Working condition segmentation method and system for battery swap station

Abstract

The application relates to a working condition segmentation method and system for a battery swap station. The working condition segmentation method comprises the following steps: constructing a time series data set related to a battery swap service task, the time series data set comprising a plurality of time window data with the same time window length, the plurality of time window data corresponding to a plurality of continuous time windows; obtaining a plurality of feature data based on the time series data set, the plurality of feature data corresponding to the plurality of continuous time windows; calculating the distance between adjacent feature data corresponding to adjacent time windows based on the plurality of feature data to obtain a distance curve describing the distance; and determining a working condition segmentation point based on an extreme value in the distance curve to obtain a working condition segmentation result. The working condition segmentation method and system can automatically segment battery swap service data according to different working conditions, thereby effectively improving the operation and maintenance efficiency of the battery swap station.

Description

Technical Field

[0001] This application relates to the field of battery swapping station technology, and more specifically, to a method and system for dividing operating conditions for battery swapping stations. Background Technology

[0002] In recent years, with the vigorous development of the new energy industry, new energy vehicles have gradually become a popular choice for people's travel. Unique battery swapping models, such as battery swapping stations, can effectively replenish the energy of new energy vehicles, solve the battery range problem, eliminate range anxiety, and thus further promote green travel.

[0003] To better serve battery-swapping vehicles and facilitate the rapid development of battery-swapping networks, maintenance of battery-swapping stations is frequently required. Current battery-swapping station maintenance systems typically include numerous sensor systems, control systems, and vibration signal systems. However, battery-swapping stations have many mechanical components and complex structures. Furthermore, as the scale of battery swapping increases, the data collected by sensors has long data cycles, large volumes of data, and significant amounts of noise. During the overall management and control of the battery-swapping station maintenance system, maintenance personnel often need to manually review large amounts of data, consuming considerable time to identify the background information of different data points, resulting in a very heavy workload. Summary of the Invention

[0004] The embodiments of this application provide a method and system for dividing operating conditions for battery swapping stations, which can automatically divide battery swapping service data of battery swapping stations according to different operating conditions, thereby effectively improving the operation and maintenance efficiency of battery swapping stations.

[0005] According to one aspect of this application, a method for segmenting operating conditions for a battery swapping station is provided. The method includes: constructing a time-series dataset concerning battery swapping service tasks, the time-series dataset including multiple time window data with the same time window length, the multiple time window data corresponding to multiple consecutive time windows; obtaining multiple characterization data based on the time-series dataset, the multiple characterization data corresponding to the multiple consecutive time windows; calculating the distance between adjacent characterization data corresponding to adjacent time windows based on the multiple characterization data to obtain a distance curve describing the distance; and determining operating condition segmentation points based on the extreme values ​​in the distance curve to obtain an operating condition segmentation result.

[0006] In some embodiments of this application, optionally, constructing a time-series dataset for battery swapping service tasks includes: receiving battery swapping service data recorded in chronological order from the battery swapping station; classifying the battery swapping service data into one or more single-service datasets according to the battery swapping service task, wherein each single-service dataset corresponds to a corresponding battery swapping service task; classifying the single-service dataset into one or more feature datasets according to a feature, wherein each feature dataset corresponds to a corresponding feature; and dividing the feature dataset into multiple time window data with the same time window length to form a corresponding time-series dataset.

[0007] In some embodiments of this application, the element may optionally include a component participating in a battery swapping service task or a shaft within the component.

[0008] In some embodiments of this application, optionally, obtaining the plurality of representation data based on the time-series dataset includes: processing the plurality of time window data to obtain input data s; and mapping the input data s from the original input space to a high-dimensional space to obtain output data f including the plurality of representation data.

[0009] In some embodiments of this application, optionally, processing the plurality of time window data to obtain the input data s includes: filtering the plurality of time window data using an exponential moving average method to obtain the filtered input data s.

[0010] Optionally, in some embodiments of this application, calculating the distance between adjacent representation data corresponding to adjacent time windows includes: for adjacent t-th time windows and t+1-th time windows, determining the representation data f corresponding to the t-th time window. t and the representation data f corresponding to the (t+1)th time window t+1 Where t is a natural number; and the adjacent representation data f is calculated using the following formula. t with f t+1 Dist between t :

[0011]

[0012] In some embodiments of this application, optionally, the plurality of characterization data are filtered to obtain updated plurality of characterization data before calculating the distance between the adjacent characterization data, so that the distance is calculated based on the updated plurality of characterization data.

[0013] In some embodiments of this application, optionally, the working condition segmentation method is implemented at least via a working condition segmentation algorithm module, wherein the working condition segmentation algorithm module includes an autoencoder submodule and a working condition segmentation submodule, the plurality of characterization data are obtained via the autoencoder submodule, and the working condition segmentation result is obtained via the working condition segmentation submodule.

[0014] Optionally, in some embodiments of this application, the autoencoder submodule includes an encoder and a decoder, wherein the encoder is configured to map the input data s from the original input space to the high-dimensional space to obtain the output data f, and the decoder is configured to map the output data f from the high-dimensional space back to the original input space to obtain reconstructed data.

[0015] Optionally, in some embodiments of this application, the mapping function that maps the input data s to the output data f is expressed by the following formula:

[0016] f = En(S) = g en (Ws+b e ),

[0017] Where Ws represents the encoding weight matrix, and b e This represents the encoding bias term; and the output data f is mapped to obtain the reconstructed data. The mapping function is expressed by the following formula:

[0018]

[0019] Where W′f represents the decoding weight matrix, and b d This indicates the decoding bias term.

[0020] In some embodiments of this application, optionally, the encoder and the decoder are based on convolutional neural networks, wherein an attention mechanism is introduced into the convolutional neural network.

[0021] In some embodiments of this application, the autoencoder submodule may optionally include a noise reduction layer configured to introduce noise values ​​into the input data s used for training during the training of the condition segmentation algorithm module.

[0022] Optionally, in some embodiments of this application, the working condition segmentation algorithm module is trained using a training method including the following steps: constructing a training dataset; generating the input data s for training based on the training dataset; and generating reconstructed data for training via the working condition segmentation algorithm module to be trained based on the input data s. The working condition segmentation algorithm module is trained so that the reconstructed data used for training... The error between the input data s used for training and the input data s is minimized.

[0023] Optionally, in some embodiments of this application, the reconstructed data for training is generated by the work condition segmentation algorithm module to be trained based on the input data s used for training. This includes: mapping the input data s used for training from the original input space to a high-dimensional space via the encoder in the work condition segmentation algorithm module to obtain output data f used for training; and mapping the output data f used for training from the high-dimensional space back to the original input space via the decoder in the work condition segmentation algorithm module to obtain reconstructed data used for training.

[0024] In some embodiments of this application, optionally, the reconstructed data used for training The minimization of the error between the input data s used for training and the input data s is expressed based on the following formula:

[0025] minJ(W,b e ,b d ,W′)=∑L(u,v)+α∑W 2 ,

[0026] Where W represents the encoding weight matrix, W′ represents the decoding weight matrix, α represents the weight coefficients, and L(u,v) is the loss function, where u = s and v = D. ec (E nc (s)).

[0027] Optionally, in some embodiments of this application, the reconstructed data for training is generated by the work condition segmentation algorithm module to be trained based on the input data s used for training. It also includes: before mapping the input data s used for training from the original input space to a high-dimensional space via the encoder in the working condition segmentation algorithm module to obtain the output data f used for training, introducing noise values ​​into the input data s used for training via the noise reduction layer in the working condition segmentation algorithm module to be trained.

[0028] According to another aspect of this application, a condition segmentation system for a battery swapping station is provided, the condition segmentation system comprising: a memory configured to store instructions; and a processor configured to execute the instructions such that the condition segmentation system performs any of the condition segmentation methods described above.

[0029] According to another aspect of this application, a battery swapping station is provided, the battery swapping station comprising any of the operating condition segmentation systems described above.

[0030] According to yet another aspect of this application, a computer-readable storage medium is provided, wherein instructions are stored therein, characterized in that, when executed by a processor, the instructions cause the processor to perform any of the working condition segmentation methods described above.

[0031] This application proposes an unsupervised learning-based operating condition segmentation method and system for battery swapping stations. It can achieve end-to-end (i.e., from the battery swapping station end to the operating condition segmentation system end) operating condition segmentation in complex battery swapping scenarios, and has good robustness, detection accuracy and inference efficiency. Attached Figure Description

[0032] The above and other objects and advantages of this application will become more fully clear from the following detailed description taken in conjunction with the accompanying drawings, wherein the same or similar elements are indicated by the same reference numerals.

[0033] Figure 1 A schematic diagram illustrating the association between a battery swapping station and a condition segmentation system according to an embodiment of this application is shown;

[0034] Figure 2 A schematic diagram of a working condition segmentation algorithm module according to an embodiment of this application is shown;

[0035] Figure 3 A schematic diagram of a working condition segmentation method according to an embodiment of this application is shown;

[0036] Figure 4 A schematic diagram of a method for training a working condition segmentation algorithm module according to an embodiment of this application is shown;

[0037] Figure 5 A schematic diagram of a working condition segmentation system according to an embodiment of this application is shown. Detailed Implementation

[0038] For the purposes of brevity and illustrativeness, this document primarily refers to exemplary embodiments thereof to describe the principles of this application. However, those skilled in the art will readily recognize that the same principles can be equivalently applied to all types of operating condition segmentation methods and systems for battery swapping stations, and that these same or similar principles can be implemented therein, without departing from the true spirit and scope of this application.

[0039] The following will combine Figure 1 This application describes the relationship between a battery swapping station 100 and a condition segmentation system 110 in one embodiment.

[0040] Figure 1 A schematic diagram showing the association between a battery swapping station 100 and a condition segmentation system 110 according to an embodiment of this application is shown.

[0041] Figure 1 The illustrated operating condition segmentation system 110 is located outside the battery swapping station 100, and the operating condition segmentation system 110 and the battery swapping station 100 can communicate with each other. Figure 1 In the illustrated embodiment, the working condition segmentation system 110 is not part of the battery swapping station 100. However, this application may also provide a battery swapping station that may include the working condition segmentation system 110. For example, the working condition segmentation system 110 may be set in the battery swapping station 100.

[0042] In some embodiments, the battery swapping station 100 can receive an order for battery swapping service from the vehicle 102 to be swapped, and after the vehicle 102 enters the battery swapping station 100, it can provide battery swapping service to the vehicle 102 based on the order. Since the vehicle 102 receiving the battery swapping service is located inside the battery swapping station 100 during the battery swapping service process, therefore... Figure 1 Vehicle 102 is shown as being in battery swapping station 100. It is understandable that, although... Figure 1 Vehicle 102 is shown as being in battery swapping station 100, but Figure 1 This does not mean that battery swapping station 100 includes vehicle 102.

[0043] like Figure 1 As shown, the battery swapping station 100 may include a sensing system 104. The sensing system 104 may include multiple sensors for collecting operating condition data within the battery swapping station 100. The operating condition data referred to herein refers to data on the operating status of the equipment under different conditions. For the battery swapping service scenario of the battery swapping station 100, the operating condition data may refer to the data related to the battery swapping service task collected by the sensing system 104, i.e., battery swapping service data. The operating condition data may include environmental parameters and load parameters. Accordingly, the battery swapping service data may include not only the operating parameters of the various components providing the battery swapping service in the battery swapping station 100, but also the operating parameters of the vehicle 102 during the battery swapping service. The battery swapping service data may include multiple timestamp data that change with the battery swapping service process, which can be used to characterize the operating condition of the battery swapping station 100 at different times. In other words, the battery swapping service data can change over time and therefore can be a series of time-series data.

[0044] In some embodiments, the interaction between the operating condition segmentation system 110 and the battery swapping station 100 may include: the battery swapping station 100 transmitting battery swapping service data collected by the sensing system 104 to the operating condition segmentation system 110. After receiving the battery swapping service data, the operating condition segmentation system 110 can utilize the operating condition segmentation algorithm module 200 (see...). Figure 2 This method analyzes and processes battery swapping service data to achieve operational condition segmentation. The main purpose of operational condition segmentation is to divide equipment data under different operating states and perform targeted signal processing and feature extraction. Commonly used operational condition segmentation variables include speed parameters, environmental parameters, and load information. Taking the operation phase of vehicle 102 during the battery swapping process as an example, vehicle 102's operation phase includes three processes: acceleration, constant speed, and deceleration. Therefore, speed features can be used as operational condition parameters to segment the data, resulting in three segments. These three segments represent two transient states and one steady state. Different feature extraction methods can be used to process different types of segments.

[0045] From an engineering perspective, battery swapping station 100 is a multi-bearing, multi-sensor system. During battery swapping, the force distribution and movement of each axis differ, resulting in a large volume of data and difficulties in retrieving heterogeneous parameterized features. Without operational condition segmentation of the battery swapping service data, maintenance personnel struggle to map observations to interpretable states and transitions unless significant manpower is invested. Therefore, the overall battery swapping station operation and maintenance system needs to improve data quality to minimize interference for maintenance personnel during data modeling, thereby ensuring efficient execution of maintenance work. Identifying the background information of different data is crucial in this process. The operational condition segmentation system 110 of this application provides an automatic operational condition segmentation algorithm for the battery swapping process. It can automatically and efficiently segment the battery swapping service data of battery swapping station 100 according to different operational conditions and standardize it, thereby facilitating subsequent feature engineering and ensuring the operational efficiency of the battery swapping station.

[0046] Next, we will combine Figure 2 Let's introduce the working condition segmentation algorithm module 200 of the working condition segmentation system 110.

[0047] Figure 2 A schematic diagram of a working condition segmentation algorithm module 200 according to an embodiment of this application is shown. Figure 2 As shown, the working condition segmentation algorithm module 200 may include an autoencoder submodule 210, a data filtering submodule 220, and a working condition segmentation submodule 230.

[0048] The autoencoder submodule 210 can process the battery swapping service data received by the operating condition segmentation system 110 using an algorithm model to obtain a basis for operating condition segmentation. In some embodiments, the autoencoder submodule 210 can identify operating condition breakpoints (i.e., abrupt change points) in complex battery swapping scenarios based on the acquired battery swapping service data, and use these breakpoints as the basis for operating condition segmentation. Therefore, the autoencoder submodule 210 can also be called an abrupt change point detection algorithm module.

[0049] In some embodiments, the autoencoder submodule 210 may include a noise-reducing variational autoencoder. For example, such as Figure 2 As shown, the autoencoder submodule 210 may include a noise reduction layer 212, an encoder 214, and a decoder 216.

[0050] The noise layer 212 can introduce noise values ​​such as extreme values ​​and random noise points during the training of the algorithm model of the working condition segmentation algorithm module 200, making the trained algorithm model more robust.

[0051] Encoder 214 can be used to map input data s from the original input space to a high-dimensional space to obtain output data f. The input data s can be determined based on battery swapping service data collected by sensing system 104, and the output data f can include multiple representation data. In some embodiments, the multiple representation data can be represented as representation vectors, where the representation vectors are further hidden vectors about the input data s. In some embodiments, encoder 214 can be a mapping input data... To output data function g en (·):

[0052] f = En(s) = g en (Ws+b e ).

[0053] Among them, Ws and b e Let g represent the weight matrix and the bias term, respectively. en (·) is a non-linear activation function.

[0054] Decoder 216 can be used to map the hidden vectors from the high-dimensional space back to the original input space to obtain reconstructed data. In some embodiments, the decoder can be a function g that maps the feature representation output of the hidden units back to the original input space. dec (·):

[0055]

[0056] Among them, decoder g dec The structure of (·) can be related to encoder g en The structure of (·) is similar, W′ and b d These represent the weight matrix and bias term of the decoder, respectively. In some embodiments, the activation function for both the encoder and decoder can be the sigmoid function.

[0057] In some embodiments, the encoder 214 and decoder 216 can be formed based on convolutional neural networks (CNNs). In some embodiments, an attention mechanism can be introduced into the CNN. Introducing the attention mechanism into the working condition segmentation algorithm module 200 can assign different weights to sequence features, using a probability allocation method to allow the algorithm model to understand which part of the information in the data is important during training, thereby enabling the algorithm model to pay close attention to this information, deepening the algorithm model's understanding of the interdependence of sequence data, and improving the accuracy of working condition segmentation. For example, by introducing the attention mechanism, the encoder 214 can assign weight values ​​to relevant data points based on the amount of degradation information contained in the data points when extracting temporal features. Optionally, when the amount of degradation information contained in the data points is large, a higher weight value is assigned to the relevant data points, thereby paying more attention to the relevant data points containing more degradation information. The "degradation information" referred to in this document refers to the working condition parameter information associated with components in the battery swapping station 100 that are prone to performance degradation or failure after a long period of battery swapping service. Introducing an attention mechanism to focus more on degradation information helps to monitor components in the swapping station 100 that are prone to failure, providing early warnings for failures in the swapping station 100, thereby ensuring the normal operation of the swapping station 100.

[0058] In other embodiments, encoder 214 and decoder 216 may be formed based on other forms of convolutional layers, including but not limited to grouped convolutions, separable convolutions, dilated convolutions, and depthwise convolutions. Optionally, layers to avoid overfitting may be added to the network architecture, including but not limited to regularization and dropout.

[0059] The data filtering submodule 220 can filter the processed time-series data (e.g., observations from different stages) during the data processing performed by the autoencoder submodule 210. For example, the data filtering submodule 220 can use the exponential moving average method to average a total of 2n+1 observations from previous and subsequent time points to obtain the filtering result for the current time point. The data filtering submodule 220 can suppress a large portion of noise when the actual data does not change significantly, making the filtered result approximate the true value. The data filtering submodule 220 in this paper can smooth the data and is therefore also called a data smoothing module. By using the data filtering submodule 220 to filter and smooth the data signal, it can effectively filter out possible sawtooth signals and more effectively preserve the variation information of the data signal.

[0060] The operating condition segmentation submodule 230 can be configured to segment the battery swapping service data received by the operating condition segmentation system 110 according to different operating conditions based on the processing results of the self-encoder submodule 210.

[0061] Next, we will combine Figure 3 Here is an embodiment of the working condition segmentation method 300 of this application.

[0062] Figure 3 A schematic diagram of a working condition segmentation method 300 according to an embodiment of this application is shown. The working condition segmentation method 300 can utilize reference... Figure 2 The described work condition segmentation algorithm module 200 is used for execution. In some embodiments, the algorithm model of the work condition segmentation algorithm module 200 used in the execution of the work condition segmentation method 300 can be trained by the work condition segmentation algorithm module 200 training method 400 (see...). Figure 4 The trained algorithm model. Using the trained algorithm model to execute the operating condition segmentation method 300 can effectively achieve operating condition segmentation inference, so that the final identification result of the battery swapping service data can be output using the operating condition segmentation strategy. For example... Figure 3 As shown, the working condition segmentation method 300 may include steps S320 to S380.

[0063] In step S320, a time-series dataset about the battery swapping service task can be constructed based on the battery swapping service data collected by the sensing system 104. In some embodiments, the time-series dataset may include data with the same time window length N. w The data consists of multiple time windows, each corresponding to a consecutive time window. The process of processing the battery swapping service data collected by the sensing system 104 and recorded in chronological order to obtain a time-series dataset about the battery swapping service task can also be referred to as preprocessing of the battery swapping service data.

[0064] In some embodiments, constructing a time-series dataset for battery swapping service tasks may include the following steps: battery swapping service data acquisition step, single service dataset acquisition step, feature dataset acquisition step, and time window data partitioning step.

[0065] The battery swapping service data acquisition step may include receiving battery swapping service data collected by the sensing system 104 and recorded in chronological order from the battery swapping station 100.

[0066] The step of obtaining a single service dataset may include: classifying the battery swapping service data into one or more single service datasets according to the battery swapping service task, wherein each single service dataset corresponds to a corresponding battery swapping service task. In some embodiments, the battery swapping service data may be classified according to order tasks, so that the data parameters contained in the same single service dataset belong to the same order task.

[0067] The feature dataset acquisition step may include: classifying the single service datasets into one or more feature datasets based on features, for each of the single service datasets, or for one or more single service datasets as needed. Data parameters belonging to the same feature dataset correspond to a specific feature, or in other words, data parameters belonging to the same feature dataset are associated with the same feature in the same battery swapping service task. In some embodiments, the feature may be a component participating in the battery swapping service task, or it may be an axis among the components participating in the battery swapping service task.

[0068] It should be understood that both the single-service dataset and the feature dataset are obtained based on the classification of battery swapping service data collected by the sensor system 104, and both are time-series datasets. In this paper, the feature dataset obtained after classifying by battery swapping service task and by feature can be labeled as a multivariate time series x. For the same battery swapping service task, the multivariate time series x can be defined as... Where N c This represents elements such as components or shafts within components involved in battery swapping services, where T represents the time length. A specific timestamp in a multivariate time series x can be used... Let represent , where i represents the sensor number and j represents the time point number. This represents the timestamp data of a certain element in battery swapping station 100 collected by the i-th sensor at the j-th time point during a specific (e.g., a particular) battery swapping service task. The multivariate time series x can be determined according to different components and different axes as needed, and the determination of the multivariate time series x is the basis for obtaining the time series dataset.

[0069] The time window data partitioning step may include: for each feature dataset in each of the single service datasets, or for one or more feature datasets in one or more single service datasets as needed, partitioning the data into equal time windows of length N. w The multivariate time series x is divided into multiple time windows to form a corresponding time series dataset, where the time window length N w This can be determined based on expert experience introduced by the working condition segmentation algorithm module 200. In some embodiments, each time window data can be represented as a vector. For example, for the t-th time window, all the data can be stacked to form a vector, and the corresponding time window data can be labeled as follows. Where t represents the ordinal number of the time window, which can be selected from any natural number.

[0070] In step S340, multiple representation data can be obtained based on the time-series dataset via the working condition segmentation algorithm module 200. In some embodiments, obtaining multiple representation data based on the time-series dataset may include an input data s acquisition step and a mapping step.

[0071] The step of obtaining input data s may include processing multiple time window data in the time series dataset to obtain input data s. In some embodiments, the input data s may be represented in vector form. Optionally, the time window data corresponding to the t-th time window may be labeled as a vector. Each time window of data is represented by a vector. Therefore, for a time series dataset containing t time windows, the input data s can be represented as s = [s1, s2, ..., s...]. t ].

[0072] In some embodiments, the input data s acquisition step may further include: (e.g., using an exponential moving average method) filtering multiple time window data in the time series dataset to obtain filtered input data s. For example, the filtering of multiple time window data can be implemented via the data filtering submodule 220 in the working condition segmentation algorithm module 200, that is, averaging a total of 2n+1 time window data before and after the multiple time window data using an exponential moving average method to obtain the filtered input data s. In this embodiment, time window data is first acquired based on the multivariate time series x, and then the time window data is filtered; in some other embodiments, the multivariate time series x can also be filtered first, and then time window data can be acquired based on the filtered multivariate time series x.

[0073] The mapping step may include mapping the input data s (e.g., filtered input data s) from the original input space to a high-dimensional space via the encoder 214 in the condition segmentation algorithm module 200 to obtain output data f including multiple representation data. In some embodiments, this can be achieved through the function g en (·) is used to map the input data s to obtain the output data f, where the function g en (·) can be expressed by the following formula:

[0074] f = En(s) = g en (Ws+b e ).

[0075] Among them, Ws and b e Let g represent the weight matrix and the bias term, respectively. en (·) is a non-linear activation function. The output data f includes multiple representation data corresponding to a high-dimensional space, where multiple representation data correspond to multiple time windows, that is, each representation data has a corresponding time window.

[0076] In step S360, a distance curve describing the distance between adjacent representation data can be obtained based on multiple representation data, wherein adjacent representation data correspond to adjacent time windows. Embodiments of this application can obtain a distance curve based on distances calculated from consecutive time windows.

[0077] In some embodiments, the distance between adjacent representation data can be calculated first, and then a distance curve can be plotted based on the distance between adjacent representation data. The step of calculating the distance between adjacent representation data may include: for adjacent time windows t and t+1, determining the representation data f corresponding to the t-th time window. t and the representation data f corresponding to the (t+1)th time window t+1 , where t is a natural number; and the adjacent representation data f is calculated using the following formula. t with f t+1 Dist between t :

[0078]

[0079] Here, the numerator is the Euclidean distance between the representation data corresponding to the continuous time window, and the denominator is used as a normalization term.

[0080] In some embodiments, before calculating the distance between adjacent representation data, multiple representation data can be filtered to obtain updated representation data. Based on this, the distance between adjacent representation data can be calculated based on the filtered updated representation data. For example, the filtering of multiple representation data can be implemented via the data filtering submodule 220 in the working condition segmentation algorithm module 200, that is, by averaging a total of 2n+1 representation data points before and after the current time using the exponential moving average method to obtain the filtered representation data.

[0081] In step S380, the operating condition segmentation result can be obtained based on the distance curve depicted in step S360. In some embodiments, the operating condition segmentation point can be determined based on the extreme values ​​in the distance curve, wherein the extreme values ​​include minimum or maximum values ​​(e.g., peaks or valleys) within a region. For example, a local maximum among all peaks in the distance curve can be selected as the detected breakpoint. The breakpoint can also be called a segmentation point to segment different operating conditions. Operating condition segmentation can be implemented by the operating condition segmentation submodule 230 in the operating condition segmentation algorithm module 200. In some embodiments, after detecting the breakpoint in the distance curve, two adjacent breakpoints can be divided first, and then the corresponding original time window data can be segmented to obtain the final operating condition segmentation result.

[0082] The end-to-end battery swapping station service operation condition segmentation method 300 proposed in this application, based on deep learning technology, can identify operation condition breakpoints in complex battery swapping scenarios. It utilizes data-driven approaches to solve engineering management problems and exhibits good operation condition segmentation accuracy and robustness. Furthermore, the embodiments of this application determine operation condition segmentation points for battery swapping service data based on a distance metric using representation vectors, eliminating the need to rely on input data models with specified prior information and demonstrating good generalization performance.

[0083] Next, we will combine Figure 4 This application describes a training method 400 for a working condition segmentation algorithm module 200 according to an embodiment of the present application.

[0084] Figure 4 A schematic diagram of a training method 400 for a working condition segmentation algorithm module 200 according to an embodiment of this application is shown. Figure 4 As shown, the training method 400 of the working condition segmentation algorithm module 200 includes steps S410 to S460.

[0085] In step S410, a training dataset is constructed. In some embodiments, the process of constructing the training dataset is similar to the preprocessing process for battery swapping service data described above (e.g., step S320). For example, the collected data used to construct the training dataset can first be classified according to battery swapping service tasks and elements such as components or shafts within components to form a multivariate time series x for constructing the training dataset, and then the multivariate time series x is divided into the same time window length N. w The data is divided into multiple time windows to form a time-series dataset (i.e., a training dataset) for training. In some embodiments, the time window data can be represented in vector form; for example, the time window data corresponding to the t-th time window can be labeled as a vector.

[0086] In some embodiments, the training dataset can be divided into a training set and a test set according to a certain ratio. The training set can be used to train the algorithm model of the working condition segmentation algorithm module 200. After the algorithm model has been trained to an appropriate level, the test set can be used to verify whether the algorithm model has completed training.

[0087] In step S420, input data s for training is generated based on the training dataset. In some embodiments, the process of generating the input data s for training is similar to the input data s acquisition step described in step S340 above. For example, the input data s for training can be expressed in vector form, i.e., s = [s1, s2, ..., s2]. t ], where s tThis represents the time window data corresponding to the t-th time window.

[0088] In step S430, noise values ​​are introduced into the input data s used for training via the noise reduction layer 212 of the working condition segmentation algorithm module 200. Because the on-site environment of the battery swapping station 100 is complex in specific battery swapping scenarios, the collected data often contains a large amount of noise and extreme values, resulting in numerous outliers. Therefore, achieving robustness of the algorithm model under such complex working conditions is challenging. In this application, embodiments actively introduce noise values ​​such as extreme values ​​and random noise points during the training process, making the trained model more robust. In some embodiments, noise values ​​can be introduced by randomly setting the values ​​in the original input data s to 0 with a certain probability distribution. After introducing noise values, the reconstruction error of the loss function can be changed through subsequent iterations, thereby optimizing the algorithm model of the working condition segmentation algorithm module 200 to easily handle working condition segmentation of battery swapping service data containing many outliers.

[0089] In step S440, the encoder 214 in the working condition segmentation algorithm module 200 maps the input data s used for training from the original input space to a high-dimensional space to obtain output data f for training. In some embodiments, the process of performing step S440 is similar to the mapping step described in step S340 above. For example, the input data s used for training can be mapped from the original input space to a high-dimensional space via the encoder 214 in the working condition segmentation algorithm module 200 to obtain output data f for training including multiple representation data, wherein the mapping function g of the encoder 214 is... en (·) can be expressed by the following formula:

[0090] f = En(s) = g en (Ws+b e ).

[0091] Among them, Ws and b e Let g represent the weight matrix and the bias term, respectively. en (·) is a non-linear activation function. Here, the multiple representation data contained in the output data f used for training correspond to multiple time windows.

[0092] In step S450, the output data f used for training is mapped back from the high-dimensional space to the original input space via the decoder 216 in the working condition segmentation algorithm module 200 to obtain the reconstructed data used for training. In some embodiments, the output data f is mapped to obtain reconstructed data. The mapping function can be expressed by the following formula:

[0093]

[0094] Wherein, W′f represents the decoding weight matrix, and b d This indicates the decoding bias term.

[0095] In step S460, the working condition segmentation algorithm module 200 is trained so that the reconstructed data used for training... Minimize the error between the input data s used for training and the input data s.

[0096] In some embodiments, the reconstructed data used for training Minimizing the error between the input data s used for training and the input data s can be expressed based on the following formula:

[0097] minJ(W,b e ,b d ,W′)=∑L(u,v)+α∑W 2 ,

[0098] Where W represents the encoding weight matrix, W′ represents the decoding weight matrix, α represents the weight coefficients, and L(u,v) is the loss function, where u = s and v = D. ec (E nc (s)).

[0099] In some embodiments, the loss function L(u,v) can be the cross-entropy loss function, and can be expressed based on the following formula:

[0100]

[0101] Alternatively, the values ​​of u and v can be substituted into the formula for minimizing the error to reconstruct the data. Minimizing the error between the input data s and the input data can be expressed by the following formula:

[0102]

[0103] It should be understood that, since T represents the length of time, N W This represents the length of the time window, hence T / N in the above expression. W This can represent the total number of time windows. In some embodiments, the overall algorithm model of the working condition segmentation algorithm module 200 can be iteratively optimized using stochastic gradient descent.

[0104] The above is about reconstructing data. The error function between the input data s and the input data s can characterize the reconstructed data obtained by the algorithm model of the working condition segmentation algorithm module 200. The loss between the input data s and the training set. In some embodiments, the algorithm model of the work condition segmentation algorithm module 200 can be trained in multiple rounds using the training set in the training dataset constructed in step S410. Training can be stopped when the loss of the algorithm model does not decrease for consecutive μ rounds when the algorithm model is validated using the test set. Here, μ can be any value greater than 20, for example, μ can be 100 or any value greater than 100. Based on the above criteria for judging the completion of the algorithm model training, the algorithm model after training can be obtained. It can be understood that the algorithm model obtained after training is the relatively optimal algorithm model. After the algorithm model is trained, the relatively optimal algorithm model can be saved for use by the work condition segmentation method 300. In some embodiments, in order to ensure the stability of training in the early stage, a warm-up strategy can be used in the setting of the training learning rate, so that the learning rate increases first and then gradually decreases.

[0105] Next, we will combine Figure 5 The present application describes a condition segmentation system 500 according to an embodiment of the present application.

[0106] Figure 5 A schematic diagram of a working condition segmentation system 110 according to an embodiment of this application is shown. Figure 5 As shown, the operating condition segmentation system 110 for a battery swapping station 100 includes a communication unit 510, a memory 520 (e.g., non-volatile memory such as flash memory, ROM, hard disk drive, disk, optical disk, etc.), and a processor 530. The communication unit 510 serves as a communication interface and is configured to establish a communication connection between the operating condition segmentation system 110 and an external device or network (e.g., the cloud, the battery swapping station 100, etc.). In some embodiments, the communication unit 510 can receive battery swapping service data from the battery swapping station 100. The memory 520 stores instructions executable by the processor 530. The processor 530 is configured to execute the instructions to implement an operating condition segmentation method for a battery swapping station according to one or more embodiments of this application.

[0107] According to another aspect of this application, a computer-readable storage medium is provided, wherein instructions are stored that, when executed by a processor, cause the processor to perform any of the above-described operating condition segmentation methods for a battery swapping station. The computer-readable medium referred to in this application includes various types of computer storage media, and can be any available medium accessible by a general-purpose or special-purpose computer. For example, the computer-readable medium may include RAM, ROM, EPROM, E... 2PROM, registers, hard disks, removable disks, CD-ROMs or other optical disc storage, magnetic disk storage or other magnetic storage devices, or any other temporary or non-temporary medium capable of carrying or storing desired units of program code in the form of instructions or data structures and accessible by a general-purpose or special-purpose computer, or a general-purpose or special-purpose processor. As used herein, disks typically magnetically copy data, while discs optically copy data using lasers. Combinations of the above should also be included within the scope of computer-readable media. An exemplary storage medium is coupled to a processor so that the processor can read and write information from / to the storage medium. In an alternative, the storage medium may be integrated into the processor. The processor and storage medium may reside in an ASIC. The ASIC may reside in a user terminal. In an alternative, the processor and storage medium may reside as discrete components in the user terminal.

[0108] In the overall operation and maintenance of battery swapping stations, maintenance engineers often need to manually review large amounts of data to identify fault characteristics. Furthermore, as the number of swaps increases, the workload for maintenance engineers becomes enormous. This application provides an automated method for segmenting operating conditions during the battery swapping process, effectively facilitating automated segmentation, thereby improving maintenance efficiency and enhancing the ability to detect anomalies and diagnose faults at battery swapping stations during subsequent maintenance.

[0109] The operating condition segmentation method in this application divides the long-term series data of each battery swapping service into different operating conditions. This enables the operation and maintenance system for battery swapping stations to conduct more targeted and in-depth research on the different mechanical motion modes of different components and different axes during the battery swapping process. In addition, each segment of the different operating condition data can be explained by a more concise model. Decomposing large datasets with redundant data into smaller components is also a key application of unsupervised learning.

[0110] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Those skilled in the art can conceive of other feasible variations or substitutions based on the technical scope disclosed in this application, and such variations or substitutions are all covered within the scope of protection of this application. Where there is no conflict, the embodiments and features described in the embodiments of this application can also be combined with each other. The scope of protection of this application is determined by the claims.

Claims

1. A method for dividing operating conditions for a battery swapping station, characterized in that, The working condition segmentation method includes: A time-series dataset for battery swapping service tasks is constructed, comprising multiple time window data with the same time window length, the multiple time window data corresponding to multiple consecutive time windows; wherein constructing the time-series dataset for battery swapping service tasks includes: receiving battery swapping service data recorded in chronological order from the battery swapping station; classifying the battery swapping service data into one or more single-service datasets according to the battery swapping service task, wherein each single-service dataset corresponds to a corresponding battery swapping service task; classifying the single-service dataset into one or more feature datasets according to features, wherein each feature dataset corresponds to a corresponding feature; and dividing the feature dataset into multiple time window data with the same time window length to form a corresponding time-series dataset; wherein the features include components participating in the battery swapping service task or shafts in the components; obtaining multiple representation data based on the time-series dataset, the multiple representation data corresponding to the multiple consecutive time windows; wherein the operating condition segmentation method is implemented at least via an operating condition segmentation algorithm module, the operating condition segmentation algorithm module including an autoencoder submodule and an operating condition segmentation submodule, the multiple representation data being obtained via the autoencoder submodule; Based on the multiple characterization data, the distance between adjacent characterization data corresponding to adjacent time windows is calculated to obtain a distance curve describing the distance; and The working condition segmentation point is determined based on the extreme value in the distance curve to obtain the working condition segmentation result, which is obtained through the working condition segmentation submodule.

2. The working condition segmentation method according to claim 1, characterized in that, The process of obtaining the multiple representation data based on the time-series dataset includes: The multiple time window data are processed to obtain input data s; and The input data s is mapped from the original input space to a high-dimensional space to obtain output data f that includes the multiple representation data.

3. The working condition segmentation method according to claim 2, characterized in that, The process of processing the multiple time window data to obtain the input data s includes: The multiple time window data are filtered using the exponential moving average method to obtain the filtered input data s.

4. The working condition segmentation method according to claim 1, characterized in that, The calculation of the distance between adjacent representation data corresponding to adjacent time windows includes: For the t-th time window and the (t+1)-th time window, determine the representation data corresponding to the t-th time window. and the representation data corresponding to the (t+1)th time window Where t is a natural number; and The following formula is used to calculate adjacent representation data. and Distance between : 。 5. The working condition segmentation method according to claim 1, characterized in that, Before calculating the distance between adjacent representation data, the plurality of representation data is filtered to obtain updated plurality of representation data, and the distance is calculated based on the updated plurality of representation data.

6. The working condition segmentation method according to claim 2, characterized in that, The autoencoder submodule includes an encoder and a decoder. The encoder is configured to map the input data s from the original input space to the high-dimensional space to obtain the output data f. The decoder is configured to map the output data f from the high-dimensional space back to the original input space to obtain reconstructed data. .

7. The working condition segmentation method according to claim 6, characterized in that, The mapping function that maps the input data s to the output data f is expressed by the following formula: ， The above Represents the encoding weight matrix, the Indicates the encoding bias term; and The output data f is mapped to obtain the reconstructed data. The mapping function is expressed by the following formula: ， The above This represents the decoding weight matrix, the... This indicates the decoding bias term.

8. The working condition segmentation method according to claim 6, characterized in that, The encoder and the decoder are both based on convolutional neural networks, wherein an attention mechanism is introduced into the convolutional neural network.

9. The working condition segmentation method according to claim 6, characterized in that, The autoencoder submodule further includes a noise reduction layer configured to introduce noise values ​​into the input data s used for training during the training of the condition segmentation algorithm module.

10. The working condition segmentation method according to claim 9, characterized in that, The working condition segmentation algorithm module is trained using a training method that includes the following steps: Build the training dataset; The input data s for training is generated based on the training dataset; The working condition segmentation algorithm module to be trained generates reconstructed data for training based on the input data s used for training. ; The working condition segmentation algorithm module is trained so that the reconstructed data used for training... The error between the input data s used for training and the input data s is minimized.

11. The working condition segmentation method according to claim 10, characterized in that, The reconstructed data for training is generated by the work condition segmentation algorithm module to be trained based on the input data s used for training. include: The input data s used for training is mapped from the original input space to a high-dimensional space by the encoder in the working condition segmentation algorithm module to be trained, so as to obtain the output data f used for training. The output data f used for training is mapped from the high-dimensional space back to the original input space via the decoder in the work condition segmentation algorithm module to be trained, so as to obtain the reconstructed data used for training. .

12. The working condition segmentation method according to claim 10, characterized in that, The reconstructed data used for training The minimization of the error between the input data s used for training and the input data s is expressed based on the following formula: ， in, Represents the encoding weight matrix. This represents the decoded weight matrix. Indicates the weighting coefficient. It is the loss function, where .

13. The working condition segmentation method according to claim 10, characterized in that, The reconstructed data for training is generated by the work condition segmentation algorithm module to be trained based on the input data s used for training. Also includes: Before the input data s used for training is mapped from the original input space to a high-dimensional space by the encoder in the working condition segmentation algorithm module to obtain the output data f used for training, noise values ​​are introduced into the input data s used for training via the noise reduction layer in the working condition segmentation algorithm module to be trained.

14. A condition segmentation system for a battery swapping station, characterized in that, The working condition segmentation system includes: Memory configured to store instructions; and A processor configured to execute the instructions to cause the load segmentation system to perform the load segmentation method as described in any one of claims 1-13.

15. A battery swapping station, the battery swapping station comprising the operating condition segmentation system as described in claim 14.

16. A computer-readable storage medium storing instructions, characterized in that, When the instruction is executed by the processor, the processor performs the condition segmentation method as described in any one of claims 1-13.

Images