequence, and determining current increase / decrease coefficients based on the distribution differences of the differential change characteristics at different time periods to construct a feature vector for the current moment; inputting the feature vector into a pre-trained prediction model to obtain the predicted output current of the photovoltaic panel at the next moment; when the predicted output current exceeds the limit, adjusting the duty cycle of the DC / DC converter in advance according to its proportional relationship with the preset current allowable value to prevent overcurrent and overvoltage of the photovoltaic inverter. This effectively ensures the operational safety and lifespan of the photovoltaic inverter.

Description

Technical Field

[0001] This application relates to the field of photovoltaic power generation technology, specifically to a method and system for overvoltage and overcurrent control of photovoltaic inverters in cold regions. Background Technology

[0002] With the development of new energy technologies, on-site photovoltaic power generation using solar energy resources in cold regions is of great significance for reducing carbon emissions, improving energy self-sufficiency, and enhancing power supply reliability in remote areas. In photovoltaic power generation systems, a photovoltaic inverter is typically used to convert the direct current generated by the photovoltaic panels into constant-frequency, constant-voltage alternating current and feed it into the power grid. To prevent damage to the inverter's internal power devices due to overload, the DC voltage and current input to the inverter must be strictly limited within safe preset ranges.

[0003] In existing technologies, internal detection circuits are typically used to monitor the output status of photovoltaic panels in real time. When the output current or voltage exceeds the safety threshold, the control system issues an overload signal and passively adjusts the duty cycle of the inverter's front-end DC / DC converter to reduce voltage and limit current.

[0004] However, this traditional "post-event feedback" control strategy faces serious lag problems in practical applications. Especially in cold environments, phenomena such as snow suddenly sliding off the surface of photovoltaic panels, strong reflections from snow and ice, or clouds moving violently with strong winds often occur, causing drastic transient changes in local light intensity within a very short time, which in turn triggers transient surges in the output current of the photovoltaic panels. Because the front end of the photovoltaic system contains a large number of energy storage components such as inductors and capacitors, there is a significant electrical inertia at the physical level. Traditional passive feedback control can only start responding after the abnormal current has been generated and collected, and it cannot adapt to such sudden large disturbances in time. This easily leads to control overshoot, causing surge currents to enter and damage the inverter. Summary of the Invention

[0005] To address the aforementioned technical problems, the purpose of this application is to provide a method and system for overvoltage and overcurrent control of photovoltaic inverters in cold regions. The specific technical solution adopted is as follows:

[0006] In a first aspect, embodiments of this application provide an overvoltage and overcurrent control method for a photovoltaic inverter in cold regions, applied to a photovoltaic power generation system. The photovoltaic power generation system includes a photovoltaic panel, a DC / DC converter, and a photovoltaic inverter connected in sequence. The method includes the following steps:

[0007] The output current of the photovoltaic panel and the light intensity of the surrounding environment are continuously acquired, and the output current is divided into multiple subsequences;

[0008] Based on the timing difference between the output current and the illumination intensity, determine the asynchronous characteristic value that characterizes the interference intensity;

[0009] Based on the overall change trend of each subsequence and the asynchronous characteristic values ​​of the corresponding time period, local fluctuation correction is performed on the subsequence to obtain a synchronous output sequence.

[0010] Extract the differential change features of the synchronous output sequence, and determine the current increase / decrease coefficient based on the distribution differences of the differential change features at different time periods, so as to construct the feature vector at the current moment;

[0011] The feature vector at the current moment is input into a pre-trained prediction model to obtain the predicted output current of the photovoltaic panel at the next moment.

[0012] When the predicted output current exceeds the limit, the duty cycle of the DC / DC converter is adjusted in advance according to its ratio with the preset current allowable value to limit the actual current input to the photovoltaic inverter and prevent the photovoltaic inverter from overcurrent and overvoltage.

[0013] In one embodiment, the sampling frequency of the output current is greater than the sampling frequency of the light intensity, and the number of subsequences divided by the output current is the same as the number of light intensity samplings.

[0014] In one embodiment, determining the asynchronous characteristic value characterizing the interference intensity includes:

[0015] Calculate the average value of the output current in each subsequence to obtain the short-time output current that is aligned with the light intensity in the corresponding time period.

[0016] The short-time output current and light intensity are normalized respectively. The synchronization difference is calculated by the overall difference between the short-time output current and light intensity in the same time period after normalization.

[0017] Based on the synchronization difference and the local differences between the normalized short-time output current and the illumination intensity in each time period, the asynchronous characteristic value for the corresponding time period is determined.

[0018] In one embodiment, the process of determining the asynchronous feature value is as follows:

[0019] For each time period, calculate the data difference between the normalized short-time output current and the normalized illuminance; use the synchronization difference to correct the absolute deviation of the data difference, and use the average level of the normalized short-time output current and the normalized illuminance to scale down the corrected absolute deviation proportionally.

[0020] The scaled values ​​are assigned corresponding signs based on the positive or negative direction of the data difference to obtain the asynchronous characteristic values ​​for each time period.

[0021] In one embodiment, the process of obtaining the synchronization output sequence is as follows:

[0022] Extreme interference data in the subsequence is removed, and trend fitting is performed on the remaining data to obtain the fitting benchmark value corresponding to each original data point in the subsequence; based on the difference between the original data point and the fitting benchmark value, the fluctuation deviation characterizing the deviation of the original data point from the overall trend is calculated.

[0023] By combining the fluctuation deviation and the asynchronous characteristic value of the corresponding time period, the correction intensity of each original data point is determined; the correction intensity is used to correct each original data point, and the corrected data points are aggregated to obtain the synchronous output sequence.

[0024] In one embodiment, determining the correction strength for each of the original data points includes:

[0025] Using the fitting benchmark value corresponding to each of the original data points as a reference, the relative deviation ratio of the fluctuation deviation is calculated; the preset base ratio, the asynchronous feature value and the relative deviation ratio are superimposed and fused to generate the correction intensity.

[0026] In one embodiment, determining the current increase / decrease coefficient includes:

[0027] The synchronous output sequence is subjected to adjacent time difference operation and the absolute amplitude is extracted. The absolute amplitude is then evenly divided into a first segment and a second segment along the time sequence.

[0028] Calculate the average value of the absolute amplitude in the first segment and the second segment respectively; determine the current increase / decrease coefficient based on the relative deviation between the average values ​​of the absolute amplitude in the first segment and the second segment.

[0029] In one embodiment, constructing the feature vector at the current moment includes:

[0030] The absolute amplitudes are arranged in chronological order to form a current change sequence. The current change sequence is then combined with the current increase / decrease coefficients to generate a feature vector for the current moment.

[0031] In one embodiment, the advance adjustment of the duty cycle of the DC / DC converter includes:

[0032] Calculate the ratio of the preset current tolerance value to the predicted output current, and use the ratio to scale the actual duty cycle of the DC / DC converter at the current moment.

[0033] Secondly, embodiments of this application also provide an overvoltage and overcurrent control system for a photovoltaic inverter in cold regions, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the steps of any of the methods described above.

[0034] This application has at least the following beneficial effects:

[0035] This application mines the time-series difference between light intensity and output current, calculates the asynchronous characteristic value that characterizes the interference intensity, and performs local fluctuation correction on the data by combining the overall change trend of each subsequence. In this way, transient electromagnetic spikes and glitches caused by extreme cold conditions are accurately removed. Without destroying the original macroscopic evolution trend of the current, a highly smooth and reliable synchronous output sequence is obtained, avoiding the misleading effect of the underlying false noise on the subsequent prediction stage, and greatly enhancing the anti-interference capability of photovoltaic inverter overvoltage and overcurrent control.

[0036] Secondly, by extracting the differential change features of the synchronous output sequence and comparing the distribution differences of these features at different times, the current increase / decrease coefficient was determined. This deeply quantified the divergence or convergence state of photovoltaic current under severe light fluctuations, accurately captured the acceleration trend of current evolution, and provided highly forward-looking prior feature inputs for subsequent prediction models. This significantly improved the model's sensitivity to abnormal surges and prediction accuracy when facing complex and severe weather.

[0037] Finally, by predicting the output current for the next time period in advance using a predictive model, and when the predicted value exceeds the limit, the duty cycle of the front-end DC / DC converter is reduced in advance based on its ratio with the preset current allowable value. The time difference brought about by the prediction offsets the response delay of the physical hardware, and can cut off the overcurrent and overvoltage hazards at the source before the destructive surge current actually invades the inverter. This fundamentally solves the problem of the lag in traditional detection and feedback mechanisms, and effectively ensures the operational safety and lifespan of the photovoltaic inverter. Attached Figure Description

[0038] Figure 1 This is a flowchart illustrating the steps of an overvoltage and overcurrent control method for a photovoltaic inverter in cold regions, provided as an embodiment of this application. Detailed Implementation

[0039] The following description, in conjunction with the accompanying drawings, details the specific scheme of the overvoltage and overcurrent control method and system for a photovoltaic inverter in cold regions provided in this application.

[0040] Please see Figure 1This document illustrates a flowchart of an overvoltage and overcurrent control method for a photovoltaic inverter in cold regions, provided in one embodiment of this application. The method is applied to a photovoltaic power generation system, which includes a photovoltaic panel, a DC / DC converter, and a photovoltaic inverter connected in sequence. The method includes the following steps:

[0041] S1, continuously acquire the output current of the photovoltaic panel and the light intensity of the surrounding environment, and divide the output current into multiple sub-sequences.

[0042] Specifically, photovoltaic cells are not nonlinear resistive elements; they possess nonlinear volt-ampere characteristics (IV curves). However, within the boundary range of normal operation and overload of a photovoltaic power generation system, its output voltage and output current exhibit a monotonic mapping relationship. When a transient surge in sunlight causes a spike in output power, overcurrent is often a direct precursor to overvoltage on the inverter's DC bus due to energy overload. Therefore, based on the mapping relationship of the photovoltaic cell's volt-ampere characteristics, effective suppression of overvoltage can be achieved simultaneously by rigorously predicting and limiting the input DC current on the photovoltaic inverter side. The specific data acquisition and processing process is as follows:

[0043] First, the photovoltaic panel's power generation current is collected using a high-frequency electrical parameter acquisition device (a smart meter in this embodiment) configured at the output end of the photovoltaic panel (i.e., the front end of the DC / DC converter). The current data acquisition frequency is set to 100Hz, and data from the most recent minute before the current moment is extracted, resulting in 6000 current data points. The collected current data is arranged in chronological order according to timestamps. Subsequently, to eliminate electromagnetic interference and high-frequency noise in the measurement, a mean filtering algorithm with a preset sliding window size (a window length of 5 in this embodiment) is used to smooth and denoise the arranged current data. The sequence of the 6000 consecutive denoised data points is recorded as the photovoltaic output current sequence, used to characterize the high-frequency output current fluctuation state of the photovoltaic panel during power generation.

[0044] Simultaneously, an environmental meteorological sensor (in this embodiment, a total solar radiation meter) positioned near the photovoltaic panel synchronously collects the solar irradiance received by the panel. The solar irradiance data collection frequency is set to 1 Hz, and data from the most recent minute preceding the current moment is extracted, resulting in a total of 60 solar irradiance data points. The collected solar irradiance data are arranged in chronological order according to their timestamps and then denoised using a corresponding matching mean filtering algorithm. The sequence of the 60 consecutive denoised data points is recorded as the solar irradiance sequence, used to characterize the low-frequency macroscopic variation trend of the current ambient light.

[0045] The physical essence of the photovoltaic effect is that photons excite semiconductor materials to generate electron-hole pairs, which then instantaneously separate under the influence of the built-in electric field of the PN junction to form a current. This energy conversion process occurs on the femtosecond to picosecond scale. Therefore, on a macroscopic scale, changes in illumination and current response are nearly instantaneously synchronized, with the system delay time approaching zero. In actual power generation, the output current intensity of the photovoltaic panel is strongly positively correlated with the ambient light intensity. Although the light intensity may experience unstable high-frequency disturbances due to factors such as rapid cloud cover, wind, and snow, within the observation window, the macroscopic evolution of the photovoltaic panel's output current still maintains a high degree of consistency with the trend of light intensity changes.

[0046] To perform synchronous correlation analysis of illumination and current at the same temporal resolution, it is first necessary to align data sequences with different acquisition frequencies. Specifically, the photovoltaic output current sequence with 6000 data points is uniformly divided into N equal-length sub-segments, denoted as photovoltaic output current sub-sequences, or simply sub-sequences. In this embodiment, N=60, meaning each sub-sequence corresponds to a 1-second time period and contains 100 current data points.

[0047] S2, based on the temporal difference between the output current of the photovoltaic panel and the light intensity of the surrounding environment, determine the asynchronous characteristic value that characterizes the interference intensity.

[0048] Specifically, the arithmetic mean of all elements within each photovoltaic output current subsequence is calculated to obtain short-time output currents that are aligned with the irradiance within the corresponding time period. The resulting N means are then arranged chronologically to obtain a downsampled short-time output current sequence. Through this aggregation operation, the number of elements in the short-time output current sequence is exactly the same as the number of data points in the irradiance sequence, thus characterizing the average output current intensity of the photovoltaic panel on a per-second scale.

[0049] Furthermore, to eliminate the difference in physical dimensions between current (A) and illumination intensity (W / m²), a maximum value normalization algorithm is used to map the values ​​of the illumination intensity sequence and the short-time output current sequence within the current observation window to the [0,1] interval, resulting in normalized illumination sequences and normalized current sequences, respectively. Based on the strong correlation at the physical level, under ideal interference-free conditions, the time-domain fluctuation trajectories of these two sets of normalized sequences should have a very high degree of overlap. If the discrete difference between the two increases significantly, it indicates that the synchronization between illumination and current is degraded due to distortion factors such as internal electromagnetic interference or extreme non-uniform shading.

[0050] Therefore, the elements of the normalized illumination sequence and the normalized current sequence at each corresponding time point are subtracted sequentially to obtain the residual sequence. Then, the arithmetic mean of the absolute values ​​of all elements in this residual sequence is calculated, and this mean is defined as the synchronization difference. This synchronization difference characterizes the overall fundamental deviation level between illumination and current changes within the current observation period, providing a background noise reference for subsequently measuring anomalous disturbances within local time periods.

[0051] Next, based on the local correlation between light intensity and output current, the asynchronous characteristic value for each time period (i.e., the time period corresponding to each photovoltaic output current subsequence, which is 1 second in this embodiment) is calculated to quantify the asynchronous distortion characteristics between current and light intensity within that time period. The specific calculation formula is as follows:

[0052]

[0053]

[0054] In the formula, This represents the asynchronous characteristic value within the time period corresponding to the i-th photovoltaic output current subsequence. This represents the i-th element in the normalized current sequence. This represents the i-th element in the normalized illumination sequence. This represents the synchronization difference calculated above. The sign value represents the i-th time period and is used to indicate the direction of data deviation. sgn() is the sign function, which takes the value 1 when the variable is positive, -1 when it is negative, and 0 when it is 0.

[0055] It should be noted that if the denominator is 0 during the calculation of asynchronous eigenvalues, the asynchronous eigenvalue is directly set to 0.

[0056] For the calculation method of asynchronous eigenvalues, the numerator first calculates the absolute deviation between illumination and current within the current time period. Then, the synchronization difference TP, which represents the inherent noise floor of the system, is subtracted to extract the true net abnormal fluctuation intensity; subsequently, it is multiplied by the sign value. The directional properties of the abnormal distortion are restored. The denominator uses the normalized mean of illumination and current within that time period as a base, and the aforementioned deviation is scaled proportionally. This dynamic benchmark scaling eliminates the interference of absolute illumination intensity on the evaluation results. The resulting asynchronous characteristic value... The larger the absolute value, the stronger the external sudden interference or electromagnetic noise that the output current is subjected to during the current time period, and the more serious the degradation of the synchronization state.

[0057] S3. Based on the overall trend of each photovoltaic output current subsequence and the asynchronous characteristic value of the corresponding time period, local fluctuation correction is performed on the photovoltaic output current subsequence to obtain the synchronous output sequence.

[0058] Specifically, in practical engineering, although the output current of photovoltaic panels may experience localized high-frequency fluctuations due to electromagnetic interference, ambient temperature drift, and sensor measurement noise, the fluctuations in external natural solar irradiance are relatively smooth within an extremely short observation window of one second, with very few instantaneous step-like abrupt changes. Therefore, within the extremely short time domain of a single photovoltaic output current subsequence, the overall evolution of the current exhibits a clear linear and continuous characteristic, with relatively small changes in its macroscopic baseline.

[0059] Based on this characteristic, the data within each photovoltaic output current subsequence is first subjected to extreme value filtering to eliminate the interference of transient spike noise on the current reference trend. Specifically, the elements within the photovoltaic output current subsequence are sorted by amplitude, and the n largest and n smallest elements are removed. Typically, to ensure fitting accuracy, the total number of extreme elements removed is approximately 10% of the total subsequence length. Given that each subsequence in this embodiment contains 100 elements, n=5 is set.

[0060] Next, the remaining elements after removing extreme values ​​are used as input data for the linear least squares method. A univariate linear regression is performed using the original time series index of each element in the subsequence as the independent variable and the current value of the element as the dependent variable to calculate the fitted linear equation (including slope and intercept) corresponding to the photovoltaic output current subsequence. Subsequently, the time series indices of all elements in the subsequence, including those that have been removed, are uniformly substituted into the fitted linear equation to obtain the fitting baseline value corresponding to each of the original data points within the photovoltaic output current subsequence. This fitting baseline value reflects the smooth evolution trend of the photovoltaic output current within this short-time window after filtering out high-frequency abnormal noise.

[0061] Subsequently, the original data points are dynamically corrected by comparing them with the fitted benchmark value. The greater the deviation between the original data point and the fitted benchmark value, the more severe the distortion caused by high-frequency interference, and the stronger the correction applied should be accordingly. Based on this, and combined with the asynchronous characteristic value calculated in step S2, the synchronous output current of each original data point in each photovoltaic output current subsequence is calculated to obtain a current sequence that filters out spurious disturbances and more accurately represents the true electrical state of the photovoltaic panel. The specific calculation formula is as follows:

[0062]

[0063] In the formula, This represents the synchronous output current of the j-th original data point in the i-th photovoltaic output current subsequence; This represents the j-th original data point in the i-th photovoltaic output current subsequence; This represents the fitted baseline value corresponding to the j-th original data point in the i-th photovoltaic output current subsequence; This represents the asynchronous characteristic value within the time period corresponding to the i-th photovoltaic output current subsequence; This is a preset zero-division adjustment factor with the same dimension as the physical unit of current (A), used to prevent division-to-zero errors caused by the fitted current value approaching 0 in extreme dark conditions. The value range is [0.1A, 1A], and it is set to 1A in this embodiment.

[0064] Regarding the formula for calculating the synchronous output current, the first term "1" within the parentheses represents the base ratio, ensuring that the current remains unchanged when there is no interference; the second term... The first term represents the asynchronous characteristic value of light and current within the current 1-second time period, used for macroscopic weighting compensation of the distortion of the entire data segment; the second term represents the relative fluctuation deviation ratio, with the numerator being... The absolute error representing the deviation of microscopic data points from linearity is represented by an adjustment factor with units of current in the denominator. The fitting benchmark value is then obtained. Therefore, by adding the base ratio, macro-weight compensation, and relative fluctuation deviation ratio, the dynamic correction strength, which characterizes the overall distortion of the original data point, can be obtained. Finally, this dimensionless correction strength is compared with the original data point... By multiplying, glitches that deviate from the reference can be adaptively corrected, resulting in a more accurate and smooth synchronous output current.

[0065] After completing the micro-correction of each original data point, the high-frequency subsequences need to be re-aggregated. Specifically, the arithmetic mean of all synchronous output currents within each photovoltaic output current subsequence is calculated. The calculated means are then reordered and reassembled according to the chronological order of the original photovoltaic output current subsequences to obtain a synchronous output sequence with a length consistent with the light intensity sequence (i.e., 60 elements in this embodiment, with the sampling frequency reduced to 1Hz). This sequence is used to characterize the macroscopic output current change trajectory of the photovoltaic panel after filtering out high-frequency noise.

[0066] S4. Extract the differential change features of the synchronous output sequence, and determine the current increase / decrease coefficient based on the distribution differences of the differential change features at different time periods, so as to construct the feature vector at the current moment.

[0067] Specifically, for the synchronous output sequence, the magnitude of the difference between data points at adjacent time points reflects the severity of the abrupt change in current evolution. A higher degree of abrupt change indicates a more significant electrical characteristic excited by unstable illumination (such as rapidly passing clouds). Therefore, the synchronous output sequence is used as the input sequence for a first-order difference algorithm. The difference between data points at adjacent time points is calculated, and the absolute value of all first-order difference values ​​output by the algorithm is taken. Subsequently, these absolute difference values ​​are arranged sequentially according to their corresponding time numbers to obtain a current change sequence of length N−1. This current change sequence effectively characterizes the transient fluctuation amplitude of the output current caused by dynamic fluctuations in the photovoltaic panel's illumination intensity.

[0068] To capture the overall evolution trend of current fluctuations within the current observation window, it is necessary to compare the distribution characteristics of the current change sequence in different time periods. If the mean fluctuations in the first and second halves of the sequence deviate slightly, it indicates that the current change trend maintains a relatively stable dynamic equilibrium; if the deviation is significant, it indicates that the current fluctuation trend has shown obvious differentiation or even runaway, and the degree of electrical disturbance is intensifying.

[0069] Given that the current change sequence extracted in this embodiment has a length of 59, in order to ensure symmetrical segmentation, the current change sequence is first truncated at the end, removing the last element to make the sequence length even. Then, the truncated sequence is strictly divided into two equal segments: a first segment (denoted as the first segment) and a second segment (denoted as the second segment).

[0070] Based on this, the arithmetic mean of all data elements in the latter half of the segment and the first half of the segment are calculated respectively. Based on the degree of deviation between the two means, the current increase / decrease coefficient of the photovoltaic panel is calculated to quantitatively characterize the divergence or convergence state of the dynamic fluctuation of the photovoltaic output current under complex illumination. The calculation formula is as follows:

[0071]

[0072] In the formula, Characterizing the current increase / decrease coefficient of a photovoltaic panel; , These represent the arithmetic mean of the elements in the latter half and the first half of the current change sequence, respectively. The zero-prevention adjustment factor is a preset value with the same dimension as the physical unit of current (A). It is used to prevent the abnormal collapse of the formula denominator when the light is absolutely stable and the first half of the current has no fluctuation. The value range is [0.1A, 1A]. In this embodiment, it is set to 1A.

[0073] The formula for calculating the current increase / decrease coefficient shows that the numerator on the right side of the equation represents the absolute deviation of the mean values ​​of the two halves, while the denominator incorporates an adjustment factor with a current unit. Using the mean of the first half as the base, the ratio result is strictly a dimensionless relative rate of change parameter. The current increase / decrease coefficient comprehensively reflects the degree of deterioration or attenuation of the current disturbance intensity near the current moment relative to historical periods, providing highly condensed trend prior features for subsequent models.

[0074] To overcome the passive control lag caused by the electrical inertia of energy storage components such as large capacitors and inductors in photovoltaic power generation systems, and to proactively regulate potential overcurrent and overvoltage in the inverter, this embodiment employs deep learning technology to predict future input current trends. Based on the aforementioned steps, the current change sequence and the current increase / decrease coefficient corresponding to different operating times are obtained. These two are then concatenated and fused to construct a feature vector characterizing the complex photoelectric fluctuation mechanism at each moment. Specifically, the current change sequence is expanded, and the current increase / decrease coefficient is added as a new dimension to the end, forming a one-dimensional feature vector.

[0075] S5 inputs the feature vector of the current moment into the pre-trained prediction model to obtain the predicted output current of the photovoltaic panel at the next moment.

[0076] Specifically, a historical dataset for supervised learning of the neural network is constructed. From the historical operation database of the photovoltaic power plant, sample data pairs corresponding to M discrete historical moments are extensively collected. Each sample data pair contains two parts: input data and label data. The input data is the feature vector extracted for that historical moment; the label data is the average value of the actual output current for the next time period corresponding to that historical moment. To ensure that the model can fully learn the surge electrical characteristics caused by transient high sunlight or drastic weather changes in cold regions, m historical records of extreme overvoltage and overcurrent conditions are specifically sampled from the collected M sample data pairs. In this embodiment, the total sample size M = 5000, and the extreme condition sample size m = 500, thereby constructing a training set with a balanced long-tail feature distribution.

[0077] Next, an LSTM (Long Short-Term Memory) neural network model is constructed. The LSTM model, with its built-in forget gate and memory unit structure, can deeply mine the long-distance temporal dependencies inherent in feature sequences. The feature vectors of each sample in the constructed training set are used as input to the LSTM neural network model, and the predicted value output by the network is compared with the corresponding true "average value of the actual output current in the next time period" label. In terms of training configuration, the Adam optimization algorithm is used as the optimizer for parameter updates, with an initial learning rate of 0.001 and the mean squared error (MSE) selected as the loss function. The prediction error is calculated through forward propagation, and backward propagation is performed along the time steps to iteratively update the connection weight parameters within the network. Training stops when the loss function on the validation set converges to a stable state or reaches the preset maximum number of iterations, thus obtaining a well-trained predictive neural network model with good generalization performance, capable of accurately predicting the average output current of the photovoltaic panel in the next time period. In this embodiment, the maximum number of iterations is set to 200.

[0078] During the online operation phase, the feature vector of the current moment is extracted in real time and input into the pre-trained predictive neural network model. The model immediately feeds forward and outputs the predicted average output current of the photovoltaic panel for the next time period. Subsequently, this predicted average output current is compared and verified with the preset current tolerance value set inside the photovoltaic inverter. When the predicted average output current is less than or equal to the preset current tolerance value, it indicates that the photovoltaic power generation system is in a safe steady state, and the photovoltaic inverter does not face the risk of overvoltage or overcurrent in the next time period, and the current control command continues to operate. Conversely, when the predicted average output current is greater than the preset current tolerance value, it indicates that a sudden change in solar radiation in cold regions is about to break through the electrical inertia of the system, and the photovoltaic inverter will face the risk of overcurrent or even overvoltage breakdown in the next time period.

[0079] S6, when the predicted output current exceeds the limit, adjust the duty cycle of the DC / DC converter in advance according to its ratio with the preset current allowable value to limit the actual current input to the photovoltaic inverter and prevent the photovoltaic inverter from overcurrent and overvoltage.

[0080] Specifically, when the predicted average output current exceeds the preset current tolerance, current limiting and voltage reduction operations need to be implemented immediately. This triggers the current limiting response of the DC / DC converter located between the photovoltaic panel output port and the photovoltaic inverter. Based on the degree of exceeding the predicted current limit, the PWM drive duty cycle of the DC / DC converter is proportionally reduced. The corrected dynamic duty cycle adjustment formula is as follows:

[0081]

[0082] In the formula, This indicates the target duty cycle of the DC / DC converter in the next time period after the current limiting adjustment; This indicates the actual duty cycle of the DC / DC converter at the current moment. This represents the average predicted output current of the photovoltaic panel for the next time period, as predicted by the model. This indicates the preset current tolerance value of the photovoltaic inverter. To ensure sufficient safety margin, this preset current tolerance value is typically set to 90% of the maximum physical current that the inverter nameplate can receive. For example, if the maximum allowable input current for a specific inverter model is 30A, then in this embodiment... The value is 27A, and the specific value can be flexibly adjusted according to the actual inverter hardware specifications installed on site.

[0083] Through the aforementioned advanced proportional scaling control, the energy conduction ratio of the DC / DC converter can be reduced in advance before the actual surge current reaches the photovoltaic inverter. This effectively offsets the physical delay caused by energy storage components such as inductors, cuts off the risk of overvoltage and overcurrent at the source, and effectively ensures the safe and stable operation of photovoltaic equipment in cold regions under extreme disturbances.

[0084] Based on the same inventive concept as the above method, this application embodiment also provides an overvoltage and overcurrent control system for a cold-region photovoltaic inverter, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of any one of the above-described methods for overvoltage and overcurrent control of a cold-region photovoltaic inverter.

Claims

1. A method for overvoltage and overcurrent control of a photovoltaic inverter in cold regions, applied to a photovoltaic power generation system, wherein the photovoltaic power generation system comprises a photovoltaic panel, a DC / DC converter, and a photovoltaic inverter connected in sequence, characterized in that, The method includes the following steps: The output current of the photovoltaic panel and the light intensity of the surrounding environment are continuously acquired, and the output current is divided into multiple subsequences; Based on the timing difference between the output current and the illumination intensity, determine the asynchronous characteristic value that characterizes the interference intensity; Based on the overall change trend of each subsequence and the asynchronous characteristic values ​​of the corresponding time period, local fluctuation correction is performed on the subsequence to obtain a synchronous output sequence. Extract the differential change features of the synchronous output sequence, and determine the current increase / decrease coefficient based on the distribution differences of the differential change features at different time periods, so as to construct the feature vector at the current moment; The feature vector at the current moment is input into a pre-trained prediction model to obtain the predicted output current of the photovoltaic panel at the next moment. When the predicted output current exceeds the limit, the duty cycle of the DC / DC converter is adjusted in advance according to its ratio with the preset current allowable value to limit the actual current input to the photovoltaic inverter and prevent the photovoltaic inverter from overcurrent and overvoltage.

2. The overvoltage and overcurrent control method for a photovoltaic inverter in cold regions as described in claim 1, characterized in that, The sampling frequency of the output current is greater than the sampling frequency of the light intensity, and the number of subsequences divided by the output current is the same as the number of light intensity samplings.

3. The overvoltage and overcurrent control method for a cold-region photovoltaic inverter as described in claim 2, characterized in that, The determination of the asynchronous characteristic value characterizing the interference intensity includes: Calculate the average value of the output current in each subsequence to obtain the short-time output current that is aligned with the light intensity in the corresponding time period. The short-time output current and light intensity are normalized respectively. The synchronization difference is calculated by the overall difference between the short-time output current and light intensity in the same time period after normalization. Based on the synchronization difference and the local differences between the normalized short-time output current and the illumination intensity in each time period, the asynchronous characteristic value for the corresponding time period is determined.

4. The overvoltage and overcurrent control method for a cold-region photovoltaic inverter as described in claim 3, characterized in that, The process for determining the asynchronous feature values ​​is as follows: For each time period, calculate the data difference between the normalized short-time output current and the normalized illuminance; use the synchronization difference to correct the absolute deviation of the data difference, and use the average level of the normalized short-time output current and the normalized illuminance to scale down the corrected absolute deviation proportionally. The scaled values ​​are assigned corresponding signs based on the positive or negative direction of the data difference to obtain the asynchronous characteristic values ​​for each time period.

5. The overvoltage and overcurrent control method for a cold-region photovoltaic inverter as described in claim 1, characterized in that, The process of obtaining the synchronous output sequence is as follows: Extreme interference data in the subsequence is removed, and trend fitting is performed on the remaining data to obtain the fitting benchmark value corresponding to each original data point in the subsequence; based on the difference between the original data point and the fitting benchmark value, the fluctuation deviation characterizing the deviation of the original data point from the overall trend is calculated. By combining the fluctuation deviation and the asynchronous characteristic value of the corresponding time period, the correction intensity of each original data point is determined; the correction intensity is used to correct each original data point, and the corrected data points are aggregated to obtain the synchronous output sequence.

6. The overvoltage and overcurrent control method for a photovoltaic inverter in cold regions as described in claim 5, characterized in that, Determining the correction strength for each of the original data points includes: Using the fitting benchmark value corresponding to each of the original data points as a reference, the relative deviation ratio of the fluctuation deviation is calculated; the preset base ratio, the asynchronous feature value and the relative deviation ratio are superimposed and fused to generate the correction intensity.

7. The overvoltage and overcurrent control method for a cold-region photovoltaic inverter as described in claim 1, characterized in that, The determination of the current increase / decrease coefficient includes: Perform adjacent timing difference operation on the synchronous output sequence and extract the absolute amplitude, then divide the absolute amplitude evenly into a first segment and a second segment along the time sequence; Calculate the average value of the absolute amplitude in the first segment and the second segment respectively; determine the current increase / decrease coefficient based on the relative deviation between the average values ​​of the absolute amplitude in the first segment and the second segment.

8. The overvoltage and overcurrent control method for a cold-region photovoltaic inverter as described in claim 7, characterized in that, The construction of the feature vector at the current moment includes: The absolute amplitudes are arranged in chronological order to form a current change sequence. The current change sequence is then combined with the current increase / decrease coefficients to generate a feature vector for the current moment.

9. The overvoltage and overcurrent control method for a cold-region photovoltaic inverter as described in claim 1, characterized in that, The advance adjustment of the duty cycle of the DC / DC converter includes: Calculate the ratio of the preset current tolerance value to the predicted output current, and use the ratio to scale the actual duty cycle of the DC / DC converter at the current moment.

10. A cold-region photovoltaic inverter overvoltage and overcurrent control system, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1-9.

Images