CN114792115B - Method, device and medium for outlier removal of telemetry signal based on deconvolution reconstruction network - Google Patents

Abstract

The embodiment of the present invention discloses a method, device and medium for removing outliers of telemetry signals based on a deconvolution reconstruction network; the method includes: constructing a deconvolution reconstruction network based on convolution construction and deconvolution reconstruction; The training data composed of the actual measured telemetry signal data and the corresponding time stamp of the measured telemetry signal data is used to train the deconvolution reconstruction network to obtain the trained deconvolution reconstruction network; the original telemetry signal data time stamp is input to the The deconvolution reconstruction network after the training is used to obtain the predicted telemetry signal data corresponding to the original telemetry signal data timestamp; and the predicted telemetry signal data corresponding to the original telemetry signal data timestamp is calculated. The distance between signal data; when the distance is greater than a set threshold, it is determined that the original telemetry signal data corresponding to the time stamp of the original telemetry signal data is an outlier value, and the outlier value is removed.

Description

Method, device and medium for outlier removal of telemetry signal based on deconvolution reconstruction network

technical field

Embodiments of the present invention relate to the technical field of spacecraft data processing, in particular to a method, device and medium for removing outliers of telemetry signals based on a deconvolution reconstruction network.

Background technique

The telemetry data of the spacecraft is an important strategic resource. Using the telemetry data, the operating status of the spacecraft can be monitored in real time and anomalies can be predicted, so as to ensure the stable operation of the spacecraft. However, due to noise interference in the space where the spacecraft operates, data loss, sensor failure, etc., the telemetry data will undergo instantaneous jumps or errors, thereby generating abnormal points, that is, outliers of the telemetry signal. Therefore, in order to obtain higher-quality spacecraft telemetry data, it is necessary to identify and eliminate the outliers of these telemetry signals.

Telemetry signal outliers of spacecraft usually include the following two types: the first type is isolated outliers, that is to say, there are jumping spikes in the early time series, and the adjacent data are much smaller or larger than the outliers; The second type belongs to isolated continuous outliers. This type of outlier is generally a large spike in the time series, and the data of the left and right neighbors of several outliers that appear continuously are much smaller or larger than wild value.

At present, conventional schemes for removing outliers of telemetry signals include recursive-based methods, statistical-based methods, and fitting-based methods; among them, methods based on recursive relations usually include forward difference algorithm, 53H algorithm, wavelet transform, and improved Kalman filtering algorithm, etc. Statistical-based methods mainly include 3-sigma criterion, Romanowski criterion, Dixon criterion, Grubbs criterion, fuzzy-based confidence method, etc. Fitting-based methods mainly include polynomial fitting, least-squares fitting, and data-driven fitting. The above three types of methods usually have certain limitations, and these limitations affect the removal effect of outliers in telemetry signals.

Contents of the invention

In view of this, the embodiment of the present invention expects to provide a method, device and medium for removing outliers of telemetry signals based on a deconvolution reconstruction network; it can improve the removal of outliers of telemetry signals when the deconvolution reconstruction network can learn quickly Effect.

The technical scheme of the embodiment of the present invention is realized like this:

In the first aspect, an embodiment of the present invention provides a method for removing outliers of telemetry signals based on a deconvolution reconstruction network, including:

Construct a deconvolution reconstruction network based on convolution construction and deconvolution reconstruction;

The deconvolution reconstruction network is trained by using the training data composed of the existing measured telemetry signal data and the corresponding time stamp of the measured telemetry signal data to obtain the deconvolution reconstruction network after training;

inputting raw telemetry signal data timestamps into the trained deconvolution reconstruction network to obtain predicted telemetry signal data corresponding to the raw telemetry signal data timestamps;

calculating the distance between the original telemetry signal data corresponding to the original telemetry signal data time stamp and the predicted telemetry signal data;

When the distance is greater than the set threshold, it is determined that the original telemetry signal data corresponding to the time stamp of the original telemetry signal data is an outlier value, and the outlier value is removed.

In the second aspect, the embodiment of the present invention provides a device for removing outliers of telemetry signals based on a deconvolution reconstruction network. part; of which,

The construction part is configured to construct a deconvolution reconstruction network based on convolution construction and deconvolution reconstruction;

The training part is configured to train the deconvolution reconstruction network through training data composed of existing measured telemetry signal data and corresponding time stamps of the measured telemetry signal data, to obtain a trained deconvolution reconstruction network ;

said deconvolution reconstruction network for inputting raw telemetry signal data timestamps to obtain predicted telemetry signal data corresponding to said raw telemetry signal data timestamps;

the calculation section configured to calculate a distance between raw telemetry signal data corresponding to a timestamp of the raw telemetry signal data and the predicted telemetry signal data;

The determination part is configured to compare the distance with a set threshold, and when the distance is greater than the set threshold, determine that the original telemetry signal data corresponding to the original telemetry signal data time stamp is an outlier , and trigger the removal part;

The removal part is configured to remove the outlier.

In a third aspect, an embodiment of the present invention provides a computing device, the computing device includes: a communication interface, a memory, and a processor, and each component is coupled together through a bus system; wherein,

The communication interface is used for receiving and sending signals during the process of sending and receiving information with other external network elements;

said memory for storing a computer program capable of running on said processor;

The processor is configured to, when running the computer program, execute the steps of the method for removing outliers of telemetry signals based on a deconvolution reconstruction network in the first aspect.

In a fourth aspect, an embodiment of the present invention provides a computer storage medium, the computer storage medium stores a program for removing outliers of telemetry signals based on a deconvolution reconstruction network, and the outlier values of telemetry signals based on a deconvolution reconstruction network When the removal program is executed by at least one processor, the steps of the method for removing outliers of telemetry signals based on the deconvolution reconstruction network described in the first aspect are realized.

Embodiments of the present invention provide a method, device, and medium for removing outliers of telemetry signals based on a deconvolution reconstruction network; the telemetry data corresponding to the time stamp is predicted by the deconvolution reconstruction network that has been trained on the measured telemetry signal data, and Determine whether it is an outlier according to the distance between the predicted value and the real value. Compared with the conventional scheme, it avoids artificial interpolation and excessive parameter setting, eliminates the limitations caused by statistical distribution and time span, and improves the removal effect of outliers when the deconvolution reconstruction network can learn quickly and reliability.

Description of drawings

FIG. 1 is a schematic flow chart of a method for removing outliers of telemetry signals based on a deconvolution reconstruction network provided by an embodiment of the present invention;

FIG. 2 is a schematic diagram of a deconvolution reconstruction network structure provided by an embodiment of the present invention;

FIG. 3 is a schematic diagram of loss iterations of training and verification provided by an embodiment of the present invention;

4 is a schematic diagram of a distance histogram and a kernel density estimation curve provided by an embodiment of the present invention;

FIG. 5 is a schematic diagram of the process of removing outliers by using the trained DRN provided by the embodiment of the present invention;

Fig. 6 is a schematic diagram of the normalized distance comparison of the semi-major axis of the orbit provided by the embodiment of the present invention;

Fig. 7 is a schematic diagram of the comparison of training and verification losses of the track semi-major axis telemetry data provided by the embodiment of the present invention;

FIG. 8 is a schematic diagram of the composition of a telemetry signal outlier removal device based on a deconvolution reconstruction network provided by an embodiment of the present invention;

FIG. 9 is a schematic composition diagram of another device for removing outliers of telemetry signals based on a deconvolution reconstruction network provided by an embodiment of the present invention;

FIG. 10 is a schematic diagram of a specific hardware structure of a computing device provided by an embodiment of the present invention.

Detailed ways

The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention.

For the current conventional outlier removal methods, in detail, first of all, the recursive-based method usually requires the processed data to be continuous, while the real satellite telemetry signal is usually discontinuous due to ground station location constraints and data packet loss. , so these methods need to interpolate the missing data before proceeding. Data with wild values will greatly affect the effect of interpolation; in addition, methods based on recursive relationships have many hyperparameters that need to be set manually, such as thresholds, the number of points participating in recursive, etc., so this type of method has certain limitations. Secondly, statistical methods have certain limitations in removing outliers of telemetry signals with large variations. This is because the statistical laws presented by such data are very diverse. It may be that the telemetry data of a certain period of time satisfies one distribution, and the next time period satisfies another distribution. In addition, the fitting-based method will be less effective in removing outliers for telemetry signals with longer time stamp spans, so local fitting is usually used.

Based on the limitations shown by the above-mentioned conventional solutions, see FIG. 1 , which shows a method for removing outliers in telemetry signals based on a Deconvolutional Reconstruction Network (DRN, Deconvolutional Reconstruction Network) provided by an embodiment of the present invention. include:

S101: Construct a deconvolution reconstruction network based on convolution construction and deconvolution reconstruction;

S102: Train the deconvolution reconstruction network by using the training data composed of the existing measured telemetry signal data and the corresponding time stamp of the measured telemetry signal data, to obtain the trained deconvolution reconstruction network;

S103: Input the time stamp of original telemetry signal data into the trained deconvolution reconstruction network, so as to obtain predicted telemetry signal data corresponding to the time stamp of the original telemetry signal data;

S104: Calculate the distance between the original telemetry signal data corresponding to the time stamp of the original telemetry signal data and the predicted telemetry signal data;

S105: When the distance is greater than a set threshold, determine that the original telemetry signal data corresponding to the time stamp of the original telemetry signal data is an outlier value, and remove the outlier value.

Through the technical solution shown in Figure 1, the telemetry data corresponding to the time stamp is predicted by the deconvolution reconstruction network trained by the measured telemetry signal data, and whether it is an outlier is determined according to the distance between the predicted value and the real value. Compared with the conventional scheme, it avoids artificial interpolation and excessive parameter setting, eliminates the limitations caused by statistical distribution and time span, and improves the removal effect and reliability of outliers.

For the technical solution shown in Figure 1, it should be noted that the embodiment of the present invention selects a deconvolution reconstruction network including a convolution construction part and a deconvolution reconstruction part, wherein the convolution construction part can input the network The data is increased in dimension, and then the high-dimensional tensor is reconstructed through deconvolution to perform dimensionality reduction operation to adapt to the telemetry signal data, and there is no latent vector in the network. In some examples, as shown in Figure 2, the deconvolution reconstruction network includes: input (Input) layer, batch normalized full connection (FC (BN), Full Connection (Batch Normalization)) layer, the first reconstruction Reshape Layer, the first convolution Conv layer, the second convolution layer (ConV(BN+ReLU)) based on batch normalization BN and the corrected linear unit activation function (ReLU, Rectified Linear Unit activation), the first deconvolution DeConV Layer, the second deconvolution layer (DeConV(BN+ReLU)) based on batch normalization BN and the modified linear unit activation function ReLU, the third deconvolution DeConV layer, the second reconstruction Reshape layer and the output of the fully connected FC (Output) layer. As can be seen from Figure 2, from the input layer to the ConV (BN+ReLU) layer belongs to the convolutional construction (Convolutional Construction) part, from the first deconvolution DeConV layer to the Output (FC) layer belongs to the deconvolutional reconstruction (Deconvolutional Reconstruction) part.

For the DRN shown in Figure 2, in some examples, the output signal of the input layer is identified as X ₀ , the output shape is 1, and the number of parameters is 0; the output signal of the FC (BN) layer is identified as X ₁ , and the output The shape is 16, the number of parameters is 96; the output signal of the first reconstruction layer is marked as X ₂ , the output shape is (4,4,1), and the number of parameters is 0; the output signal of the first convolutional layer is marked as X ₃ , the output shape is (4,4,16), and the number of parameters is 32; the output signal of the second convolutional layer based on the batch normalization and modified linear unit activation function is identified as X ₄ , and the output shape is (4, 4,32), the number of parameters is 672; the output signal of the first deconvolution layer is identified as X ₅ , the output shape is (4,4,32), and the number of parameters is 1056; the line type based on batch normalization and correction The output signal of the second deconvolution layer of the unit activation function is marked as X ₆ , the output shape is (4,4,16), and the number of parameters is 592; the output signal of the third deconvolution layer is marked as X ₇ , and the output The shape is (4,4,1), the number of parameters is 17; the output signal of the second reconstruction layer is marked as X ₈ , the output shape is 16, and the number of parameters is 0; the output signal of the fully connected output layer is marked as X ₉ , the output shape is 1, and the number of parameters is 17. From the above example, it can be seen that the total number of parameters of the entire DRN is not large because it is aimed at satellite telemetry signal data, so the structure of the DRN is relatively simple and fast training can be achieved.

For the technical solution shown in Fig. 1, in some examples, the deconvolution reconstruction network is trained by using the training data consisting of the existing measured telemetry signal data and the corresponding time stamp of the measured telemetry signal data to obtain The trained deconvolution reconstruction network includes:

For the initialized deconvolution reconstruction network, the time stamp of the measured telemetry signal data is input to the deconvolution reconstruction network as input data;

According to the output of the deconvolution reconstruction network and the measured telemetry signal data corresponding to the time stamp of the measured telemetry signal data, the deconvolution reconstruction network is trained, and each layer included in the deconvolution reconstruction network is determined. Corresponding parameters to obtain the trained deconvolution reconstruction network.

For the above example, it needs to be explained that in the existing measured telemetry signal data, outliers only account for a very small part of the entire data. Therefore, the measured telemetry signal data is suitable for DRN network training and does not need to be eliminated DRN is trained on telemetry data after outliers.

Based on the above example, during the training process of the deconvolution reconstruction network, the parameters for the DRN network training are set to: learning rate = 0.01, the optimization method is the Adam algorithm, and the learning rate decay rule is that if the training is not updated for 5 generations Then the learning rate is reduced to half of the original.

For the technical solution shown in Fig. 1, in some examples, the method further includes: according to the subdivision histogram and kernel density estimation curve of the distance between the outlier and the normal telemetry signal data, determine the The threshold of .

For the above example, it should be noted that for the trained DRN, for the normal data in the telemetry signal data, the distance between its input and output can be reflected in the prediction error of the DRN network; and for the outlier value, Its distance will not meet the training rules of normal data, that is, it will be much larger than the prediction error. Therefore, the threshold for distinguishing normal data from outliers can be determined by analyzing the distribution levels of all differences. Regarding the difference distribution level, the kernel density estimation is used to estimate the unknown density function, which is one of the non-parametric test methods. The kernel density estimation includes multiple histogram indicators. In the embodiment of the present invention, the distance of the normal data is at the threshold = within 0.03.

For the technical solution shown in Figure 1, in some examples, for the satellite telemetry signal data, the distance between the two data can be considered as the difference between the two data, therefore, the calculation of the original telemetry signal data time Stamping the distance between the original telemetry signal data corresponding to the predicted telemetry signal data, comprising: calculating the difference between the original telemetry signal data corresponding to the timestamp of the original telemetry signal data and the predicted telemetry signal data .

Based on the foregoing technical solution, the embodiment of the present invention performs a simulation experiment on 6 different types of telemetry signal data of a certain type of hyperspectral satellite. The telemetry signal data in the simulation experiment is represented by the J2000 coordinate system. These 6 types include: X-axis position (X position) in J2000 coordinate system, Y-axis position (Yposition) in J2000 coordinate system, Z-axis position (Z position) in J2000 coordinate system, X-axis velocity (XVelocity) in J2000 coordinate system, The Y-axis velocity (Y Velocity) in the J2000 coordinate system and the Z-axis velocity (ZVelocity) in the J2000 coordinate system. Because this type of satellite does not have the function of delaying the download of data, these telemetry signal data will be limited by the location of the ground observation station and will appear in the form of discrete data points over a period of time. For the training data set, each batch of training data is 32, and the ratio of the training training set in the training data to the verification validation set is 3:1, and the verification set is verified after each training round epoch. For each set of data, Train for 20 rounds of epochs respectively. The details of the training data used in the embodiment of the present invention are shown in Table 1:

name number of training sets Number of validation sets total training data Number of outliers X-axis position (X position) 46677 15558 62235 123 Y-axis position (Y position) 46677 15558 62235 243 Z position 46677 15558 62235 204 X-axis speed (X Velocity) 46677 15558 62235 121 Y-axis speed (Y Velocity) 46677 15558 62235 146 Z Velocity 46677 15558 62235 133

Table 1

The iterative situation of loss (loss) for training and verification is shown in Figure 3. From Figure 3, it can be seen that during the training process of DRN, the training and verification losses rarely fluctuate, and can be quickly achieved within 20 rounds. convergence. In addition, for the distance histogram and kernel density estimation curve used to determine the threshold, as shown in Figure 4, it can be seen from the histograms and kernel density estimation curves of the six types of data that the threshold is preferably 0.03.

In this simulation experiment, the process of removing outliers by using the DRN after training is shown in Figure 5. In Figure 5, (a1) to (a6) respectively represent the X, Y, and Z axis positions in the J2000 coordinate system And the original discrete signal data of the X, Y, and Z-axis speeds in the J2000 coordinate system; (b1) to (b6) represent the X, Y, and Z-axis positions in the J2000 coordinate system and the X, Y, and The distance between the normalized predicted value corresponding to the Z-axis speed and the normalized original discrete signal data, it can be seen from (b1) to (b6) that the outlier value can be removed by the threshold of 0.03. The discrete signal data corresponding to the positions of X, Y, and Z axes in the J2000 coordinate system and the speed of the X, Y, and Z axes in the J2000 coordinate system after removing outliers are shown in (c1) to (c6), respectively, from Figure 5 It can be seen from the figure that the distance of the outliers is far greater than the distance of the normal data. In extreme cases, if the DRN has fully learned the real orbital dynamics and control laws, it can perfect the normal data in the original data and you. Then the normal data distance will be equal to 0.

In addition to the above data simulation, in order to further reflect the superiority of the technical solution proposed by the embodiment of the present invention, the semi-axis telemetry data of the track is used, and the method for removing the outlier value of the telemetry signal based on the deconvolution reconstruction network proposed by the aforementioned technical solution ( Hereinafter referred to as DRN), compared with MLP autoencoder network (MLP-AE) and convolutional autoencoder network (CAE), the simulation conditions are consistent with the aforementioned data simulation experiments. The normalized distance of the semi-major axis of the track after the training of the three networks is shown in Figure 6. It can be seen from Figure 6 that although the three networks can recognize most of the outliers, the performance of DRN is better. This is because DRN can find more outliers than MLP-AE and CAE. In addition, as shown in Figure 7, DRN has lower training and verification losses than MLP-AE and CAE on orbit semi-major axis telemetry data, which means that DRN has better training effect than the other two networks, so it can Find more outliers.

Referring to FIG. 8 , it shows a telemetry signal outlier removal device 80 based on a deconvolution reconstruction network provided by an embodiment of the present invention. The device 80 includes: a construction part 801, a deconvolution reconstruction network 802, a training part 803, calculation part 804, determination part 805 and removal part 806; wherein,

The construction part 801 is configured to construct a deconvolution reconstruction network 802 based on convolution construction and deconvolution reconstruction;

The training part 803 is configured to train the deconvolution reconstruction network 802 through the training data composed of the existing measured telemetry signal data and the corresponding time stamp of the measured telemetry signal data, and obtain the deconvolution reconstruction network 802 after training. rebuild network 802;

The deconvolution reconstruction network 802 is configured to input raw telemetry signal data timestamps to obtain predicted telemetry signal data corresponding to the raw telemetry signal data timestamps;

The calculation part 804 is configured to calculate the distance between the original telemetry signal data corresponding to the original telemetry signal data time stamp and the predicted telemetry signal data;

The determination part 805 is configured to compare the distance with a set threshold, and when the distance is greater than the set threshold, determine that the original telemetry signal data corresponding to the time stamp of the original telemetry signal data is wild value, and trigger the removal part 806;

The removing part 806 is configured to remove the outliers.

In the above scheme, the deconvolution reconstruction network 802 includes: an input layer, a batch normalized fully connected FC (BN) layer, a first reconstruction layer, a first convolution layer, and a linear unit activation based on batch normalization and correction The second convolutional layer of the function, the first deconvolutional layer, the second deconvolutional layer based on batch normalization and modified linear unit activation function, the third deconvolutional layer, the second reconstruction layer and the output of the full connection layer.

In the above scheme, the output shape of the input layer is 1, and the number of parameters is 0; the output shape of the FC (BN) layer is 16, and the number of parameters is 96; the output shape of the first reconstruction layer is (4,4 ,1), the number of parameters is 0; the output shape of the first convolutional layer is (4,4,16), and the number of parameters is 32; the second convolutional layer based on batch normalization and modified linear unit activation function The output shape of the first deconvolution layer is (4,4,32), and the number of parameters is 672; the output shape of the first deconvolution layer is (4,4,32), and the number of parameters is 1056; the line type based on batch normalization and correction The output shape of the second deconvolution layer of the unit activation function is (4,4,16), and the number of parameters is 592; the output shape of the third deconvolution layer is (4,4,1), and the number of parameters is 17; the output shape of the second reconstruction layer is 16, and the number of parameters is 0; the output shape of the fully connected output layer is 1, and the number of parameters is 17.

In the above solution, the training part 803 is configured to:

For the initialized deconvolution reconstruction network 802, the time stamp of the measured telemetry signal data is input to the deconvolution reconstruction network 802 as input data;

According to the output of the deconvolution reconstruction network 802 and the measured telemetry signal data corresponding to the time stamp of the measured telemetry signal data, the deconvolution reconstruction network 802 is trained, and the deconvolution reconstruction network 802 is determined to include parameters corresponding to each layer to obtain the trained deconvolution reconstruction network 802 .

In the above scheme, during the training process of the deconvolution reconstruction network 802, the network training parameters are set as: learning rate=0.01, the optimization method is the Adam algorithm, and the learning rate decay rule is that if the training set is not updated for 5 generations Then the learning rate is reduced to half of the original.

In the above solution, referring to Fig. 9, the device further includes: a determining part 807 configured to determine the The threshold for value determination.

In the above solution, the calculation part 804 is configured to calculate the difference between the original telemetry signal data corresponding to the time stamp of the original telemetry signal data and the predicted telemetry signal data.

Understandably, in this embodiment, a "part" may be a part of a circuit, a part of a processor, a part of a program or software, etc., and of course it may also be a unit, or a module or non-modular.

In addition, each component in this embodiment may be integrated into one processing unit, each unit may exist separately physically, or two or more units may be integrated into one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software function modules.

If the integrated unit is implemented in the form of a software function module and is not sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this embodiment is essentially or It is said that the part that contributes to the prior art or the whole or part of the technical solution can be embodied in the form of a software product, the computer software product is stored in a storage medium, and includes several instructions to make a computer device (which can It is a personal computer, a server, or a network device, etc.) or a processor (processor) that executes all or part of the steps of the method described in this embodiment. The aforementioned storage medium includes: various media that can store program codes such as U disk, mobile hard disk, read-only memory (ROM, Read Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or optical disk.

Therefore, this embodiment provides a computer storage medium, the computer storage medium is stored with a telemetry signal outlier removal program based on a deconvolution reconstruction network, and the telemetry signal outlier removal program based on a deconvolution reconstruction network is At least one processor implements the steps of the method for removing outliers of telemetry signals based on the deconvolution reconstruction network described in the above technical solution when executed.

According to the above-mentioned telemetry signal wild value removal device 80 based on the deconvolution reconstruction network and the computer storage medium, see FIG. The specific hardware structure of the computing device 100 of the value removal device 80, which may be a wireless device, a mobile or cellular phone (including a so-called smart phone), a personal digital assistant (PDA), a video game console (including a video display, mobile video game devices, mobile video conferencing units), laptop computers, desktop computers, television set-top boxes, tablet computing devices, e-book readers, fixed or mobile media players, etc. The computing device 100 includes: a communication interface 1001 , a memory 1002 and a processor 1003 ; each component is coupled together through a bus system 1004 . It can be understood that the bus system 1004 is used to realize connection and communication between these components. In addition to the data bus, the bus system 1004 also includes a power bus, a control bus and a status signal bus. However, the various buses are labeled as bus system 1004 in FIG. 10 for clarity of illustration. in,

The communication interface 1001 is used for receiving and sending signals during the process of sending and receiving information with other external network elements;

The memory 1002 is configured to store computer programs that can run on the processor 1003;

The processor 1003 is configured to, when running the computer program, execute the steps of the method for removing outliers of telemetry signals based on the deconvolution reconstruction network 802 in the above technical solution.

It can be understood that the memory 1002 in this embodiment of the present invention may be a volatile memory or a nonvolatile memory, or may include both volatile and nonvolatile memories. Among them, the non-volatile memory can be read-only memory (Read-Only Memory, ROM), programmable read-only memory (Programmable ROM, PROM), erasable programmable read-only memory (Erasable PROM, EPROM), electronically programmable Erase Programmable Read-Only Memory (Electrically EPROM, EEPROM) or Flash. The volatile memory can be Random Access Memory (RAM), which acts as external cache memory. By way of illustration and not limitation, many forms of RAM are available, such as Static Random Access Memory (Static RAM, SRAM), Dynamic Random Access Memory (Dynamic RAM, DRAM), Synchronous Dynamic Random Access Memory (Synchronous DRAM, SDRAM), double data rate synchronous dynamic random access memory (Double DataRate SDRAM, DDRSDRAM), enhanced synchronous dynamic random access memory (Enhanced SDRAM, ESDRAM), synchronous connection dynamic random access memory (Synchlink DRAM, SLDRAM) and Direct Memory Bus Random Access Memory (Direct Rambus RAM, DRRAM). Memory 1002 of the systems and methods described herein is intended to include, but is not limited to, these and any other suitable types of memory.

The processor 1003 may be an integrated circuit chip with signal processing capability. In the implementation process, each step of the above method may be completed by an integrated logic circuit of hardware in the processor 1003 or instructions in the form of software. The above-mentioned processor 1003 may be a general-purpose processor, a digital signal processor (Digital Signal Processor, DSP), an application specific integrated circuit (Application Specific Integrated Circuit, ASIC), a field programmable gate array (Field Programmable Gate Array, FPGA) or other Programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. Various methods, steps and logic block diagrams disclosed in the embodiments of the present invention may be implemented or executed. A general-purpose processor may be a microprocessor, or the processor may be any conventional processor, or the like. The steps of the methods disclosed in the embodiments of the present invention may be directly implemented by a hardware decoding processor, or implemented by a combination of hardware and software modules in the decoding processor. The software module can be located in a mature storage medium in the field such as random access memory, flash memory, read-only memory, programmable read-only memory or electrically erasable programmable memory, register. The storage medium is located in the memory 1002, and the processor 1003 reads the information in the memory 1002, and completes the steps of the above method in combination with its hardware.

It should be understood that the embodiments described herein may be implemented by hardware, software, firmware, middleware, microcode or a combination thereof. For hardware implementation, the processing unit can be implemented in one or more application specific integrated circuits (Application Specific Integrated Circuits, ASIC), digital signal processor (Digital Signal Processing, DSP), digital signal processing device (DSP Device, DSPD), programmable logic Device (Programmable Logic Device, PLD), Field-Programmable Gate Array (Field-Programmable Gate Array, FPGA), general-purpose processor, controller, microcontroller, microprocessor, other electronic units for performing the functions described in this application or a combination thereof.

For a software implementation, the techniques described herein can be implemented through modules (eg, procedures, functions, and so on) that perform the functions described herein. Software codes can be stored in memory and executed by a processor. Memory can be implemented within the processor or external to the processor.

It can be understood that the above-mentioned exemplary technical solutions of the deconvolution reconstruction network-based telemetry signal outlier removal device 80 and the computing device 100 belong to the same concept as the aforementioned technical solutions of the deconvolution reconstruction network-based telemetry signal outlier removal method Therefore, for the details not described in detail in the technical solution of the above-mentioned telemetry signal outlier removal device 80 based on the deconvolution reconstruction network and the technical solution of the computing device 100, you can refer to the above-mentioned telemetry signal outlier removal method based on the deconvolution reconstruction network A description of the technical solution. This embodiment of the present invention does not describe it in detail.

It should be noted that: the technical solutions described in the embodiments of the present invention can be combined arbitrarily if there is no conflict.

The above is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Anyone skilled in the art can easily think of changes or substitutions within the technical scope disclosed in the present invention. Should be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention should be determined by the protection scope of the claims.

Claims (8)

1. A method for removing outliers of telemetry signals based on deconvolution reconstruction network, characterized in that, comprising: Construct a deconvolution reconstruction network based on convolution construction and deconvolution reconstruction; The deconvolution reconstruction network is trained by using the training data composed of the existing measured telemetry signal data and the corresponding time stamp of the measured telemetry signal data to obtain the deconvolution reconstruction network after training; inputting raw telemetry signal data timestamps into the trained deconvolution reconstruction network to obtain predicted telemetry signal data corresponding to the raw telemetry signal data timestamps; calculating the distance between the original telemetry signal data corresponding to the original telemetry signal data time stamp and the predicted telemetry signal data; When the distance is greater than a set threshold, determine that the original telemetry signal data corresponding to the time stamp of the original telemetry signal data is an outlier value, and remove the outlier value; Wherein, the deconvolution reconstruction network includes: an input layer, a batch normalized fully connected FC (BN) layer, a first reconstruction layer, a first convolutional layer, a second linear unit activation function based on batch normalization and correction. Convolution layer, first deconvolution layer, second deconvolution layer based on batch normalization and modified linear unit activation function, third deconvolution layer, second reconstruction layer and fully connected output layer; The deconvolution reconstruction network is trained by using the training data composed of the existing measured telemetry signal data and the corresponding time stamp of the measured telemetry signal data to obtain the deconvolution reconstruction network after training, including: For the initialized deconvolution reconstruction network, the time stamp of the measured telemetry signal data is input to the deconvolution reconstruction network as input data; According to the output of the deconvolution reconstruction network and the measured telemetry signal data corresponding to the time stamp of the measured telemetry signal data, the deconvolution reconstruction network is trained, and each layer included in the deconvolution reconstruction network is determined. Corresponding parameters to obtain the trained deconvolution reconstruction network. 2. The method according to claim 1, wherein the output shape of the input layer is 1, and the number of parameters is 0; the output shape of the FC (BN) layer is 16, and the number of parameters is 96; the first The output shape of the reconstruction layer is (4,4,1), and the number of parameters is 0; the output shape of the first convolutional layer is (4,4,16), and the number of parameters is 32; the line based on batch normalization and correction The output shape of the second convolutional layer of the type unit activation function is (4,4,32), and the number of parameters is 672; the output shape of the first deconvolution layer is (4,4,32), and the number of parameters is 1056; the output shape of the second deconvolution layer based on batch normalization and modified linear unit activation function is (4,4,16), and the number of parameters is 592; the output shape of the third deconvolution layer is (4 ,4,1), the number of parameters is 17; the output shape of the second reconstruction layer is 16, and the number of parameters is 0; the output shape of the fully connected output layer is 1, and the number of parameters is 17. 3. The method according to claim 1, characterized in that, in the process of training the deconvolution reconstruction network, the network training parameters are set to: learning rate=0.01, optimization method is Adam algorithm, learning rate decay The rule is that if the training set is not updated for 5 generations, the learning rate will be reduced to half of the original. 4. The method according to claim 1, characterized in that, the method further comprises: according to the subdivision histogram and the kernel density estimation curve of the distance between the outlier and the normal telemetry signal data, determine to remove the outlier The threshold of the decision. 5. The method according to claim 1, wherein the calculating the distance between the original telemetry signal data corresponding to the time stamp of the original telemetry signal data and the predicted telemetry signal data comprises: calculating a difference between raw telemetry signal data corresponding to the raw telemetry signal data timestamp and the predicted telemetry signal data. 6. A device for removing outliers of telemetry signals based on a deconvolution reconstruction network, characterized in that the device includes: a construction part, a deconvolution reconstruction network, a training part, a calculation part, a judgment part and a removal part; wherein, The construction part is configured to construct a deconvolution reconstruction network based on convolution construction and deconvolution reconstruction; The training part is configured to train the deconvolution reconstruction network through training data composed of existing measured telemetry signal data and corresponding time stamps of the measured telemetry signal data, to obtain a trained deconvolution reconstruction network ; said deconvolution reconstruction network for inputting raw telemetry signal data timestamps to obtain predicted telemetry signal data corresponding to said raw telemetry signal data timestamps; the calculation section configured to calculate a distance between raw telemetry signal data corresponding to a timestamp of the raw telemetry signal data and the predicted telemetry signal data; The determination part is configured to compare the distance with a set threshold, and when the distance is greater than the set threshold, determine that the original telemetry signal data corresponding to the original telemetry signal data time stamp is an outlier , and trigger the removal part; The removal part is configured to remove the outlier value; Wherein, the deconvolution reconstruction network includes: an input layer, a batch normalized fully connected FC (BN) layer, a first reconstruction layer, a first convolutional layer, a second linear unit activation function based on batch normalization and correction. Convolution layer, first deconvolution layer, second deconvolution layer based on batch normalization and modified linear unit activation function, third deconvolution layer, second reconstruction layer and fully connected output layer; The deconvolution reconstruction network is trained by using the training data composed of the existing measured telemetry signal data and the corresponding time stamp of the measured telemetry signal data to obtain the deconvolution reconstruction network after training, including: For the initialized deconvolution reconstruction network, the time stamp of the measured telemetry signal data is input to the deconvolution reconstruction network as input data; According to the output of the deconvolution reconstruction network and the measured telemetry signal data corresponding to the time stamp of the measured telemetry signal data, the deconvolution reconstruction network is trained, and each layer included in the deconvolution reconstruction network is determined. Corresponding parameters to obtain the trained deconvolution reconstruction network. 7. A computing device, characterized in that the computing device comprises: a communication interface, a memory and a processor, and each component is coupled together through a bus system; wherein, The communication interface is used for receiving and sending signals during the process of sending and receiving information with other external network elements; said memory for storing a computer program capable of running on said processor; The processor is configured to, when running the computer program, execute the steps of the method for removing outliers of telemetry signals based on a deconvolution reconstruction network according to any one of claims 1 to 5. 8. A computer storage medium, characterized in that, the computer storage medium is stored with a telemetry signal outlier removal program based on a deconvolution reconstruction network, and the telemetry signal outlier removal program based on a deconvolution reconstruction network is at least A processor implements the steps of implementing the method for removing outliers of telemetry signals based on the deconvolution reconstruction network according to any one of claims 1 to 5.

Images