A wind farm active power optimization control method and system

Abstract

The application provides a wind farm active power optimization control method and system, which comprises the following steps: collecting and processing meteorological parameters and wind power data; taking the meteorological parameters as the input of a prediction model and taking the wind power data as the output of the prediction model to perform model training; inputting the meteorological parameters into the prediction model to obtain the prediction value of the wind power after the training is completed; generating wind power merging data; performing traversal search on the wind power merging data in a fixed time window to obtain the inflection point of data form change; calculating the active power change amount in each rising or falling form of the wind power; identifying the index out-of-limit interval according to the maximum value of the active power change amount in the time window; and performing active power optimization control according to the identification result. The wind farm active power optimization control is realized by combining the wind farm short-term wind power prediction, the wind farm minimum active power output, the active power change limit value and other factors.

Description

Technical Field

[0001] This invention relates to the field of power system technology, and more specifically to a method and system for optimizing active power control in wind farms. Background Technology

[0002] Wind power, as an increasingly mature form of renewable energy generation, has become a new type of power energy with the largest potential for development and industrialization globally. While wind power alleviates the energy crisis and environmental pressures to some extent, large-scale wind power grid connection also brings some adverse effects to the power system. Because wind power heavily relies on wind energy, but the intensity and direction of wind energy are random and uncertain, the controllability of wind farm output power decreases. Large-scale wind power integration will bring power surges to the grid, affecting the safe and stable operation of the power system and increasing the difficulty of power system dispatch. The volatility of active power from wind farms poses a significant challenge to the safe operation of the power grid. From a development trend perspective, controlling the reliable and stable operation of wind farms and achieving active power control are development requirements for the safe grid connection and operation of large-scale wind farms.

[0003] In existing technologies, automatic active power control includes three main control parameters: power control cycle, power dead zone, and control step size. These parameters are manually adjusted to achieve a given value for the rate of change of active power at the power plant. Wind farms are affected by cloudy weather, cloud cover, and gusts during operation, leading to sharp drops or rises in power output. Existing methods for active power control in wind farms often only consider the economics of active power output and the scenarios of power-limited dispatching, without effectively limiting the rate of change of active power. This results in many wind farms having maximum active power variations exceeding relevant national standards, making the current technology less than ideal for controlling active power in wind farms. Therefore, it is necessary to study active power control methods for wind farms based on grid connection standards to ensure the safe and stable operation of the power grid.

[0004] The invention patent with application number 201410120142.1 discloses a wind farm active power optimization control method that takes into account the maximum value of active power variation. By combining the calculation of ultra-short-term wind power prediction, the maximum value of active power variation, and real-time active power, the active power of the wind farm can be stably controlled. Under the premise that the maximum value of active power variation of the wind farm in one minute and ten minutes meets the relevant requirements of GB / T19963-2011 "Technical Regulations for Wind Farm Access to Power System", the real-time active power of the wind farm is controlled in real time according to the grid dispatch requirements to reduce the impact of active power fluctuations on the grid.

[0005] Patent application number 202111501370.X discloses a method and device for controlling the active power change rate of a centralized renewable energy power station. This method can determine the station's operating mode based on self-generation control commands. When the station's operating mode is free generation, it calculates the change in active power within a preset time window. If the change in active power exceeds a threshold, the active power command is adjusted based on the change in active power and the threshold, resulting in an adjusted active power command with a set effective duration. The adjusted active power command and the effective duration are then sent to the station. This method improves the operational stability of renewable energy power stations.

[0006] Both of the above methods calculate active power changes using traditional methods. If the analysis doesn't differentiate between increases and decreases in active power, it can lead to inaccurate active power monitoring information. Therefore, it is essential to strengthen the monitoring of wind farm output power changes through both early warning and alarm mechanisms to reduce power frequency fluctuations caused by wind power grid connection and ensure stable grid operation.

[0007] Therefore, it can be seen that there are several problems with the existing active power optimization control technology for wind farms. First, it does not consider using short-term wind power prediction technology, so it is impossible to predict the short-term trend of active power in wind farms, and therefore impossible to predict whether there will be phenomena of power fluctuation amplitude that does not meet the requirements. Second, the measurement of active power change is inaccurate. The method of using the range of power values ​​within a fixed time window as the maximum value of active power change is questionable. Obviously, it does not distinguish well whether the power situation is rising or falling, and it cannot provide accurate start and end time information when the change amplitude exceeds the limit. Summary of the Invention

[0008] To address the problems existing in the prior art, the present invention aims to provide a method and system for optimizing the active power control of wind farms. By combining factors such as short-term wind power forecasting, minimum active power output of wind farms, and active power variation limits, the method achieves optimized control of active power in wind farms, ensuring that the maximum active power variation meets the requirements of relevant national standards and technical specifications through advance scheduling of active power in wind farms.

[0009] To achieve the above objectives, the present invention employs the following technical solution:

[0010] A method for optimizing active power control in a wind farm includes the following steps:

[0011] S1: Collect meteorological parameters and wind power data, identify abnormal data in the collected data, perform numerical replacement processing, perform numerical normalization preprocessing on the processed data, and use the generated normalized data for the construction of the BP neural network wind power prediction model.

[0012] S2: Use meteorological parameters as input to the BP neural network wind power prediction model and wind power data as output to train the BP neural network wind power prediction model; after training, input the meteorological parameters for the next 4 hours obtained through weather forecasts into the BP neural network wind power prediction model to obtain the predicted value of wind power for the next 4 hours.

[0013] S3: Merge the wind power data of the most recent 4 hours after anomaly identification and numerical replacement with the predicted wind power values ​​for the next 4 hours to generate merged wind power data; then iterate through the merged wind power data in a fixed time window to search for inflection points where the data form changes.

[0014] S4: Based on the inflection point of data transformation, the active power change during each rising or falling phase of wind power is calculated using the time window traversal method.

[0015] S5: Calculate the maximum value of the change in active power within the time window, and determine whether the maximum threshold is exceeded based on the preset threshold of the maximum change in active power within the time window; if it occurs, record the time of the first exceedance and the time of the last exceedance.

[0016] S6: Based on the recorded first and last over-limit times, optimize the control of active power.

[0017] Furthermore, step S1 specifically includes the following steps:

[0018] S11: Obtain historical meteorological parameter data for the past two years based on the specific location information of the wind farm, with a data interval of 15 minutes; read historical wind power parameter data for the past two years from the SCADA database, with a data interval of 6 seconds.

[0019] S12: Use the box plot method to identify and calculate abnormal data in the historical data of acquired meteorological parameters and wind power parameters. The specific calculation method is as follows:

[0020] Arrange the data in ascending order;

[0021] The first quartile Q1 of the data is calculated using the formula Q1 = B2 (25th percentile);

[0022] The third quartile Q3 of the data is calculated using the formula Q3 = B2 (75th quartile);

[0023] The interquartile range (IQR) of the data is calculated using the formula IOR = Q3 - Q1.

[0024] According to the formula minval = Q1 - 2.5 × IQR

[0025] maxval = Q3 + 2.5 × IQR

[0026] Calculate the historical data upper limit (maxval) and lower limit (minval) of wind power parameters;

[0027] S13: Use linear interpolation to perform empty data interpolation and fill in the data after abnormal data identification and calculation;

[0028] The linear interpolation formula used is as follows:

[0029] f(x) = ωf(x0) + (1-ω)f(x1)

[0030] Where x0 and x1 are two known points, and ω is the ratio of the distance from x0 to x to the distance from x0 to x1;

[0031] S14: The historical data of meteorological parameters and wind power parameters after the empty data interpolation and filling operation are divided into training set and evaluation set by using k-fold cross-validation. The training set data is used to construct the BP neural network wind power prediction model, and the evaluation set data is used to evaluate the quality and prediction effect of the BP neural network wind power prediction model.

[0032] Furthermore, step S2 includes:

[0033] Meteorological parameters every 4 hours are used as input to the BP neural network wind power prediction model. Since the data time interval is 15 minutes, the total number of input layers of the BP neural network is E = 4 * 4 * 4 = 64; since the prediction duration is 4 hours, the number of neurons in the output layer of the BP neural network is F = 16.

[0034] Through formula Calculate the number of hidden layer nodes H in the BP neural network, where a is a constant between [0, 10].

[0035] The parameters of the BP neural network wind power prediction model were optimized using k-fold cross-validation, with k set to 10. The standard error formula was used as the evaluation criterion for model parameter selection, and the model with the smallest standard error was selected as the optimal BP neural network wind power prediction model.

[0036]

[0037] Where n is the number of data points in the evaluation set, x i x represents the actual value output by the model. i The predicted value output by the model is used to obtain the predicted value of wind power output for the next 4 hours by inputting the meteorological parameters for the next 4 hours obtained from the weather forecast into the BP neural network wind power prediction model.

[0038] Furthermore, step S3 includes the following steps:

[0039] S31: A traversal scan of the wind power merged data trend inflection point identification method is performed using a 1-hour time window. Within the time window, the first and last points of the data segment are connected to form a line, and then fitted to a line. The deviation degree D is identified by the vertical distance deviation between the actual value and the fitted line. If the maximum deviation degree Dmax exceeds the set threshold T, the corresponding maximum value position will be taken as an inflection point, and the inflection point position will be saved. Then, the same operation is performed on the data segment of the next time window starting from this inflection point. If the maximum deviation degree Dmax does not exceed the set threshold T, the same operation is performed on the data segment of the next time window starting from the last moment of this window. After the traversal is completed, all wind power trend inflection point positions will be obtained.

[0040] S32: Merge all inflection points and the first and last positions, and use the wind power data segments between every two positions as the target sequence data to determine whether the data pattern is rising, falling, or stable.

[0041] Furthermore, in step S32, the wind power data segments between every two locations are used as target sequence data to determine whether the data pattern is rising, falling, or stable, including:

[0042] The target sequence data is divided into 5 equal parts, and then the data is combined into 5 data segments of unequal length.

[0043] The matrix solution formula using the least squares method is θ=(X T X) -1 X T Y calculates the slope k of 5 data segments of unequal length; where X is the input vector and Y is the output vector; the variable θ includes the slope k and the intercept l;

[0044] Use the slope threshold TL to determine the shape of 5 data segments of unequal length; if the slope is greater than the slope threshold TL, it is determined to be an upward trend, and its value is set to 1; if the slope is less than or greater than the opposite of the slope threshold TL, it is determined to be a downward trend, and its value is set to -1; otherwise, it is determined to be a stationary trend, and its value is set to 0.

[0045] The trend value of each data segment is calculated using a weighted method; if the trend value is greater than 0.5, the overall trend of the data segment is upward; if the trend value is less than -0.5, the overall trend of the data segment is downward; if the trend value is between -0.5 and 0.5, the overall trend of the data segment is stable.

[0046] Furthermore, step S4 includes:

[0047] The maximum value of active power change is calculated by scanning each data segment in the rising or falling pattern within a 10-minute time window. The maximum value of active power change is calculated by subtracting the first number from the last number and comparing it with the change threshold Tp. The time point when the maximum value of active power change exceeds the change threshold Tp is recorded.

[0048] Furthermore, in step S5, the maximum threshold for the change in active power within the preset time window includes:

[0049] The threshold for the maximum value of the change in active power over 10 minutes is set to 0.3 times the installed capacity;

[0050] The maximum threshold for the change in active power over 1 minute is set to 0.1 times the installed capacity.

[0051] Furthermore, step S6 includes:

[0052] If the active power exceeds the limit during the rising phase, the recommended control index is to reduce the step size of the active power limit to slow down the change in active power.

[0053] If the active power falls below the minimum active power output during the decline phase, reduce non-stop operation of the wind turbine units.

[0054] Furthermore, meteorological parameters include: wind speed, wind direction, temperature, and air pressure.

[0055] Accordingly, the present invention also discloses a wind farm active power optimization control system, comprising:

[0056] The data acquisition and processing module is used to collect meteorological parameters and wind power data, identify abnormal data in the collected data, perform numerical replacement processing, perform numerical normalization preprocessing on the processed data, and use the generated normalized data for the construction of the BP neural network wind power prediction model.

[0057] The ultra-short-term wind power prediction module is used to train the BP neural network wind power prediction model by taking meteorological parameters as input and wind power data as output. After training, the meteorological parameters for the next 4 hours obtained through weather forecasts are input into the BP neural network wind power prediction model to obtain the predicted value of wind power for the next 4 hours.

[0058] The data trend identification and segmentation module is used to merge the wind power data of the most recent 4 hours after abnormal data identification and numerical replacement processing with the predicted value of wind power for the next 4 hours to generate wind power merged data; then, it traverses the wind power merged data in a fixed time window to search for inflection points where the data form changes.

[0059] The active power change calculation module is used to calculate the active power change during each rising or falling phase of wind power based on the inflection point of data change and by traversing the time window method.

[0060] The indicator limit exceedance interval identification module is used to count the maximum change in active power within a time window, and determine whether the maximum threshold has been exceeded based on the preset maximum threshold for active power change within the time window; if it has occurred, the first exceedance time and the last exceedance time are recorded.

[0061] The active power optimization control module is used to optimize active power control based on the recorded first and last over-limit times.

[0062] Compared with existing technologies, the advantages of this invention are as follows: This invention provides a method and system for optimizing the active power control of wind farms. It achieves short-term wind power prediction by constructing a BP neural network prediction model. By automatically classifying and identifying three operating modes of power data, it calculates the time range and maximum change range of active power changes exceeding the limit in the one-minute and ten-minute active power changes during the rising or falling modes. This allows for the scheduling of wind farm power output, ensuring that the maximum value of active power changes in the one-minute and ten-minute periods meets the relevant national standards and specifications when receiving power limiting commands or when the wind suddenly rises.

[0063] This invention uses a time-series slope-based morphological judgment algorithm to automatically identify the operating morphology of historical and predicted power data. By investigating and calculating the active power change amplitude within one minute and ten minutes during the rise or fall of active power, it can accurately capture the time range and maximum amplitude of active power change exceeding the limit.

[0064] Therefore, it is evident that the present invention has outstanding substantive features and significant progress compared with the prior art, and the beneficial effects of its implementation are also obvious. Attached Figure Description

[0065] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0066] Figure 1 This is a flowchart illustrating a specific embodiment of the present invention.

[0067] Figure 2 This is a topology diagram of the BP neural network according to a specific embodiment of the present invention.

[0068] Figure 3 This is a diagram illustrating the effect of identifying the maximum value of active power change exceeding the limit in a specific embodiment of the present invention.

[0069] Figure 4 This is a system structure diagram of a specific embodiment of the present invention.

[0070] In the diagram, 1. Data acquisition and processing module; 2. Wind power ultra-short-term forecasting module; 3. Data trend identification and classification module; 4. Active power change calculation module; 5. Index limit violation identification module; 6. Active power optimization control module. Detailed Implementation

[0071] The specific embodiments of the present invention will now be described with reference to the accompanying drawings.

[0072] like Figure 1 The active power optimization control method for a wind farm shown includes the following steps:

[0073] S1: Collect meteorological parameters and wind power data, identify abnormal data in the collected data, perform numerical replacement processing, perform numerical normalization preprocessing on the processed data, and use the generated normalized data to construct the BP neural network wind power prediction model.

[0074] The purpose of this step is to collect and process data. Since wind power mainly depends on the geographical environment and meteorological conditions of the wind farm, with wind speed being the most important, and considering the convenience of data collection, relevant meteorological parameters such as wind speed, wind direction, temperature, air pressure, and wind power are collected as historical data sources required to build the BP neural network wind power prediction model. The collected data is first used to identify outliers and perform numerical replacement processing. The data after outlier processing is then used to train the BP neural network wind power prediction model in step S2.

[0075] It should be noted that the meteorological parameters include: wind speed, wind direction, temperature, and air pressure.

[0076] As an example, this step specifically includes the following steps:

[0077] S11: Obtain historical meteorological parameter data for the past two years based on the specific location information of the wind farm, with a data interval of 15 minutes; read historical wind power parameter data for the past two years from the SCADA database, with a data interval of 6 seconds.

[0078] For example, based on the specific location information of the wind farm, historical data for four meteorological parameters—wind speed, wind direction, temperature, and air pressure—from January 2020 to December 2021 were obtained, with a data interval of 15 minutes. The meteorological data selected were wind speed, wind direction, temperature, and sea-level air pressure at locations close to the wind turbine hub height as basic variables. Simultaneously, historical data for the total wind power parameters of the wind farm over the past 24 months were retrieved from the SCADA database, with a data interval of 6 seconds, totaling 10.51 million data entries.

[0079] S12: Use the box plot method to identify and calculate abnormal data in the historical data of acquired meteorological parameters and wind power parameters. Taking the historical data of wind power parameters as an example, the specific calculation method is as follows:

[0080] Arrange the data in ascending order;

[0081] The first quartile Q1 of the data is calculated using the formula Q1 = B2 (25th percentile);

[0082] The third quartile Q3 of the data is calculated using the formula Q3 = B2 (75th quartile);

[0083] The interquartile range (IQR) of the data is calculated using the formula IOR = Q3 - Q1.

[0084] According to the formula minval = Q1 - 2.5 × IQR

[0085] maxval = Q3 + 2.5 × IQR

[0086] Calculate the upper limit (maxval) and lower limit (minval) of historical data for wind power parameters.

[0087] The data at the 25th percentile is called the first quartile (Q1), and the data at the 75th percentile is called the third quartile (Q3). The difference between the third quartile and the first quartile is the interquartile range (IQR).

[0088] In addition, the other four meteorological parameter data were all subjected to abnormal data identification and calculation according to the same process as step S12.

[0089] S13: Use linear interpolation to perform empty data interpolation and fill in the data after abnormal data identification and calculation;

[0090] The linear interpolation formula used is as follows:

[0091] f(x) = ωf(x0) + (1-ω)f(x1)

[0092] Where x0 and x1 are two known points, and ω is the ratio of the distance from x0 to x to the distance from x0 to x1.

[0093] S14: The historical data of meteorological parameters and wind power parameters after the empty data interpolation and imputation operation are divided into training and evaluation sets using k-fold cross-validation. The training set data is used to construct the BP neural network wind power prediction model, and the evaluation set data is used to evaluate the quality and prediction effect of the BP neural network wind power prediction model. Considering the significant impact of seasonal climate on four meteorological parameters (wind speed, wind direction, temperature, and air pressure) in northern regions, this invention employs k-fold cross-validation to optimize the model parameters.

[0094] S2: Use meteorological parameters as input to the BP neural network wind power prediction model and wind power data as output to train the BP neural network wind power prediction model; after training, input the meteorological parameters for the next 4 hours obtained through weather forecasts into the BP neural network wind power prediction model to obtain the predicted value of wind power for the next 4 hours.

[0095] Specifically, the purpose of this step is to achieve ultra-short-term wind power forecasting. First, meteorological data such as wind speed, wind direction, temperature, and air pressure are used as inputs to the prediction model, while wind power data (15-minute average) is used as the output. Then, a backpropagation (BP) neural network (BP) wind power forecasting model is trained using sufficient historical training data, following the BP neural network training procedures. Finally, the predicted wind speed, wind direction, temperature, and air pressure for the next four hours are input into the trained BP neural network wind power forecasting model to obtain the predicted wind power values ​​for the next four hours.

[0096] It should be noted that the BP neural network is a feedforward neural network based on the backpropagation learning algorithm, and its topology is as follows: Figure 2As shown, this is a multilayer perceptron structure with one hidden layer, one input layer, and one output layer. In the figure, X is the E-dimensional input node vector, Z is the H-dimensional hidden layer node vector, Y is the F-dimensional output node vector, Wij is the connection weight between the i-th neuron in the input layer and the j-th neuron in the hidden layer, Wjk is the connection weight between the j-th neuron in the hidden layer and the k-th neuron in the output layer, W1 and W2 are the connection weights from the input layer to the hidden layer and from the hidden layer to the output layer, respectively, and b1 and b2 represent the thresholds of the hidden layer and the output layer, respectively. During the network learning and training process, the main idea of ​​the BP algorithm is that the working signal propagates forward from the input layer to the output layer, and the error signal propagates backward from the output layer to the input layer. The weights and thresholds of the network are adjusted and corrected to reduce the error between the actual output and the expected output of the network. The weight adjustment amount ΔW = n·δ·v (n is the learning rate, δ is the local gradient, and v is the output signal of the previous layer).

[0097] As an example, this step specifically includes:

[0098] Meteorological parameters every 4 hours are used as input to the BP neural network wind power prediction model. Since the data time interval is 15 minutes, the total number of neurons in the BP neural network input layer is E = 4 * 4 * 4 = 64; since the prediction duration is 4 hours, the number of neurons in the BP neural network output layer is F = 16; using the formula... Calculate the number of hidden layer nodes H in the BP neural network, where a is a constant between [0, 10]. The BP neural network uses the error returned by the model parameters each time as feedback for the next network training.

[0099] The parameters of the BP neural network wind power prediction model were optimized using k-fold cross-validation, with k set to 10. The standard error formula was used as the evaluation criterion for model parameter selection, and the model with the smallest standard error was selected as the optimal BP neural network wind power prediction model.

[0100]

[0101] Where n is the number of data points in the evaluation set, x i x represents the actual value output by the model. i These are the predicted values ​​output by the model.

[0102] Using the optimal BP neural network wind power prediction model obtained through training, the meteorological parameters for the next 4 hours obtained through weather forecasts are input into the optimal BP neural network wind power prediction model to obtain the predicted value of wind power for the next 4 hours.

[0103] S3: Merge the wind power data of the most recent 4 hours after anomaly identification and numerical replacement with the predicted wind power values ​​for the next 4 hours to generate merged wind power data; then, traverse the merged wind power data in a fixed time window to search for inflection points where the data form changes.

[0104] This invention uses a pattern judgment method based on the slope of segmented time series to identify data patterns, which mainly include three states: rising, falling, and stable.

[0105] As an example, this step specifically includes the following steps:

[0106] S31: A traversal scan of the combined wind power data is performed using a 1-hour time window to identify trend inflection points. Within the time window, the first and last points of each data segment are connected by a line, which is then fitted to a line. The deviation D is identified by the vertical distance between the actual value and the fitted line. If the maximum deviation Dmax exceeds a set threshold T, the corresponding maximum value location is taken as an inflection point, and the inflection point location is saved. The same operation is then performed on the next time window data segment starting from this inflection point. If the maximum deviation Dmax does not exceed the set threshold T, the same operation is performed on the next time window data segment starting from the last moment of the current window. After the traversal is complete, all wind power trend inflection point locations are obtained. Here, the threshold T is set to 5.

[0107] S32: Determine the data trend pattern. Specifically, merge all inflection points and the first and last positions, and use the wind power data segments between every two positions as the target sequence data to determine whether the data pattern is rising, falling, or stable. The specific process for determining the data pattern is as follows:

[0108] First, the target sequence data is divided into 5 equal parts, and then the data is combined into 5 segments of unequal length. The combination method is to start from the last segment, with the last segment as the first segment, the last two segments combined as the second segment, the last three segments combined as the third segment, the last four segments combined as the fifth segment, and finally the entire target sequence data as the fifth segment.

[0109] Then, using the matrix solution formula of the least squares method, θ=(X T X) -1 X T Y calculates the slope k of 5 data segments of unequal length. Here, X is the input vector, Y is the output vector, and the variable θ includes the slope k and the intercept l.

[0110] At this point, the slope threshold TL is used to determine the shape of the 5 data segments of unequal length; those greater than the slope threshold TL are judged as an upward trend, and their value is set to 1; those less than or greater than the opposite of the slope threshold TL are judged as a downward trend, and their value is set to -1; the rest are judged as a stable trend, and their value is set to 0.

[0111] Finally, a weighted method is used to determine the overall shape of the data. The shape combination is weighted and judged. The weight value is defined as the ratio of the data length to the overall length. The trend value is the product of the weight of each segment and the shape value, and then the products are accumulated. After obtaining the trend value, one valid data point is retained. If the trend value is greater than 0.5, the overall shape of the data segment is rising; if the trend value is less than -0.5, the overall shape of the data segment is falling; if the trend value is between -0.5 and 0.5, the overall shape of the data segment is stable.

[0112] After the above data trend identification and segmentation operations, the 24-hour wind power data can be divided into 30 data segments, including 7 stable data segments, 12 rising data segments, and 11 falling data segments.

[0113] S4: Based on the inflection point of the data transformation, the active power change during each rising or falling phase of wind power is calculated using the time window traversal method.

[0114] This step calculates the change in active power. Specifically, it uses a time window traversal method to calculate the change in active power during each rise or fall phase of wind power. The time window duration includes 1 minute and 10 minutes. During the traversal, it is necessary to count the maximum value of the change in active power within the time window.

[0115] As an example, the specific implementation of step S4 is explained here using the calculation of the maximum value index of active power change over 10 minutes as an example:

[0116] The maximum value of active power change is calculated by scanning the entire data segment of each rise or fall within a 10-minute time window. The maximum value of active power change is calculated by subtracting the first number from the last number and comparing it with the change threshold Tp. The time point when the maximum value of active power change exceeds the change threshold Tp is recorded.

[0117] The maximum value of the change in active power over one minute is also calculated in the same way as described above.

[0118] S5: Calculate the maximum value of the change in active power within the time window, and determine whether the maximum threshold is exceeded based on the preset threshold of the maximum change in active power within the time window; if it occurs, record the first time the limit is exceeded and the last time the limit is exceeded.

[0119] As an example, this method sets the maximum threshold for active power change over 10 minutes to 0.3 times the installed capacity, and sets the maximum threshold for active power change over 1 minute to 0.1 times the installed capacity, according to national technical standards. During the iteration process, the maximum value of active power change within the time window needs to be identified and judged in real time to determine whether the above-mentioned maximum value threshold for active power change has been exceeded. If it has, the time of the first and last time the index has exceeded the limit is recorded.

[0120] For example, since the installed capacity of the wind farm is 99MW, the maximum threshold for active power change over 10 minutes is set at 33MW, and the maximum threshold for active power change over 1 minute is set at 9.9MW. Five instances of exceeding these limits were detected during March 2022. (Details are as follows...) Figure 3 As shown, the duration of the out-of-limit events ranged from 6 seconds to 5 minutes. All of these occurred during the active power increase phase.

[0121] S6: Based on the recorded first and last over-limit times, optimize the control of active power.

[0122] Specifically, if there is a possibility of exceeding the limit during the wind power increase phase, a control recommendation indicator is issued to immediately reduce the step size of the active power limit, so that the change in active power is slowed down; if there is a possibility of the wind power falling below the minimum active power output during the wind power decrease phase, intervention is carried out to reduce the non-stop of wind turbine units, so that the active power control of the wind farm meets the requirements of national technical specifications and grid dispatch.

[0123] Therefore, the wind farm active power optimization control method disclosed in this invention optimizes the active power control of the wind farm based on information such as the real-time power and predicted power of each wind turbine generator. It determines the expected output of each wind turbine generator based on the expected active power value set by the dispatch center and ultra-short-term wind power forecast, allowing each controllable unit to generate power according to the expected output value, thus achieving power control of the wind farm. Simultaneously, if the active power change in the wind farm during the one-minute and ten-minute periods shows a potential for exceeding the limit during the rising or falling phases, the control system immediately reduces the step size of the active power limit, slowing down the change in active power and ensuring that the active power control of the wind farm meets the requirements of national technical specifications and grid dispatch.

[0124] Correspondingly, such as Figure 4 As shown, the present invention also discloses a wind farm active power optimization control system, including: a data acquisition and processing module 1, a wind power ultra-short-term prediction module 2, a data trend identification and classification module 3, an active power change calculation module 4, an index limit exceedance identification module 5, and an active power optimization control module 6.

[0125] The data acquisition and processing module 1 is used to collect meteorological parameters and wind power data, identify abnormal data in the collected data, perform numerical replacement processing, perform numerical normalization preprocessing on the processed data, and use the generated normalized data for the construction of the BP neural network wind power prediction model.

[0126] The wind power ultra-short-term prediction module 2 is used to train the BP neural network wind power prediction model by taking meteorological parameters as input and wind power data as output. After training, the meteorological parameters for the next 4 hours obtained through weather forecasts are input into the BP neural network wind power prediction model to obtain the predicted value of wind power for the next 4 hours.

[0127] The data trend identification and segmentation module 3 is used to merge the wind power data of the most recent 4 hours after abnormal data identification and numerical replacement processing with the predicted value of wind power for the next 4 hours to generate wind power merged data; then, it traverses the wind power merged data in a fixed time window to search for inflection points where the data form changes.

[0128] The active power change calculation module 4 is used to calculate the active power change during each rising or falling phase of wind power by using a time window traversal method based on the inflection point of data change.

[0129] The indicator limit exceedance interval identification module 5 is used to count the maximum value of the change in active power within a time window, and to determine whether the maximum threshold of the change in active power within a preset time window has been exceeded; if it has occurred, the first exceedance time and the last exceedance time are recorded.

[0130] The active power optimization control module 6 is used to optimize the control of active power based on the recorded first and last over-limit times.

[0131] Those skilled in the art will clearly understand that the techniques in the embodiments of the present invention can be implemented using software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solutions in the embodiments of the present invention, or the parts that contribute to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium such as a USB flash drive, mobile hard drive, read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk, or any other medium capable of storing program code. It includes several instructions to cause a computer terminal (which may be a personal computer, server, or a second terminal, network terminal, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. Similar or identical parts between the various embodiments in this specification can be referred to mutually. In particular, for the terminal embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and relevant parts can be referred to the description in the method embodiments.

[0132] In the embodiments provided by this invention, it should be understood that the disclosed systems, methods, and approaches can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between systems or units may be electrical, mechanical, or other forms.

[0133] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0134] In addition, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each module can exist physically separately, or two or more modules can be integrated into one unit.

[0135] Similarly, in the various embodiments of the present invention, each processing unit can be integrated into a functional module, or each processing unit can exist physically, or two or more processing units can be integrated into a functional module.

[0136] The present invention will be further described in conjunction with the accompanying drawings and specific embodiments. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. Furthermore, it should be understood that after reading the teachings of this invention, those skilled in the art can make various alterations or modifications to the invention, and these equivalent forms also fall within the scope defined in this application.

Claims

1. A method for optimizing active power control in a wind farm, characterized in that, The steps include the following: S1: Collect meteorological parameters and wind power data, identify abnormal data in the collected data, perform numerical replacement processing, perform numerical normalization preprocessing on the processed data, and use the generated normalized data for the construction of the BP neural network wind power prediction model. S2: Use meteorological parameters as input to the BP neural network wind power prediction model and wind power data as output to train the BP neural network wind power prediction model; after training, input the meteorological parameters for the next 4 hours obtained through weather forecasts into the BP neural network wind power prediction model to obtain the predicted value of wind power for the next 4 hours. S3: Merge the wind power data of the most recent 4 hours after anomaly identification and numerical replacement with the predicted wind power values ​​for the next 4 hours to generate merged wind power data; then iterate through the merged wind power data in a fixed time window to search for inflection points where the data form changes. S4: Based on the inflection point of data transformation, the active power change during each rising or falling phase of wind power is calculated using the time window traversal method. S5: Calculate the maximum value of the change in active power within the time window, and determine whether the maximum threshold is exceeded based on the preset threshold of the maximum change in active power within the time window; if it is exceeded, record the time of the first exceedance and the time of the last exceedance. S6: Based on the recorded first and last over-limit times, optimize the control of active power; Step S3 includes the following steps: S31: A traversal scan of the combined wind power data trend inflection point identification method is performed using a 1-hour time window. Within the time window, the first and last points of the data segment are connected to form a line, which is then fitted to a line. The deviation degree D is identified by the vertical distance deviation between the actual value and the fitted line. If the maximum deviation degree Dmax exceeds the set threshold T, the corresponding maximum value position will be taken as an inflection point, and the inflection point position will be saved. Then, the same operation is performed on the data segment of the next time window starting from this inflection point. If the maximum deviation degree Dmax does not exceed the set threshold T, the same operation is performed on the data segment of the next time window starting from the last moment of this window. After the traversal is completed, all wind power trend inflection point positions will be obtained. S32: Merge all inflection points and the first and last positions, and use the wind power data segments between every two positions as the target sequence data to determine whether the data pattern is rising, falling or stable. In step S32, the wind power data segment between every two locations is used as the target sequence data to determine whether the data pattern is rising, falling, or stable, including: The target sequence data is divided into 5 equal parts, and then the data is combined into 5 data segments of unequal length. Matrix solution formula using the least squares method Calculate the slope k of 5 data segments of unequal length; Where X is the input vector and Y is the output vector; variables It includes the slope k and the intercept l; Use the slope threshold TL to determine the shape of 5 data segments of unequal length; if the slope is greater than the slope threshold TL, it is determined to be an upward trend, and its value is set to 1; if the slope is less than or greater than the opposite of the slope threshold TL, it is determined to be a downward trend, and its value is set to -1; otherwise, it is determined to be a stationary trend, and its value is set to 0. The trend value of each data segment is calculated using a weighted method; if the trend value is greater than 0.5, the overall trend of the data segment is upward; if the trend value is less than -0.5, the overall trend of the data segment is downward; if the trend value is between -0.5 and 0.5, the overall trend of the data segment is stable.

2. The wind farm active power optimization control method according to claim 1, characterized in that, Step S1 specifically includes the following steps: S11: Obtain historical meteorological parameter data for the past two years based on the specific location information of the wind farm, with a data interval of 15 minutes; read historical wind power parameter data for the past two years from the SCADA database, with a data interval of 6 seconds. S12: Use the box plot method to identify and calculate abnormal data in the historical data of acquired meteorological parameters and wind power parameters. The specific calculation method is as follows: Arrange the data in ascending order; Based on the sorted data, the data in the 25th percentile position is taken as the first quartile Q1 of the data; Based on the sorted data, the data in 75% of the positions are taken as the third quartile Q3 of the data; The interquartile range (IQR) of the data is calculated using the formula IOR = Q3 - Q1. According to the formula minval = Q1 - 2.5 × IQR maxval = Q3 + 2.5 × IQR Calculate the historical data upper limit (maxval) and lower limit (minval) of wind power parameters; S13: Use linear interpolation to perform empty data interpolation and fill in the data after abnormal data identification and calculation; The linear interpolation formula used is as follows: in, and Given two points that exist, for arrive distance and arrive The ratio of distances; S14: The historical data of meteorological parameters and wind power parameters after the empty data interpolation and filling operation are divided into training set and evaluation set by using k-fold cross-validation. The training set data is used to construct the BP neural network wind power prediction model, and the evaluation set data is used to evaluate the quality and prediction effect of the BP neural network wind power prediction model.

3. The wind farm active power optimization control method according to claim 2, characterized in that, Step S2 includes: Meteorological parameters every 4 hours are used as input to the BP neural network wind power prediction model. Since the data time interval is 15 minutes, the total number of input layers of the BP neural network is E=4*4*4=64; since the prediction duration is 4 hours, the number of neurons in the output layer of the BP neural network is F=16. Through formula Calculate the number of hidden layer nodes H in the BP neural network, where a is a constant between [0, 10]. The parameters of the BP neural network wind power prediction model were optimized using k-fold cross-validation, with k set to 10. The standard error formula was used as the evaluation criterion for model parameter selection, and the model with the smallest standard error was selected as the optimal BP neural network wind power prediction model. in, To evaluate the number of data sets, The actual value output by the model. The predicted value output by the model; Using the optimal BP neural network wind power prediction model obtained through training, the meteorological parameters for the next 4 hours obtained through weather forecasts are input into the BP neural network wind power prediction model to obtain the predicted value of wind power for the next 4 hours.

4. The wind farm active power optimization control method according to claim 1, characterized in that, Step S4 includes: The maximum value of active power change is calculated by scanning each data segment in the rising or falling pattern within a 10-minute time window. The maximum value of active power change is calculated by subtracting the first number from the last number and comparing it with the change threshold Tp. The time point when the maximum value of active power change exceeds the change threshold Tp is recorded.

5. The wind farm active power optimization control method according to claim 4, characterized in that, In step S5, the maximum threshold for the change in active power within the preset time window includes: The threshold for the maximum value of the change in active power over 10 minutes is set to 0.3 times the installed capacity; The maximum threshold for the change in active power over 1 minute is set to 0.1 times the installed capacity.

6. The wind farm active power optimization control method according to claim 5, characterized in that, Step S6 includes: If the active power exceeds the limit during the rising phase, the recommended control index is to reduce the step size of the active power limit to slow down the change in active power. If the active power falls below the minimum active power output during the decline phase, reduce non-stop operation of the wind turbine units.

7. The wind farm active power optimization control method according to claim 1, characterized in that, The meteorological parameters include: wind speed, wind direction, temperature, and air pressure.

8. A wind farm active power optimization control system, characterized in that, include: The data acquisition and processing module is used to collect meteorological parameters and wind power data, identify abnormal data in the collected data, perform numerical replacement processing, perform numerical normalization preprocessing on the processed data, and use the generated normalized data for the construction of the BP neural network wind power prediction model. The ultra-short-term wind power prediction module is used to train the BP neural network wind power prediction model by taking meteorological parameters as input and wind power data as output. After training, the meteorological parameters for the next 4 hours obtained through weather forecasts are input into the BP neural network wind power prediction model to obtain the predicted value of wind power for the next 4 hours. The data trend identification and segmentation module is used to merge the wind power data of the most recent 4 hours after abnormal data identification and numerical replacement processing with the predicted value of wind power for the next 4 hours to generate wind power merged data; then, it traverses the wind power merged data in a fixed time window to search for inflection points where the data form changes. The active power change calculation module is used to calculate the active power change during each rising or falling phase of wind power based on the inflection point of data change and by traversing the time window method. The indicator limit exceedance interval identification module is used to count the maximum change in active power within a time window, and determine whether the maximum threshold is exceeded based on the preset maximum threshold for the change in active power within the time window. If it occurs, record the time of the first time the limit is exceeded and the time of the last time the limit is exceeded; The active power optimization control module is used to optimize active power control based on the recorded first and last over-limit times. The data trend identification and segmentation module is specifically used to perform the following steps: S31: A traversal scan of the combined wind power data trend inflection point identification method is performed using a 1-hour time window. Within the time window, the first and last points of the data segment are connected to form a line, which is then fitted to a line. The deviation degree D is identified by the vertical distance deviation between the actual value and the fitted line. If the maximum deviation degree Dmax exceeds the set threshold T, the corresponding maximum value position will be taken as an inflection point, and the inflection point position will be saved. Then, the same operation is performed on the data segment of the next time window starting from this inflection point. If the maximum deviation degree Dmax does not exceed the set threshold T, the same operation is performed on the data segment of the next time window starting from the last moment of this window. After the traversal is completed, all wind power trend inflection point positions will be obtained. S32: Merge all inflection points and the first and last positions, and use the wind power data segments between every two positions as the target sequence data to determine whether the data pattern is rising, falling or stable. In step S32, the wind power data segment between every two locations is used as the target sequence data to determine whether the data pattern is rising, falling, or stable, including: The target sequence data is divided into 5 equal parts, and then the data is combined into 5 data segments of unequal length. Matrix solution formula using the least squares method Calculate the slope k of 5 data segments of unequal length; Where X is the input vector and Y is the output vector; variables It includes the slope k and the intercept l; Use the slope threshold TL to determine the shape of 5 data segments of unequal length; if the slope is greater than the slope threshold TL, it is determined to be an upward trend, and its value is set to 1; if the slope is less than or greater than the opposite of the slope threshold TL, it is determined to be a downward trend, and its value is set to -1; otherwise, it is determined to be a stationary trend, and its value is set to 0. The trend value of each data segment is calculated using a weighted method; if the trend value is greater than 0.5, the overall trend of the data segment is upward; if the trend value is less than -0.5, the overall trend of the data segment is downward; if the trend value is between -0.5 and 0.5, the overall trend of the data segment is stable.

Images