Method and apparatus for processing time sequence signal, device and readable medium

Abstract

The application provides a kind of processing method and device of time sequence signal, equipment and readable medium.Its method includes: according to the time sequence signal that multiple sensors are collected in real time, generates prediction sample data;The prediction sample data is carried out feature extraction processing, and prediction sample feature is obtained;The prediction sample feature is input to the signal processing model that has been trained to obtain the output result of the signal processing model.In the technical scheme of the application, not only the time sequence of each sensor is considered, but also the spatial relationship of multiple sensors is considered, the spatial relationship of multiple sensors is effectively retained, and the accuracy and processing efficiency of time sequence signal processing can be effectively improved.

Description

[0001] This application is a divisional application of the patent application with application number 201910838480.1, titled "Processing method and device of time sequence signal, equipment and readable medium", and filed on September 5, 2019. TECHNICAL FIELD

[0002] The present application relates to the technical field of computer application, and in particular relates to a processing method and device of time sequence signal, equipment and readable medium. BACKGROUND

[0003] In the prior art, the signals generated in many scenarios are time sequence signals. Time sequence signal processing, such as anomaly detection, classification or numerical prediction, is a very important work in the research of this kind of signals.

[0004] In the prior art, there are usually many sensors distributed in a system, and the signal of each sensor can be regarded as a time sequence signal. The method of the prior art mainly processes the signal on the time axis of a single sensor signal to make some related prediction and judgment, but the single sensor signal fluctuates randomly with time, so the judgment is not accurate. SUMMARY

[0005] The present application provides a processing method and device of time sequence signal, equipment and readable medium, which is used to improve the accuracy of time sequence signal processing.

[0006] The present application provides a processing method of time sequence signal, comprising:

[0007] generating prediction sample data according to time sequence signals collected in real time by multiple sensors;

[0008] performing feature extraction processing on the prediction sample data to obtain prediction sample features;

[0009] inputting the prediction sample features into a signal processing model trained to obtain an output result of the signal processing model.

[0010] The present application also provides a training method of a signal processing model, comprising:

[0011] generating a training sample data set according to time sequence signals collected by multiple sensors in a target environment and corresponding historical state information in the target environment;

[0012] performing feature extraction processing on the training sample data set to obtain a training sample feature set;

[0013] training a signal processing model based on the training sample feature set, the collected historical state information and a machine learning algorithm.

[0014] The application further provides a processing device of a time sequence signal, comprising:

[0015] a generating module configured to generate prediction sample data according to time sequence signals collected in real time by a plurality of sensors;

[0016] an extracting module configured to perform feature extraction processing on the prediction sample data to obtain prediction sample features;

[0017] a processing module configured to input the prediction sample features into a signal processing model trained to obtain an output result of the signal processing model.

[0018] The application further provides a training device of a signal processing model, comprising:

[0019] a generating module configured to generate a training sample data set according to time sequence signals collected by a plurality of sensors in a target environment and corresponding historical state information in the target environment;

[0020] an extracting module configured to perform feature extraction processing on the training sample data set to obtain a training sample feature set;

[0021] a training module configured to train a signal processing model based on the training sample feature set, the collected historical state information and a machine learning algorithm.

[0022] The application further provides a computing device, comprising:

[0023] a processor; and

[0024] a memory having executable code stored thereon, the executable code, when executed by the processor, causing the processor to perform the method according to any one of the preceding method embodiments.

[0025] The application further provides a non-transitory machine readable storage medium having executable code stored thereon, the executable code, when executed by a processor of an electronic device, causing the processor to perform the method according to any one of the preceding method embodiments.

[0026] The processing method and device of a time sequence signal, the device and the readable medium of the application can use a pre-trained signal processing model to process time sequence signals of a plurality of sensors in real time. Compared with the prior art, the technical solution of the application not only considers the time sequence of each sensor, but also considers the spatial relationship of the plurality of sensors, effectively retains the spatial relationship of the plurality of sensors, and further effectively improves the accuracy and processing efficiency of time sequence signal processing. BRIEF DESCRIPTION OF DRAWINGS

[0027] The above and other objects, features and advantages of the present disclosure will become more apparent from the following detailed description when taken in conjunction with the accompanying drawings in which like reference characters refer to like parts throughout the figures, and wherein:

[0028] Figure 1 Flow chart of the embodiment of the method for processing time series signals of the present application.

[0029] Figure 2 Flow chart of the embodiment of the method for training signal processing model of the present application.

[0030] Figure 3 Structure diagram of the CNN model framework used in the present application.

[0031] Figure 4 Graphical diagram of two functions Sigmoid and Tanh provided by the present application.

[0032] Figure 5 Code logic diagram of the batch normalization layer in the CNN model of the present application.

[0033] Figure 6 Operation diagram of the pooling layer in the CNN model of the present application.

[0034] Figure 7 Comparison diagram of the Dropout layer and the previous fully connected layer in the CNN model of the present application.

[0035] Figure 8 Structure diagram of the embodiment of the device for processing time series signals of the present application.

[0036] Figure 9 Structure diagram of the embodiment of the device for training signal processing model of the present application.

[0037] Figure 10 Structure diagram of the computing device used to implement the above method according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0038] The preferred embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the preferred embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided so that the present disclosure is more thorough and complete, and the scope of the present disclosure is fully conveyed to those skilled in the art.

[0039] The inventors of this application have discovered that existing time-series signal analysis and classification methods typically utilize traditional statistical regression models (such as ARIMA) or custom-extracted time windows to aggregate features. These methods can analyze the historical distribution of a single time-series signal to predict the future. However, in reality, a system usually contains many time-series signal sensors, and the signals of individual sensors fluctuate randomly over time. Current methods cannot effectively utilize the spatial relationships between adjacent sensor signals, thus losing one dimension of information and often leading to inaccurate judgments.

[0040] Figure 1 This is a flowchart illustrating an embodiment of the timing signal processing method of the present invention. Figure 1 As shown, the timing signal processing method of this embodiment may specifically include the following steps:

[0041] S100. Generate prediction sample data based on the time-series signals collected in real time by multiple sensors;

[0042] The execution subject of the timing signal processing method in this embodiment is a timing signal processing device. The timing signal processing device uses a trained signal processing model to process the timing signals collected by multiple sensors.

[0043] For example, the processing of time-series signals in this embodiment may include anomaly detection, classification, or numerical prediction of time-series signals. The classification of time-series signals may be binary or multi-class classification.

[0044] Regardless of the processing method applied to the timing signals, the principle of the timing signal processing method in this embodiment can be implemented. In this embodiment, the signals from each of the multiple sensors have a temporal relationship, and spatial relationships can also be preset between different sensors.

[0045] Specifically, in practical applications, to process time-series signals, multiple sensors can be set up in the detection environment to collect these signals in real time. For example, in a scenario where rain is expected in the future, multiple environmental sensors can be set up, such as at least two of humidity sensors, temperature sensors, wind speed sensors, and barometric pressure sensors. A pre-trained signal processing model can then be used to predict whether rain will occur based on the time-series signals from multiple sensors. In this case, the output of the signal processing model is the environmental detection result.

[0046] In detecting whether a specified machine has a problem of running too hot, temperature sensors can be arranged on each part of the specified machine, and then a pre-trained signal processing model is used to determine whether the specified machine has a problem of running too hot according to the time series signals of the temperature sensors of each part. At this time, the output result of the corresponding signal processing model is the determination result of whether the specified machine is running too hot.

[0047] In actual application, for each sensor in the plurality of sensors, the sensor can be set as a 500-sampling-rate sensor, and 500 values per second can be obtained as the signal of this second. In actual sampling, the signal in the sampling period can be selected as the signal collected in real time by the sensor according to the preset sampling period. The sampling period of the embodiment can be 5 seconds, 3 seconds or other preset time length according to actual needs, which will not be repeated here. The signal collected by each sensor in the preset time length is the time series signal collected in real time. The time series signal of each sensor in the preset time length is a real-time prediction sample data. For example, when the preset time length is 5 seconds, the 500-sampling-rate corresponds to 2500 sampling data of each sensor in the prediction sample data.

[0048] The above is the signal representation of each sensor, but in the embodiment, at least two or even multiple sensors are involved, and the spatial order of these sensors can be configured in advance. Thus, the signal collected by the plurality of sensors in each preset time length can be arranged according to the spatial order configured in advance for the plurality of sensors to generate a two-dimensional signal array corresponding to time and space as a prediction sample data. For example, if each sensor in the n sensors collects m time series signal values in the preset time length in the embodiment, the corresponding two-dimensional signal array is n*m.

[0049] S101, performing feature extraction processing on the prediction sample data to obtain prediction sample features;

[0050] The prediction sample data of the embodiment includes all sampling data of a preset time length collected by multiple sensors in real time, and the data is relatively rich. A part of features can be extracted from all the sampling data, and then the extracted part of features is statistically processed, and the prediction sample features are generated based on the values of the processed statistics. In the embodiment, in order to enrich the prediction sample features, the values of two or more statistics can be obtained. For example, in the embodiment, the step S101 "performing feature extraction processing on the prediction sample data to obtain prediction sample features" can specifically include: counting the values of k statistics of each time window in adjacent d2 time windows of each sensor in adjacent d1 sensors in the two-dimensional signal array, taking the values of the statistics as pixel values to obtain k d1*d2 signal pictures as prediction sample features, wherein d1, d2 and k are all positive integers greater than 1.

[0051] The preset time length in the embodiment can be 10s, 20s or other time lengths continuously, and the time window can be a time unit less than the preset time length, for example, in the embodiment, one second can be taken as a time window.

[0052] When the two-dimensional signal in the embodiment is n*m, d1 is a positive integer greater than 1 and less than n, and the range of d2 in the embodiment is greater than 1 and less than the maximum number of time windows included in the preset time length. For example, in actual application, one second can be taken as a time window.

[0053] The k statistics in the embodiment can include part of signal range, signal variance, signal mean, signal maximum difference and the like.

[0054] According to each piece of prediction sample data, the values of k statistics of each time window in adjacent d2 time windows of each sensor in adjacent d1 sensors in the corresponding two-dimensional signal array can be counted, and k d1*d2 signal pictures are obtained by taking the values of the statistics as pixel values. In this way, k d1*d2 pictures of the same time space can be constructed as prediction sample features.

[0055] For example, in a certain scenario, for a signal graph with n*m prediction sample data, k=3 is selected, and the signal range, signal variance and signal mean are taken respectively. At this time, three d1*d2 signal pictures can be obtained. In one of the d1*d2 signal pictures, the pixel value of each position is the value of the signal range of the sensor corresponding to the horizontal coordinate in the corresponding position within the time window corresponding to the vertical coordinate in the corresponding position; in another d1*d2 signal picture, the pixel value of each position is the value of the signal variance of the sensor corresponding to the horizontal coordinate in the corresponding position within the time window corresponding to the vertical coordinate in the corresponding position; in the last d1*d2 signal picture, the pixel value of each position is the signal mean of the sensor corresponding to the horizontal coordinate in the corresponding position within the time window corresponding to the vertical coordinate in the corresponding position.

[0056] S102, input the prediction sample feature into the trained signal processing model to obtain the output result of the signal processing model.

[0057] The above obtained k d1*d2 signal pictures can be input to the signal processing model as k channels of d1*d2 signal, that is, a d1*d2*k three-dimensional signal picture is input to the signal processing model. In this embodiment, for each prediction sample feature, there is also a corresponding label, which can be identified by the time period corresponding to the prediction sample feature. For example, whether the machine part near a certain position at time t is overheated is dangerous, at this time, the time sequence signal of a plurality of sensors near the certain position at time t is needed to be collected as prediction sample data, and the prediction sample feature at time t is obtained by processing. According to the above method of this embodiment, a k channel d1*d2 prediction sample feature is obtained. Finally, the k channel d1*d2 prediction sample feature is input to the pre-trained signal processing model. The signal processing model can input whether the machine part near a certain position at time t is overheated, such as the output value which can be a probability value between 0 and 1. For example, it can be set that the closer to 1, the more dangerous, the closer to 0, the safer, and vice versa. At this time, the signal processing model of this embodiment is pre-trained as an overheating danger detection model. The signal processing model of this embodiment can be trained based on a convolutional neural network (CNN) model for machine learning.

[0058] In this embodiment, the number of k channels and the specific statistical quantity used in different use scenarios can be different, which can be configured according to the requirements of the scene in advance.

[0059] The processing method of the timing signal of the embodiment can adopt the pre-trained signal processing model to process the timing signals of the plurality of sensors in real time. Compared with the prior art, the technical scheme of the embodiment not only considers the timing of each sensor, but also considers the spatial relationship of the plurality of sensors, effectively retains the spatial relationship of the plurality of sensors, and can effectively improve the accuracy and processing efficiency of the timing signal processing.

[0060] Figure 2 The flowchart of the training method of the signal processing model of the embodiment. As shown in the figure, the training method of the signal processing model of the embodiment can specifically include the following steps: Figure 2

[0061] S200, according to the timing signal collected by the plurality of sensors in the target environment and the corresponding historical state information in the target environment, generate a training sample data set;

[0062] The execution subject of the training method of the signal processing model of the embodiment is a training device of the signal processing model, which is used to train the signal processing model of the embodiment shown in the figure. Figure 1

[0063] First, the generation process of the training sample data set in the step S200 can specifically include the following steps: cutting the timing signal collected by the plurality of sensors into a plurality of signal segments according to a preset time length; for each signal segment, arranging the timing signals of the plurality of sensors in the signal segment according to the spatial order preconfigured for the plurality of sensors, generating a two-dimensional signal array corresponding to time and space as a sample, and taking the historical state in the corresponding time of the signal segment as a label, obtaining a training sample data and putting it into the training sample data set.

[0064] Each training sample data in the training sample data set of the embodiment is different from the prediction sample data of the embodiment shown in the figure, and further includes the corresponding historical state, so as to train the signal processing model based on the two-dimensional signal array in each training sample data and the corresponding historical state. Figure 1

[0065] ​​​The historical state of the embodiment can be different based on different signal processing models to be trained. For example, if the signal processing model of the embodiment is an anomaly detection model of a time series signal, the corresponding historical state can be whether the state at the corresponding time is abnormal. If the signal processing model is a classification model of a time series signal, the corresponding historical state can be the classification to which the corresponding time belongs. If the signal processing model is a numerical prediction model of a time series signal, the corresponding historical state is the historical numerical value at the corresponding time. In practical applications, the signal processing model of the embodiment can also be other processing models of time series signals, and the corresponding historical state can also be other parameters, which will not be repeated here.

[0066] In addition, the two-dimensional signal array in the training sample data of the embodiment is generated in the same way as the two-dimensional signal array in the prediction sample data of the embodiment described above, and details can be referred to the related description of the embodiment described above, which will not be repeated here. Figure 1 Figure 1 The generation method of the two-dimensional signal array in the prediction sample data of the embodiment is the same as that of the two-dimensional signal array in the training sample data of the embodiment described above, and details can be referred to the related description of the embodiment described above, which will not be repeated here.

[0067] S201, performing feature extraction processing on the training sample data set to obtain a training sample feature set;

[0068] For example, when the step S201 is implemented, it can specifically include: for each training sample data in the training sample data set, counting the value of the k statistics of each time window in the adjacent d2 time windows of each sensor in the adjacent d1 sensors in the two-dimensional signal array, taking the value of the statistics as the pixel value to obtain k signal pictures of d1*d2 as the training sample features, where d1, d2 and k are all positive integers greater than 1.

[0069] In the embodiment, based on each training sample data, the corresponding training sample features are extracted, and the process of performing feature extraction processing on the prediction sample data to obtain the prediction sample features in the step S101 of the embodiment described above is completely the same, and details can be referred to the related description of the embodiment described above, which will not be repeated here. Figure 1 Figure 1 The generation method of the two-dimensional signal array in the prediction sample data of the embodiment is the same as that of the two-dimensional signal array in the training sample data of the embodiment described above, and details can be referred to the related description of the embodiment described above, which will not be repeated here.

[0070] Similarly, the statistics of the embodiment can include signal range, signal variance, signal mean or signal maximum difference.

[0071] Similarly, if the embodiment trains a signal processing model to predict whether it will rain in the future, the plurality of sensors in the corresponding target environment can include at least two of a humidity sensor, a temperature sensor, a wind speed sensor and a barometric pressure sensor, and the output result of the signal processing model is an environmental detection result.

[0072] ​​Alternatively, if this embodiment trains a signal processing model to predict whether a specified machine is overheating, the multiple sensors in the target environment can include multiple temperature sensors for the various components of the specified machine. In this case, the output of the signal processing model is a judgment result regarding whether the specified machine is overheating.

[0073] Similarly, the signal processing model in this embodiment can also be other models for performing time-series signal processing, which will not be listed in detail here.

[0074] S202. Train the signal processing model based on the training sample feature set, the collected historical state information, and the machine learning algorithm.

[0075] After the above steps, a training sample feature set can be obtained, which may include multiple training sample features. Furthermore, since each training sample feature corresponds to a signal segment from multiple sensors in the target environment, and each signal segment corresponds to a historical state information, each training sample feature can also correspond to a historical state information. Before training, each training sample feature and its corresponding historical state information are used as a training data point. Before training, an untrained signal processing model, such as a CNN model, can be obtained, and its parameters are assigned initial values. During training, a training data point is taken, and its training sample features are input into the signal processing model. The signal processing model outputs a predicted state. Then, it is determined whether the predicted state is consistent with the acquired historical state. If they are inconsistent, the parameters of the signal processing model are adjusted so that the predicted state is consistent with the acquired historical state. Using all the training sample features and corresponding historical state information from the acquired training sample feature set, the signal processing model is continuously trained according to the above training method until the predicted state is consistent with the acquired historical state. Training ends, the parameters of the signal processing model are determined, and thus the signal processing model is finalized.

[0076] In this embodiment, the more training samples in the collected training sample dataset, the more accurate the trained signal processing model will be. For example, in practical applications, the number of training samples in the collected training sample dataset can reach millions or more.

[0077] The training process of the signal processing model in this embodiment is the same as described above. Figure 1 The processing principle of the timing signals in the illustrated embodiments is similar, and details can also be found above. Figure 3 The relevant steps of the illustrated embodiment will not be repeated here.

[0078] The CNN model adopted by the signal processing model of the embodiment can define different learning goals according to different problem definitions after taking a two-dimensional time sequence signal as input, so as to solve binary classification, multi-classification, regression and other problems. For a classification problem, the result output by the model is the probability of each class; for a regression problem, the model result is the predicted value. Since the CNN model structure has good learning and anti-overfitting ability for picture type data, the CNN model is adopted as a prototype in the application to train the corresponding signal processing model.

[0079] During specific training, after the picture sample, that is, the training sample feature, is constructed, the signal picture of multiple channels is input to the CNN model for training.

[0080] For example, when the input sample dimension is 16, the channel number is 2, and a three-classification problem is solved, the corresponding CNN model framework structure can be as shown in Figure 3 .

[0081] According to the model structure shown in Figure 4 , it can be seen that the main components of the model at this time include a two-dimensional convolution layer (conv2D), an activation layer (Activation), a batch normalization layer (BatchNormalization), a pooling layer (MaxPooling) and a DropOut layer. The calculation logic and function of each component are as follows:

[0082] 2D convolution layer: divide the values in the input picture sample into small windows and perform convolution operation, and the convolution formula is as follows:

[0083]

[0084] The convolution operation can extract local information in the image through parameter sharing and small window operation, and enhance the local perception ability.

[0085] Activation layer: the role of the activation layer is to map the output result of network calculation to another function, so that it is more conducive to model learning. Common functions include Sigmoid, Tanh, ReLU and the like. Figure 5 The graphical diagram of the two functions Sigmoid and Tanh provided by the embodiment is shown in

[0086] Batch normalization layer: the calculation logic of batch normalization is to normalize each batch of samples entering the model training to a standard normal distribution, which aims to resist overfitting, improve training speed and reduce the requirement for initialization data. The code logic of the corresponding batch normalization layer can be as shown in Figure 6 .

[0087] Pooling layer: like convolution, pooling also has a sliding kernel, which can be called a sliding window,Figure 7 Max Pooling: In this case, the size of the sliding window is 2x2, and the stride is 2. For each region that is slid over, the maximum value is taken as the output. The purpose of this operation is to abstract some global features and reduce the dimensionality of the data, speeding up the computation.

[0088] DropOut layer: During training, some network neurons are discarded to enhance the robustness of the model and resist overfitting. For example, Figure 8 On the right is the comparison between the network with DropOut and the previous fully connected network.

[0089] The training method of the signal processing model can train a precise signal processing model applied in a target environment with multiple sensors, and then process the time series signals collected by the multiple sensors in the target environment based on the signal processing model. Compared with the prior art, not only the time series of each sensor is considered, but also the spatial relationship of multiple sensors is considered, the spatial relationship of multiple sensors is effectively retained, and the accuracy and processing efficiency of time series signal processing can be effectively improved.

[0090] Figure 8 The structure diagram of the time series signal processing device embodiment of the application is shown in FIG. Figure 9 As shown in the figure, the time series signal processing device of the embodiment can specifically include:

[0091] The generation module 10 is configured to generate prediction sample data according to the time series signals collected by the multiple sensors in real time.

[0092] The extraction module 11 is configured to perform feature extraction processing on the prediction sample data generated by the generation module 10 to obtain prediction sample features.

[0093] The processing module 12 is configured to input the prediction sample features extracted by the extraction module 11 into the trained signal processing model to obtain an output result of the signal processing model.

[0094] Further optionally, the generation module 10 is configured to:

[0095] Arrange the signals collected by the multiple sensors in real time within each preset time length according to the spatial order preconfigured for the multiple sensors, generate a two-dimensional signal array corresponding to time and space as a piece of prediction sample data.

[0096] Further optionally, the extraction module 11 is configured to:

[0097] The values of the k statistics of each time window of the adjacent d2 time windows of each sensor of the adjacent d1 sensors in the statistical two-dimensional signal array are counted, and k d1*d2 signal pictures are obtained by taking the values of the statistics as pixel values, as the prediction sample features, wherein d1, d2 and k are all positive integers greater than 1.

[0098] Further optionally, the statistics include signal range, signal variance, signal mean value or signal maximum difference.

[0099] Further optionally, the processing of the time series signal includes anomaly detection of the time series signal, classification of the time series signal or numerical prediction of the time series signal.

[0100] Further optionally, the plurality of sensors includes at least two of a humidity sensor, a temperature sensor, a wind speed sensor and a barometric pressure sensor, and the output result of the signal processing model is an environmental detection result.

[0101] Alternatively, the plurality of sensors are a plurality of temperature sensors respectively used to detect the temperature of each component of a specified machine, and the output result of the signal processing model is a judgment result about whether the specified machine is running overheated.

[0102] The processing device of the time series signal of the embodiment, by adopting the above-mentioned module, realizes the processing of the time series signal, and has the same implementation principle and technical effects as the related method embodiment, and details can be referred to the related description of the method embodiment, which will not be repeated here.

[0103] Figure 9 The structural diagram of the training device embodiment of the signal processing model of the application is shown in FIG. 1. Figure 10 As shown in the figure, the training device of the signal processing model of the embodiment can specifically include:

[0104] The generating module 20 is configured to generate a training sample data set according to the time series signals historically collected by the plurality of sensors in the target environment and the corresponding historical state information in the target environment.

[0105] The extracting module 21 is configured to perform feature extraction processing on the training sample data set generated by the generating module 20 to obtain a training sample feature set.

[0106] The training module 22 is configured to train the signal processing model based on the training sample feature set extracted by the extracting module 21, the historical state information collected in the training sample data set generated by the generating module 20 and a machine learning algorithm.

[0107] Further optionally, the generating module 20 is configured to:

[0108] The time series signals historically collected by the plurality of sensors are cut into a plurality of signal segments according to a preset time length.

[0109] For each signal segment, the time series signals of the plurality of sensors in the signal segment are arranged according to the spatial order pre-configured for the plurality of sensors to generate a two-dimensional signal array corresponding to time and space as a sample, and the historical state in the corresponding time of the signal segment is taken as a label to obtain a training sample data and put it into a training sample data set.

[0110] Further optionally, the extraction module 21 is configured to:

[0111] For each training sample data in the training sample data set, the values of the k statistics of each time window in the adjacent d2 time windows of each sensor in the adjacent d1 sensors in the two-dimensional signal array are counted, and the values of the statistics are taken as pixel values to obtain k signal pictures of d1*d2 as training sample features, where d1, d2 and k are positive integers greater than 1.

[0112] Further optionally, the statistics include signal range, signal variance, signal mean or signal maximum difference.

[0113] Further optionally, the signal processing model includes an anomaly detection model of time series signals, a classification model of time series signals or a numerical prediction model of time series signals.

[0114] Further optionally, the plurality of sensors includes at least two of a humidity sensor, a temperature sensor, a wind speed sensor and a barometric pressure sensor, and the output result of the signal processing model is an environmental detection result.

[0115] Alternatively, the plurality of sensors are a plurality of temperature sensors respectively used to detect the temperature of each component of a specified machine, and the output result of the signal processing model is a judgment result about whether the specified machine is running overheated.

[0116] The training device of the signal processing model of the above embodiment realizes the training of the signal processing model by using the above module, and has the same implementation principle and technical effects as the above related method embodiment. For details, refer to the related description of the above method embodiment, which will not be repeated here.

[0117] Figure 10 A structural schematic diagram of a computing device according to an embodiment of the present application is shown, which can be used to implement the above method. For example, the computing device can be used to implement the above time series signal processing method or the signal processing model training method.

[0118] Referring to ​ , the computing device 1000 includes a memory 1010 and a processor 1020.

[0119] The processor 1020 can be a single core processor or a multiple core processor. In some embodiments, the processor 1020 can include a general purpose processor and one or more special purpose co-processors such as graphics processing units (GPUs), digital signal processors (DSPs), and so on. In some embodiments, the processor 1020 can be implemented using a custom circuit, such as an application specific integrated circuit (ASIC) or a field programmable gate array (FPGA).

[0120] The memory 1010 can include various types of memory units, such as a system memory, a read-only memory (ROM), and a permanent storage device. The ROM can store static data or instructions that the processor 1020 or other modules of the computer need. The permanent storage device can be a read-and-write memory device. The permanent storage device can be a non-volatile memory device that does not lose stored instructions and data even if the computer is powered off. In some embodiments, the permanent storage device employs a mass storage device (such as a magnetic or optical disk, a flash memory) as the permanent storage device. In other embodiments, the permanent storage device can be a removable storage device (such as a floppy disk, an optical disk). The system memory can be a read-and-write memory device or a volatile read-and-write memory device, such as a dynamic random access memory. The system memory can store some or all of the instructions and data that the processor needs at runtime. In addition, the memory 1010 can include a combination of any or all of the above-mentioned memory units, including various types of semiconductor memory chips (DRAM, SRAM, SDRAM, flash memory, programmable read-only memory), magnetic disks and / or optical disks. In some embodiments, the memory 1010 can include a read-and / or write removable storage device, such as a compact disc (CD), a read-only digital versatile disc (e.g., DVD-ROM, dual-layer DVD-ROM), a read-only Blu-ray disc, an ultra-density disc, a flash memory card (e.g., SD card, min SD card, Micro-SD card, and so on), a magnetic floppy disk, and so on. The computer readable storage media does not include carrier waves and transitory electronic signals propagating through a wire or cable or a wireless transmission.

[0121] The memory 1010 stores executable code that, when processed by the processor 1020, causes the processor 1020 to perform the above-mentioned method of processing a timing signal or the above-mentioned method of training a signal processing model.

[0122] The method of processing a timing signal or the method of training a signal processing model according to the present application has been described in detail above with reference to the accompanying drawings.

[0123] Furthermore, the method according to the present application can also be implemented as a computer program or computer program product comprising computer program code instructions for executing the above steps defined in the above method of the present application.

[0124] Alternatively, the present application can also be implemented as a non-transitory machine readable storage medium (or computer readable storage medium, or machine readable storage medium) having stored thereon executable code (or computer program, or computer instruction code) which, when executed by a processor of an electronic device (or computing device, server, etc.), causes the processor to perform the steps of the above method according to the present application.

[0125] Those skilled in the art will further appreciate that the various illustrative logical blocks, modules, circuits, and algorithm steps described in connection with the disclosure herein can be implemented as electronic hardware, computer software, or combinations of both.

[0126] The flow diagrams and block diagrams in the drawings are representative of the architecture, functionality, and operation of possible implementations of systems and methods according to the present application. In this regard, each block in the flow diagrams and block diagrams can represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions noted in the blocks can occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently or the blocks can sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and / or flowchart illustrations, and combinations thereof, can be implemented by special purpose hardware-based systems that perform the specified functions or operations, or combinations of special purpose hardware and computer instructions.

[0127] Embodiments of the present application have been described above, and the description is intended to be illustrative of the embodiments of the present application and not exhaustive, and is not limited to the embodiments disclosed. Numerous modifications and adaptations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The choice of words in the specification is intended to best explain the principles of the embodiments, practical application, or improvement to the art, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A method for processing time series signals, comprising: generating prediction sample data from time series signals collected in real time by a plurality of sensors, including: arranging signals collected in real time by the plurality of sensors within each preset time length in a spatial order preconfigured for the plurality of sensors to generate a two-dimensional signal array corresponding to time and space as a piece of prediction sample data; performing feature extraction processing on the prediction sample data to obtain prediction sample features; inputting the prediction sample features into a trained signal processing model to obtain an output result of the signal processing model; wherein the plurality of sensors include at least two of a humidity sensor, a temperature sensor, a wind speed sensor, and a barometric pressure sensor, and the output result of the signal processing model is an environmental detection result; or, the plurality of sensors are a plurality of temperature sensors respectively used to detect the temperature of each component of a specified machine, and the output result of the signal processing model is a judgment result about whether the specified machine is running overheated.

2. The method of claim 1, wherein, performing feature extraction processing on the prediction sample data to obtain prediction sample features includes: statistically obtaining the value of k statistics of each of the time windows in the adjacent d2 time windows of each of the adjacent d1 sensors in the two-dimensional signal array, and obtaining k signal pictures of d1*d2 as the prediction sample features, with the value of the statistics as the pixel value, where d1, d2, and k are all positive integers greater than 1.

3. The method of claim 2, wherein, The statistics include signal range, signal variance, signal mean, or signal maximum difference.

4. The method of claim 1, wherein, The processing of the time series signals includes anomaly detection of time series signals, classification of time series signals, or numerical prediction of time series signals.

5. A method for training a signal processing model, comprising: generating a training sample data set from time series signals collected by a plurality of sensors in a target environment and corresponding historical state information in the target environment, including: dividing the time series signals collected by the plurality of sensors into a plurality of signal segments according to a preset time length; for each signal segment, arranging the time series signals of the plurality of sensors in the signal segment in a spatial order preconfigured for the plurality of sensors to generate a two-dimensional signal array corresponding to time and space as a sample, and obtaining the historical state corresponding to the time of the signal segment as a label to obtain a piece of training sample data and put it into the training sample data set; performing feature extraction processing on the training sample data set to obtain a training sample feature set; training a signal processing model based on the training sample feature set, the collected historical state information, and a machine learning algorithm; wherein the plurality of sensors include at least two of a humidity sensor, a temperature sensor, a wind speed sensor, and a barometric pressure sensor, and the output result of the signal processing model is an environmental detection result; Alternatively, the plurality of sensors are a plurality of temperature sensors respectively configured to detect temperatures of respective components of a designated machine, and the output result of the signal processing model is a result of determining whether the designated machine is running overheat.

6. The method of claim 5, wherein, The training sample data set is subjected to feature extraction processing to obtain a training sample feature set, including: For each piece of training sample data in the training sample data set, the values of k statistics of each of the time windows in the adjacent d2 time windows of each of the adjacent d1 sensors in the two-dimensional signal array are counted, and k signal pictures of d1*d2 are obtained by taking the values of the statistics as pixel values, as the training sample features, where d1, d2 and k are all positive integers greater than 1.

7. The method of claim 6, wherein, The statistics include signal range, signal variance, signal mean or signal maximum difference.

8. The method of claim 6, wherein, The signal processing model includes an abnormality detection model of time series signals, a classification model of time series signals, or a numerical prediction model of time series signals.

9. A time series signal processing device, comprising: A generating module configured to generate prediction sample data according to time series signals collected in real time by a plurality of sensors, the generating module being configured to: arrange signals collected in real time by the plurality of sensors within each preset time length according to a spatial order pre-configured for the plurality of sensors to generate a two-dimensional signal array corresponding to time and space as a piece of prediction sample data; An extracting module configured to perform feature extraction processing on the prediction sample data to obtain prediction sample features; A processing module configured to input the prediction sample features into a signal processing model that has been trained to obtain an output result of the signal processing model; The plurality of sensors include at least two of a humidity sensor, a temperature sensor, a wind speed sensor, and a barometric pressure sensor, and the output result of the signal processing model is an environmental detection result. Alternatively, the plurality of sensors are a plurality of temperature sensors respectively configured to detect temperatures of respective components of a designated machine, and the output result of the signal processing model is a result of determining whether the designated machine is running overheat.

10. The apparatus of claim 9, wherein, The extracting module is configured to: count values of k statistics of each of the time windows in the adjacent d2 time windows of each of the adjacent d1 sensors in the two-dimensional signal array, and obtain k signal pictures of d1*d2 by taking the values of the statistics as pixel values, as the prediction sample features, where d1, d2 and k are all positive integers greater than 1.

11. The apparatus of claim 10, wherein, The statistics include signal range, signal variance, signal mean or signal maximum difference.

12. The apparatus of claim 9, wherein, The time series signal processing includes abnormality detection of time series signals, classification of time series signals, or numerical prediction of time series signals.

13. A signal processing model training device, comprising: A generating module configured to generate a training sample data set according to time series signals collected historically by a plurality of sensors in a target environment and corresponding historical state information in the target environment, the generating module being configured to: The time series signals collected by the plurality of sensors are divided into a plurality of signal segments according to a preset time length; for each signal segment, the time series signals of the plurality of sensors in the signal segment are arranged according to a spatial order pre-configured for the plurality of sensors to generate a two-dimensional signal array corresponding to time and space as a sample, and a historical state in the corresponding time of the signal segment is taken as a label to obtain a training sample data and put it into the training sample data set; an extraction module configured to perform feature extraction processing on the training sample data set to obtain a training sample feature set; a training module configured to train a signal processing model based on the training sample feature set, the collected historical state information, and a machine learning algorithm; wherein the plurality of sensors include at least two of a humidity sensor, a temperature sensor, a wind speed sensor, and a barometric pressure sensor, and an output result of the signal processing model is an environmental detection result; Alternatively, the plurality of sensors are a plurality of temperature sensors respectively configured to detect temperatures of respective components of a specified machine, and an output result of the signal processing model is a judgment result about whether the specified machine is running overheated.

14. The apparatus of claim 13, wherein, The extraction module is configured to: for each training sample data in the training sample data set, count values of k statistics of each of a plurality of time windows in a plurality of adjacent sensors in the two-dimensional signal array, to obtain k signal pictures of d1*d2 as the training sample feature, where d1, d2, and k are all positive integers greater than 1.

15. The apparatus of claim 14, wherein, The statistics include signal range, signal variance, signal mean, or signal maximum difference.

16. The apparatus of claim 14, wherein, The signal processing model includes an anomaly detection model of time series signals, a classification model of time series signals, or a numerical prediction model of time series signals.

17. A computing device comprising: a processor; and a memory having stored thereon executable code that, when executed by the processor, causes the processor to perform the method of any of claims 1-4; or perform the method of any of claims 5-8.

18. A non-transitory machine-readable storage medium having stored thereon executable code that, when executed by a processor of an electronic device, causes the processor to perform the method of any of claims 1-4; or perform the method of any of claims 5-8. ​

Images