A fan energy management method and system based on multi-dimensional monitoring

Abstract

The application discloses a kind of fan energy management method and system based on multidimensional monitoring, it is related to fan data monitoring and control technical field, obtain the sample set of multidimensional time series data in historical monitoring data construction;Build power generation prediction model, real-time acquisition data to be predicted, import into power generation prediction model, obtain overall predicted power generation;Get wind farm operation area and constraint wind speed, with constraint wind speed as screening condition, screen node fan in wind farm operation area, obtain the node fan set corresponding to wind farm operation area, using the multidimensional attribute of each node fan, calculate load balancing value;Based on load balancing value, calculate the load distribution coefficient value of each fan, according to load distribution coefficient value and overall predicted power generation, load adjustment is carried out to node fan.The application can effectively realize the prediction of overall power generation and the load adjustment of each fan, so as to ensure the stability of power generation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fan data monitoring and control, and more particularly to a fan energy management method and system based on multi-dimensional monitoring. BACKGROUND

[0002] Under the background of global energy transformation, wind power as a clean and sustainable energy utilization method occupies an increasingly important position in the power supply system. The efficient operation and optimized management of wind farms are of great significance to improve the competitiveness of wind power and promote the large-scale development of renewable energy.

[0003] With the increasing attention to environmental protection and sustainable development, countries are committed to promoting the transformation of energy structure to renewable energy. Wind power, due to its abundant resources and no pollution, has become one of the important ways to achieve this goal. The scale of wind farms is expanding, and the installed capacity is growing continuously. The operation and management level of wind farms directly affects the position and role of wind power in the energy market. For example, in some European countries, wind power has become one of the main sources of electricity, and the operation efficiency and reliability of its wind farms are crucial to the security of national energy.

[0004] However, the intermittency and volatility of wind power pose challenges to the stable operation of the power system. The change of wind speed is not under human control, which may cause large fluctuations in the power generation of wind turbines at any time. This uncertainty makes it difficult for wind farms to coordinate with the power grid in operation, such as difficulty in meeting the real-time power balance requirements of the power grid, and may cause voltage and frequency fluctuations in the power grid. Therefore, how to effectively manage wind farms to improve the stability and predictability of power generation is a problem that needs to be solved by those skilled in the art. SUMMARY

[0005] Therefore, the present application provides a fan energy management method and system based on multi-dimensional monitoring, which can effectively realize the prediction of overall power generation and the load adjustment of each fan, thereby ensuring the stability of power generation.

[0006] In order to achieve the above purpose, the present application adopts the following technical solutions:

[0007] A fan energy management method based on multi-dimensional monitoring, comprising:

[0008] Collecting various types of historical monitoring data of wind farms, and screening multi-dimensional time series data, constructing a sample set according to the multi-dimensional time series data and the power generation at each sampling time;

[0009] The neural network model is constructed, the neural network model is trained by using the sample set, a full-precision model is obtained, parameters of the neural network model are binary quantized to obtain a quantization model, the quantization model is trained based on the full-precision model, an error between models is added in a loss function of the quantization model, and a power generation prediction model is obtained; real-time acquisition of to-be-predicted data is performed, the to-be-predicted data is input into the power generation prediction model, and overall predicted power generation is obtained.

[0010] An operating area of a wind farm and a constraint wind speed are obtained, the operating area of the wind farm at least including a node wind turbine, each node wind turbine having a multi-dimensional attribute for representing a power generation state of a corresponding unit area in the operating area of the wind farm;

[0011] The node wind turbines in the operating area of the wind farm are screened based on the constraint wind speed, a node wind turbine set corresponding to the operating area of the wind farm is obtained, and a load balancing value of each node wind turbine is calculated based on the multi-dimensional attribute of each node wind turbine in the node wind turbine set.

[0012] Based on the load balancing value of each node wind turbine, a load distribution coefficient value of each node wind turbine is calculated, and the node wind turbines are adjusted based on the load distribution coefficient value and the overall predicted power generation.

[0013] Preferably, the sample set is constructed based on the multi-dimensional time sequence data and the power generation at each sampling time, specifically including: grouping the multi-dimensional data at the same sampling time into a piece of sampling data to obtain a sampling data sequence; dividing the sampling data sequence based on a preset time window to obtain a plurality of time window samples and put the time window samples into the sample set; and taking the power generation at the last sampling time of each time window sample as a label of the time window sample.

[0014] Preferably, in the construction of the neural network model, the neural network model includes a composite attention module and a time sequence convolution module; the composite attention module includes a trend segment preprocessing layer, a pooling layer and a composite perception attention layer; the trend segment preprocessing layer is used to divide each input sample from a time sequence dimension into a plurality of trend samples and then transmit the trend samples to the pooling layer; the pooling layer is used to respectively perform maximum pooling and average pooling on the trend samples, add maximum pooling features and average pooling features to obtain pooled features, and then transmit the pooled features to the composite perception attention layer; the composite perception attention layer is used to learn the pooled features to obtain a composite attention vector of the last layer; the time sequence convolution module includes a time sequence convolution layer and a multilayer perception layer; the time sequence convolution layer is used to receive an output result of the composite attention module, implement time sequence information transmission by using one-dimensional convolution with different expansion values of each layer, and transmit the time sequence information to the multilayer perception layer; and the multilayer perception layer learns a prediction result by using multilayer perception.

[0015] Preferably, the binary quantization of the neural network model parameters obtains a quantization model, which is the binary quantization of the weights and biases of the neural network model to obtain the binary weights and biases as initial parameters of the quantization model; the training of the quantization model based on the full-precision model comprises:

[0016] Obtain the mean value of each layer weight in the full-precision model as the equalization factor of each layer;

[0017] In the forward propagation, the last layer of the compound attention module in the quantization model uses a binary Sigmoid activation function to calculate a compound attention vector, and the binary weights of other layers are multiplied by the activation vector of the previous layer and added with a bias, and then a Relu activation function is used to obtain an initial activation vector of each layer. After the initial activation vector of each layer is binary quantized and multiplied by the equalization factor of the corresponding layer, a final activation vector of each layer is obtained, which is used for the next layer;

[0018] In the backward propagation, the gradient of the binary weight is calculated according to the loss function, and the gradient is clipped as the gradient value of the floating-point gradient for weight update.

[0019] Preferably, the calculation of the load balancing value of each node fan based on the multi-dimensional attributes of each node fan in the node fan set comprises: calculating the state deviation degree of each attribute of each node fan in the node fan set based on the multi-dimensional attributes of each node fan; and calculating the load balancing value of each node fan based on the state deviation degree and the multi-dimensional attributes of each node fan.

[0020] Preferably, the calculation of the load balancing value of each node fan based on the multi-dimensional attributes of each node fan in the node fan set comprises: calculating the state deviation degree of each attribute of each node fan in the node fan set based on the multi-dimensional attributes of each node fan; and calculating the load balancing value of each node fan based on the state deviation degree and the multi-dimensional attributes of each node fan.

[0021] Preferably, the calculation of the load balancing value of each node fan based on the multi-dimensional attributes of each node fan in the node fan set comprises:

[0022]

[0023] Wherein, g is the load distribution coefficient value, b ji is the i-th node fan in the j-th wind farm operation area, b j is the load mean value of the j-th wind farm operation area, m is the maximum value of the number of node fans in the j-th wind farm operation area, and k is the maximum value of the wind farm operation area.

[0024] Preferably, the multi-dimensional attributes include power coefficient, fan vibration frequency and component temperature.

[0025] The node fan in the wind farm operation area is screened with the constraint wind speed as a screening condition, to obtain a node fan set corresponding to the wind farm operation area, including: obtaining the fan vibration frequency of each node fan in the wind farm operation area; and screening out the node fan whose fan vibration frequency meets the screening condition to form the node fan set.

[0026] A fan energy management system based on multi-dimensional monitoring, comprising:

[0027] A sample set construction module acquires various types of historical monitoring data of a wind farm, screens out multi-dimensional time series data, and constructs a sample set according to the multi-dimensional time series data and power generation at each sampling time;

[0028] A power generation prediction module constructs a neural network model, trains the neural network model using the sample set to obtain a full-precision model, performs binary quantization on parameters of the neural network model to obtain a quantized model, trains the quantized model based on the full-precision model, adds an inter-model error in a loss function of the quantized model, and obtains a power generation prediction model; real-time acquisition of to-be-predicted data is performed, the to-be-predicted data is input into the power generation prediction model, and overall predicted power generation is obtained.

[0029] A constraint acquisition module acquires a wind farm operation area and a constraint wind speed, the wind farm operation area containing at least one node fan, each node fan having multi-dimensional attributes for representing the power generation state of a corresponding unit area in the current wind farm operation area;

[0030] A load balancing value calculation module screens the node fan in the wind farm operation area with the constraint wind speed as a screening condition, to obtain a node fan set corresponding to the wind farm operation area, and calculates a load balancing value of each node fan using the multi-dimensional attributes of each node fan in the node fan set;

[0031] A load adjustment module calculates a load distribution coefficient value of each node fan based on the load balancing value of each node fan, and adjusts the load of the node fan according to the load distribution coefficient value and the overall predicted power generation.

[0032] Compared with the prior art, the fan energy management method and system based on multi-dimensional monitoring provided by the application can comprehensively integrate data information of different dimensions and different time points, so that subsequent model training can be based on rich and relevant data, the existing data resources are used to the greatest extent, and a foundation is laid for accurate energy management; the constructed neural network model comprises a composite attention module and a time series convolution module, the composite attention module can process input samples from multiple angles through the collaborative work of a trend segment preprocessing layer, a pooling layer and a composite perception attention layer, and better capture trend characteristics and key information in the data; the time series convolution module realizes time series information transmission by using one-dimensional convolution with different expansion values, and then combines a multi-layer perception layer to learn a prediction result, so that the overall architecture enables the model to have strong learning and prediction ability for time series related data such as fan power generation, and can improve the accuracy of power generation prediction; the node fan set is screened out by taking the constraint wind speed as the screening condition, then the load balancing value of each node fan is calculated, and on this basis, the load distribution coefficient value is calculated, and finally the node fan is adjusted according to the load distribution coefficient value and the overall predicted power generation; this way of implementing load adjustment based on multi-dimensional state evaluation can ensure power generation while making the load of each fan more reasonable and balanced, avoiding the situation that some fans are overloaded or insufficiently loaded, improving the power generation efficiency of the entire wind farm, and also helping to prolong the service life of the fan and reduce equipment maintenance costs. BRIEF DESCRIPTION OF DRAWINGS

[0033] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description only constitute a part of the embodiments of the present application, and for those skilled in the art, other drawings can be obtained without creative labor based on the provided drawings.

[0034] Figure 1 The method steps provided by the present application are shown in the following figure:

[0035] Figure 2 The structure schematic diagram provided by the present application is shown in the following figure: DETAILED DESCRIPTION

[0036] The technical solutions in the embodiments of the present application will be described clearly and completely in the following with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments only constitute a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0037] The embodiment of the application discloses a fan energy management method based on multi-dimensional monitoring, which comprises the following steps: Figure 1 As shown in the figure, comprising:

[0038] Collecting various types of historical monitoring data of the wind farm, screening multi-dimensional time series data, and constructing a sample set according to the multi-dimensional time series data and the power generation capacity of each sampling time;

[0039] Constructing a neural network model, training the neural network model by using the sample set to obtain a full-precision model, quantizing the parameters of the neural network model to obtain a quantized model, training the quantized model based on the full-precision model, adding an inter-model error in the loss function of the quantized model, and obtaining a power generation capacity prediction model; collecting real-time prediction data and inputting the data into the power generation capacity prediction model to obtain the overall predicted power generation capacity;

[0040] Obtaining the running area of the wind farm and the constraint wind speed, the running area of the wind farm containing at least one node fan, each node fan having multi-dimensional attributes for representing the power generation state of the corresponding unit area in the current wind farm running area;

[0041] Screening the node fans in the running area of the wind farm by taking the constraint wind speed as a screening condition to obtain a node fan set corresponding to the running area of the wind farm, and calculating the load balancing value of each node fan by using the multi-dimensional attributes of each node fan in the node fan set;

[0042] Based on the load balancing value of each node fan, the load distribution coefficient value of each node fan is calculated, and the node fan is adjusted according to the load distribution coefficient value and the overall predicted power generation capacity.

[0043] The historical monitoring data includes different types of long-time continuous data, and most of these data are different in magnitude, unit, form, etc. In addition, there are redundant and bad features in the data characteristics collected by multiple sensors, which have no contribution to the prediction accuracy of the model. Therefore, the historical monitoring data needs to be preprocessed as follows to screen out multi-dimensional time series data, reduce the environmental factor interference during sensor data collection, and improve the prediction accuracy:

[0044] Removing noise features: using the Savitzky-Golay smoothing filter method to remove abnormal data in each type of monitoring data respectively, and supplementing data according to the obtained polynomial to obtain denoised data;

[0045] Standardizing data features: standardizing each type of monitoring data after denoising according to the mean and standard deviation;

[0046] Screening data features: curve fitting is performed on each type of standardized monitoring data, regression evaluation index values are calculated, monitoring data types with regression evaluation index values greater than a screening threshold are obtained, and data features for predicting overall power generation are obtained.

[0047] In one specific embodiment, constructing a sample set according to multi-dimensional time series data and power generation at each sampling time point specifically includes: grouping multi-dimensional data at the same sampling time point into a sampling data to obtain a sampling data sequence; dividing the sampling data sequence according to a preset time window to obtain a plurality of time window samples and put them into the sample set; and the power generation at the last sampling time point of each time window sample is taken as the label of the time window sample.

[0048] Specifically, each sampling data is represented as {d1, d2,..., d M} M×C There are M sampling time points in total, and C characteristic dimensions of monitoring data are screened out at each sampling time point. For example, wind speed, wind direction, air temperature, air pressure, and fan maintenance record data are screened out from various types of historical monitoring data, and C = 5.

[0049] The time window length is an empirical value, which is adjusted according to the training result in the training process. The time window length is set to 30, that is, the monitoring data of every 30 sampling time points are grouped into a time window sample.

[0050] In one specific embodiment, in constructing the neural network model, the neural network model includes a composite attention module and a time series convolution module; the composite attention module includes a trend segment preprocessing layer, a pooling layer, and a composite perception attention layer; the trend segment preprocessing layer is used to divide each input sample from the time series dimension into multiple trend segments and then transmit them into the pooling layer; the pooling layer is used to respectively perform maximum pooling and average pooling on each trend segment, add the maximum pooled features and the average pooled features to obtain pooled features, and then transmit the pooled features into the composite perception attention layer; the composite perception attention layer is used to learn the pooled features and obtain a composite attention vector of the last layer; the time series convolution module includes a time series convolution layer and a multilayer perception layer; the time series convolution layer is used to receive the output result of the composite attention module, implement time series information transmission by using one-dimensional convolution with different expansion values of each layer, and transmit the time series information into the multilayer perception layer; and the multilayer perception layer obtains a prediction result by multilayer perception learning.

[0051] In one specific embodiment, the length of the composite attention vector of the last layer is consistent with the feature dimension of the input sample; the input sample is weighted according to the composite attention vector to obtain the output result of the composite attention module.

[0052] Since each time window sample is time series data, in order to accurately obtain the trend change caused by the change of data over time, the input sample is again divided into N trend samples in the trend segment preprocessing layer, which is expressed as follows:

[0053]

[0054] wherein Split(·) represents the time period division operation, Hconcat(·) represents the time series dimension splicing operation, D H×C represents the input sample, that is, each time window sample in the sample set; each trend sample has samples.

[0055] Specifically, in the pooling layer, each trend sample is input into the maximum pooling layer and the average pooling layer respectively to obtain the maximum pooling feature and the average pooling feature, thereby generating two different spatial context descriptions. Then, the maximum pooling feature and the average pooling feature are added according to the following formula to obtain the pooling feature:

[0056]

[0057] wherein pool(·) represents the pooling operation, Avg(·) represents the average pooling operation, Max(·) represents the maximum pooling operation, represents the obtained pooling feature.

[0058] In one specific embodiment, the neural network model parameters are binary quantized to obtain a quantization model, which is to binary quantize the weights and biases of the neural network model to obtain binary weights and biases as initial parameters of the quantization model; training the quantization model based on the full-precision model includes:

[0059] Obtain the mean of each layer weight in the full-precision model as the equalization factor of each layer;

[0060] In the forward propagation, the last layer of the compound attention module in the quantization model uses a binary Sigmoid activation function to calculate a compound attention vector, and the binary weights of other layers are multiplied by the activation vector of the previous layer, and then added with a bias, and then a Relu activation function is used to obtain an initial activation vector of each layer. After binary quantization of the initial activation vector of each layer, the equalization factor of the corresponding layer is multiplied to obtain the final activation vector of each layer, which is used for the next layer;

[0061] In the backward propagation, the gradient of the binary weight is calculated according to the loss function, and then the gradient is clipped as the gradient value of the floating-point gradient for weight update.

[0062] The pooled features are input into a composite perception attention layer to calculate a composite attention vector. The composite perception attention layer is constructed based on a multi-layer perception (MLP) and can include an input layer, multiple hidden layers, and a fully connected layer as output, and can only include a fully connected layer. In the last layer, the weights are multiplied by the pooled features of the previous layer, and the composite attention vector is obtained after activation. It should be noted that if there is a bias in the last layer, the weights are multiplied by the pooled features of the previous layer and then added to the bias, and the composite attention vector is obtained after activation.

[0063] The length of the composite attention vector is consistent with the feature dimension C of the input sample, representing the weight distribution of the multi-dimensional features. Finally, the input sample is weighted according to the composite attention vector to obtain the output result of the composite attention module, i.e., the monitoring data of each feature dimension at each sampling time in the input sample is multiplied by the composite attention vector of the same feature dimension to obtain the weighted sample as the output result of the composite attention module, which is input into the time convolution module.

[0064] The time convolution module includes a time convolution layer and a multi-layer perception layer. The time convolution layer is constructed according to a temporal convolutional network (TCN) and is used to receive the output result of the composite attention module, to realize time sequence information transmission by one-dimensional convolution with different expansion values in each layer, and to transmit the time sequence information to the multi-layer perception layer. The multi-layer perception layer is constructed according to a multi-layer perception (MLP) and learns through multiple layers of perception. The entire time sequence information is captured by a single neuron in the last layer, and a prediction result is output.

[0065] The binary Sigmoid activation function includes: when the activation value obtained by using the Sigmoid activation function is greater than or equal to a preset threshold, the activation is 1; otherwise, the inhibition is 0; and the binary quantization of the weights, the bias, and the initial activation vector is to convert the values into two numerical values of 1 and -1.

[0066] The inter-model error is added in the loss function of the quantization model, which is the sum of the mean square error of the prediction result of the quantization model and the label of the power generation, and the information entropy of the weights of each layer of the full-precision model and the quantization model and the mean square error of the output of each layer.

[0067] Considering that in the process of binary quantization of neural network model parameters, the model parameters are converted from high-bit expression form (such as 32-bit floating-point number type) to binary low-bit quantization form, the memory occupation of the model is obviously reduced, but the accuracy loss is obvious, and it is difficult to maintain the prediction performance of the model. Therefore, the sample set is used to perform two training processes on the constructed neural network model: the first time, the model parameters are not quantized, and the model parameters are all floating-point numbers. The mean error of the overall predicted power generation and the label of the power generation is used as the loss function for training. The training is stopped after the maximum number of iterations or the prediction accuracy is reached, and the trained model is obtained as the full-precision model. The second time, the neural network model parameters of the same structure are quantized to obtain a quantized model, and the same sample set is used to optimize the model parameters during the training process. In order to improve the prediction accuracy and improve the prediction performance, the balance factor is constructed according to the first trained full-precision model, and the information entropy and output error of each layer between the full-precision model and the second trained quantized model are added to the loss function to optimize the quantized model parameters. The second trained quantized model is used as the final power generation prediction model.

[0068] Specifically, the binary quantization of the neural network model parameters obtains a quantized model, which is the binary quantization of the weights and biases of the neural network model, and the binary weights and biases are used as the initial parameters of the quantized model. Then, the quantized model is trained based on the full-precision model, including:

[0069] ①Obtain the mean value of each layer weight in the full-precision model as the balance factor of each layer.

[0070] It should be noted that the model parameters in the full-precision model are all floating-point number type data, and the mean value of each layer weight in the full-precision model is obtained, including: for the convolution layer, the mean value of all convolution kernel weights in the current layer is obtained; for the fully connected layer, the mean value of the weights connecting all neurons in the current layer is obtained.

[0071] ②In the forward propagation, the last layer of the compound attention module in the quantized model uses a binary Sigmoid activation function to calculate the compound attention vector, and the binary weights of the other layers are multiplied by the activation vector of the previous layer and then added to the bias, and then a Relu activation function is used to obtain the initial activation vector of each layer. After binary quantization of the initial activation vector of each layer, the balance factor of the corresponding layer is multiplied to obtain the final activation vector of each layer, which is used for the next layer.

[0072] It should be noted that the weights in the neural network model are used to represent the connection strength between neurons in each layer; the bias is used to correctly classify samples and ensure that the output value cannot be activated at will; the activation function plays a role in nonlinear mapping, which can limit the output amplitude of the neuron within a certain range.

[0073] The binary quantization of the weights, biases and initial activation vectors is performed by a sign function that converts values to two values, 1 and -1, as follows:

[0074]

[0075] where x1 represents a value in the weights, biases or initial activation vectors, the last layer of the compound attention module adopts a Sigmoid activation function, and the other layers adopt a Relu activation function.

[0076] The binary Sigmoid activation function is a binary quantization of the activation value based on the Sigmoid activation function, and if the activation value is greater than or equal to a preset threshold, the activation is 1; otherwise, the inhibition is 0, as follows:

[0077]

[0078] where x2 represents a value to be activated, i.e., the value of the last layer of the binary weight multiplied by the value of the previous layer of the pooled feature.

[0079] The loss function Loss of the quantization model includes three parts: the mean square error of the prediction result and the remaining life label, the information entropy of the weights of each layer of the full-precision model and the quantization model, and the mean square error of the output of each layer.

[0080] In one specific embodiment, the load balancing value of each node fan is calculated using the multi-dimensional attributes of each node fan in the node fan set, which includes: calculating the state deviation of each attribute of each node fan using the multi-dimensional attributes of each node fan in the node fan set; and calculating the load balancing value of each node fan according to the state deviation and multi-dimensional attributes of each node fan.

[0081] The state deviation is used to measure the deviation between the current actual operating state of the fan and the ideal operating state. The ideal operating state refers to a state in which the fan can achieve the highest efficiency, the longest service life and the most stable operation under various constraint conditions. For example, for a fan, its ideal operating state may be at a rated wind speed, with an output power of rated power, health indicators such as vibration and temperature within a normal range, and a power factor meeting the requirements of the power grid. The state deviation quantifies the gap between the current state of the fan and this ideal state by considering the multi-dimensional attributes of the fan, such as the power coefficient, fan vibration frequency and component temperature.

[0082] b ji =w ji *X ji ;

[0083] where b jiis the state deviation degree of the i-th node wind turbine in the j-th wind farm operation area, w ji is the deviation coefficient, X ji is the attribute value corresponding to the i-th node wind turbine in the j-th wind farm operation area.

[0084] The step of calculating the deviation coefficient needs to calculate the deviation value, the deviation value proportion, the deviation mean value and the limited value, which are calculated as follows:

[0085]

[0086] wherein y ji is the deviation value, representing the absolute difference between the true attribute value and the average attribute value, p avg is the average value of the power coefficient, the wind turbine vibration frequency and the component temperature of all node wind turbines a j in the j-th wind farm operation area, p ji is the attribute value corresponding to the i-th node wind turbine in the j-th wind farm operation area, and max(p)-min(p) is the difference between the maximum and minimum values of the same attribute value, so that all deviation values {y j1 ,y j2 ,...,y ji}(i=1,...,m, m represents the number of node wind turbines corresponding to the j-th wind farm operation area) in the node wind turbine set a j in the j-th wind farm operation area are calculated. After the deviation value is calculated, the deviation value proportion is further calculated:

[0087]

[0088] wherein p ji is the deviation value proportion of the attribute values (q ji , v ji and c ji ) of each node wind turbine in a j .

[0089] After obtaining the deviation value proportion of the attribute values of each node wind turbine, the deviation mean value of each attribute (q j , v j and c j ) in the node wind turbine set in the j-th business area is calculated:

[0090]

[0091] wherein e ji is the deviation mean value.

[0092] The deviation mean value of each attribute (q j , vj and c j ) of the limited value q ji :

[0093]

[0094] The limited value q ji is calculated, and the deviation coefficient of each attribute in the i th node wind turbine in the j th wind farm operation area is calculated respectively.

[0095]

[0096] Wherein, m represents the number of node wind turbines corresponding to the j th wind farm operation area, is the sum of the limited values of the node wind turbines in the wind farm operation area.

[0097] In one specific embodiment, the calculation of the load distribution coefficient value of each node wind turbine specifically includes: calculating the load balancing value of each node wind turbine in the node wind turbine set corresponding to each wind farm operation area; performing average processing on the load balancing values in the corresponding wind farm operation area to obtain the load mean value of the wind farm operation area; and calculating the load distribution coefficient value of each node wind turbine by using the load mean value of each wind farm operation area.

[0098] In one specific embodiment, the calculation of the load distribution coefficient value of each node wind turbine by using the load mean value of each wind farm operation area specifically includes:

[0099]

[0100] Wherein, g is the load distribution coefficient value, b ji is the i th node wind turbine in the j th wind farm operation area, b j ' is the load mean value of the j th wind farm operation area, m is the maximum value of the number of node wind turbines in the j th wind farm operation area, and k is the maximum value of the wind farm operation area.

[0101] In one specific embodiment, the multi-dimensional attributes include power coefficient, wind turbine vibration frequency and component temperature.

[0102] The node wind turbines in the wind farm operation area are screened by taking the constraint wind speed as the screening condition to obtain the node wind turbine set corresponding to the wind farm operation area, including: obtaining the wind turbine vibration frequency of each node wind turbine in the wind farm operation area; and screening out the node wind turbines whose wind turbine vibration frequencies meet the screening condition to form the node wind turbine set.

[0103] A wind turbine energy management system based on multi-dimensional monitoring, as shown in Figure 2 , includes:

[0104] The sample set construction module collects various types of historical monitoring data of the wind farm, screens multi-dimensional time series data, and constructs a sample set according to the multi-dimensional time series data and power generation at each sampling time;

[0105] The power generation prediction module constructs a neural network model, trains the neural network model using the sample set to obtain a full-precision model, quantizes parameters of the neural network model to obtain a quantized model, trains the quantized model based on the full-precision model, adds an inter-model error in a loss function of the quantized model, and obtains a power generation prediction model; real-time acquisition of to-be-predicted data is performed, the to-be-predicted data is input into the power generation prediction model, and overall predicted power generation is obtained.

[0106] The constraint acquisition module acquires a wind farm operation area and a constraint wind speed, the wind farm operation area contains at least one node wind turbine, each node wind turbine has multi-dimensional attributes for representing a power generation state of a corresponding unit area in the current wind farm operation area;

[0107] The load balancing value calculation module screens node wind turbines in the wind farm operation area according to the constraint wind speed, obtains a node wind turbine set corresponding to the wind farm operation area, and calculates a load balancing value of each node wind turbine by using multi-dimensional attributes of each node wind turbine in the node wind turbine set.

[0108] The load adjustment module calculates a load distribution coefficient value of each node wind turbine based on the load balancing value of each node wind turbine, and adjusts the node wind turbine according to the load distribution coefficient value and the overall predicted power generation.

[0109] In one specific embodiment, adjusting the node wind turbine according to the load distribution coefficient value and the overall predicted power generation specifically includes: distributing a load value to a wind turbine that does not meet the screening condition according to a maximum wind speed that the wind turbine can withstand, adjusting related components (such as adjusting a blade pitch angle or a rotating speed) in the wind turbine that affect a wind speed, and adjusting a corresponding load of a node wind turbine in the node wind turbine set according to the load distribution coefficient value and the overall predicted power generation.

[0110] The embodiments in the specification are described in a progressive manner, and each embodiment focuses on the difference from other embodiments. The same or similar parts of each embodiment can be referred to each other. For the device disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple, and the related parts can be referred to the method part.

[0111] The foregoing description of the disclosed embodiments enables a person skilled in the art to make or use the application. Modifications of these embodiments will occur to persons of skill in the art, and, while certain modifications are discussed, it is desired to be protected in accordance with the spirit and scope of the application. Therefore, the application is not limited to the specific embodiments shown and described, but only by the scope of the appended claims, unless otherwise specified.

Claims

1. A fan energy management method based on multi-dimension monitoring, characterized in that, The application relates to a wind power generation prediction method and device. The method comprises the following steps: collecting various historical monitoring data of a wind power plant, screening multi-dimensional time sequence data, and constructing a sample set according to the multi-dimensional time sequence data and power generation capacity at each sampling time; A neural network model is constructed, the neural network model is trained by using the sample set, a full-precision model is obtained, the parameters of the neural network model are binary quantized to obtain a quantized model, the quantized model is trained based on the full-precision model, an inter-model error is added to a loss function of the quantized model, and a power generation capacity prediction model is obtained; real-time acquisition of to-be-predicted data is performed, the to-be-predicted data is input into the power generation capacity prediction model, and overall predicted power generation capacity is obtained; An operating area of the wind power plant and a constraint wind speed are obtained, at least one node wind turbine is contained in the operating area of the wind power plant, each node wind turbine has multi-dimensional attributes, and the multi-dimensional attributes are used for representing the power generation state of a corresponding unit area in the operating area of the wind power plant; The node wind turbines in the operating area of the wind power plant are screened according to the constraint wind speed, a node wind turbine set corresponding to the operating area of the wind power plant is obtained, and the multi-dimensional attributes of each node wind turbine in the node wind turbine set are used to calculate a load balancing value of each node wind turbine. Based on the load balancing value of each node wind turbine, a load distribution coefficient value of each node wind turbine is calculated, and the node wind turbines are adjusted in load according to the load distribution coefficient value and the overall predicted power generation capacity. The calculation of the load distribution coefficient value of each node wind turbine specifically comprises the following steps: the load balancing value of each node wind turbine in the node wind turbine set corresponding to each operating area of the wind power plant is calculated; the load balancing values corresponding to the operating area of the wind power plant are averaged to obtain a load average value of the operating area of the wind power plant; and the load average value of each operating area of the wind power plant is used to calculate the load distribution coefficient value of each node wind turbine. The calculation of the load distribution coefficient value of each node wind turbine based on the load average value of each operating area of the wind power plant specifically comprises the following steps: ； in, This is the load distribution coefficient value. For the first The first wind farm operation area Deviation of state of individual wind turbine nodes For the first Average load of each wind farm operating area For the first The maximum number of wind turbines in a single wind farm operating area. This represents the maximum value for the wind farm's operating area. The multi-dimensional attributes comprise a power coefficient, a wind turbine vibration frequency and a component temperature. The node wind turbines in the operating area of the wind power plant are screened according to the constraint wind speed, a node wind turbine set corresponding to the operating area of the wind power plant is obtained, and the multi-dimensional attributes of each node wind turbine in the node wind turbine set are used to calculate a load balancing value of each node wind turbine.

2. The fan energy management method based on multi-dimension monitoring according to claim 1, characterized in that, The construction of the sample set based on the multi-dimensional time sequence data and the power generation capacity at each sampling time specifically comprises the following steps: multi-dimensional data at the same sampling time are combined to form a sampling data, and a sampling data sequence is obtained; the sampling data sequence is divided according to a preset time window, a plurality of time window samples are obtained, and the time window samples are put into the sample set; and the power generation capacity corresponding to the last sampling time of each time window sample is used as a label of the time window sample.

3. The fan energy management method based on multi-dimension monitoring according to claim 1, characterized in that, In the construction of the neural network model, the neural network model comprises a composite attention module and a time sequence convolution module; the composite attention module comprises a trend segment preprocessing layer, a pooling layer and a composite perception attention layer. The trend segment preprocessing layer is used for dividing each input sample from a time sequence dimension into multiple trend segments and then transmitting the trend segments into the pooling layer; the pooling layer is used for respectively performing maximum pooling and average pooling on the trend segments, adding the maximum pooled features and the average pooled features to obtain pooled features, and transmitting the pooled features into the composite perception attention layer; the composite perception attention layer is used for learning the pooled features to obtain a composite attention vector of the last layer; the time sequence convolution module comprises a time sequence convolution layer and a multilayer perception layer; the time sequence convolution layer is used for receiving an output result of the composite attention module, performing time sequence information transmission by using one-dimensional convolution with different expansion values of each layer, and transmitting the time sequence information into the multilayer perception layer; and the multilayer perception layer learns to obtain a prediction result.

4. The fan energy management method based on multi-dimension monitoring according to claim 1, characterized in that, The binary quantization of the neural network model parameters to obtain the quantization model is binary quantization of weights and biases of the neural network model, and the binary weights and biases are obtained as initial parameters of the quantization model. The training of the quantization model based on the full-precision model comprises: obtaining a mean value of each layer weight in the full-precision model as a balancing factor of each layer; In the forward propagation, the last layer of the composite attention module in the quantization model adopts a binary Sigmoid activation function to calculate a composite attention vector, and the binary weights of other layers are multiplied by an activation vector of a previous layer and added to a bias, and then a Relu activation function is adopted to obtain an initial activation vector of each layer, and the initial activation vector of each layer is multiplied by the balancing factor of the corresponding layer after binary quantization to obtain a final activation vector of each layer for a next layer; In the backward propagation, the gradient of the binary weight is calculated according to a loss function, and the gradient is clipped as a gradient value of a floating-point gradient for weight updating.

5. The fan energy management method based on multi-dimension monitoring according to claim 1, characterized in that, The load balancing value of each node fan is calculated by using the multi-dimensional attributes of each node fan in the node fan set, which comprises: calculating the state deviation of each attribute of each node fan by using the multi-dimensional attributes of each node fan in the node fan set; and calculating the load balancing value of each node fan according to the state deviation and the multi-dimensional attributes of each node fan.

6. A fan energy management system based on multi-dimension monitoring, applying the fan energy management method based on multi-dimension monitoring of any one of claims 1-5, characterized in that, It comprises: a sample set construction module that collects various types of historical monitoring data of a wind farm, filters out multi-dimensional time sequence data, and constructs a sample set according to the multi-dimensional time sequence data and power generation at each sampling time; a power generation prediction module that constructs a neural network model, trains the neural network model by using the sample set to obtain a full-precision model, performs binary quantization on parameters of the neural network model to obtain a quantization model, trains the quantization model based on the full-precision model, adds an inter-model error in a loss function of the quantization model, and obtains a power generation prediction model; and collects real-time to-be-predicted data, transmits the data into the power generation prediction model, and obtains an overall predicted power generation; a constraint acquisition module that acquires a wind farm operation area and a constraint wind speed, wherein the wind farm operation area contains at least one node fan, and each node fan has multi-dimensional attributes for representing a power generation state of a corresponding unit area in the current wind farm operation area. The load balancing value calculation module screens the node wind turbines in the wind farm operation area by taking the constraint wind speed as a screening condition, obtains a node wind turbine set corresponding to the wind farm operation area, and calculates a load balancing value of each node wind turbine by using multi-dimensional attributes of each node wind turbine in the node wind turbine set; The load adjustment module calculates a load distribution coefficient value of each node wind turbine based on the load balancing value of each node wind turbine, and adjusts the load of the node wind turbine according to the load distribution coefficient value and the overall predicted power generation.

Images