Automobile variable completion method and system based on destabilized convolution self-attention

By applying a destationary convolutional self-attention mechanism in the operation data of new energy vehicles, the problem of failure to fully consider time dependence and multivariate relationships in the existing technology is solved, and a more efficient data completion effect is achieved. Improves the accuracy and reliability of missing values.

CN118820220BActive Publication Date: 2024-11-19AUTOMOTIVE DATA OF CHINA (TIANJIN) CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411305496.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-19
Publication Date
2024-11-19
Estimated Expiration
2044-09-19

AI Technical Summary

Technical Problem

When processing missing values ​​in the operating data of new energy vehicles, the prior art fails to fully consider the complex relationship between time dependence and multivariables, resulting in poor data completion effect.

Method used

The automotive multivariable time sequence data missing value completion method (MDCformer) based on destationary convolutional self-attention is adopted. Through multi-time processing and destationary convolutional self-attention mechanism, the fusion characteristics of the automobile operation data are extracted and the data is carried out. Complete.

Benefits of technology

In experiments on two real vehicle data sets, MDCformer significantly outperforms the multivariable timing data filling method in recent years, improving the accuracy and reliability of missing values.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118820220B_ABST
    Figure CN118820220B_ABST
Patent Text Reader

Abstract

The present application relates to the field of data control technology, and to a method and system for completing automobile variables based on destabilized convolutional self-attention, which obtains original vehicle variables, completes and synchronizes the vehicle component variable sequence to be verified of the fused vehicle component to the original vehicle variables; the fused vehicle component obtains the vehicle component variable sequence to be verified of other similar vehicle components from the original vehicle variables as a missing warning reference for the fused vehicle component; when the fused vehicle component is missing, the vehicle component variables of the fused vehicle component are queried to see if there is a vehicle component variable sequence to be verified of the stabilized vehicle component that meets the first similarity condition, and the vehicle component variable sequence to be verified of the stabilized vehicle component is obtained as a reference to be verified for the fused vehicle component; the present invention shows that the MDCformer model has better performance in experiments on nine missing rates of two real vehicle operation data sets, and is suitable for completing multivariate time series data of automobile operation.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of data control technology, and in particular to an automobile variable completion method and system based on destabilized convolutional self-attention. Background Art

[0002] With the popularity of new energy vehicles, people are paying more and more attention to their safety. At the same time, the large amount of data generated by new energy vehicles has become the key to research and management. These data include vehicle performance indicators, driving behavior, battery status and other aspects, which are of great significance for evaluating vehicle safety, optimizing driving behavior, and improving energy efficiency. However, in the operation data of new energy vehicles, data missing is often encountered due to sensor failure, data transmission errors, equipment damage or human operation. Missing values ​​will reduce the amount of information that can be mined in the sequence and increase the challenge of modeling the time series operation data of automobiles. Therefore, how to effectively fill the missing operation data is an urgent problem to be solved. Traditional filling methods include mean filling, median filling, nearest neighbor interpolation, etc., which ignore the characteristics of irregular periodicity, local global correlation and inherent non-stationarity of operation data.

[0003] At present, the methods for processing missing values ​​include deletion, statistical methods, and machine learning methods. The deletion method may delete valuable information, thereby affecting the error of the experimental results. Traditional statistical methods such as mean completion and exponential smoothing attempt to estimate missing values ​​based on the general statistical characteristics of global or local time series, but ignore the time dependency between time series. Traditional machine learning methods usually use complex techniques to obtain statistical information of time series, such as K-nearest neighbor (KNN), matrix factorization (MF), etc.

[0004] However, in multivariate time series analysis, it is necessary to consider temporal dependencies, which are mostly not adequately considered by these methods.

[0005] Traditional methods for handling missing values, such as mean imputation, exponential smoothing, and K-nearest neighbor (KNN), often fail to fully consider the time dependence in time series and the complex relationship between multiple variables.

[0006] To address this problem, the present invention combines destabilization and convolutional self-attention and proposes a missing value completion method for automobile multivariate time series data based on multi-cycle destabilization convolutional self-attention (Multi-DestabilizationConvolution Transformer, MDCformer). This method can better extract the fusion features of automobile operation data through the multi-time processing method of data and the destabilization convolutional self-attention mechanism. On two real vehicle data, the method of the present invention is superior to the multivariate time series data filling method in recent years. Summary of the invention

[0007] To achieve the above objectives, this application provides the following technical solutions:

[0008] According to a first aspect of the present invention, the present invention claims a method for completing automobile variables based on destabilized convolutional self-attention, which is used for preventing missing data of multiple vehicle components, comprising the steps of:

[0009] Obtaining original vehicle variables, each vehicle component having a vehicle component variable, the original vehicle variables having a to-be-verified variable record of each vehicle component variable, and the plurality of vehicle component variables respectively having a to-be-verified recorded vehicle component variable sequence of a fused vehicle component and a stabilized vehicle component;

[0010] The plurality of vehicle component variables complete and synchronize the sequence of vehicle component variables to be verified of the fused vehicle component to the original vehicle variables according to the first position sequence;

[0011] The fused vehicle component obtains the to-be-verified recorded vehicle component variable sequences of other similar vehicle components from the original vehicle variables according to the second position sequence as a missing warning reference for the fused vehicle component;

[0012] When the fused vehicle component is missing, query whether there is a sequence of vehicle component variables to be verified of the stabilized vehicle component that meets the first similarity condition in the vehicle component variables of the fused vehicle component;

[0013] When there is a to-be-verified recorded vehicle component variable sequence of the stabilized vehicle component that meets the first similarity condition, obtaining the to-be-verified recorded vehicle component variable sequence of the stabilized vehicle component as a to-be-verified reference of the fused vehicle component;

[0014] When there is no vehicle component variable sequence to be verified of the stabilized vehicle component that meets the first similarity condition, vehicle component variables of other reference vehicle components that meet the first similarity condition and are missing are obtained from the original vehicle variables, and the vehicle component variable sequence to be verified of the reference vehicle component is extracted as the reference to be verified of the fused vehicle component, and the vehicle component variable sequence to be verified of the stabilized vehicle component of the vehicle component variables of the fused vehicle component is synchronized;

[0015] When there is a missing fused vehicle component and the vehicle component variable sequence of the to-be-verified recorded vehicle component of the stabilized vehicle component does not exist in the vehicle component variable of the fused vehicle component, the location of the missing component is obtained, the vehicle component variable sequence of the to-be-verified recorded vehicle component of each vehicle component is obtained from the original vehicle variable, and the missing vehicle component that meets the second similarity condition is used as the stabilized vehicle component of the fused vehicle component and stored in the vehicle component variable of the fused vehicle component.

[0016] Further, the original vehicle variables are obtained, the original vehicle variables include a plurality of vehicle component variables, the original vehicle variables have variable records to be verified for each of the vehicle component variables, and the plurality of vehicle component variables respectively have vehicle component variable sequences to be verified for the fused vehicle component and the stabilized vehicle component, specifically including:

[0017] Each vehicle component obtains a missing existence record, encapsulates it into a missing history tuple, and uses a set of multiple missing history tuples as a vehicle component variable sequence to be verified for the fused vehicle component;

[0018] Acquiring vehicle component variables of the fused vehicle component according to the sequence of recorded vehicle component variables to be verified of the fused vehicle component;

[0019] Opening up independent storage space in the vehicle component variables for subsequent synchronization of the sequence of vehicle component variables to be verified and recorded for stabilizing the vehicle components;

[0020] The vehicle component variables of each vehicle component upload their own sequence of recorded vehicle component variables to be verified to the distributed central data, and the distributed center obtains the original vehicle variables with the summary of the sequence of recorded vehicle component variables to be verified uploaded by each vehicle component as the basis for sharing the information to be verified of each vehicle component.

[0021] Further, when the plurality of vehicle component variables complete and synchronize the sequence of vehicle component variables to be verified of the fused vehicle component to the original vehicle variables according to the first position sequence, it includes:

[0022] Multi-period missing data fusion features, static normalization of sequence data, and improved destabilization convolutional self-attention layer in the encoding layer;

[0023] For time length The time column is processed into multiple periods according to days, weeks, months and years, and connected with the masked sequence of the original sequence without the time column to generate ;

[0024] Based on the sequence mean and standard deviation, After normalization, we get ;

[0025] After normalization, the sequence , is input into an encoder for encoding, wherein the encoder includes a destabilizing convolutional self-attention layer;

[0026] Decode and denormalize the encoded data to get the padded data .

[0027] Further, the plurality of vehicle component variables complete and synchronize the sequence of vehicle component variables to be verified of the fused vehicle component to the original vehicle variables according to the first position sequence, specifically including:

[0028] After the fusion vehicle component has built the vehicle component variable, the first sensor of the missing cycle is started, and when the first position sequence is reached, the fusion variable monitors whether there is a new missing in the vehicle component variable in the latest cycle;

[0029] When there is a new missing in the latest cycle, the missing cycle sensor is reset and the existing new missing data is used as the vehicle component variable sequence to be verified and recorded for the fused vehicle component;

[0030] When there is no new loss in the latest cycle, the first sensor of the loss cycle is adjusted based on the first ratio, and when the third position sequence is reached, the fusion variable monitors whether there is a new loss in the latest cycle of the vehicle component variable.

[0031] Further, the fused vehicle component obtains the to-be-verified recorded vehicle component variable sequence of other similar vehicle components from the original vehicle variables according to the second position sequence as a missing warning reference for the fused vehicle component, specifically including: after the fused vehicle component has built the vehicle component variable, starting the missing cycle second sensor, and when the second position sequence is reached, obtaining the summary of the to-be-verified recorded vehicle component variable sequence in the current position and the original vehicle variables;

[0032] When the summary indicates that there is a missing vehicle component within a preset range that matches the current position, obtaining a vehicle component variable sequence of the to-be-verified record of the vehicle component as a missing warning reference for the fused vehicle component;

[0033] When the summary indicates that there are no missing vehicle components within a preset range that matches the current position, the second sensor of the missing period is adjusted based on the first ratio, and when the fourth position sequence is reached, a summary of the current position and the sequence of recorded vehicle component variables to be verified in the original vehicle variables is obtained.

[0034] Furthermore, when the fused vehicle component is missing, querying whether there is a sequence of vehicle component variables to be verified of the stabilized vehicle component that meets the first similarity condition in the vehicle component variables of the fused vehicle component specifically includes:

[0035] When there is a missing component in the fused vehicle, obtaining the missing location and position;

[0036] The vehicle component variable sequence to be verified of the recorded vehicle component of the smoothed vehicle component in the fused variable is matched based on the missing existing position and the existing position.

[0037] Further, when there is a variable sequence of recorded vehicle components to be verified of a stabilized vehicle component that meets the first similarity condition, obtaining the variable sequence of recorded vehicle components to be verified of the stabilized vehicle component as a reference to be verified of the fused vehicle component specifically includes:

[0038] When the missing existing position and the existing position match the to-be-verified recorded vehicle component variable sequence of the stabilized vehicle component in the fusion variable, obtaining the to-be-verified recorded vehicle component variable sequence of the stabilized vehicle component;

[0039] The prevention and control measures in the variable sequence of the vehicle component to be verified of the stabilized vehicle component are extracted as the reference to be verified of the fused vehicle component.

[0040] Further, when there is no vehicle component variable sequence to be verified of the stabilized vehicle component that meets the first similarity condition, vehicle component variables of other reference vehicle components that meet the first similarity condition and are missing are obtained from the original vehicle variables, and the vehicle component variable sequence to be verified of the reference vehicle component is extracted as the reference to be verified of the fused vehicle component and the stabilized vehicle component variable sequence to be verified of the vehicle component variables of the fused vehicle component is synchronized, specifically including:

[0041] When the missing existing position and the existing position do not match the to-be-verified recorded vehicle component variable sequence of the stabilized vehicle component in the fused variable, issuing an access request to the original vehicle variable;

[0042] The original vehicle variable searches the summary of the to-be-verified recorded vehicle component variable sequence to determine whether there is a vehicle component that matches the missing existence position and existence location;

[0043] When it exists, the original vehicle variable issues a vehicle component variable access request to the matching vehicle component, and the vehicle component variable shares the to-be-verified recorded vehicle component variable sequence to the fused vehicle component;

[0044] The fused vehicle component synchronizes the vehicle component variables of the fused vehicle component according to the vehicle component variable sequence to be verified and stabilizes the vehicle component variable sequence to be verified.

[0045] Further, when there is a missing fused vehicle component and there is no vehicle component variable sequence to be verified of the stabilized vehicle component in the vehicle component variable of the fused vehicle component, the missing position is obtained, the vehicle component variable sequence to be verified of each vehicle component is obtained from the original vehicle variable, and the missing vehicle component that meets the second similarity condition is used as the stabilized vehicle component of the fused vehicle component and stored in the vehicle component variable of the fused vehicle component, which specifically includes:

[0046] After the fusion vehicle component completes the construction of the vehicle component variable, the independent storage space is empty, and the vehicle component variable sequence to be verified of the stabilized vehicle component does not exist in the vehicle component variable;

[0047] When there is a missing part in the fused vehicle component, the missing part's location is obtained, and a sequence of vehicle component variables to be verified of each vehicle component is obtained from the original vehicle variables;

[0048] Extract the existence position and the existence position of the variable sequence of the vehicle component to be verified, and match them with the existence position of the fused vehicle component and the existence position of the missing component;

[0049] The matched vehicle components are used as the stabilized vehicle components of the fused vehicle components and the corresponding vehicle component variables of the fused vehicle components are stored in the sequence of recorded vehicle components variables to be verified.

[0050] According to the second aspect of the present invention, the present invention claims a vehicle variable completion system based on destabilized convolutional self-attention, which is used for missing data prevention of multiple vehicle components, including original vehicle variables located at the center of a distributed architecture, each vehicle component having a vehicle component variable; the original vehicle variable has a to-be-verified variable record of each of the vehicle component variables, and the multiple vehicle component variables respectively have a to-be-verified recorded vehicle component variable sequence of a fused vehicle component and a stabilized vehicle component; there is a communication line between the fused vehicle component and the original vehicle variable;

[0051] The automobile variable completion system based on destabilizing convolutional self-attention is used to execute the automobile variable completion method based on destabilizing convolutional self-attention.

[0052] The present application relates to the field of data control technology, and to a method and system for completing automobile variables based on destabilized convolutional self-attention, which obtains original vehicle variables, completes and synchronizes the vehicle component variable sequence to be verified of the fused vehicle component to the original vehicle variables; the fused vehicle component obtains the vehicle component variable sequence to be verified of other similar vehicle components from the original vehicle variables as a missing warning reference for the fused vehicle component; when the fused vehicle component is missing, the vehicle component variables of the fused vehicle component are queried to see if there is a vehicle component variable sequence to be verified of the stabilized vehicle component that meets the first similarity condition, and the vehicle component variable sequence to be verified of the stabilized vehicle component is obtained as a reference to be verified for the fused vehicle component; the present invention shows that the MDCformer model has better performance in experiments on nine missing rates of two real vehicle operation data sets, and is suitable for completing multivariate time series data of automobile operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 A workflow diagram of a method for completing automobile variables based on destabilized convolutional self-attention as claimed in an embodiment of the present application;

[0054] Figure 2 A second workflow diagram of a method for completing automobile variables based on destabilized convolutional self-attention as claimed in an embodiment of the present application;

[0055] Figure 3 A third workflow diagram of a method for completing automobile variables based on destabilized convolutional self-attention as claimed in an embodiment of the present application;

[0056] Figure 4 A fourth workflow diagram of a method for completing automobile variables based on destabilized convolutional self-attention as claimed in an embodiment of the present application;

[0057] Figure 5 The fifth workflow diagram of a method for automobile variable completion based on destabilized convolutional self-attention requested for protection in an embodiment of the present application. DETAILED DESCRIPTION

[0058] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0059] The terms "first", "second" and "third" in this application are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features. Therefore, the features defined as "first", "second" and "third" can explicitly or implicitly include at least one of the features. In the description of this application, the meaning of "multiple" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined. All directional indications (such as up, down, left, right, front, back...) in the embodiments of this application are only used to explain the relative position relationship, movement, etc. between the components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication also changes accordingly. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but optionally also includes steps or units that are not listed, or optionally also includes other steps or units inherent to these processes, methods, products or devices.

[0060] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0061] According to the first aspect of the present invention, referring to the attached Figure 1 The present invention claims a method for completing automobile variables based on destabilized convolutional self-attention, which is used for preventing and controlling missing data of multiple vehicle components, and is characterized in that it includes the steps of:

[0062] Obtaining original vehicle variables, each vehicle component having a vehicle component variable, the original vehicle variables having a to-be-verified variable record of each vehicle component variable, and the plurality of vehicle component variables respectively having a to-be-verified recorded vehicle component variable sequence of a fused vehicle component and a stabilized vehicle component;

[0063] The plurality of vehicle component variables complete and synchronize the sequence of vehicle component variables to be verified of the fused vehicle component to the original vehicle variables according to the first position sequence;

[0064] The fused vehicle component obtains the to-be-verified recorded vehicle component variable sequences of other similar vehicle components from the original vehicle variables according to the second position sequence as a missing warning reference for the fused vehicle component;

[0065] When the fused vehicle component is missing, query whether there is a sequence of vehicle component variables to be verified of the stabilized vehicle component that meets the first similarity condition in the vehicle component variables of the fused vehicle component;

[0066] When there is a to-be-verified recorded vehicle component variable sequence of the stabilized vehicle component that meets the first similarity condition, obtaining the to-be-verified recorded vehicle component variable sequence of the stabilized vehicle component as a to-be-verified reference of the fused vehicle component;

[0067] When there is no vehicle component variable sequence to be verified of the stabilized vehicle component that meets the first similarity condition, vehicle component variables of other reference vehicle components that meet the first similarity condition and are missing are obtained from the original vehicle variables, and the vehicle component variable sequence to be verified of the reference vehicle component is extracted as the reference to be verified of the fused vehicle component, and the vehicle component variable sequence to be verified of the stabilized vehicle component of the vehicle component variables of the fused vehicle component is synchronized;

[0068] When there is a missing fused vehicle component and the vehicle component variable sequence of the to-be-verified recorded vehicle component of the stabilized vehicle component does not exist in the vehicle component variable of the fused vehicle component, the location of the missing component is obtained, the vehicle component variable sequence of the to-be-verified recorded vehicle component of each vehicle component is obtained from the original vehicle variable, and the missing vehicle component that meets the second similarity condition is used as the stabilized vehicle component of the fused vehicle component and stored in the vehicle component variable of the fused vehicle component.

[0069] Specifically, refer to Figure 2 , the obtaining of original vehicle variables, the original vehicle variables including a plurality of vehicle component variables, the original vehicle variables having a to-be-verified variable record of each of the vehicle component variables, the plurality of vehicle component variables respectively having a to-be-verified recorded vehicle component variable sequence of a fused vehicle component and a stabilized vehicle component, specifically including:

[0070] Each vehicle component obtains a missing existence record, encapsulates it into a missing history tuple, and uses a set of multiple missing history tuples as a vehicle component variable sequence to be verified for the fused vehicle component;

[0071] Acquiring vehicle component variables of the fused vehicle component according to the sequence of recorded vehicle component variables to be verified of the fused vehicle component;

[0072] Opening up independent storage space in the vehicle component variables for subsequent synchronization of the sequence of vehicle component variables to be verified and recorded for stabilizing the vehicle components;

[0073] The vehicle component variables of each vehicle component upload their own sequence of recorded vehicle component variables to be verified to the distributed central data, and the distributed center obtains the original vehicle variables with the summary of the sequence of recorded vehicle component variables to be verified uploaded by each vehicle component as the basis for sharing the information to be verified of each vehicle component.

[0074] In this embodiment, the specific step of obtaining vehicle component variables also includes:

[0075] Step S1: generating a shared label of the vehicle component variables of the fused vehicle component to be acquired according to a target search parameter common to the vehicle component variables of the fused vehicle component.

[0076] It can be understood that the shared tag in step S1 is a type of parameter tag, indicating a tag that may be used by all parameters.

[0077] Step S2: Generate a privacy label for the vehicle component variable of each fused vehicle component according to the target search parameter specific to the vehicle component variable of the fused vehicle component to be acquired.

[0078] It can be understood that the privacy label in step S2 is also a type of parameter label, which represents a specific label of a certain type of parameter.

[0079] Step S3: Assign the data of the target search parameter to its corresponding shared tag or private tag.

[0080] It is understandable that the number of the shared tags and private tags is 1 or more.

[0081] Step S4: Obtain the distributed variables of the fused vehicle components according to the shared tags, the private tags, and the preset search paths of the shared tags and the private tags.

[0082] From the above description, it can be seen that the present invention provides a method for obtaining distributed variables of fused vehicle components. Based on the fact that the vehicle component business of each vehicle component in the fused vehicle component may have vehicle component differences, a privacy label is set, and a common label is obtained according to the parameters shared by all vehicle components, so as to obtain the distributed variables of the fused vehicle components. Specifically: first, the shared label of the vehicle component variable of the fused vehicle component is generated according to the target search parameter common to the vehicle component variables of the fused vehicle component to be obtained; and the privacy label of the vehicle component variable of the fused vehicle component is generated according to the target search parameter unique to the vehicle component variable of each fused vehicle component to be obtained; then, the data of the target search parameter is assigned to the corresponding shared label or privacy label, and finally the distributed variables of the fused vehicle component are obtained. It can be seen from the above content that the method for obtaining distributed variables of fused vehicle components provided by the present invention realizes unified parameter storage, high compatibility, flexible and changeable parameter mechanism and access mechanism, and maximizes the decoupling between parameter structure and access rules and application code. It not only avoids repeated modification of application code due to parameter adjustment, but also reduces the development complexity of application code and improves fault tolerance. It meets the requirement of quickly responding to business needs through parameter configuration while changing as little or no code as possible.

[0083] Step S4 specifically includes:

[0084] Step S41: obtaining the vehicle component variables of the fused vehicle component according to the shared tags, the private tags and the preset search paths of the shared tags and the private tags.

[0085] Step S42: Obtain the distributed variables of the fused vehicle component according to the vehicle component variables of the plurality of fused vehicle components.

[0086] The method also includes:

[0087] Step S5: storing the shared tag in a thread-level cache.

[0088] It is understandable that the parameters with higher frequency of use are set as shared tags, and the corresponding cache is designed for reading. Due to the relationship between the search path, there is often more than one search node when the parameters are actually used. The amount of parameter data determines that the system cannot load all parameters into the cache at one time, but it is necessary to clarify whether the parameters exist in the cache. Therefore, the acquisition of parameters is actually divided into two steps: first determine whether a certain search node has parameters, and then read the parameter data with this node. The data in the cache is actually divided into two categories: parameter label sets and parameter data. The parameter label set can be preloaded when the system starts (it can also be loaded by type when it is used), while the parameter data will only load a certain record into the cache when it is used. If this parameter record is used again in the future, it only needs to be read through the cache.

[0089] The method also includes:

[0090] Step S6: When the shared tag and the private tag change, the parameter data corresponding to the shared tag and the private tag are synchronized in the vehicle component variables of the fused vehicle component, and the shared tag in the thread-level cache is synchronized.

[0091] In the case of distributed variables, parameter management and maintenance cannot be dispersed to each cluster, and will only be managed uniformly in a certain cluster (hereinafter referred to as the parameter cluster). For example: using the method of fusion cache + central cache + remote service to obtain parameters. When a certain parameter needs to be obtained, read the fusion cache first. If the fusion cache does not exist, read the central cache. If the central cache does not exist, read the parameter cluster. When the parameters of the parameter cluster change, the changes will also be synchronized to the central cache and the fusion cache of each cluster. The cache is actually from the class data: parameter label set, parameter data. Therefore, when the parameter label set changes (such as parameter addition, parameter deletion), in addition to synchronizing the cache of the parameter data, it is also necessary to synchronize the cache of the parameter label set to ensure that all clusters have a complete and accurate parameter label set.

[0092] Specifically, when the plurality of vehicle component variables complete and synchronize the sequence of vehicle component variables to be verified of the fused vehicle component to the original vehicle variables according to the first position sequence, it includes:

[0093] Multi-period missing data fusion features, static normalization of sequence data, and improved destabilization convolutional self-attention layer in the encoding layer;

[0094] For time length The time column is processed into multiple periods according to days, weeks, months and years, and connected with the masked sequence of the original sequence without the time column to generate ;

[0095] Based on the sequence mean and standard deviation, After normalization, we get ;

[0096] After normalization, the sequence , is input into an encoder for encoding, wherein the encoder includes a destabilizing convolutional self-attention layer;

[0097] Decode and denormalize the encoded data to get the padded data .

[0098] In this embodiment, given an incomplete multivariate time series dataset Indicates the length of Dimensional variables, The vector of moments is defined as ,in To indicate where the value is missing in a multivariate time series, given a binary mask matrix , Time variable The mask is ,in ,if If it can be observed, is 1, and the others are 0.

[0099] Multi-period missing data fusion features:

[0100] For time length The time column is processed into multiple periods according to days, weeks, months and years, and connected with the masked sequence of the original sequence without the time column to generate .

[0101] Stationary series:

[0102] First, the mean and standard deviation of non-stationary series changes are processed by sequence normalization, and finally the model output is inversely transformed from the original statistical data by denormalizing the sequence data.

[0103] Normalization: Normalize in the time dimension through sliding window. For each input sequence , and change it by translation and scaling operations to obtain The normalization module formula is as follows:

[0104] ,

[0105] in , , represents element-wise division, Represents element-wise product.

[0106] Denormalization: In the basic model After filling in the missing values, denormalization is used to output the model as ,and and Transform and get the final completion result. The formula is as follows:

[0107]

[0108] Through the two-stage transformation, the basic model will receive smooth inputs that follow a stable distribution and are easy to generalize. This design makes the model equivalent to translation and scaling perturbations of the time series, which is beneficial for multivariate series completion of vehicle operation data.

[0109] Encoder layer:

[0110] (1) Convolutional self-attention mechanism

[0111] In order to prevent missing values ​​from affecting the calculation of weight assignment in the non-stationary Transformer self-attention layer, a convolutional layer is added to capture the information around each feature. The specific calculation process is as follows:

[0112]

[0113] in is obtained by a one-dimensional convolutional layer. The state information of the vector near the moment vector machine is ,

[0114]

[0115] (2) Destabilized Convolutional Self-Attention Layer

[0116] In order to solve the problem of over-stationarity caused by sequence stabilization, that is, the disappearance of the inherent non-stationary information of vehicle operation data, this paper introduces a de-stationary attention mechanism to obtain un-stationary attention. The formula for self-attention is:

[0117] ,

[0118] in They are query, The keys and values ​​of the dimensions, It is done row by row. Assuming that the embedding layer and the feedforward layer are performed separately at each time point, and assuming that each variable of the sequence has the same variance, then is a scalar, , yes The average value along the time dimension. , The same is true. When there is no series stationarization, The input self-attention is , after the sequence is static for:

[0119] ,

[0120] ,

[0121] because and Respectively Repeat this operation for each column and element of The same shift in the row dimension of the input is invariant, so

[0122] ,

[0123] The above formula obtains attention , except from the stationary convolution sequence In addition, the non-stationary information of the sequence static elimination is also obtained . Let the approximate positive scale scalar and displacement vector , through MLP directly never stabilizes Studying Alone and :

[0124] ,

[0125] ,

[0126] That and Shared by all layers of non-stationary attention, the non-stationary attention mechanism is derived from the stationary convolution sequence and non-stationary series learn the temporal correlation and multiply by , and therefore benefits from the predictability of the stationary series and the inherent time dependence of the original series.

[0127] In order to verify the effectiveness of the proposed MDCformer in filling missing values ​​in new energy vehicle operation data, we conducted experiments on a real vehicle operation dataset. The operating vehicles include common models on the market and a development direction of future renewable energy transportation models, namely pure electric vehicles (BEV) and fuel cell vehicles (FCV). 51,807 samples were sampled at an interval of 30 seconds for pure electric vehicles, and 224,000 samples were sampled at an interval of 1 second for fuel cell vehicles. We compared with six state-of-the-art deep learning models on two real new energy vehicle operation datasets. Transformer, Informer, Timesnet, FEDformer, Autoformer, LSTM use recurrent neural networks to model the temporal relationship of data and predict missing values. Experimental results show that the proposed filling model outperforms other advanced benchmark models in terms of accuracy and reliability.

[0128] Specifically, refer to Figure 3 , the plurality of vehicle component variables complete and synchronize the sequence of vehicle component variables to be verified of the fused vehicle component to the original vehicle variables according to the first position sequence, specifically including:

[0129] After the fusion vehicle component has built the vehicle component variable, the first sensor of the missing cycle is started, and when the first position sequence is reached, the fusion variable monitors whether there is a new missing in the vehicle component variable in the latest cycle;

[0130] When there is a new missing in the latest cycle, the missing cycle sensor is reset and the existing new missing data is used as the vehicle component variable sequence to be verified and recorded for the fused vehicle component;

[0131] When there is no new loss in the latest cycle, the first sensor of the loss cycle is adjusted based on the first ratio, and when the third position sequence is reached, the fusion variable monitors whether there is a new loss in the latest cycle of the vehicle component variable.

[0132] In this embodiment, the first period is set according to the probability of missing onset of the fused vehicle component, and the first period is set to 10 days;

[0133] Adjusting the missing period based on the first ratio The first sensor is to compress the period length based on a certain ratio, because when the missing is not sent in one period, the probability of existence in the next period will be greatly increased. In order to deal with it in advance, the compression period length is set;

[0134] In this embodiment, the first ratio is 10%, so the third position sequence is 9 days.

[0135] For details, please refer to the attached Figure 4The fused vehicle component obtains the to-be-verified recorded vehicle component variable sequence of other similar vehicle components from the original vehicle variables according to the second position sequence as a missing warning reference for the fused vehicle component, specifically including:

[0136] After the fused vehicle component has built the vehicle component variable, the missing period second sensor is started, and when the second position sequence is reached, a summary of the current position and the vehicle component variable sequence to be verified in the original vehicle variable is obtained;

[0137] When the summary indicates that there is a missing vehicle component within a preset range that matches the current position, obtaining a vehicle component variable sequence of the to-be-verified record of the vehicle component as a missing warning reference for the fused vehicle component;

[0138] When the summary indicates that there are no missing vehicle components within a preset range that matches the current position, the second sensor of the missing period is adjusted based on the first ratio, and when the fourth position sequence is reached, a summary of the current position and the sequence of recorded vehicle component variables to be verified in the original vehicle variables is obtained.

[0139] In this embodiment, the second period is set according to the severity of the loss episode of the fusion vehicle component, and the first period is set to 20 days;

[0140] Adjust the missing period based on the first ratio. The second sensor is to compress the period length based on a certain ratio. Because if a serious missing is not sent in one period, the probability of it existing in the next period will be greatly increased. In order to deal with it in advance, the compression period length is set;

[0141] In this embodiment, the first ratio is 10%, so the fourth position sequence is 18 days.

[0142] Specifically, when the fused vehicle component is missing, querying whether there is a sequence of vehicle component variables to be verified of the stabilized vehicle component that meets the first similarity condition in the vehicle component variables of the fused vehicle component specifically includes:

[0143] When there is a missing component in the fused vehicle, obtaining the missing location and position;

[0144] The vehicle component variable sequence to be verified of the recorded vehicle component of the smoothed vehicle component in the fused variable is matched based on the missing existing position and the existing position.

[0145] Specifically, when there is a to-be-verified recorded vehicle component variable sequence of a stabilized vehicle component that meets the first similarity condition, obtaining the to-be-verified recorded vehicle component variable sequence of the stabilized vehicle component as a to-be-verified reference for the fused vehicle component specifically includes:

[0146] When the missing existing position and the existing position match the to-be-verified recorded vehicle component variable sequence of the stabilized vehicle component in the fusion variable, obtaining the to-be-verified recorded vehicle component variable sequence of the stabilized vehicle component;

[0147] The prevention and control measures in the variable sequence of the vehicle component to be verified of the stabilized vehicle component are extracted as the reference to be verified of the fused vehicle component.

[0148] A plurality of recorded vehicle component variable sequences to be verified of stabilized vehicle components are obtained by pre-splitting stabilized vehicle components based on preset missing existence positions and existence position groups, and missing data of fused vehicle components are obtained, wherein the recorded vehicle component variable sequences to be verified of each stabilized vehicle component corresponds to a preset missing existence position and existence position group.

[0149] Determine the first tuple in the variable sequence of recorded vehicle components to be verified of the stabilized vehicle component with the greatest similarity to the missing data of the fused vehicle component; calculate the similarity between the variable sequence of recorded vehicle components to be verified of the stabilized vehicle component and the missing data of the fused vehicle component, and obtain a score for characterizing the degree of similarity between the variable sequence of recorded vehicle components to be verified of the stabilized vehicle component and the missing data of the fused vehicle component. The higher the score, the greater the similarity between the variable sequence of recorded vehicle components to be verified of the stabilized vehicle component and the missing data of the fused vehicle component. Among them, the variable sequence of recorded vehicle components to be verified of the stabilized vehicle component corresponding to the highest score is the first tuple with the greatest similarity to the missing data of the fused vehicle component.

[0150] The target missing existence position and existence position group are adjusted, and based on the adjusted target missing existence position and existence position group, the stabilized vehicle component is split to obtain a second tuple. The initial values ​​of the target missing existence position and existence position group are: the missing existence position and existence position group corresponding to the first tuple.

[0151] Compute the similarity between the second tuple and the missing data of the fused vehicle component.

[0152] Determine whether the similarity meets the preset similarity condition. If not, return to the step of adjusting the target missing location and location group. If yes, determine the second tuple as the result of missing data alignment between the stabilized vehicle component and the fused vehicle component.

[0153] Specifically, when there is no vehicle component variable sequence to be verified of the stabilized vehicle component that meets the first similarity condition, vehicle component variables of other reference vehicle components that meet the first similarity condition and are missing are obtained from the original vehicle variables, and the vehicle component variable sequence to be verified of the reference vehicle component is extracted as a reference to be verified of the fused vehicle component, and the stabilized vehicle component variable sequence to be verified of the vehicle component variables of the fused vehicle component is synchronized, specifically including:

[0154] When the missing existing position and the existing position do not match the to-be-verified recorded vehicle component variable sequence of the stabilized vehicle component in the fused variable, issuing an access request to the original vehicle variable;

[0155] The original vehicle variable searches the summary of the to-be-verified recorded vehicle component variable sequence to determine whether there is a vehicle component that matches the missing existence position and existence location;

[0156] When it exists, the original vehicle variable issues a vehicle component variable access request to the matching vehicle component, and the vehicle component variable shares the to-be-verified recorded vehicle component variable sequence to the fused vehicle component;

[0157] The fused vehicle component synchronizes the vehicle component variables of the fused vehicle component according to the vehicle component variable sequence to be verified and stabilizes the vehicle component variable sequence to be verified.

[0158] Specifically, when there is a missing fusion vehicle component and there is no vehicle component variable sequence to be verified of the stabilized vehicle component in the vehicle component variable of the fusion vehicle component, the missing position is obtained, the vehicle component variable sequence to be verified of each vehicle component is obtained from the original vehicle variable, and the missing vehicle component that meets the second similarity condition is used as the stabilized vehicle component of the fusion vehicle component and stored in the vehicle component variable of the fusion vehicle component, which specifically includes:

[0159] After the fusion vehicle component completes the construction of the vehicle component variable, the independent storage space is empty, and the vehicle component variable sequence to be verified of the stabilized vehicle component does not exist in the vehicle component variable;

[0160] When there is a missing part in the fused vehicle component, the missing part's location is obtained, and a sequence of vehicle component variables to be verified of each vehicle component is obtained from the original vehicle variables;

[0161] Extract the existence position and the existence position of the variable sequence of the vehicle component to be verified, and match them with the existence position of the fused vehicle component and the existence position of the missing component;

[0162] The matched vehicle components are used as the stabilized vehicle components of the fused vehicle components and the corresponding vehicle component variables of the fused vehicle components are stored in the sequence of recorded vehicle components variables to be verified.

[0163] According to the second embodiment of the present invention, the present invention claims a vehicle variable completion system based on destabilized convolutional self-attention, which is used for missing data prevention of multiple vehicle components, characterized in that it includes original vehicle variables located at the center of the distributed architecture, each vehicle component has a vehicle component variable; the original vehicle variable has a to-be-verified variable record of each of the vehicle component variables, and the multiple vehicle component variables respectively have a to-be-verified recorded vehicle component variable sequence of a fused vehicle component and a stabilized vehicle component; there is a communication line between the fused vehicle component and the original vehicle variable;

[0164] The automobile variable completion system based on destabilizing convolutional self-attention is used to execute the automobile variable completion method based on destabilizing convolutional self-attention.

[0165] In the several embodiments provided in the present application, it should be understood that the disclosed systems, systems and methods can be implemented in other ways. For example, the system embodiments described above are only schematic. For example, the division of units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of systems or units, which can be electrical, mechanical or other forms.

[0166] In addition, each functional unit in each embodiment of the present application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The above integrated unit can be implemented in the form of hardware or in the form of software functional units. The above is only an implementation method of the present application, and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the specification and drawings of this application, or directly or indirectly used in other related technical fields, is also included in the patent protection scope of the present application.

[0167] The specific implementation methods of the invention are described in detail above, but they are only examples, and the present application is not limited to the specific implementation methods described above. For those skilled in the art, any equivalent modification or substitution of the invention is also within the scope of the present application, and therefore, the equalization, modification, and improvement made without departing from the spirit and principle of the present application should be included in the scope of the present application.

Claims

1. A method for automobile variable completion based on destabilized convolutional self-attention for preventing missing data of multiple vehicle components, characterized in that: Includes steps: Acquire original vehicle variables, each vehicle component has a vehicle component variable, the original vehicle variables have a to-be-verified variable record of each vehicle component variable, and the plurality of vehicle component variables respectively have a to-be-verified recorded vehicle component variable sequence of a fused vehicle component and a stabilized vehicle component; The plurality of vehicle component variables complete and synchronize the sequence of vehicle component variables to be verified of the fused vehicle component to the original vehicle variables according to the first position sequence; The fused vehicle component obtains the to-be-verified recorded vehicle component variable sequences of other similar vehicle components from the original vehicle variables according to the second position sequence as a missing warning reference for the fused vehicle component; When the fused vehicle component is missing, query whether there is a sequence of vehicle component variables to be verified of the stabilized vehicle component that meets the first similarity condition in the vehicle component variables of the fused vehicle component; When there is a to-be-verified recorded vehicle component variable sequence of the stabilized vehicle component that meets the first similarity condition, obtaining the to-be-verified recorded vehicle component variable sequence of the stabilized vehicle component as a to-be-verified reference of the fused vehicle component; When there is no vehicle component variable sequence to be verified of the stabilized vehicle component that meets the first similarity condition, vehicle component variables of other reference vehicle components that meet the first similarity condition and are missing are obtained from the original vehicle variables, and the vehicle component variable sequence to be verified of the reference vehicle component is extracted as the reference to be verified of the fused vehicle component, and the vehicle component variable sequence to be verified of the stabilized vehicle component of the vehicle component variables of the fused vehicle component is synchronized; When there is a missing fused vehicle component and the vehicle component variable sequence of the to-be-verified recorded vehicle component of the stabilized vehicle component does not exist in the vehicle component variable of the fused vehicle component, the location of the missing component is obtained, the vehicle component variable sequence of the to-be-verified recorded vehicle component of each vehicle component is obtained from the original vehicle variable, and the missing vehicle component that meets the second similarity condition is used as the stabilized vehicle component of the fused vehicle component and stored in the vehicle component variable of the fused vehicle component.

2. The automobile variable completion method based on destabilized convolution self-attention according to claim 1, characterized in that: The obtaining of original vehicle variables, wherein the original vehicle variables include a plurality of vehicle component variables, the original vehicle variables have variable records to be verified for each of the vehicle component variables, and the plurality of vehicle component variables respectively have vehicle component variable sequences to be verified for fused vehicle components and stabilized vehicle components, specifically includes: Each vehicle component obtains a missing existence record, encapsulates it into a missing history tuple, and uses a set of multiple missing history tuples as a vehicle component variable sequence to be verified for the fused vehicle component; Acquiring vehicle component variables of the fused vehicle component according to the sequence of recorded vehicle component variables to be verified of the fused vehicle component; Opening up independent storage space in the vehicle component variables for subsequent synchronization of the sequence of vehicle component variables to be verified and recorded for stabilizing the vehicle components; The vehicle component variables of each vehicle component upload their own sequence of recorded vehicle component variables to be verified to the distributed central data, and the distributed center obtains the original vehicle variables with the summary of the sequence of recorded vehicle component variables to be verified uploaded by each vehicle component as the basis for sharing the information to be verified of each vehicle component.

3. The automobile variable completion method based on destabilized convolution self-attention as claimed in claim 2, characterized in that: When the plurality of vehicle component variables complete and synchronize the sequence of vehicle component variables to be verified of the fused vehicle component to the original vehicle variables according to the first position sequence, it includes: Multi-period missing data fusion features, static normalization of sequence data, and improved destabilization convolutional self-attention layer in the encoding layer; For time length The time column is processed into multiple periods according to days, weeks, months and years, and connected with the masked sequence of the original sequence without the time column to generate ; Based on the sequence mean and standard deviation, After normalization, we get ; After normalization, the sequence , is input into an encoder for encoding, wherein the encoder includes a destabilizing convolutional self-attention layer; Decode and denormalize the encoded data to get the padded data .

4. The automobile variable completion method based on destabilized convolution self-attention according to claim 1, characterized in that: The plurality of vehicle component variables complete and synchronize the sequence of vehicle component variables to be verified of the fused vehicle component to the original vehicle variables according to the first position sequence, specifically including: After the fusion vehicle component has built the vehicle component variable, the first sensor of the missing cycle is started, and when the first position sequence is reached, the fusion variable monitors whether there is a new missing in the vehicle component variable in the latest cycle; When there is a new missing in the latest cycle, the missing cycle sensor is reset and the existing new missing data is used as the vehicle component variable sequence to be verified and recorded for the fused vehicle component; When there is no new loss in the latest cycle, the first sensor of the loss cycle is adjusted based on the first ratio, and when the third position sequence is reached, the fusion variable monitors whether there is a new loss in the latest cycle of the vehicle component variable.

5. The automobile variable completion method based on destabilized convolution self-attention as claimed in claim 2, characterized in that: The fusion vehicle component obtains the to-be-verified recorded vehicle component variable sequence of other similar vehicle components from the original vehicle variables according to the second position sequence as a missing warning reference of the fusion vehicle component, specifically including: after the fusion vehicle component has built the vehicle component variable, starting the missing cycle second sensor, and when the second position sequence is reached, obtaining the summary of the to-be-verified recorded vehicle component variable sequence in the current position and the original vehicle variables; When the summary indicates that there is a missing vehicle component within a preset range that matches the current position, obtaining a vehicle component variable sequence of the to-be-verified record of the vehicle component as a missing warning reference for the fused vehicle component; When the summary indicates that there are no missing vehicle components within a preset range that matches the current position, the second sensor of the missing period is adjusted based on the first ratio, and when the fourth position sequence is reached, a summary of the current position and the sequence of recorded vehicle component variables to be verified in the original vehicle variables is obtained.

6. The automobile variable completion method based on destabilized convolution self-attention as claimed in claim 1, characterized in that: When the fused vehicle component is missing, querying whether there is a sequence of vehicle component variables to be verified of the stabilized vehicle component that meets the first similarity condition in the vehicle component variables of the fused vehicle component specifically includes: When there is a missing component in the fused vehicle, obtaining the missing location and position; The vehicle component variable sequence to be verified of the recorded vehicle component of the smoothed vehicle component in the fused variable is matched based on the missing existing position and the existing position.

7. The method for automobile variable completion based on destabilized convolution self-attention as claimed in claim 6, characterized in that: When there is a to-be-verified recorded vehicle component variable sequence of the stabilized vehicle component that meets the first similarity condition, obtaining the to-be-verified recorded vehicle component variable sequence of the stabilized vehicle component as a to-be-verified reference for the fused vehicle component specifically includes: When the missing existing position and the existing position match the to-be-verified recorded vehicle component variable sequence of the stabilized vehicle component in the fusion variable, obtaining the to-be-verified recorded vehicle component variable sequence of the stabilized vehicle component; The prevention and control measures in the variable sequence of the vehicle component to be verified of the stabilized vehicle component are extracted as the reference to be verified of the fused vehicle component.

8. The method for automobile variable completion based on destabilized convolution self-attention as claimed in claim 2, characterized in that: When there is no vehicle component variable sequence to be verified of the stabilized vehicle component that meets the first similarity condition, vehicle component variables of other reference vehicle components that meet the first similarity condition and are missing are obtained from the original vehicle variables, the vehicle component variable sequence to be verified of the reference vehicle component is extracted as the reference to be verified of the fused vehicle component, and the stabilized vehicle component variable sequence to be verified of the vehicle component variables of the fused vehicle component is synchronized, specifically including: When the missing existing position and the existing position do not match the to-be-verified recorded vehicle component variable sequence of the stabilized vehicle component in the fused variable, issuing an access request to the original vehicle variable; The original vehicle variable searches the summary of the to-be-verified recorded vehicle component variable sequence to determine whether there is a vehicle component that matches the missing existence position and existence location; When it exists, the original vehicle variable issues a vehicle component variable access request to the matching vehicle component, and the vehicle component variable shares the to-be-verified recorded vehicle component variable sequence to the fused vehicle component; The fused vehicle component synchronizes the vehicle component variables of the fused vehicle component according to the vehicle component variable sequence to be verified and stabilizes the vehicle component variable sequence to be verified.

9. The method for automobile variable completion based on destabilized convolution self-attention as claimed in claim 2, characterized in that: When there is a missing fusion vehicle component and there is no vehicle component variable sequence to be verified of the stabilized vehicle component in the vehicle component variable of the fusion vehicle component, obtaining the location of the missing fusion vehicle component, obtaining the vehicle component variable sequence to be verified of each vehicle component from the original vehicle variable, and taking the missing vehicle component that meets the second similarity condition as the stabilized vehicle component of the fusion vehicle component and storing it in the vehicle component variable of the fusion vehicle component, specifically includes: After the fusion vehicle component completes the construction of the vehicle component variable, the independent storage space is empty, and the vehicle component variable sequence to be verified of the stabilized vehicle component does not exist in the vehicle component variable; When there is a missing part in the fused vehicle component, the missing part's location is obtained, and a sequence of vehicle component variables to be verified of each vehicle component is obtained from the original vehicle variables; Extract the existence position and the existence position of the variable sequence of the vehicle component to be verified, and match them with the existence position of the fused vehicle component and the existence position of the missing component; The matched vehicle components are used as the stabilized vehicle components of the fused vehicle components and the corresponding vehicle component variables of the fused vehicle components are stored in the sequence of recorded vehicle components variables to be verified.

10. An automotive variable completion system based on destabilized convolutional self-attention for missing data prevention of multiple vehicle components, characterized in that: It includes an original vehicle variable located at the center of the distributed architecture, each vehicle component has a vehicle component variable; the original vehicle variable has a variable record to be verified of each vehicle component variable, and multiple vehicle component variables respectively have a vehicle component variable sequence to be verified of a fused vehicle component and a stabilized vehicle component; a communication line is provided between the fused vehicle component and the original vehicle variable; The automobile variable completion system based on destabilizing convolutional self-attention is used to execute an automobile variable completion method based on destabilizing convolutional self-attention as described in any one of claims 1-9.