Training method and device of information determination model, and environment information determination method and device

Abstract

This disclosure provides a training method for an information determination model, relating to the field of artificial intelligence, and particularly to deep learning technology. The specific implementation scheme is as follows: the information determination model includes a first sub-model and a second sub-model. The method includes: inputting at least one sequence feature information corresponding to a first preset time period into the first sub-model to obtain first predicted information; obtaining second predicted information based on the at least one sequence feature information and the second sub-model; obtaining an information sequence corresponding to a target time period based on the first and second predicted information; and training the information determination model based on the label sequence and information sequence corresponding to the target time period. This disclosure also provides an environmental information determination method, apparatus, electronic device, and storage medium.

Description

Technical Field

[0001] This disclosure relates to the field of artificial intelligence technology, and more particularly to deep learning technology. More specifically, this disclosure provides a method for training an information determination model, a method for determining environmental information, an apparatus, an electronic device, and a storage medium. Background Technology

[0002] Numerical simulation or time series analysis techniques can be used to determine short-term or ultra-short-term information after a specified time. Numerical simulation can determine information for a small region based on information from a large region. Time series analysis methods can fit relationships between time series data and then determine the information based on the fitting results. Summary of the Invention

[0003] This disclosure provides a training method for an information determination model, a method for determining environmental information, an apparatus, a device, and a storage medium.

[0004] According to the first aspect, a training method for an information determination model is provided, the information determination model including a first sub-model and a second sub-model, the method comprising: inputting at least one sequence feature information corresponding to a first preset time period into the first sub-model to obtain first prediction information; obtaining second prediction information based on the at least one sequence feature information and the second sub-model; obtaining an information sequence corresponding to a target time period based on the first prediction information and the second prediction information; and training the information determination model based on a label sequence corresponding to the target time period and the information sequence.

[0005] According to the second aspect, a method for determining environmental information is provided. This method includes: inputting at least one sequence feature information corresponding to a third preset time period into an information determination model to obtain environmental information corresponding to a target time period. The aforementioned information determination model is trained according to the method provided in this disclosure.

[0006] According to a third aspect, a training apparatus for an information determination model is provided, the information determination model including a first sub-model and a second sub-model. The apparatus includes: a first obtaining module, configured to input at least one sequence feature information corresponding to a first preset time period into the first sub-model to obtain first prediction information; a second obtaining module, configured to obtain second prediction information based on the at least one sequence feature information and the second sub-model; a third obtaining module, configured to obtain an information sequence corresponding to a target time period based on the first prediction information and the second prediction information; and a training module, configured to train the information determination model based on a label sequence corresponding to the target time period and the information sequence.

[0007] According to a fourth aspect, an environmental information determination apparatus is provided, the apparatus comprising: a determination module, configured to input at least one sequence feature information corresponding to a third preset time period into an information determination model to determine environmental information corresponding to a target time period; wherein the information determination model is trained based on the apparatus provided in this disclosure.

[0008] According to a fifth aspect, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform a method provided according to the present disclosure.

[0009] According to a sixth aspect, a non-transitory computer-readable storage medium is provided storing computer instructions for causing a computer to perform the methods provided in this disclosure.

[0010] According to a seventh aspect, a computer program product is provided, comprising a computer program that, when executed by a processor, implements the method provided in this disclosure.

[0011] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0012] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein:

[0013] Figure 1 This is a flowchart of a training method for an information determination model according to an embodiment of the present disclosure;

[0014] Figure 2 This is a flowchart of a method for determining a model training method based on information from another embodiment of this disclosure;

[0015] Figure 3 This is a flowchart of a method for determining a model training method based on information from another embodiment of this disclosure;

[0016] Figure 4 The schematic diagram of the model is determined based on information from one embodiment of this disclosure;

[0017] Figure 5 The schematic diagram of the model is determined based on information from another embodiment of this disclosure;

[0018] Figure 6 This is a flowchart of an environmental information determination method according to an embodiment of the present disclosure;

[0019] Figure 7 This is a block diagram of a model training apparatus determined based on information from one embodiment of the present disclosure;

[0020] Figure 8 This is a block diagram of an environmental information determination apparatus according to an embodiment of the present disclosure; and

[0021] Figure 9 This is a block diagram of an electronic device according to an embodiment of the present disclosure that can apply information to determine a model training method and / or an environmental information determination method. Detailed Implementation

[0022] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0023] For example, wind speeds in the very short term can be determined using numerical simulation or time series analysis techniques. These very short-term wind speeds can be wind speeds occurring within minutes or even less of the current moment. In the field of wind power generation, the power generation capacity in the very short term can be determined based on these wind speeds, thus providing a reference for grid dispatch.

[0024] In regions with frequent weather changes, wind turbines are susceptible to meteorological fluctuations. During sudden wind changes, the wind turbine control system exhibits a degree of lag, which can lead to excessive load on the turbine. This can cause the turbine to tip over, resulting in significant economic losses.

[0025] A method for determining wind speed based on numerical simulation technology determines wind speed information for a small area based on a numerical simulation model and wind speed information for a large area. The large area can be a global region, while the small area can be the region where the wind turbine is located. This method requires significant equipment and time costs. Furthermore, the numerical simulation model includes numerous uncertainties, making it difficult to accurately determine wind speeds in the very short term.

[0026] An information determination method based on time series technology divides historical wind speed information into multiple wind speed information sequences. Then, based on the relationships between these sequences, a fitting process is performed to determine the wind speed in the very short term. However, the historical wind speed information used in this method is mostly based on a longer time scale, resulting in lower accuracy and making it difficult to determine the wind speed within the next few minutes.

[0027] Figure 1 This is a flowchart of a method for determining a model training method based on information from one embodiment of this disclosure.

[0028] like Figure 1 As shown, the method 100 may include operations S110 to S140. The information determination model includes a first sub-model and a second sub-model.

[0029] In operation S110, at least one sequence feature information corresponding to the first preset time period is input into the first sub-model to obtain the first prediction information.

[0030] For example, the first preset time period can be one or more hours.

[0031] For example, the first preset time period can be the time period between time T and time T-1. In one example, time T can be 12 o'clock on a certain day, time T-1 can be 11 o'clock on the same day, and the first preset time period can be one hour between 11 o'clock and 12 o'clock on that day.

[0032] For example, sequence feature information can be obtained from historical data sequences. In one example, a data sequence corresponding to a first preset time period can be obtained based on historical data within that time period. Then, sequence feature information corresponding to the first preset time period can be obtained based on that data sequence.

[0033] In operation S120, second prediction information is obtained based on at least one sequence feature information and the second sub-model.

[0034] For example, at least one sequence feature can be input into the second sub-model to obtain the second prediction information.

[0035] In operation S130, an information sequence corresponding to the target time period is obtained based on the first prediction information and the second prediction information.

[0036] For example, the first and second prediction information can be added together to obtain an information sequence corresponding to the target time period.

[0037] For example, the target time period could be the period following time T as described above. In one example, time T could be 12:00 noon on a certain day, and the target time period could be 10 minutes after 12:00 noon. In one example, the first prediction information could be (V1_1, ..., V1_20). The second prediction information could be (V2_1, ..., V2_20). Adding the two together yields the information sequence (V_1, ..., V_20). For example, V_1 = V1_1 + V2_1, V_20 = V1_20 + V2_20.

[0038] In operation S140, the information determination model is trained based on the label sequence and information sequence corresponding to the target time period.

[0039] For example, various loss functions can be used to calculate loss values ​​based on the label sequence and the information sequence, in order to train the model. In one example, the loss function could be the MSE (Mean Squared Error) function.

[0040] Through the embodiments of this disclosure, information in the ultra-short term can be accurately determined by utilizing the linear and nonlinear components of sequence feature information. Furthermore, information in the ultra-short term can be further accurately determined by utilizing the relationship between at least one sequence feature information.

[0041] For example, wind speeds in the very short term can be accurately determined.

[0042] Figure 2 This is a flowchart of a method for determining a model training method based on information from another embodiment of this disclosure.

[0043] like Figure 2 As shown, method 220 can obtain second prediction information based on at least one sequence feature information and a second sub-model. The following will provide a detailed explanation in conjunction with operations S221 to S222.

[0044] The first preset time period includes sub-time periods.

[0045] In operation S221, at least one sequence sub-feature information corresponding to the sub-time period is obtained based on at least one sequence feature information.

[0046] In this embodiment of the disclosure, the length of the sub-time period is less than or equal to the length of the first preset time period.

[0047] For example, the length of the first preset time period is 1 hour. The length of the sub-time period can be 10 minutes.

[0048] In this embodiment of the disclosure, the sub-time period corresponds to multiple consecutive time points within the first preset time period.

[0049] For example, the first preset time period includes 120 time points. The interval between each time point is 30 seconds. A sub-time period can consist of multiple consecutive time points within the first preset time period. In one example, a sub-time period can consist of the last 20 time points within the first preset time period.

[0050] For example, sequence feature information can be represented by a vector. In one example, this vector can include data in 120 dimensions. Each dimension corresponds to a point in time. If a sub-period consists of the last 20 time points of a first preset period, the sequence sub-feature information is composed of the data in the dimensions corresponding to these last 20 time points.

[0051] In operation S222, at least one sequence sub-feature information is input into the second sub-model to obtain the second prediction information.

[0052] For example, the sequence sub-feature information, which consists of data from the dimensions corresponding to the last 20 time points in the sequence feature information, can be input into the second sub-model to obtain the second prediction information.

[0053] By utilizing sequence sub-feature information corresponding to sub-time periods through the embodiments of this disclosure, the computational resources required for model training can be reduced. However, the sequence feature information may include some inaccurate or erroneous information; training with all the information in the sequence feature information would reduce the accuracy of the information determined by the trained model. Therefore, utilizing sequence sub-feature information can further improve model performance, enabling more accurate determination of information in the ultra-short term.

[0054] In some embodiments, sub-time periods can be adjusted based on the label sequence and information sequence to train the information determination model. For example, the information determination model can be trained over multiple periods based on at least one sequence feature information corresponding to a first preset time period. In one example, in the current training period of multi-period training, the sub-time period consists of the last 20 time points of the first preset time period described above. After the current training period ends, the sub-time period is adjusted. In the next training period, the sub-time period consists of the 17th to 36th time points of the first preset time period. Training can be stopped after reaching a preset number of training periods or after the change in the loss value described above is less than a preset change threshold. For another example, the length of the sub-time period can be adjusted.

[0055] In some embodiments, the label sequence includes J data points, and the information sequence includes J data points. J is an integer greater than or equal to 2. Training the information determination model based on the label sequence and information sequence corresponding to the target time period includes: obtaining a first difference based on the first I data points of the information sequence and the first I data points of the label sequence, where I is less than or equal to J; obtaining a second difference based on the last JI data points of the information sequence and the last JI data points of the label sequence; and calculating a loss value based on the first and second differences to train the information determination model.

[0056] For example, the loss value can be calculated based on the label sequence and the information sequence using the following formula to determine the model using training information:

[0057]

[0058] For example, Loss is the loss value, i is a positive integer less than or equal to I, j is a positive integer less than or equal to J, and V i It is one of the first I data items in the information sequence. V is one of the first I data points in the label sequence. j It is one of the last JI data in the information sequence. ω is one of the last JI data in the label sequence. a As the first weight, ω b It is the second weight. For the first difference, This is the second difference.

[0059] In one example, the information sequence could be, for instance, (V_1, ..., V_20) as described above, V i The value of can be V_1, ..., V_10, V j The values ​​can be V_11, ..., V_20. The label sequence can be (V^_1, ..., V^_20). The value of can be V^_1, ..., V^_10. The value of can be V^_11, ..., V^_20.

[0060] In one example, ω a =0.6, ω b =0.4.

[0061] In one example, I = 10, J = 20.

[0062] In some embodiments, the parameters of the first sub-model and / or the second sub-model can be adjusted based on the loss value described above to determine the model using training information.

[0063] Figure 3 This is a flowchart of a method for determining a model training method based on information from another embodiment of this disclosure.

[0064] like Figure 3 As shown, method 300 can be executed before method 100. At least one sequence feature information corresponds to at least one object. Method 300 can perform operations S301 to S303 for each object.

[0065] In operation S301, the data sequence corresponding to each object is obtained.

[0066] For example, an object can be a device. In one example, an object can be a wind turbine. In another example, at least one object can be at least one wind turbine located within the same geographical area.

[0067] For example, a data sequence can be obtained from historical data corresponding to each object.

[0068] In one example, each wind turbine assembly is equipped with a detection module. This module can detect data such as wind speed, air pressure, temperature, and wind direction at preset intervals. The preset interval could be, for example, 30 seconds.

[0069] In one example, the data sequence could be a wind speed data sequence, an air pressure data sequence, and a temperature data sequence, etc.

[0070] For example, each data point in a data sequence corresponds to a specific point in time.

[0071] In operation S302, the data sequence is divided into M sub-data sequences.

[0072] In this embodiment of the disclosure, the M sub-data sequences correspond one-to-one with the M time periods. M is an integer greater than or equal to 1.

[0073] For example, the length of each time period can be the same as the first preset time period. In one example, each time period includes 120 time points.

[0074] In this embodiment of the disclosure, the data sequence can be segmented to obtain N sub-data sequences.

[0075] For example, N sub-data sequences are obtained by directly splitting the data sequence. For example, N is an integer greater than or equal to 1, and N is an integer less than or equal to M.

[0076] In this embodiment of the disclosure, N sub-data sequences are oversampled to obtain K oversampled sub-data sequences.

[0077] For example, SMOTE (Synthetic minority class oversampling technique) can be used to oversample N sub-data sequences to obtain K oversampled sub-data sequences. Each oversampled sub-data sequence corresponds to a time period. The length of the oversampled sub-data sequence can be the same as the length of the sub-data sequence.

[0078] For example, an oversampled sub-data sequence can be an oversampled sub-wind speed data sequence, an oversampled sub-temperature data sequence, an oversampled sub-barometric pressure data sequence, etc. In one example, the oversampled sub-wind speed data sequence Seq_V, the oversampled sub-temperature data sequence Seq_Temp, and the oversampled sub-barometric pressure data sequence Seq_Pa correspond to the time period Time_Interval.

[0079] For example, K is an integer greater than or equal to 1.

[0080] In this embodiment of the disclosure, H oversampled sub-data sequences are obtained based on the preset threshold corresponding to each second preset time period and the second preset time period in which each oversampled sub-data sequence is located.

[0081] For example, there are multiple second preset time periods.

[0082] In one example, there can be 12 second preset time periods, each corresponding to a calendar month. In another example, a second preset time period can correspond to the first calendar month (January) of the Gregorian calendar.

[0083] For example, the average of multiple data points in an oversampled sub-data sequence can be compared to a preset threshold. If the average is greater than or equal to the preset threshold, the oversampled sub-data sequence can be retained. Otherwise, it can be deleted. Using a similar method, H oversampled sub-data sequences can be obtained from K oversampled sub-data sequences. For example, H is an integer less than or equal to K. H is an integer greater than or equal to 1.

[0084] In this embodiment of the disclosure, M sub-data sequences are obtained based on N sub-data sequences and H oversampled sub-data sequences.

[0085] For example, H oversampled sub-data sequences can be used as H sub-data sequences to obtain M sub-data sequences. In one example, M = H + N. Oversampling techniques increase the number of sequences and reduce the distance between them, thereby improving model performance.

[0086] In operation S303, based on M sub-data sequences, sequence feature information corresponding to the first preset time period is obtained.

[0087] In this embodiment of the disclosure, a first feature extraction is performed on M sub-data sequences to obtain M time-domain sub-feature information.

[0088] For example, for each sub-data sequence, one or more of the following can be calculated: median, mean, maximum, minimum, and quantiles of multiple data points in each sub-data sequence, to perform the first feature extraction. In one example, the quantiles could be the quartiles or tetrameters of multiple data points in each sub-data sequence, etc.

[0089] In this embodiment of the disclosure, a second feature extraction is performed on M sub-data sequences to obtain M first frequency domain sub-feature information.

[0090] For example, a second feature can be extracted from each sub-data sequence based on Fourier transform and / or wavelet transform to obtain a first frequency domain sub-feature information.

[0091] In this embodiment of the disclosure, M second frequency domain sub-feature information are obtained based on M first frequency domain sub-feature information.

[0092] For example, a second frequency domain sub-feature information can be obtained based on each first frequency domain sub-feature information and the high-frequency and low-frequency information in the first frequency domain sub-feature information.

[0093] In this embodiment of the disclosure, M directional information are determined based on the position information of each object.

[0094] For example, location information could be the geographical region where each object is located. For instance, an object might be a wind turbine generator. Directional information could be wind direction. The prevailing wind direction for the geographical region where the wind turbine generator is located can be obtained in different natural months or seasons to determine the directional information for each sub-data sequence. In one example, the prevailing wind direction can be converted into a wind direction value according to a preset mapping relationship to obtain directional information. The wind direction value can be a decimal less than 1.

[0095] In this embodiment of the disclosure, sequence feature information corresponding to M time periods is obtained based on M sub-data sequences, M time-domain sub-feature information, M first frequency-domain sub-feature information, M second frequency-domain sub-feature information, and M direction information.

[0096] For example, each sub-data sequence, each time-domain sub-feature information, each first-frequency-domain sub-feature information, each second-frequency-domain sub-feature information, and each direction information can be concatenated to obtain sequence feature information corresponding to a time period.

[0097] In this embodiment of the disclosure, sequence feature information corresponding to the first preset time period is obtained based on sequence feature information corresponding to M time periods.

[0098] For example, in one round of training, the sequence feature information corresponding to one of the M time periods can be used as the sequence feature information corresponding to the first preset time period for multi-cycle training. In the next round of training, the sequence feature information corresponding to another of the M time periods can be used as the sequence feature information corresponding to the first preset time period for multi-cycle training again.

[0099] In some embodiments, there may be no overlap between the M sub-data sequences. For example, sub-data sequence Seq_1 contains data corresponding to time points 1 to 120. Sub-data sequence Seq_2 contains data corresponding to time points 121 to 240.

[0100] In some embodiments, unlike method 300, sequence feature information corresponding to the M time periods can be obtained based on the M sub-data sequences, M time-domain sub-feature information, M first frequency-domain sub-feature information, M second frequency-domain sub-feature information, M time period information, object identification information, and M direction information. Time period information and object identification information are added to each sequence feature information to further improve the performance of the trained model.

[0101] In some embodiments, unlike method 300, the data sequence can be segmented to obtain N' sub-data sequences. From these N' sub-data sequences, N sub-data sequences with a valid data ratio greater than or equal to a preset ratio threshold are selected. For example, during detection, due to malfunctions or other reasons, the detection module described above may miss detections, resulting in some missing data in the sub-data sequences. Too much missing data leads to fewer valid data, which in turn causes a decrease in the performance of the trained model. In one example, the preset ratio threshold is 90%.

[0102] In some embodiments, the first sub-model includes a convolutional network, a recurrent network, and an attention network. Inputting at least one sequence feature information corresponding to a first preset time period into the first sub-model to obtain first prediction information includes: inputting at least one sequence feature information into the convolutional network to obtain first output feature information; inputting the first output feature information into the recurrent network to obtain second output feature information; and inputting the second output feature information into the attention network to obtain the first prediction information.

[0103] For example, a convolutional network can be a CNN (Convolutional Neural Network) model. A recurrent network can be an LSTM (Long Short-Term Memory) model. An attention network can be a TPA-LSTM (Temporal Pattern Attention Long Short-Term Memory) model. Furthermore, a recurrent network can also be an RNN (Recurrent Neural Network) model. And, for example, a recurrent network can also be a GRU (Gate Recurrent Unit) model.

[0104] Figure 4 This is a schematic diagram of a model determined based on information from one embodiment of this disclosure.

[0105] like Figure 4 As shown, the information determination model 400 includes a first sub-model 410 and a second sub-model 420.

[0106] The first sub-model 410 may include one or more neural network models. The first sub-model 410 may process at least one sequence feature information of the input to output first prediction information.

[0107] Take sequence feature information F_1 as an example. Sequence feature information F_1 corresponds to a first preset time period TP_1. In one example, the first preset time period TP_1 can be one hour between 11:00 and 12:00 on a certain day.

[0108] The sequence feature information F_1 corresponds to a wind turbine generator. The sequence feature information F_1 can be obtained based on multiple wind speed data and the identification information of the wind turbine generator within the first preset time period TP_1. For example, various feature extraction methods can be used to extract features from multiple wind speed data within the first preset time period TP_1 to obtain various feature information related to multiple wind speed data, so as to obtain the sequence feature information F_1.

[0109] In one example, the first prediction information could be (V1_1, ..., V1_20) as described above.

[0110] The second sub-model 420 can be an AR model (AutoRegressive Model). The second sub-model 420 can process at least one sequence feature information of the input to output second prediction information. In one example, the second prediction information can be (V2_1, ..., V2_20) as described above.

[0111] Next, the first and second prediction information can be added together to obtain the information sequence. The information sequence can be (V_1, ..., V_20) as described above. For example, in the information sequence, V_1 can be the sum of V1_1 and V2_1.

[0112] Figure 5 This is a schematic diagram of a model determined based on information from one embodiment of this disclosure.

[0113] like Figure 5 As shown, the information determination module 500 may include a first sub-model 510 and a second sub-module 520. The difference between the information determination model 400 and the first sub-model 510 of the information determination model 500 is that the first sub-model 510 includes a convolutional network 511, a recurrent network 512, and an attention network 513.

[0114] The convolutional network 511 can be a CNN model, the recurrent network 512 can be an LSTM model, and the attention network 513 can be a TPA-LSTM model.

[0115] Since the input to the first sub-model 510 is at least one sequence feature information, and each sequence feature information corresponds to a wind turbine generator, the convolutional network 511 can find the correlation features between the generators and output the first output feature information. The first output feature information can be a fusion of the above-mentioned correlation features. Next, the recurrent network 512 can simulate the temporal variation pattern and output the second output feature information. The attention network 513 can output the first prediction information based on the self-attention mechanism.

[0116] In some embodiments, the attention network 513 may include a fully connected layer whose output is the first prediction information.

[0117] In some embodiments, the information determination model differs from information determination model 400 or information determination model 500 in that, in this embodiment, the information determination model is built upon LSTNet (Long and Short-term Time-series Net). LSTNet includes convolutional components, recurrent components, and autoregressive components. LSTNet may also include recurrent-skip components and / or temporal attention components. The recurrent-skip component can process information based on the periodicity of the input information. The input to the temporal attention component can be the output of the recurrent component. The input to the recurrent-skip component can also be the output of the recurrent component.

[0118] In this embodiment, the length of the first preset time period is relatively short, and the sequence feature information may not contain periodic information. Therefore, the information determination model can be an LSTNet with the recurrent-skip component removed, and the input of the attention component of the information determination model can be the output of the recurrent component.

[0119] It should be noted that, in the embodiments of this disclosure, wind speed is used as an example to describe in detail the training method of the information determination model of this disclosure. However, the information determination model of this disclosure can also be trained based on data such as water flow speed, vehicle flow speed, temperature, and air pressure, and this disclosure does not impose any limitations on it.

[0120] Figure 6 This is a flowchart of an environmental information determination method according to an embodiment of the present disclosure.

[0121] like Figure 6 As shown, the method 600 may include operation S610.

[0122] In operation S610, at least one sequence feature information corresponding to the third preset time period is input into the information determination model to determine the environmental information corresponding to the target time period.

[0123] For example, the information determination model is trained according to the methods provided in this disclosure.

[0124] For example, the third preset time period can be the time period between the current time T_Cur and the time T_Cur-1. In one example, the current time T_Cur could be 12:00 noon, and the time T_Cur-1 could be 11:00 noon. The third preset time period could be one hour between 11:00 noon and 12:00 noon.

[0125] For example, the target time period can be a period of time after the current time. In one example, the current time T_Cur could be 12:00 noon, and the target time period could be 10 minutes after 12:00 noon.

[0126] For example, environmental information can be one or more of the following: wind speed, water flow speed, vehicle speed, temperature, air pressure, etc.

[0127] In some embodiments, inputting at least one sequence feature information corresponding to a third preset time period into an information determination model to determine environmental information corresponding to a target time period includes: inputting at least one sequence feature information into a first sub-model to obtain first environmental information; obtaining second environmental information based on at least one sequence feature information and a second sub-model; and determining environmental information corresponding to the target time period based on the first environmental information and the second environmental information.

[0128] Figure 7 This is a block diagram of a training apparatus for determining a model based on information from one embodiment of the present disclosure.

[0129] like Figure 7 As shown, the device 700 may include a first acquisition module 710, a second acquisition module 720, a third acquisition module 730, and a training module 740.

[0130] The above information determines that the model includes a first sub-model and a second sub-model.

[0131] The first acquisition module 710 is used to input at least one sequence feature information corresponding to the first preset time period into the first sub-model to obtain the first prediction information.

[0132] The second obtaining module 720 is used to obtain second prediction information based on at least one sequence feature information and the second sub-model.

[0133] The third acquisition module 730 is used to obtain an information sequence corresponding to the target time period based on the first prediction information and the second prediction information mentioned above.

[0134] Training module 740 is used to train the above information determination model based on the label sequence corresponding to the target time period and the above information sequence.

[0135] In some embodiments, the first preset time period includes sub-time periods, and the second obtaining module includes: a first obtaining sub-module, configured to obtain at least one sequence sub-feature information corresponding to the sub-time period based on the at least one sequence feature information; and a first obtaining sub-module, configured to input the at least one sequence sub-feature information into a second sub-model to obtain second prediction information.

[0136] In some embodiments, the at least one sequence feature information corresponds one-to-one with at least one object, and the device further includes: an execution module, configured to perform related operations for each object through the following sub-modules: a first acquisition sub-module, configured to acquire a data sequence corresponding to each object; a segmentation module, configured to segment the data sequence to obtain M sub-data sequences, wherein the M sub-data sequences correspond one-to-one with M time periods, and M is an integer greater than or equal to 1; and a second acquisition sub-module, configured to obtain sequence feature information corresponding to the first preset time period based on the M sub-data sequences.

[0137] In some embodiments, the above-mentioned segmentation module includes: a segmentation unit for segmenting the data sequence to obtain N sub-data sequences; an oversampling unit for oversampling the N sub-data sequences to obtain K oversampled sub-data sequences; a first obtaining unit for obtaining H oversampled sub-data sequences based on a preset threshold corresponding to each second preset time period and the second preset time period in which each oversampled sub-data sequence is located, wherein there are multiple second preset time periods; and a second obtaining unit for obtaining M sub-data sequences based on the N sub-data sequences and the H oversampled sub-data sequences, wherein N is an integer less than or equal to M, H is an integer less than or equal to K, and K is an integer greater than or equal to 1.

[0138] In some embodiments, the second obtaining submodule includes: a first feature extraction unit, configured to perform first feature extraction on the M sub-data sequences to obtain M time-domain sub-feature information; a second feature extraction unit, configured to perform second feature extraction on the M sub-data sequences to obtain M first frequency-domain sub-feature information; a third obtaining unit, configured to obtain M second frequency-domain sub-feature information based on the M first frequency-domain sub-feature information; a determining unit, configured to determine M direction information based on the position information of each object; a fourth obtaining unit, configured to obtain sequence feature information corresponding to M time periods based on the M sub-data sequences, the M time-domain sub-feature information, the M first frequency-domain sub-feature information, the M second frequency-domain sub-feature information, and the M direction information; and a fifth obtaining unit, configured to obtain sequence feature information corresponding to the first preset time period based on the sequence feature information corresponding to the M time periods.

[0139] In some embodiments, the first sub-model includes a convolutional network, a recurrent network, and an attention network, and the first obtaining module includes: a third obtaining sub-module, used to input the at least one sequence feature information into the convolutional network to obtain first output feature information; a fourth obtaining sub-module, used to input the first output feature information into the recurrent network to obtain second output feature information; and a fifth obtaining sub-module, used to input the second output feature information into the attention network to obtain first prediction information.

[0140] In some embodiments, the training module includes: a first training module, configured to adjust the sub-time period according to the label sequence and the information sequence to train the information determination model.

[0141] In some embodiments, the label sequence includes J data points, the information sequence includes J data points, where J is an integer greater than or equal to 2, and the training module includes: a sixth obtaining submodule, configured to obtain a first difference based on the first I data points of the information sequence and the first I data points of the label sequence, wherein I is less than or equal to J; a seventh obtaining submodule, configured to obtain a second difference based on the last JI data points of the information sequence and the last JI data points of the label sequence; and a second training submodule, configured to calculate a loss value based on the first difference and the second difference to train the information determination model.

[0142] Figure 8 This is a block diagram of an environmental information determination apparatus according to another embodiment of the present disclosure.

[0143] like Figure 8 As shown, the device 800 may include a determining module 810.

[0144] The determination module 810 is used to input at least one sequence feature information corresponding to the third preset time period into the information determination model to determine the environmental information corresponding to the target time period.

[0145] For example, the above information determines that the model was trained using the apparatus provided in this disclosure.

[0146] In some embodiments, the determining module includes: a sixth obtaining submodule, configured to input the at least one sequence feature information into the first sub-model to obtain first environmental information; a seventh obtaining submodule, configured to obtain second environmental information based on the at least one sequence feature information and the second sub-model; and a determining submodule, configured to determine environmental information corresponding to the target time period based on the first environmental information and the second environmental information.

[0147] The collection, storage, use, processing, transmission, provision, and disclosure of user personal information involved in the technical solution disclosed herein comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0148] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0149] Figure 9 A schematic block diagram of an example electronic device 900 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0150] like Figure 9 As shown, device 900 includes a computing unit 901, which can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) 902 or a computer program loaded into random access memory (RAM) 903 from storage unit 908. RAM 903 may also store various programs and data required for the operation of device 900. The computing unit 901, ROM 902, and RAM 903 are interconnected via bus 904. Input / output (I / O) interface 905 is also connected to bus 904.

[0151] Multiple components in device 900 are connected to I / O interface 905, including: input unit 906, such as keyboard, mouse, etc.; output unit 907, such as various types of monitors, speakers, etc.; storage unit 908, such as disk, optical disk, etc.; and communication unit 909, such as network card, modem, wireless transceiver, etc. Communication unit 909 allows device 900 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0152] The computing unit 901 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 901 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 901 performs the various methods and processes described above, such as methods for training information determination models and / or methods for determining environmental information. For example, in some embodiments, the methods for training information determination models and / or methods for determining environmental information can be implemented as computer software programs tangibly contained in a machine-readable medium, such as storage unit 908. In some embodiments, part or all of the computer program can be loaded and / or installed on device 900 via ROM 902 and / or communication unit 909. When the computer program is loaded into RAM 903 and executed by the computing unit 901, one or more steps of the methods for training information determination models and / or methods for determining environmental information described above can be performed. Alternatively, in other embodiments, the computing unit 901 may be configured by any other suitable means (e.g., by means of firmware) to perform training methods for determining information about the model and / or methods for determining environmental information.

[0153] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0154] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0155] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0156] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0157] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with embodiments of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0158] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other.

[0159] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0160] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A training method for an information determination model, the information determination model comprising a first sub-model and a second sub-model, the first sub-model comprising a convolutional network, a recurrent network, and an attention network, the attention network being a temporal pattern attention-long short-term memory network, and the second sub-model being an autoregressive model, the method comprising: At least one sequence feature information corresponding to the first preset time period is input into the convolutional network to obtain the first output feature information; The first output feature information is input into the recurrent network to obtain the second output feature information; The second output feature information is input into the attention network to obtain the first prediction information. The at least one sequence feature information corresponds one-to-one with at least one wind turbine generator set. The first preset time period is one or more hours. Based on the at least one sequence feature information and the second sub-model, the second prediction information is obtained; Based on the first prediction information and the second prediction information, an information sequence corresponding to the target time period is obtained; as well as The information determination model is trained based on the label sequence corresponding to the target time period and the information sequence. The method further includes: For each wind turbine, perform the following operations: Obtain the wind speed data sequence corresponding to each wind turbine generator set; The data sequence is divided into N sub-data sequences; The N sub-data sequences are oversampled to obtain K oversampled sub-data sequences; Based on the preset threshold corresponding to each second preset time period and the second preset time period in which each oversampled sub-data sequence is located, H oversampled sub-data sequences are obtained, wherein there are multiple second preset time periods; the average value of multiple data in each of the H oversampled sub-data sequences is greater than or equal to the preset threshold corresponding to the second preset time period in which the oversampled sub-data sequence is located, and each second preset time period corresponds to a natural month; Based on the N sub-data sequences and the H oversampled sub-data sequences, M sub-data sequences are obtained, where N is an integer less than or equal to M, H is an integer less than or equal to K, and K is an integer greater than or equal to 1. Each of the M sub-data sequences corresponds one-to-one with one of the M time periods. The first feature is extracted from the M sub-data sequences to obtain M time-domain sub-feature information; The second feature is extracted from the M sub-data sequences to obtain M first frequency domain sub-feature information; Based on the M first frequency domain sub-feature information, M second frequency domain sub-feature information are obtained; Based on the location information of each wind turbine, determine M wind direction information, including: obtaining the prevailing wind direction of the geographical area where the wind turbine is located in different natural months, so as to determine the wind direction information of each sub-data sequence; Based on the M sub-data sequences, M time-domain sub-feature information, M first frequency-domain sub-feature information, M second frequency-domain sub-feature information, and M wind direction information, sequence feature information corresponding to the M time periods is obtained; Based on the sequence feature information corresponding to the M time periods, the sequence feature information corresponding to the first preset time period is obtained.

2. The method of claim 1, wherein, The first preset time period includes sub-time periods. The step of obtaining the second prediction information based on the at least one sequence feature information and the second sub-model includes: Based on the at least one sequence feature information, at least one sequence sub-feature information corresponding to the sub-time period is obtained; and The at least one sequence sub-feature information is input into the second sub-model to obtain the second prediction information.

3. The method of claim 2, wherein, Training the information determination model based on the label sequence corresponding to the target time period and the information sequence includes: The sub-time period is adjusted based on the label sequence and the information sequence to train the information determination model.

4. The method according to any one of claims 1 to 3, wherein, The tag sequence includes J data points, and the information sequence includes J data points, where J is an integer greater than or equal to 2. Training the information determination model based on the label sequence corresponding to the target time period and the information sequence includes: Based on the first I data points of the information sequence and the first I data points of the tag sequence, a first difference is obtained, where I is less than or equal to J; The second difference is obtained based on the last JI data points of the information sequence and the last JI data points of the tag sequence; and Based on the first difference and the second difference, a loss value is calculated to train the model using the information.

5. A method for determining environmental information, comprising: Input at least one sequence feature information corresponding to the third preset time period into the information determination model to determine the environmental information corresponding to the target time period; The information determination model is trained according to the method described in any one of claims 1 to 4.

6. The method of claim 5, wherein, The step of inputting at least one sequence feature information corresponding to the third preset time period into the information determination model to determine the environmental information corresponding to the target time period includes: The at least one sequence feature information is input into the first sub-model to obtain the first environmental information; Based on the at least one sequence feature information and the second sub-model, second environmental information is obtained; and Based on the first environmental information and the second environmental information, determine the environmental information corresponding to the target time period.

7. A training apparatus for an information determination model, the information determination model comprising a first sub-model and a second sub-model, the first sub-model comprising a convolutional network, a recurrent network, and an attention network, the attention network being a temporal pattern attention-long short-term memory network, and the second sub-model being an autoregressive model, the apparatus comprising: The third submodule is used to input at least one sequence feature information corresponding to the first preset time period into the convolutional network to obtain the first output feature information. The fourth submodule is used to input the first output feature information into the recurrent network to obtain the second output feature information; The fifth submodule is used to input the second output feature information into the attention network to obtain the first prediction information. The at least one sequence feature information corresponds one-to-one with at least one wind turbine generator set. The first preset time period is one or more hours. The second obtaining module is used to obtain second prediction information based on the at least one sequence feature information and the second sub-model; The third acquisition module is used to obtain an information sequence corresponding to the target time period based on the first prediction information and the second prediction information. as well as The training module is used to train the information determination model based on the label sequence corresponding to the target time period and the information sequence. The device further includes: The execution module is used to perform relevant operations for each wind turbine generator set through the following sub-modules: The first acquisition submodule is used to acquire the wind speed data sequence corresponding to each wind turbine generator set; A segmentation unit is used to segment the data sequence to obtain N sub-data sequences; An oversampling unit is used to oversample the N sub-data sequences to obtain K oversampled sub-data sequences; The first obtaining unit is used to obtain H oversampled sub-data sequences based on a preset threshold corresponding to each second preset time period and the second preset time period in which each oversampled sub-data sequence is located, wherein there are multiple second preset time periods; the average value of multiple data in each of the H oversampled sub-data sequences is greater than or equal to the preset threshold corresponding to the second preset time period in which the oversampled sub-data sequence is located, and each second preset time period corresponds to a natural month; The second obtaining unit is used to obtain M sub-data sequences based on the N sub-data sequences and the H oversampled sub-data sequences, where N is an integer less than or equal to M, H is an integer less than or equal to K, and K is an integer greater than or equal to 1. The M sub-data sequences correspond one-to-one with the M time periods. The first feature extraction unit is used to perform first feature extraction on the M sub-data sequences to obtain M time-domain sub-feature information; The second feature extraction unit is used to perform second feature extraction on the M sub-data sequences to obtain M first frequency domain sub-feature information; The third obtaining unit is used to obtain M second frequency domain sub-feature information based on the M first frequency domain sub-feature information; The determining unit is used to determine M wind direction information based on the location information of each wind turbine generator set, including: obtaining the prevailing wind direction of the geographical area where the wind turbine generator set is located in different natural months, so as to determine the wind direction information of each sub-data sequence; The fourth obtaining unit is used to obtain sequence feature information corresponding to the M time periods based on the M sub-data sequences, the M time-domain sub-feature information, the M first frequency-domain sub-feature information, the M second frequency-domain sub-feature information, and the M wind direction information; The fifth obtaining unit is used to obtain the sequence feature information corresponding to the first preset time period based on the sequence feature information corresponding to the M time periods.

8. The apparatus of claim 7, wherein, The first preset time period includes sub-time periods. The second obtaining module includes: The first acquisition submodule is configured to acquire at least one sequence sub-feature information corresponding to the sub-time period based on the at least one sequence feature information; and The first acquisition submodule is used to input the at least one sequence sub-feature information into the second sub-model to obtain the second prediction information.

9. The apparatus of claim 8, wherein, The training module includes: The first training module is used to adjust the sub-time period according to the label sequence and the information sequence in order to train the information determination model.

10. The apparatus of any one of claims 7 to 9, wherein, The tag sequence includes J data points, and the information sequence includes J data points, where J is an integer greater than or equal to 2. The training module includes: The sixth obtaining submodule is used to obtain a first difference based on the first I data of the information sequence and the first I data of the tag sequence, wherein I is less than or equal to J; The seventh obtaining submodule is used to obtain a second difference based on the last JI data points of the information sequence and the last JI data points of the tag sequence; and The second training submodule is used to calculate a loss value based on the first difference and the second difference in order to train the information to determine the model.

11. An environmental information determination device, comprising: The determination module is used to input at least one sequence feature information corresponding to the third preset time period into the information determination model to determine the environmental information corresponding to the target time period; The information determination model is trained using the apparatus according to any one of claims 7 to 10.

12. The apparatus of claim 11, wherein, The determining module includes: The sixth submodule is used to input the at least one sequence feature information into the first sub-model to obtain the first environmental information; The seventh obtaining submodule is used to obtain second environmental information based on the at least one sequence feature information and the second sub-model; and The determination submodule is used to determine the environmental information corresponding to the target time period based on the first environmental information and the second environmental information.

13. An electronic device, comprising: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1 to 6.

14. A non-transitory computer readable storage medium having stored thereon computer instructions, wherein, The computer instructions are used to cause the computer to perform the method according to any one of claims 1 to 6.

15. A computer program product comprising a computer program that, when executed by a processor, implements the method according to any one of claims 1 to 6.

Images