An EMD-LSTM-based wind speed prediction method fusing spatial features

Abstract

The application relates to an EMD-LSTM-based wind speed prediction method fusing spatial features. The method obtains time series and spatial data from a monitoring site, pre-processes and constructs a data set, establishes a time-space prediction model to predict wind speed, fuses two prediction results by adopting weight Bayesian optimization, and obtains a prediction result. In the time model, EMD decomposition is performed on the wind speed rate feature, multiple LSTMs are input, the prediction components of each LSTM are accumulated to obtain a wind speed prediction value considering only the time series feature. In the space model, the wind speed of the adjacent station is vector decomposed and interpolated, the wind speed vector is restored to obtain the interpolation of the wind direction and the wind speed rate, and the distance from the actual value is calculated to obtain the spatial independence weight. In the model fusion, the weight Bayesian optimization is performed according to the spatial independence weight to obtain the wind speed value prediction fusing the time-space features. The application simultaneously considers the influence of the related time features and the spatial features in the wind speed prediction task, and the model structure and the prediction method are more reasonable.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of weather prediction, and particularly relates to a wind speed prediction method based on EMD-LSTM and fusing spatial features. BACKGROUND

[0002] The transmission and diffusion of urban atmospheric pollutants are significantly affected by wind direction and wind speed. The wind direction determines the direction of the transmission and diffusion of pollutants, and the wind speed directly affects the speed of the transmission and diffusion. Therefore, if an accurate wind speed prediction model is constructed based on wind speed and related environmental data sequences, the diffusion law of atmospheric pollutants in the city can be better understood, the distribution and concentration of pollutants can be evaluated, and important reference bases can be provided for environmental supervision departments to cope with urban pollution problems.

[0003] In the wind speed prediction task, the classical method usually includes a basic model based on statistics and a composite algorithm fusing multiple basic models. The basic model includes a moving average model, an autoregressive model, etc., and the composite algorithm includes an autoregressive moving average model (ARMA), an autoregressive integrated moving average model (ARIMA), etc. Although these classical models perform well in simple tasks, they are usually only suitable for univariate time series prediction and are difficult to capture nonlinear relationships in data. With the rapid development of deep learning, deep recurrent neural networks represented by GRU and LSTM are concerned in the time series prediction field due to their strong versatility and high prediction accuracy. However, in the wind speed prediction task, the relevant spatial features also have an important influence on the prediction results. Therefore, a single recurrent neural network model usually performs poorly in the wind speed prediction task. SUMMARY

[0004] In order to better support the air pollution supervision work of the ecological environment department and quickly predict and evaluate before and after pollution, the application provides a wind speed prediction method based on EMD-LSTM and fusing spatial features, which can more accurately predict urban wind speed, thereby providing scientific protection for the air quality safety of the people.

[0005] The technical solution adopted by the application to achieve the above purpose is as follows:

[0006] A wind speed prediction method based on EMD-LSTM and fusing spatial features, comprising the following steps:

[0007] Collecting time series data and spatial data in the target monitoring area, marking the actual observation value D of the wind direction at the current time l , the actual observation value V of the wind speed t ; data filling is performed on missing values and abnormal values, and a time series data set and a spatial direction component are constructed;

[0008] The EMD decomposition is performed on the wind speed time series in the time series data to extract wind speed components of different time scales; all time series features are combined as input, an LSTM time series model is iteratively trained to output prediction components, and after inverse normalization processing, the wind speed prediction value in the time series space is obtained by accumulation; the trained LSTM time series model is used to predict the wind direction time series prediction value D t and the wind speed rate time series prediction value Vl;

[0009] The interpolation is calculated for the spatial direction component, and the wind speed rate is vector restored to obtain the wind direction interpolation prediction value D p and the wind speed rate interpolation prediction value V p ; the spatial independence weight W d_a and W v_a is calculated;

[0010] The spatial independence weight is optimized by spatio-temporal weight Bayesian optimization; the optimized spatial independence weight W a_z , W d_z is obtained, and a fusion prediction model is established to predict the time and space wind speed vector;

[0011] The final wind direction prediction value D z and the final wind speed rate prediction value V z are output.

[0012] The time series data includes monitoring time, wind speed, wind direction, air temperature, air pressure, and humidity; the spatial data includes site number, latitude and longitude.

[0013] The construction of the sequence data set and the spatial component includes:

[0014] The time series data set: according to the monitoring time, the corresponding seasonal features are generated, one-hot code processing is performed, and normalization operation is performed;

[0015] The spatial component: the k nearest neighboring stations with the closest Euclidean distance to the target monitoring station are selected, and the wind speed vector data of all neighboring stations at this moment is further decomposed into east-west and north-south wind speed rate components.

[0016] The EMD decomposition of different time scale wind speed sequence components includes short-term fluctuation characteristics, periodic characteristics, or local characteristics further divided in fine granularity.

[0017] The loss function of the iterative training of the LSTM model is defined as:

[0018] The mean square error loss function is used to calculate the wind speed loss, the cycle loss function is used to process the periodicity of the angle to calculate the wind direction loss, and the total loss is the weighted sum of the wind speed loss and the wind direction loss.

[0019] The interpolation of the spatial direction component is calculated by IDW inverse distance weighting.

[0020] Wind direction spatial independence weight W a Wind speed spatial independence weight W v_a W d_a = |D t -D p |, W v_a = |V t -V p |.

[0021] The two spatial independence weights W d_ a and W v_a are respectively brought into the W a of the mapping function for spatio-temporal weight Bayesian optimization, to obtain the optimized spatial independence weights W d_z , W d_z ;

[0022]

[0023] Among them, W a represents the independence weight term, which is used to adjust the weight of the time series prediction value and the interpolation prediction value, to reflect the degree of influence of the target station by the surrounding monitoring points; A m represents the lower limit of the difference value range, A n represents the upper limit of the difference value range; W m represents the upper limit of the weight distribution proportion, W n represents the upper limit of the weight distribution proportion;

[0024] In the above formula, if W z <1, W z is not adjusted; otherwise, if W z >1, it means that the spatial characteristics of the surrounding monitoring points are not considered, then set W z =1.

[0025] The fusion prediction model is:

[0026] D z = (1-W d_z )*D p +W d_z *D l

[0027] V z = (1-W v_z )*V p +W v_z *V l

[0028] Among them, D zV z is the final wind speed prediction value.

[0029] The present application has the following advantages and benefits:

[0030] 1. The algorithm is highly automated. For a certain specific monitoring site, the present application only needs one-time LSTM training and spatio-temporal weight search, and after the end, it can automatically perform wind speed prediction without human intervention, which can save a lot of labor costs.

[0031] 2. The algorithm is highly versatile. The idea of the present application has good versatility, and no additional adaptation is needed for different time and regional scenarios, which can be easily extended to wind speed prediction tasks in various scenarios.

[0032] 3. The algorithm is highly accurate. The present application performs feature decomposition and extraction on multiple scales of meteorology, uses the LSTM model as the basic model, and simultaneously weights and fuses the spatial features between monitoring sites, which is a more reasonable new wind speed prediction method. This method has higher accuracy than previous prediction methods that only consider time sequence features. BRIEF DESCRIPTION OF DRAWINGS

[0033] Figure 1 is a structure diagram of the wind speed prediction method based on EMD-LSTM of the present application; DETAILED DESCRIPTION

[0034] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation method of the present application will be described in detail below in combination with the drawings. In the following description, a large number of specific details are set forth in order to facilitate a full understanding of the present application. However, the present application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without departing from the spirit of the present application, so the present application is not limited to the specific implementations disclosed below.

[0035] Step 1: Data acquisition. The environmental monitoring station is equipped with various environmental data monitoring equipment (such as an anemometer, a thermometer, a barometer, a hygrometer, etc.), and the data collected by these stations is stored in a central database. By retrieving from this database, the time series data recorded by the target environmental monitoring station is obtained: including monitoring time, wind speed, wind direction, air temperature, air pressure, humidity, and the actual observation value of the current moment wind direction is marked as D t , and the actual observation value of the wind speed is V t ; at the same time, the spatial data of all environmental monitoring stations is obtained, including station number, latitude and longitude.

[0036] Step 2: Data cleaning. The time series data and spatial data of the monitoring site are preprocessed: mainly including missing value and abnormal value processing.

[0037] Step 3: Perform feature conversion on the time series data of the monitoring sites, including generating corresponding seasonal features according to the monitoring time and performing one-hot processing on the seasonal features; normalizing the data; and dividing the data into training and test sets.

[0038] Step 4: Filter the neighboring subsets of the target monitoring station, select the k neighboring stations with the closest Euclidean distance to the target monitoring station, and decompose the wind speed vector data of all neighboring stations at that moment into the wind speed in the east-west direction and the wind speed in the north-south direction.

[0039] Step 5: Based on the time series data obtained in step 3, decompose the wind speed time series to extract the short-term fluctuation characteristics, periodic characteristics or other local characteristics of the wind speed series.

[0040] Step 6: LSTM time series model training and testing. The decomposed feature sequences and the remaining features (wind direction, temperature, air pressure, and humidity) are combined and fed into the LSTM as input features for training. The extracted features are fully learned, and the predicted values ​​are denormalized. The predicted value components of each model are then accumulated to obtain the prediction results, which are then tested and verified.

[0041] Step 7: LSTM time series model prediction. Use the trained LSTM model to predict and obtain the wind direction time series prediction value D at the next moment in the future. l and wind speed time series prediction value V l .

[0042] Step 8: Using the wind speed data obtained in step 4, interpolate the k nearest neighboring stations in the east-west and north-south directions using IDW inverse distance weighted interpolation, and perform vector reduction on the wind speeds in the two directions to obtain the wind direction interpolation prediction value D at the current moment. p and wind speed interpolation prediction value V p .

[0043] Step 9: Extract the independence weights of monitoring sites. Evaluate the predicted values ​​obtained in step 8 to obtain the spatial independence weights of wind direction and wind speed.

[0044] Step 10: Based on the interpolation prediction value obtained in step 8 and the spatial independence weight obtained in step 10, combined with the time series prediction value obtained in step 7, perform spatiotemporal weight Bayesian optimization to obtain a fusion prediction model.

[0045] Step 11: Obtain the prediction output based on the fusion prediction model optimized in step 10, and obtain the final wind direction prediction value D of the fusion spatial features. z and the final wind speed prediction value Vz .

[0046] The data cleaning specifically includes: using a k-nearest neighbor algorithm for outlier detection; for short-time missing data, forward fill, backward fill and mean fill are used simultaneously to ensure filling coverage, and smoothing processing is performed.

[0047] The season feature one-hot processing specifically includes: converting season information into a numerical form available for the model; and using one-hot encoding to represent each season as a vector, the length of the vector being equal to the number of seasons, and having a 1 at the corresponding season position and 0 at other positions.

[0048] The normalization operation is to map the data to [0, 1] using min-max standardization.

[0049]

[0050] In the above formula, f old is the original feature value, f min is the minimum value of the feature in the sample set, and f max is the maximum value of the feature in the sample set.

[0051] The selection of the nearest k neighboring stations: a suitable k value should be selected according to the size of the number of stations and the density of the stations, and the embodiment selects the k value as 12 through a control experiment.

[0052] The decomposition of the wind speed time series specifically includes: using EMD to decompose wind speed data with non-stationarity and non-linearity into multiple relatively stationary intrinsic mode functions (IMFs) and residual quantities (RES), each IMF representing fluctuations on different time scales. Through the EMD method, the non-stationarity of the wind speed sequence can be removed, the short-term fluctuation characteristics, periodic characteristics or other local characteristics of the wind speed sequence can be extracted, thereby providing a more fine-grained prediction result.

[0053] The training process of the LSTM time series prediction is as follows:

[0054] 1) Input the time step to obtain multiple prediction value components. The prediction value components of each LSTM model are accumulated to obtain a wind speed time series prediction result V l , and the wind direction D l is also predicted.

[0055] 2) The embodiment can achieve the best prediction effect by selecting the time step as 20, the hidden layer size as 256, and using the AdamW method for model optimization through continuous improvement of the model parameters.

[0056] 3) Definition of loss function: the wind speed loss is calculated using the mean square error loss function, and the wind direction loss is calculated using the cyclic loss function to handle the periodicity of the angle, and the total loss is the weighted sum of the wind speed loss and the wind direction loss, where the wind speed loss weight is 1.0 and the wind direction loss weight is 0.5. The training is stopped when the total loss converges.

[0057] The decomposition and restoration of the wind speed vector are specifically operated as follows:

[0058] 1) In the decomposition stage: according to the angle value of the wind direction, the wind speed vector is decomposed into east-west direction component (marked as U) and north-south direction component (marked as V).

[0059] 2) In the restoration stage: the arctangent function is used to calculate the final wind direction angle value using the interpolated wind speed rate in the two directions, and the complete wind speed information is calculated.

[0060] The extraction of the independence weight of the monitoring site is specifically operated as follows: the absolute value of the difference between the wind direction interpolation prediction value D p , the wind speed rate interpolation prediction value Vp and the corresponding wind direction actual observation value D t , and the wind speed rate actual observation value V t is taken as an additional weight item W d_a and W v_a , which can be expressed as:

[0061] W d_a = |D t -D p |

[0062] W v_a = |V t -V p |

[0063] According to the characteristics that the independence weight items W d_a and W v_a are negatively correlated with the degree of influence of the target site by the wind direction / speed rate of the surrounding monitoring points, mapping functions are respectively set, and the unified form of the formula is as follows:

[0064]

[0065] Wherein, W a represents the independence weight item, that is, the difference, which is used to adjust the weight of the time series prediction value and the interpolation prediction value to reflect the degree of influence of the target site by the surrounding monitoring points. A m represents the lower limit of the difference range, A n represents the upper limit of the difference range, and the upper and lower limits of the weight distribution ratio refer to two boundary values that determine the range of the weight distribution ratio. W mThe representative weight distribution proportion upper limit is W n The upper and lower limits of the difference range are set to limit the value range of the difference and ensure that the weight distribution proportion is adjusted within the set range. In the above formula: if W z > 1 is calculated, the representative does not consider the spatial characteristics of the surrounding monitoring points, and W z = 1 is set.

[0066] Through Bayesian optimization and comparative verification, a reasonable mapping range is set to determine the weight distribution proportion of the timing prediction value and the interpolation prediction value, so as to more accurately represent the spatial correlation between the target station and the adjacent stations, obtain a fusion prediction model, and the model weight distribution is as follows:

[0067] D z = (1-W d_z )*D p +W d_z *D l

[0068] V z = (1-W v _ z )*V p +W v_z *V l

[0069] The specific operation of obtaining the prediction output is: according to the fusion prediction model obtained by step 10 optimization, input the wind direction timing prediction value and the wind direction interpolation prediction value into the model to obtain the final wind direction prediction value; input the wind speed rate timing prediction value and the wind speed rate interpolation prediction value into the model to obtain the final wind speed rate prediction value.

[0070] The above is only a preferred embodiment of the present application, and does not limit the present application. Any simple modification, change and equivalent structure change according to the technical essence of the present application to the above embodiment are still within the protection scope of the technical solution of the present application.

Claims

1. An EMD-LSTM-based wind speed prediction method fusing spatial features, characterized in that, The method comprises the following steps: Collecting time series data and spatial data in the target monitoring area, marking the actual observation value D of the wind direction at the current time t , the actual observation value V of the wind speed t ; data filling is performed on missing values and abnormal values to construct a time series data set and a spatial direction component; EMD decomposition is performed on the wind speed time series in the time series data to extract wind speed components of different time scales; The LSTM time series model is iteratively trained with all time series features as input to output predicted components, and the wind speed prediction value in the time series space is obtained after inverse normalization processing and accumulation; predicting a wind direction time series prediction value D for the next time instant using the trained LSTM time series model l and a wind speed time series prediction value V l ; The interpolation is calculated for the spatial direction component, and the wind direction interpolation prediction value D at the current time is obtained by respectively performing vector reduction on the wind speed rate p and the wind speed rate interpolation prediction value V p ; the spatial independence weight is calculated: the wind direction spatial independence weight W d_a and the wind speed rate spatial independence weight W v_a ; spatial independence weight is subjected to spatiotemporal weight Bayesian optimization; an optimized spatial independence weight is obtained: an optimized wind direction spatial independence weight W d_z , an optimized wind speed spatial independence weight W v_z , and a fusion prediction model is established to predict the wind speed vector in time and space. outputting a final wind direction prediction value D z and a final wind speed prediction value V z .

2. The wind speed prediction method based on EMD-LSTM fusion spatial features according to claim 1, characterized in that, The time series data includes monitoring time, wind speed, wind direction, air temperature, air pressure, and humidity; the spatial data includes site number, latitude, and longitude. 3.The wind speed prediction method based on EMD-LSTM fusion spatial features according to claim 1, wherein, The construction of the time series data set and the spatial direction component comprises: The time series data set: generate corresponding seasonal features according to the monitoring time, perform one-hot code processing, and perform normalization operation; The spatial direction component: filter out k nearest neighboring stations with the closest Euclidean distance to the target monitoring station, and further decompose the wind speed vector data of all neighboring stations at the moment into east-west and north-south wind speed component.

4. The wind speed prediction method based on EMD-LSTM fusion spatial features according to claim 1, characterized in that, The wind speed sequence component features of different time scales in the EMD decomposition are short-term fluctuation features, periodic features, or local features further divided in fine granularity.

5. The wind speed prediction method based on EMD-LSTM fusion spatial features according to claim 1, characterized in that, The loss function of the iteratively trained LSTM time series model is defined as: The mean square error loss function is used to calculate the wind speed loss, the cyclic loss function is used to process the periodicity of the angle to calculate the wind direction loss, and the total loss is the weighted sum of the wind speed loss and the wind direction loss.

6. The wind speed prediction method based on EMD-LSTM fusion spatial features according to claim 1, characterized in that, The interpolation calculation of the spatial direction component is performed by IDW inverse distance weighting.

7. The EMD-LSTM-based wind speed prediction method fusing spatial features according to claim 1, characterized in that, Wind direction spatial independence weight W d_a Wind speed spatial independence weight W v_a W d_a = |D t -D p |, W v_a = |V t -V p | 8.The wind speed prediction method based on EMD-LSTM fusion spatial features according to claim 1, wherein, is to bring two space independence weights: wind direction space independence weight W d_a and wind speed space independence weight W v_a , into the W a mapping function respectively, and carry out space-time weight Bayesian optimization to obtain the optimized space independence weights: the optimized wind direction space independence weight W d_z , and the optimized wind speed space independence weight W v_z ; wherein, W a represents an independent weight term, used to adjust the weight of the time series prediction value and the interpolation prediction value, to reflect the degree of overall influence of the surrounding monitoring points on the target site; A m represents the lower limit of the difference range, A n represents the upper limit of the difference range; W m represents the upper limit of the weight distribution proportion, W n represents the upper limit of the weight distribution proportion; In the above equation, if W z <1, no adjustment is made to W z ; otherwise, if W z >1, it means that the spatial characteristics of the surrounding monitoring points are not considered, and W z =1 is set. 9.The wind speed prediction method based on EMD-LSTM fusion spatial features according to claim 1, wherein, The fusion prediction model is D z = (1 - W d_z ) * D p + W d_z * D l V z = (1 - W v_z )*V p + W v_z *V l where D z is the final wind direction prediction value, V z is the final wind speed prediction value, D l is the wind direction time series prediction value for the next time instant, V l is the wind speed time series prediction value for the next time instant.

Images