Offshore wind power ramp prediction method and system

Abstract

The present invention proposes a method and system for predicting the ramping of offshore wind power, which relates to the technical field of power system prediction and control. The method includes: obtaining a multi-source data set of an offshore wind farm; preprocessing the multi-source data set to obtain meteorological driving factors, timestamps of wind power and geographical feature information of the offshore wind farm; organizing the meteorological driving factors, wind power, timestamps and geographical feature information into ramping data in a unified format, expanding the ramping data, and obtaining a ramping data set; constructing a multi-dimensional collaborative graph perception network for predicting wind power ramping events; training the multi-dimensional collaborative graph perception network using the ramping data set to obtain a trained multi-dimensional collaborative graph perception network; collecting the multi-source data set to be predicted of the offshore wind farm, inputting the multi-source data set to be predicted into the trained multi-dimensional collaborative graph perception network, and outputting the prediction result of the wind power ramping event. The present invention effectively improves the prediction accuracy of wind power ramping events.

Description

Technical Field

[0001] The present invention relates to the technical field of power system prediction and control, and in particular to a method and system for predicting offshore wind power ramping. Background Art

[0002] Due to the clean, abundant and environmentally friendly characteristics of wind energy, its proportion in the energy development of many countries in the world is increasing. Offshore wind farms have the advantages of not consuming land resources, high wind speed and zero dust emission. Compared with onshore wind farms, offshore wind farms have 20%-40% higher energy efficiency and are more suitable for large-scale development. However, offshore wind power generation is easily affected by many factors, resulting in uncertainty in offshore wind power generation, which will bring some adverse effects to the dispatching and operation of the power system. Therefore, accurate offshore wind power prediction is of great significance to the safe and stable operation of the power system.

[0003] Existing offshore wind power ramp prediction methods are mostly based on statistical models or time series analysis models. These methods have certain advantages in capturing single-dimensional features. For example, time series analysis can reveal the trend of wind power changes over time. However, the change of wind power is a complex process, which involves the interaction of multiple dimensional features such as wind speed, wind direction, temperature, and humidity. Existing prediction methods often find it difficult to fully reflect the complex interactions between multi-dimensional features, which limits the improvement of prediction accuracy. In addition, wind power ramp events are a special case of wind power changes. It refers to the phenomenon that wind power changes significantly in a short period of time. Due to the low frequency of ramp events, the relevant sample data is relatively scarce. This makes it difficult to predict offshore wind power ramp events. The problem of insufficient data, which in turn affects the prediction accuracy and robustness of wind power ramp events. Summary of the invention

[0004] In order to solve the problem of low wind power ramp event prediction accuracy in the above-mentioned prior art, the present invention proposes an offshore wind power ramp event prediction method and system, which effectively improves the wind power ramp event prediction accuracy.

[0005] In order to achieve the above technical effects, the technical solution of the present invention is as follows:

[0006] A method for predicting offshore wind power ramping includes the following steps:

[0007] S1. Obtain multi-source datasets for offshore wind farms;

[0008] S2. Preprocessing the multi-source data set to obtain meteorological driving factors, wind power timestamps and geographic feature information of offshore wind farms;

[0009] S3. Arrange the meteorological driving factor, the wind power, the timestamp and the geographic feature information into a unified format of climbing data, expand the climbing data, and obtain a climbing data set;

[0010] S4. Construct a multi-dimensional collaborative graph perception network for predicting wind power ramp events;

[0011] S5. Using the climbing data set to train the multi-dimensional collaborative graph perception network to obtain a trained multi-dimensional collaborative graph perception network;

[0012] S6. Collect the multi-source data set to be predicted of the offshore wind farm, input the multi-source data set to be predicted into the trained multi-dimensional collaborative graph perception network, and output the prediction result of the wind power ramp event.

[0013] Preferably, the multi-source data set includes wind power time series data, weather forecast numerical data and geographical condition data, the weather forecast numerical data includes wind speed, wind direction, air pressure, humidity, temperature and precipitation, and the geographical condition data includes the longitude and latitude and altitude of each wind turbine.

[0014] Preferably, the preprocessing of the multi-source data set includes:

[0015] S21. The Pearson correlation coefficient method is used to calculate the correlation between each meteorological feature in the multi-source data set and the target value of the climbing event, and the calculation expression of the Pearson correlation coefficient is obtained as follows:

[0016]

[0017] Among them, r represents the Pearson correlation coefficient corresponding to the current meteorological parameters, Indicates c The normalized value of a meteorological characteristic at a time point, Indicates c The wind power change rate value corresponding to the time point is: Indicates meteorological characteristics The average value of represents the mean value of wind power change rate, and N represents the total number of data points;

[0018] S22. According to the Pearson correlation coefficient, the meteorological factor data with high correlation with wind power in the weather forecast numerical data are extracted, namely, the meteorological driving factor M, the timestamp T of the wind power time series data, and the geographical feature information G of the offshore wind farm closely related to wind power.

[0019] Preferably, the step of expanding the climbing data to obtain a climbing data set includes:

[0020] S31. Arrange the meteorological driving factor, the wind power, the timestamp and the geographic feature information into a unified format of climbing data in the format of timestamp T-meteorological driving factor M-geographic feature information G-wind power P, align the climbing data according to the timestamp T, so that each row contains the meteorological driving factor M, geographic feature information G and wind power P at a certain timestamp T at the same time, and obtain the expression of the data matrix as follows:

[0021]

[0022] Among them, T n Indicates the timestamp of the nth row, M n represents the meteorological driving factor of the nth row, G n Indicates the geographic feature information of the nth row, P n represents the wind power of the nth row;

[0023] S32. Input the data matrix into a preset adversarial network TimeGAN for time series generation, wherein the adversarial network TimeGAN includes a generator and a discriminator, wherein the generator receives the data matrix, generates new time series data and inputs it into the discriminator, and the discriminator receives the time series data and outputs a discriminant result of whether the time series data conforms to the distribution and regularity of real data;

[0024] S33. Vertically splicing the discrimination result and the slope climbing data in a specific ratio to obtain the slope climbing data set.

[0025] Preferably, the multi-dimensional collaborative graph perception network includes a network input layer, an intra-node attention learning layer, an inter-node attention learning layer, a feature integration layer, and a multi-layer perceptron. The network input layer is connected to the input end of the intra-node attention learning layer, the output end of the intra-node attention learning layer is connected to the input end of the inter-node attention learning layer, the output end of the inter-node attention learning layer is connected to the input end of the feature integration layer, the output end of the feature integration layer is connected to the input end of the multi-layer perceptron, and the output end of the multi-layer perceptron outputs the prediction result of the wind power ramp event.

[0026] Preferably, the construction of a multi-dimensional collaborative graph perception network for predicting wind power ramp events includes:

[0027] S41. Use the climbing data to form a multidimensional collaborative graph, the multidimensional collaborative graph includes four large nodes and dynamic edges, the four large nodes are timestamp T, elephant driving factor M, geographic feature information G and wind power P, each large node includes a number of small nodes, each small node is used to represent the specific sub-features of the large node, the dynamic edge is composed of intra-node edges and inter-node edges, the intra-node edges represent the relationship between small and medium nodes in the same large node, and the inter-node edges represent the relationship between different large nodes, and the multidimensional collaborative graph is input into the intra-node attention learning layer through the network input layer;

[0028] S42. The node-in-attention learning layer adjusts the node-in-edge weights of the multi-dimensional collaborative graph through node-in-attention learning to obtain After a round of information transmission, the big node The eigenvector of :

[0029]

[0030] in, represents the softmax activation function, Indicates that the path Next small node Pass to the big node attention, Indicates that the path Next big node Small and medium nodes A collection of Represents a large node Small and medium nodes Intra-node attention features;

[0031] S43. According to the feature vector The node-to-node attention learning layer adjusts the node-to-node edge weights of the multi-dimensional collaborative graph through node-to-node attention learning to obtain a node feature vector set :

[0032]

[0033] in, P Indicates the path L The total number of Indicates p Path L， Representative in the p Path L The set of all large nodes under represents the node feature vector set, represents the normalized inter-node attention;

[0034] S44. The feature integration layer collects node feature vectors Integrate into a global graph to represent and input into the multi-layer perceptron;

[0035] S45. The multi-layer perceptron receives the global graph, maps the connections between large and small nodes with wind power ramp events, and outputs the prediction results of wind power ramp events.

[0036] Preferably, the multilayer perceptron includes an input layer, a hidden layer and an output layer connected in sequence, and the calculation formula from the input layer to the hidden layer is:

[0037]

[0038]

[0039] in, Represents the input layer The output of a neuron, express Activation function, Represents the input layer The input of a neuron, Represents the hidden layer The neurons in the input layer The weights between neurons, Represents the input layer The output of a neuron, Represents the hidden layer The bias of each neuron;

[0040] The calculation formula from the hidden layer to the output layer is:

[0041]

[0042] in, Indicates that the path L The lower output layer outputs The probability that a sample belongs to a wind power ramp event, L Indicates the path, Indicates that the path L Lower input layer The input of a neuron, Indicates that the path L Lower input layer k The input of a neuron, K represents the total number of neurons in the output layer, Represents the activation function.

[0043] Preferably, the criteria for determining the prediction results are as follows:

[0044] Calculate specific time intervals The actual wind power consumption change rate as follows:

[0045]

[0046] in, Indicates time The actual wind power consumed, Indicates time The actual wind power consumed;

[0047] Determine the wind power change rate Is it greater than or equal to the first change threshold? and less than or equal to If so, determine the time interval The event occurring within the time interval is not a wind power ramp event; otherwise, the time interval is determined to be The event that occurred within is a wind power ramp event.

[0048] Preferably, the step of training the multi-dimensional collaborative graph perception network using the hill climbing data set includes:

[0049] S51. Dividing the climbing data set into a training set and a test set according to a specific ratio;

[0050] S52. Use the training set to train the multidimensional collaborative graph perception network, and use the test set to test the effectiveness of the multidimensional collaborative graph perception network, until the prediction accuracy of the multidimensional collaborative graph perception network for wind power ramping events reaches a preset accuracy threshold, thereby obtaining a trained multidimensional collaborative graph perception network.

[0051] The present invention also proposes an offshore wind power ramp prediction system, comprising:

[0052] An acquisition module for acquiring multi-source datasets for offshore wind farms;

[0053] A preprocessing module, used to preprocess the multi-source data set to obtain meteorological driving factors, timestamps of wind power and geographical feature information of offshore wind farms;

[0054] A data expansion module, used for arranging the meteorological driving factor, the wind power, the timestamp and the geographic feature information into a uniformly formatted slope data, expanding the slope data, and obtaining a slope data set;

[0055] A multi-dimensional collaborative graph perception network building module, which is used to build a multi-dimensional collaborative graph perception network for predicting wind power ramping events;

[0056] A training module, used to train the multi-dimensional collaborative graph perception network using the climbing data set to obtain a trained multi-dimensional collaborative graph perception network;

[0057] The prediction module is used to collect the multi-source data set to be predicted of the offshore wind farm, input the multi-source data set to be predicted into the trained multi-dimensional collaborative graph perception network, and output the prediction result of the wind power ramp event.

[0058] Compared with the prior art, the technical solution of the present invention has the following beneficial effects:

[0059] The present invention provides an offshore wind power ramp prediction method and system. First, a multi-source data set of an offshore wind farm is acquired, and the multi-source data set is preprocessed to obtain meteorological driving factors, timestamps of wind power, and geographic feature information of the offshore wind farm. Then, the meteorological driving factors, wind power, timestamps, and geographic feature information are sorted into ramp data in a unified format for sample expansion to generate a ramp data set with richer data. Then, a multi-dimensional collaborative graph perception network is constructed and trained to obtain a trained multi-dimensional collaborative graph perception network, and the trained multi-dimensional collaborative graph perception network is used to output the prediction results of wind power ramp events. The present invention combines the multi-source data set with the multi-dimensional collaborative graph perception network, effectively improves the prediction accuracy and real-time performance of wind power ramp events, provides safety protection for the operation of offshore wind farms, and has important engineering value. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 A flowchart showing a method for predicting offshore wind power ramping proposed in an embodiment of the present invention;

[0061] Figure 2 A structural block diagram showing a multi-dimensional collaborative graph perception network proposed in an embodiment of the present invention;

[0062] Figure 3 A diagram showing a node attention layer structure of a multi-dimensional collaborative graph perception network proposed in an embodiment of the present invention;

[0063] Figure 4 A diagram showing the recognition effect of a wind power ramping event proposed in an embodiment of the present invention;

[0064] Figure 5 A structural block diagram of an offshore wind power ramp prediction system proposed in an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0065] The drawings are for illustrative purposes only and should not be construed as limiting the present patent;

[0066] It is understandable to those skilled in the art that some well-known contents may be omitted in the drawings;

[0067] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0068] Example 1

[0069] like Figure 1 As shown, this embodiment proposes a method for predicting offshore wind power ramping, comprising the following steps:

[0070] S1. Obtain multi-source datasets for offshore wind farms;

[0071] The multi-source data set includes wind power time series data, weather forecast numerical data and geographical condition data. The weather forecast numerical data includes wind speed, wind direction, air pressure, humidity, temperature and precipitation. The geographical condition data includes the longitude, latitude and altitude of each wind turbine.

[0072] S2. Preprocessing the multi-source data set to obtain meteorological driving factors, wind power timestamps and geographic feature information of offshore wind farms;

[0073] The preprocessing of the multi-source data set includes:

[0074] S21. The Pearson correlation coefficient method is used to calculate the correlation between each meteorological feature of the weather forecast numerical data in the multi-source data set and the target value of the climbing event, and the calculation expression of the Pearson correlation coefficient is obtained as follows:

[0075]

[0076] Among them, r represents the Pearson correlation coefficient corresponding to the current meteorological parameters, Indicates c The normalized value of a meteorological characteristic at a time point, Indicates c The wind power change rate value corresponding to the time point is: Indicates meteorological characteristics The average value of represents the mean of the wind power change rate, and N represents the total number of data points; the next step is to determine the importance of meteorological characteristics based on the size of the Pearson correlation coefficient;

[0077] S22. According to the Pearson correlation coefficient, the meteorological factor data with high correlation with wind power in the weather forecast numerical data are extracted, namely, the meteorological driving factor M, the timestamp T of the wind power time series data, and the geographical feature information G of the offshore wind farm closely related to wind power.

[0078] In S22, first, the correlation between each meteorological feature and the wind power change rate is calculated using the Pearson correlation coefficient formula to obtain a series of Pearson correlation coefficients; then, according to the size of these Pearson correlation coefficients, the meteorological features that are highly correlated with wind power are screened out as meteorological driving factors M; at the same time, the timestamp T is directly extracted from the wind power time series data; finally, by analyzing the relationship between the geographical characteristics of the area where the offshore wind farm is located and the wind power, combined with expert knowledge and statistical analysis methods, the geographical feature information G that is closely related to wind power is determined.

[0079] S3. Arrange the meteorological driving factor, the wind power, the timestamp and the geographic feature information into a unified format of climbing data, expand the climbing data, and obtain a climbing data set;

[0080] The step of expanding the climbing data to obtain a climbing data set includes:

[0081] S31. Arrange the meteorological driving factor, the wind power, the timestamp and the geographic feature information into a unified format of climbing data in the format of timestamp T-meteorological driving factor M-geographic feature information G-wind power P, align the climbing data according to the timestamp T, so that each row contains the meteorological driving factor M, geographic feature information G and wind power P at a certain timestamp T at the same time, and obtain the expression of the data matrix as follows:

[0082]

[0083] Among them, T n Indicates the timestamp of the nth row, M n represents the meteorological driving factor of the nth row, G n Indicates the geographic feature information of the nth row, P n represents the wind power of the nth row;

[0084] S32. Input the data matrix into a preset adversarial network TimeGAN for time series generation, wherein the adversarial network TimeGAN includes a generator and a discriminator, wherein the generator receives the data matrix, generates new time series data and inputs it into the discriminator, and the discriminator receives the time series data and outputs a discriminant result of whether the time series data conforms to the distribution and regularity of real data;

[0085] In S32, the generator and discriminator in the adversarial network TimeGAN are continuously optimized through adversarial training. The generator generates new time series data based on the input data; the discriminator evaluates the generated data to determine whether it conforms to the distribution and rules of real data. The generator and discriminator are repeatedly trained until the generated new data is similar enough to the original data in statistical features and time series patterns and cannot be distinguished by the discriminator.

[0086] S33. Vertically splice the discrimination result and the climbing data in a specific ratio of 2:1 to increase the number of samples in the data set, thereby expanding the climbing data in the format of "timestamp T-meteorological driving factor M-geographic information G-wind power P" to obtain the climbing data set.

[0087] S4. Construct a multi-dimensional collaborative graph perception network for predicting wind power ramp events;

[0088] S5. Using the climbing data set to train the multi-dimensional collaborative graph perception network to obtain a trained multi-dimensional collaborative graph perception network;

[0089] S6. Collect the multi-source data set to be predicted of the offshore wind farm, input the multi-source data set to be predicted into the trained multi-dimensional collaborative graph perception network, and output the prediction result of the wind power ramp event.

[0090] In this embodiment, a multi-source data set of an offshore wind farm is first obtained, and the multi-source data set is preprocessed to obtain meteorological driving factors, timestamps of wind power, and geographical feature information of the offshore wind farm. Then, the meteorological driving factors, wind power, timestamps, and geographical feature information are organized into climbing data in a unified format for sample expansion to generate a climbing data set with richer data. Then, a multi-dimensional collaborative graph perception network is constructed and trained to obtain a trained multi-dimensional collaborative graph perception network, and the trained multi-dimensional collaborative graph perception network is used to output the prediction results of wind power climbing events. The present invention combines the multi-source data set with the multi-dimensional collaborative graph perception network, effectively improves the prediction accuracy and real-time performance of wind power climbing events, provides safety protection for the operation of offshore wind farms, and has important engineering value.

[0091] Example 2

[0092] See also Figure 2 and Figure 3The multi-dimensional collaborative graph perception network includes a network input layer, an intra-node attention learning layer, an inter-node attention learning layer, a feature integration layer, and a multi-layer perceptron MLP. The network input layer is connected to the input end of the intra-node attention learning layer, the output end of the intra-node attention learning layer is connected to the input end of the inter-node attention learning layer, the output end of the inter-node attention learning layer is connected to the input end of the feature integration layer, the output end of the feature integration layer is connected to the input end of the multi-layer perceptron, and the output end of the multi-layer perceptron outputs the prediction result of the wind power ramp event.

[0093] The multi-dimensional collaborative graph perception network for predicting wind power ramp events is constructed, including:

[0094] S41. The climbing data is used to form a multidimensional collaborative graph, which includes four large nodes and dynamic edges. The four large nodes are timestamp T, meteorological driving factor M, geographic feature information G and wind power P. Each large node includes a number of small nodes, each of which is used to represent the specific sub-features of the large node. The dynamic edge is composed of intra-node edges and inter-node edges. The intra-node edge represents the relationship between small and medium nodes in the same large node, which is used to establish the association within the node dimension; the inter-node edge represents the relationship between different large nodes, and the multidimensional collaborative graph is input into the node attention learning layer through the network input layer to establish cross-dimensional associations; the small nodes of meteorological driving factor M are temperature, humidity, and wind speed; the small nodes of timestamp T are timestamp and periodic seasonal characteristics; the small nodes of geographic information G are longitude and latitude; the small nodes of wind power P are historical power value and power increment;

[0095] S42. The node-in-attention learning layer adjusts the node-in-edge weights of the multi-dimensional collaborative graph through node-in-attention learning to obtain After a round of information transmission, the big node The eigenvector of :

[0096]

[0097] in, represents the softmax activation function, Indicates that the path Next small node Pass to the big node attention, Indicates that the path Next big node Small and medium nodes A collection of Represents a large node Small and medium nodes Intra-node attention features;

[0098] The calculation expression is as follows:

[0099]

[0100]

[0101] Among them, k represents the set The kth element of and Represents the linear transformation matrix for training the multi-dimensional collaborative graph-aware network, represents the softmax activation function, Represents a large node The intra-node attention feature of Represents a large node The initial eigenvector of The main function of is to improve the fitting ability. The function is to adjust the shape;

[0102] The attention of the small nodes in each large node is used as the weight for weighted summation calculation to obtain the After a round of information transmission, the big node The eigenvector of .

[0103] S43. According to the feature vector The inter-node attention learning layer adjusts the inter-node edge weights of the multi-dimensional collaborative graph through inter-node attention learning to obtain a node feature vector set :

[0104]

[0105] in, P Indicates the path L The total number of Indicates p Path L， Representative in the p Path L The set of all large nodes under represents the node feature vector set, represents the normalized inter-node attention;

[0106] go through Inter-node attention after activation function normalization The calculation expression is as follows:

[0107] ;

[0108] represents the averaging function, The calculation expression is as follows:

[0109]

[0110] In the formula, Represents all nodes, , , It is the network parameter that needs to be trained in the multi-dimensional collaborative graph perception network. This formula represents the average value after linearly transforming the large nodes and adjusting them to one-dimensional scalars after calculating the intra-node attention of all large nodes.

[0111] Finally, perform inter-node attention The weighted sum of , thereby obtaining the node feature vector set .

[0112] S44. The feature integration layer collects node feature vectors Integrate into a global graph for representation and input into the multi-layer perceptron; in the feature integration layer, the network integrates all updated node features into a global graph for representation using a global pooling method;

[0113] S45. The multi-layer perceptron receives the global graph, maps the connections between large and small nodes with wind power ramp events, and outputs the prediction results of wind power ramp events.

[0114] The global graph is used as input, and a multilayer perceptron MLP is used as a connection layer. The multilayer perceptron MLP includes an input layer, a hidden layer, and an output layer connected in sequence. The cross entropy loss function is used as the loss function, and the gradient descent method is used to optimize the hyperparameters in the network. The multilayer perceptron MLP as the connection layer maps the connection between large and small nodes with the wind power ramp event, and finally outputs the predicted probability of the wind power ramp event. If the output predicted probability is greater than 0.75, it is determined to be a wind power ramp event, otherwise, it is determined to be a non-wind power ramp event.

[0115] The calculation formula from the input layer to the hidden layer is:

[0116]

[0117]

[0118] in, Represents the input layer The output of a neuron, express Activation function, Represents the input layer The input of a neuron, Represents the hidden layer The neurons in the input layer The weights between neurons, Represents the input layer The output of a neuron, Represents the hidden layer The bias of each neuron;

[0119] The calculation formula from the hidden layer to the output layer is:

[0120]

[0121] in, Indicates that the path L The lower output layer outputs The probability that a sample belongs to a wind power ramp event, L Indicates the path, Indicates that the path L Lower input layer The input of a neuron, Indicates that the path L Lower input layer k The input of a neuron, K represents the total number of neurons in the output layer, Represents the activation function.

[0122] L The calculation expression is as follows:

[0123] ,

[0124] In the formula, m is the number of samples, For the The true labels of samples, For the multilayer perceptron to predict The probability that a sample belongs to the positive class.

[0125] The criteria for judging the prediction results are as follows:

[0126] Calculate specific time intervals The actual wind power consumption change rate as follows:

[0127]

[0128] in, Indicates time The actual wind power consumed, Indicates time The actual wind power consumed;

[0129] Determine the wind power change rate Is it greater than or equal to the first change threshold? and less than or equal to If so, determine the time interval The event occurring within the time interval is not a wind power ramp event; otherwise, the time interval is determined to be The event that occurred within is a wind power ramp event.

[0130] in, The calculation expression is as follows:

[0131]

[0132] The calculation expression is as follows:

[0133]

[0134] In the formula, , Respectively The ramp-up and ramp-down rates of the generators, , They are respectively a collection of generators that do not have regulation capability or have reached their regulation limit;

[0135] The step of training the multi-dimensional collaborative graph perception network using the hill climbing data set includes:

[0136] S51. Divide the ramp data set into a training set and a test set according to a specific ratio; the training set is used for the subsequent multi-dimensional collaborative graph perception network training process, and the test set is used for the performance evaluation of the trained multi-dimensional collaborative graph perception network; label the samples according to the definition of wind power ramp events as follows:

[0137] The collected data are manually labeled according to whether it is a wind power ramp event. If it is judged to be a wind power ramp event, it is marked as 1, otherwise it is marked as 0.

[0138] S52. Use the training set to train the multidimensional collaborative graph perception network, and use the test set to test the effectiveness of the multidimensional collaborative graph perception network, until the prediction accuracy of the multidimensional collaborative graph perception network for wind power ramping events reaches a preset accuracy threshold, thereby obtaining a trained multidimensional collaborative graph perception network.

[0139] In S52, the multidimensional collaborative graph perception network is first trained using the training set, and the multidimensional collaborative graph perception network is iteratively trained using the gradient descent method. The prediction results are calculated by forward propagation, and the network weights are updated by back propagation in combination with the cross entropy loss function so that it can learn the characteristics and associations of multi-source data; then the multidimensional collaborative graph perception network after training is evaluated using the test set to determine the prediction accuracy of the multidimensional collaborative graph perception network for wind power ramping events. The prediction accuracy is calculated by comparing the consistency between the prediction results of the multidimensional collaborative graph perception network and the true labels in the test set. The specific calculation method is: accuracy = (number of correctly predicted samples / total number of samples in the test set) × 100%; then when the prediction accuracy of the multidimensional collaborative graph perception network does not reach the preset accuracy threshold of 95%, the multidimensional collaborative graph perception network is tuned according to the evaluation results, and the multidimensional collaborative graph perception network is continuously trained until the prediction accuracy of the multidimensional collaborative graph perception network reaches the preset accuracy threshold of 95%; finally, the trained multidimensional collaborative graph perception network is used to discriminate wind power ramping events.

[0140] In this embodiment, by integrating multi-source data into a multi-dimensional collaborative graph structure, using a graph neural network combined with an attention mechanism to capture the complex interactive relationship between multi-dimensional features, and then building an efficient multi-dimensional collaborative graph perception network, an innovative solution is provided for the identification of offshore wind power ramping events. The present invention not only improves the recognition accuracy and real-time performance, but also provides important technical guarantees for the safe operation of offshore wind farms.

[0141] This embodiment further verifies the effectiveness of the offshore wind power ramp prediction method proposed in the present invention. First, in step S1, a multi-source data set of an offshore wind farm in Guangdong Province in the time period of 2022 / 1 / 1 / 0:00~2022 / 12 / 31 / 23:59 is obtained, including wind power data, weather forecast numerical data, and geographical conditions. Then, the processed and expanded node data is input into the multidimensional collaborative graph perception network, and the number of network layers is set to 5, including a network input layer, an intra-node attention learning layer, an inter-node attention learning layer, a feature integration layer, and a multi-layer perceptron. Then, the offshore wind power ramp prediction method proposed in the present invention is used to obtain the following: Figure 4 The offshore wind power prediction effect diagram shown in the figure uses an offshore wind power ramp prediction method proposed by the present invention to effectively improve the prediction accuracy of wind power ramp events.

[0142] Example 3

[0143] See also Figure 5 This embodiment proposes an offshore wind power ramp prediction system, including:

[0144] An acquisition module for acquiring multi-source datasets for offshore wind farms;

[0145] A preprocessing module, used to preprocess the multi-source data set to obtain meteorological driving factors, timestamps of wind power and geographical feature information of offshore wind farms;

[0146] A data expansion module, used for arranging the meteorological driving factor, the wind power, the timestamp and the geographic feature information into a uniformly formatted slope data, expanding the slope data, and obtaining a slope data set;

[0147] A multi-dimensional collaborative graph perception network building module, which is used to build a multi-dimensional collaborative graph perception network for predicting wind power ramping events;

[0148] A training module, used to train the multi-dimensional collaborative graph perception network using the climbing data set to obtain a trained multi-dimensional collaborative graph perception network;

[0149] The prediction module is used to collect the multi-source data set to be predicted of the offshore wind farm, input the multi-source data set to be predicted into the trained multi-dimensional collaborative graph perception network, and output the prediction result of the wind power ramp event.

[0150] In this embodiment, a multi-source data set of an offshore wind farm is first obtained, and the multi-source data set is preprocessed to obtain meteorological driving factors, timestamps of wind power, and geographical feature information of the offshore wind farm. Then, the meteorological driving factors, wind power, timestamps, and geographical feature information are organized into climbing data in a unified format for sample expansion to generate a climbing data set with richer data. Then, a multi-dimensional collaborative graph perception network is constructed and trained to obtain a trained multi-dimensional collaborative graph perception network, and the trained multi-dimensional collaborative graph perception network is used to output the prediction results of wind power climbing events. The present invention combines the multi-source data set with the multi-dimensional collaborative graph perception network, effectively improves the prediction accuracy and real-time performance of wind power climbing events, provides safety protection for the operation of offshore wind farms, and has important engineering value.

[0151] Obviously, the above embodiments of the present invention are only examples for clearly illustrating the present invention, and are not intended to limit the implementation methods of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all implementation methods here. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the claims of the present invention.

Claims

1. A method for predicting offshore wind power ramping, characterized in that: The following steps are involved: S1. Obtain multi-source datasets for offshore wind farms; S2. Preprocessing the multi-source data set to obtain meteorological driving factors, wind power timestamps and geographic feature information of offshore wind farms; S3. Arrange the meteorological driving factor, the wind power, the timestamp and the geographic feature information into a unified format of climbing data, expand the climbing data, and obtain a climbing data set; S4. Construct a multi-dimensional collaborative graph perception network for predicting wind power ramp events, wherein the multi-dimensional collaborative graph perception network includes a network input layer, an intra-node attention learning layer, an inter-node attention learning layer, a feature integration layer, and a multi-layer perceptron; S5. Use the climbing data set to train the multi-dimensional collaborative graph perception network to obtain a trained multi-dimensional collaborative graph perception network; S6. Collect the multi-source data set to be predicted of the offshore wind farm, input the multi-source data set to be predicted into the trained multi-dimensional collaborative graph perception network, and output the prediction result of the wind power ramp event; Among them, S4 includes: S41. Use the climbing data to form a multidimensional collaborative graph, the multidimensional collaborative graph includes four large nodes and dynamic edges, the four large nodes are timestamp T, meteorological driving factor M, geographic feature information G and wind power P, each large node includes a number of small nodes, each small node is used to represent a specific sub-feature of the large node, the dynamic edge is composed of intra-node edges and inter-node edges, the intra-node edges represent the relationship between small and medium nodes in the same large node, and the inter-node edges represent the relationship between different large nodes, and the multidimensional collaborative graph is input into the intra-node attention learning layer through the network input layer; S42. The node-in-attention learning layer adjusts the node-in-edge weights of the multi-dimensional collaborative graph through node-in-attention learning to obtain After a round of information transmission, the big node The eigenvector of : in, represents the softmax activation function, Indicates that the path Next small node Pass to the big node attention, Indicates that the path Next big node Small and medium nodes A collection of Represents a large node Small and medium nodes Intra-node attention features; S43. According to the feature vector The inter-node attention learning layer adjusts the inter-node edge weights of the multi-dimensional collaborative graph through inter-node attention learning to obtain a node feature vector set : in, P Indicates the path L The total number of Indicates p Path L， Representative in the p Path L The set of all large nodes under represents the node feature vector set, represents the normalized inter-node attention; S44. The feature integration layer collects node feature vectors Integrate into a global graph to represent and input into the multi-layer perceptron; S45. The multi-layer perceptron receives the global graph, maps the connections between large and small nodes with wind power ramp events, and outputs the prediction results of wind power ramp events.

2. The offshore wind power ramp prediction method according to claim 1, characterized in that: The multi-source data set includes wind power time series data, weather forecast numerical data and geographical condition data. The weather forecast numerical data includes wind speed, wind direction, air pressure, humidity, temperature and precipitation. The geographical condition data includes the longitude, latitude and altitude of each wind turbine.

3. The offshore wind power ramp prediction method according to claim 2, characterized in that: The preprocessing of the multi-source data set includes: S21. The Pearson correlation coefficient method is used to calculate the correlation between each meteorological feature in the multi-source data set and the target value of the climbing event, and the calculation expression of the Pearson correlation coefficient is obtained as follows: Among them, r represents the Pearson correlation coefficient corresponding to the current meteorological parameters, Indicates c The normalized value of a meteorological characteristic at a time point, Indicates c The wind power change rate value corresponding to the time point is: Indicates meteorological characteristics The average value of represents the mean value of wind power change rate, and N represents the total number of data points; S22. According to the Pearson correlation coefficient, the meteorological factor data with high correlation with wind power in the weather forecast numerical data are extracted, namely, the meteorological driving factor M, the timestamp T of the wind power time series data, and the geographical feature information G of the offshore wind farm closely related to wind power.

4. The offshore wind power ramp prediction method according to claim 3, characterized in that: The step of expanding the climbing data to obtain a climbing data set includes: S31. Arrange the meteorological driving factor, the wind power, the timestamp and the geographic feature information into a unified format of climbing data in the format of timestamp T-meteorological driving factor M-geographic feature information G-wind power P, align the climbing data according to the timestamp T, so that each row contains the meteorological driving factor M, geographic feature information G and wind power P at a certain timestamp T at the same time, and obtain the expression of the data matrix as follows: Among them, T n Indicates the timestamp of the nth row, M n represents the meteorological driving factor of the nth row, G n Indicates the geographic feature information of the nth row, P n represents the wind power of the nth row; S32. Input the data matrix into a preset adversarial network TimeGAN for time series generation, wherein the adversarial network TimeGAN includes a generator and a discriminator, wherein the generator receives the data matrix, generates new time series data and inputs it into the discriminator, and the discriminator receives the time series data and outputs a discriminant result of whether the time series data conforms to the distribution and regularity of real data; S33. Vertically splicing the discrimination result and the slope climbing data in a specific ratio to obtain the slope climbing data set.

5. The offshore wind power ramp prediction method according to claim 4, characterized in that: The network input layer is connected to the input end of the intra-node attention learning layer, the output end of the intra-node attention learning layer is connected to the input end of the inter-node attention learning layer, the output end of the inter-node attention learning layer is connected to the input end of the feature integration layer, the output end of the feature integration layer is connected to the input end of the multi-layer perceptron, and the output end of the multi-layer perceptron outputs the prediction result of the wind power ramp event.

6. The offshore wind power ramp prediction method according to claim 5, characterized in that: The multilayer perceptron includes an input layer, a hidden layer and an output layer connected in sequence, and the calculation formula from the input layer to the hidden layer is: in, Represents the input layer The output of a neuron, express Activation function, Represents the input layer The input of a neuron, Represents the hidden layer The neurons in the input layer The weights between neurons, Represents the input layer The output of a neuron, Represents the hidden layer The bias of each neuron; The calculation formula from the hidden layer to the output layer is: in, Indicates that the path L The lower output layer outputs The probability that a sample belongs to a wind power ramp event, L Indicates the path, Indicates that the path L Lower input layer The input of a neuron, Indicates that the path L Lower input layer k The input of a neuron, K represents the total number of neurons in the output layer, Represents the activation function.

7. The offshore wind power ramp prediction method according to claim 1, characterized in that: The criteria for judging the prediction results are as follows: Calculate specific time intervals The actual wind power consumption change rate as follows: in, Indicates time The actual wind power consumed, Indicates time The actual wind power consumed; Determine the wind power change rate Is it greater than or equal to the first change threshold? and less than or equal to , if so, determine the time interval The event occurring within the time interval is not a wind power ramp event; otherwise, the time interval is determined to be The event that occurred within is a wind power ramp event.

8. The offshore wind power ramp prediction method according to claim 1, characterized in that: The step of training the multi-dimensional collaborative graph perception network using the hill climbing data set includes: S51. Dividing the climbing data set into a training set and a test set according to a specific ratio; S52. Use the training set to train the multidimensional collaborative graph perception network, and use the test set to test the effectiveness of the multidimensional collaborative graph perception network, until the prediction accuracy of the multidimensional collaborative graph perception network for wind power ramping events reaches a preset accuracy threshold, thereby obtaining a trained multidimensional collaborative graph perception network.

9. An offshore wind power ramp prediction system, characterized in that: include: An acquisition module for acquiring multi-source datasets for offshore wind farms; A preprocessing module, used to preprocess the multi-source data set to obtain meteorological driving factors, timestamps of wind power and geographical feature information of offshore wind farms; A data expansion module, used for arranging the meteorological driving factor, the wind power, the timestamp and the geographic feature information into a uniformly formatted slope data, expanding the slope data, and obtaining a slope data set; A multi-dimensional collaborative graph perception network building module, which is used to build a multi-dimensional collaborative graph perception network for predicting wind power ramping events; A training module, used to train the multidimensional collaborative graph perception network using the climbing data set to obtain a trained multidimensional collaborative graph perception network, wherein the multidimensional collaborative graph perception network includes a network input layer, an intra-node attention learning layer, an inter-node attention learning layer, a feature integration layer, and a multi-layer perceptron; A prediction module is used to collect a multi-source data set to be predicted of an offshore wind farm, input the multi-source data set to be predicted into a trained multi-dimensional collaborative graph perception network, and output a prediction result of a wind power ramp event; The multi-dimensional collaborative graph perception network for predicting wind power ramp events is constructed, including: S41. Use the climbing data to form a multidimensional collaborative graph, the multidimensional collaborative graph includes four large nodes and dynamic edges, the four large nodes are timestamp T, meteorological driving factor M, geographic feature information G and wind power P, each large node includes a number of small nodes, each small node is used to represent a specific sub-feature of the large node, the dynamic edge is composed of intra-node edges and inter-node edges, the intra-node edges represent the relationship between small and medium nodes in the same large node, and the inter-node edges represent the relationship between different large nodes, and the multidimensional collaborative graph is input into the intra-node attention learning layer through the network input layer; S42. The node-in-attention learning layer adjusts the node-in-edge weights of the multi-dimensional collaborative graph through node-in-attention learning to obtain After a round of information transmission, the big node The eigenvector of : in, represents the softmax activation function, Indicates that the path Next small node Pass to the big node attention, Indicates that the path Next big node Small and medium nodes A collection of Represents a large node Small and medium nodes Intra-node attention features; S43. According to the feature vector The inter-node attention learning layer adjusts the inter-node edge weights of the multi-dimensional collaborative graph through inter-node attention learning to obtain a node feature vector set : in, P Indicates the path L The total number of Indicates p Path L， Representative in the p Path L The set of all large nodes under represents the node feature vector set, represents the normalized inter-node attention; S44. The feature integration layer collects node feature vectors Integrate into a global graph to represent and input into the multi-layer perceptron; S45. The multi-layer perceptron receives the global graph, maps the connections between large and small nodes with wind power ramp events, and outputs the prediction results of wind power ramp events.

Images