Calibration parameter completion method and device, equipment and storage medium

Abstract

The invention discloses a calibration parameter complementing method and device, equipment and a storage medium. According to the scheme, the method comprises the steps of performing segmentation and vectorization processing on a target calibration table to be complemented to obtain a target input tensor representing the target calibration table to be complemented; and inputting the target input tensor as time sequence data into a preset neural network model for completion processing to obtain a completed target calibration table. Therefore, the calibration table is automatically complemented through the preset neural network model, the workload required by calibration is reduced, and the calibration cost is reduced; the calibration period is shortened, the calibration efficiency is improved, and the vehicle model updating speed is increased.

Description

Technical Field

[0001] The present application relates to the technical field of parameter calibration, and in particular to a method, apparatus, device and storage medium for completing calibration parameters. Background Art

[0002] Vehicle calibration is a critical link in the automotive development process. Its technical quality directly determines the coordinated optimization effect of the powertrain system and vehicle performance, and has a decisive impact on core indicators such as vehicle acceleration performance, fuel efficiency, emission compliance and driving smoothness. At the same time, it profoundly shapes the user's perception of sensory experiences such as vehicle control response and noise control.

[0003] Traditional calibration engineering, a key technical pillar of automotive R&D, still faces significant operational constraints. The industry's current, widely adopted experience-driven calibration method relies on experienced engineers' intuitive understanding of vehicle characteristics to adjust parameters. This development model, heavily reliant on manual experience, has significant flaws. Engineers must verify the feasibility of parameters through repeated test drives, a process that resembles searching for the optimal solution in a fog. Calibration cycles are lengthy, often lasting months or even longer, resulting in high costs and low efficiency. Summary of the Invention

[0004] In view of this, the present application is committed to providing a calibration parameter completion method, device, equipment and storage medium, which can reduce the calibration workload and repetitive work, improve the efficiency of cross-model parameter migration, and reduce R&D costs.

[0005] According to a first aspect of the present application, a method for completing calibration parameters is provided, comprising:

[0006] Obtaining a target calibration table to be completed; the target calibration table is used to determine calibration parameters corresponding to operating condition parameters by a table lookup method;

[0007] Slicing the target calibration table to obtain target table segments of the target calibration table;

[0008] performing vectorization processing on the target table segment to obtain a target input vector corresponding to the target table segment and a target input tensor composed of the target input vector; the target input vector is used to represent sequence data composed of calibration parameters in the target table segment;

[0009] Inputting the target input tensor into a preset neural network model to obtain an output matrix for the target calibration table;

[0010] According to the output matrix, a completed target calibration table is obtained.

[0011] Optionally, the preset neural network model includes a recurrent neural network module, a fully connected layer and a Transformer module in sequence;

[0012] The step of inputting the target input tensor into a preset neural network model to obtain an output matrix for the target calibration table includes:

[0013] Inputting the target input tensor into the recurrent neural network module to obtain an intermediate input tensor corresponding to the target input tensor;

[0014] Input the intermediate input tensor into the fully connected layer to obtain a query matrix, a key matrix, and a value matrix corresponding to the target input tensor;

[0015] The query matrix, the key matrix and the value matrix are input into the Transformer module to obtain an output matrix for the target calibration table.

[0016] Optionally, before inputting the target input tensor into a preset neural network model, the method further includes:

[0017] For each sample calibration table in the training set, the sample calibration table is segmented to obtain a sample table segment corresponding to the sample calibration table; the sample calibration table is of the same type as the target calibration table;

[0018] Performing the vectorization process on the sample table segments to obtain an initial sample vector corresponding to each sample table segment in the sample calibration table;

[0019] A target sample matrix is ​​constructed based on the initial sample vectors corresponding to all sample table fragments; the initial model is trained based on the target sample matrix, with the goal of minimizing the value of the loss function, and the model parameters of the initial model are adjusted to obtain a preset neural network model; the initial model includes a recurrent neural network module, a fully connected layer and a Transformer module in sequence; the loss function is used to represent the difference between the actual value and the predicted value of the calibration parameter.

[0020] Optionally, constructing a target sample matrix according to the initial sample vectors corresponding to all sample table fragments includes:

[0021] Performing stacking processing on the initial sample vectors to form a first sample matrix;

[0022] performing a normalization process on the first sample matrix based on the normalization parameter of the first sample matrix to obtain a second sample matrix;

[0023] Based on the projection matrix, a dimensionality reduction process is performed on the second sample matrix to obtain a target sample matrix; the projection matrix is ​​obtained by performing principal component analysis on the second sample matrix.

[0024] Optionally, the vectorizing the target table fragment to obtain a target input vector corresponding to the target table fragment and a target input tensor composed of the target input vector includes:

[0025] Performing vectorization processing on the target table fragment to obtain an initial input vector and an initial input tensor composed of the initial input vector; the initial input vector is sequence data composed of calibration parameters contained in the target table fragment;

[0026] Based on a standardization parameter, the initial input tensor is standardized to obtain a second input tensor; the standardization parameter is obtained by statistically analyzing a sample calibration table in a training set of the preset neural network model;

[0027] Based on the projection matrix, the second input tensor is subjected to dimensionality reduction processing to obtain the target input tensor; the projection matrix is ​​obtained by performing principal component analysis on the sample calibration table in the training set of the preset neural network model.

[0028] Optionally, obtaining a completed target calibration table according to the output matrix includes:

[0029] Based on the projection matrix, performing dimension recovery processing on the first output matrix output by the preset neural network model to obtain a second output matrix;

[0030] Based on the normalization parameters, the second output matrix is ​​subjected to inverse normalization processing to obtain a completed target calibration table; the inverse normalization processing is an inverse process of the normalization processing.

[0031] Optionally, inputting the target input tensor into the recurrent neural network module to obtain an intermediate input tensor corresponding to the target input tensor includes:

[0032] For each target input vector in the target input tensor, input the target input vector into the recurrent neural network module to obtain an intermediate vector corresponding to the target input vector; the intermediate vector is the output vector of the last time step of the recurrent neural network module;

[0033] According to the intermediate vector corresponding to each target input vector, the intermediate input tensor corresponding to the target input vector is obtained.

[0034] Optionally, the target calibration table is a motor calibration table of a target motor, the operating condition parameters include a speed value and a torque value of the target motor, and the calibration parameters include a control current value of the target motor;

[0035] The target calibration table is segmented to obtain target table segments of the target calibration table, including:

[0036] Slicing the motor calibration table according to a preset speed interval to obtain target table segments of the motor calibration table;

[0037] The vectorizing the target table fragment includes:

[0038] For each target table segment, the control current values ​​are arranged in order of rotational speed priority.

[0039] According to a second aspect of the present application, a calibration parameter completion device is provided, comprising:

[0040] An acquisition module is used to acquire a target calibration table to be completed; the target calibration table is used to determine calibration parameters corresponding to operating condition parameters by a table lookup method;

[0041] a segmentation module, configured to segment the target calibration table to obtain target table segments of the target calibration table;

[0042] a vectorization module, configured to perform vectorization processing on the target table segment to obtain a target input vector corresponding to the target table segment and a target input tensor composed of the target input vector; the target input vector is used to represent sequence data composed of calibration parameters in the target table segment;

[0043] A completion module, configured to input the target input tensor into a preset neural network model to obtain an output matrix for the target calibration table;

[0044] The output module is used to obtain a completed target calibration table according to the output matrix.

[0045] According to a third aspect of the present application, an electronic device is provided, comprising: a processor; a memory for storing instructions executable by the processor; and the processor is configured to execute the method described in any one of the above embodiments.

[0046] According to a fourth aspect of the present application, a computer-readable storage medium is provided, wherein the storage medium stores a computer program, and the computer program is used to execute the method described in any one of the above embodiments.

[0047] This application provides a calibration parameter completion method, apparatus, device, and storage medium. The method comprises: segmenting and vectorizing a target calibration table to be completed to obtain a target input tensor representing the target calibration table to be completed; treating the target input tensor as time series data and inputting it into a preset neural network model for completion processing to obtain a completed target calibration table. Automatically completing the calibration table through the preset neural network model reduces the workload required for calibration and lowers calibration costs. This method shortens the calibration cycle, improves calibration efficiency, and contributes to increasing the speed of vehicle model upgrades. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 The figure shows a flow chart of a calibration parameter completion method provided in one embodiment of the present application.

[0049] Figure 2 Shown is a schematic structural diagram of the preset neural network model provided in an embodiment of the present application.

[0050] Figure 3 The figure shows a block diagram of a calibration parameter completion device provided by one embodiment of the present application.

[0051] Figure 4 Shown is a structural block diagram of an electronic device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0052] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0053] Application Overview

[0054] In automotive control systems, the Look-Up Table (LUT) method, as a core parameter mapping tool, is widely used to quickly obtain parameters for complex nonlinear relationships or experimental calibration. Its principle is to use a pre-generated calibration table to index the corresponding calibration parameters based on real-time operating parameters. The calibration table is generated through calibration and is essentially constructed from a large amount of calibration experimental data. Its core function is to convert complex nonlinear relationships into discrete data structures, thereby achieving efficient and stable parameter call.

[0055] The experience-driven calibration method currently used in the industry relies on experienced engineers to adjust parameters based on their intuitive understanding of vehicle characteristics. This development model, which relies heavily on manual experience, has significant drawbacks:

[0056] Engineers must adjust and verify the feasibility of each parameter individually, a process often requiring repeated test drives. During the parameter space exploration phase, engineers must conduct extensive trial and error, and the verification of each parameter requires the support of actual road testing or bench testing. This calibration model results in extremely low calibration efficiency, with calibration cycles often lasting several months. To complete this work, companies must deploy a large number of experienced calibration engineers, significantly increasing labor costs. Furthermore, the frequent test drives lead to a sharp increase in test site rental fees, vehicle wear and tear costs, and energy consumption, creating a high-cost development bottleneck.

[0057] More critically, existing parameter optimization methods lack self-learning capabilities and are unable to effectively tap into the physical laws and empirical knowledge hidden in historical calibration data. This makes it difficult to effectively transfer calibration knowledge between different vehicle platforms. When developing new models, engineering teams are forced to repeat basic calibration work, unable to transform valid patterns in historical data into reusable calibration strategies. This severely restricts vehicle development efficiency and makes it difficult to meet the market's responsiveness to vehicle upgrades.

[0058] To address the aforementioned issues, the present embodiment segments and vectorizes the target calibration table to be completed, obtaining a target input tensor representing the target calibration table to be completed. This target input tensor is then treated as time series data and input into a preset neural network model for completion, resulting in a completed target calibration table. This approach automatically completes the calibration table using the preset neural network model, reducing the workload and cost of calibration, shortening the calibration cycle, and improving calibration efficiency, thereby accelerating vehicle model upgrades.

[0059] After introducing the basic principles of the present application, various non-limiting embodiments of the present application will be described in detail with reference to the accompanying drawings.

[0060] Exemplary Methods

[0061] Figure 1 This is a flow chart of a method for completing calibration parameters provided in one embodiment of the present application. Figure 1 The method is performed by a computing device (e.g., a server), but the present invention is not limited thereto. The server may be a single server, or may be composed of several servers, or may be a virtualization platform, or may be a cloud computing service center, and the present invention is not limited thereto. Figure 1 As shown, the method includes the following contents:

[0062] Step S110: obtaining a target calibration table to be completed; the target calibration table is used to determine calibration parameters corresponding to operating condition parameters by a table lookup method.

[0063] In the embodiments of the present application, the target calibration table is a structured data table used to record the correspondence between operating parameters and calibration parameters. Specifically, it can be various calibration tables related to vehicles, industrial control, consumer electronics, and other equipment. For example, a motor calibration table is used to determine the required motor current based on speed and torque lookup tables. An engine ignition advance angle calibration table is used to select the optimal ignition advance angle based on speed and load lookup tables. These are not specifically limited here.

[0064] In the embodiment of the present application, the operating condition parameters are parameters that reflect the operating conditions of the equipment, such as the speed, torque, load, etc. during the operation of the vehicle. Different values ​​of the operating condition parameters will result in different required calibration parameters.

[0065] In an embodiment of the present application, the calibration parameters are parameter setting values ​​obtained from the target calibration table according to the operating condition parameters through a table lookup method, which are used for stable operation of the equipment or to achieve specific performance indicators, such as the operating current and voltage of the motor, the ignition advance angle of the engine, etc.

[0066] Step S120: Segment the target calibration table to obtain target table segments of the target calibration table.

[0067] In the embodiments of the present application, the dimension of the target calibration table refers to the number of operating parameters included. For example, one-dimensional calibration tables such as temperature compensation curves and S-curve soft start tables, as well as the motor calibration table and engine ignition advance angle calibration table mentioned above, are two-dimensional calibration tables. Furthermore, the target calibration table can also be a multidimensional calibration table for multivariable control, without specific limitation.

[0068] In an embodiment of the present application, the target calibration table can be regarded as a tensor, and the dimension of the target calibration table is the number of coordinate axes in the tensor. Each coordinate axis is used to represent an independent operating parameter dimension in the calibration table, and its scale interval is set to linear or nonlinear distribution according to the calibration accuracy requirements.

[0069] In an embodiment of the present application, the target calibration table is segmented, which may refer to segmenting the target calibration table according to the working condition parameters (i.e., the coordinate axis direction of the tensor) in the direction of the coordinate axis to obtain a target table segment of the target calibration table; specifically, the multidimensional parameter matrix may be divided into a plurality of sub-regions along a vertical plane of at least one selected coordinate axis. For example, the motor calibration table to be completed is segmented into a plurality of segments according to the rotational speed. For another example, the three-dimensional calibration table may be layered and sliced ​​along the Z-axis direction so that each table segment retains the parameter mapping relationship within the XY plane.

[0070] In this embodiment, the table segments are subtables obtained by segmenting the target calibration table. They contain structured data for some operating parameter dimensions or some parameter value ranges in the original table. For example, if a motor calibration table is segmented at certain speed intervals, each table segment covers a specific speed range and fully contains the current corresponding to all torques within that speed range.

[0071] Step S130: vectorize the target table fragment to obtain a target input vector corresponding to the target table fragment and a target input tensor composed of the target input vector; the target input vector is used to represent sequence data composed of calibration parameters in the target table fragment.

[0072] In an embodiment of the present application, the vectorization process is used to convert the calibration parameters in the table fragment into a vector form to facilitate subsequent processing by the neural network model. Specifically, if the table fragment can be obtained by segmenting the target calibration table according to the direction of a specified coordinate axis, the vectorization process can refer to arranging the calibration parameters in the table fragment according to the priority order of the specified coordinate axis.

[0073] In an embodiment of the present application, the target input vector may include sequence data composed of calibration parameters in the target calibration table, or a vector after preprocessing the sequence data, and the preprocessing may include one or more of mask processing, zero-padding processing, standardization processing, and PCA dimensionality reduction processing.

[0074] In practical applications, time series can reflect the dynamic characteristics and trend information of parameter values ​​changing over time; the recurrent neural network module (RNN) can process time series and extract the dynamic change rules of parameter values ​​at continuous timestamps; in the embodiment of the present application, the target input vector obtained by vectorizing the table fragment can reflect the change trend of the calibration parameters along with the working condition parameters of the target calibration table. Based on the time series processing capability of the recurrent neural network module (RNN), the change trend of the calibration parameters along with the working condition parameters is extracted, and the evolution rules of the calibration parameters at different working condition parameter positions are captured.

[0075] In an embodiment of the present application, the target input tensor is formed by stacking the target input vectors corresponding to each table segment, and is used to integrate the vector information of all table segments into a unified input format for processing by the preset neural network model. In some cases, the target input tensor can be obtained by stacking in the order of the specified coordinate axes.

[0076] In some cases, the order of the splitting and vectorization steps can be reversed, that is, the target calibration table is first vectorized and then the vector is split to obtain the target input vector.

[0077] Step S140: inputting the target input tensor into a preset neural network model to obtain an output matrix for the target calibration table.

[0078] In an embodiment of the present application, the preset neural network model may include a neural network model that can be used for time series completion, that is, converting the target calibration table to be completed into sequence data, treating it as time series data, and completing it using the preset neural network model.

[0079] In an embodiment of the present application, the sample feature vector of the preset neural network model is generated based on a sample calibration table of the same type as the target calibration table.

[0080] In an embodiment of the present application, the output matrix is ​​the processing result of the preset neural network model on the target calibration table.

[0081] Step S150: Obtaining a completed target calibration table according to the output matrix.

[0082] In an embodiment of the present application, since the target calibration table is segmented, vectorized, and other processes are performed on the target input tensor during the generation process, the output matrix also needs to be inversely processed to obtain the completed target calibration table.

[0083] In the embodiment of the present application, the calibration parameters missing from the completed target calibration table are reasonably filled in to improve the correspondence between the operating condition parameters and the calibration parameters.

[0084] In an embodiment of the present application, the target calibration table to be completed is segmented to obtain target table segments of the target calibration table; the target table segments are vectorized to obtain target input vectors corresponding to the target table segments and target input tensors composed of the target input vectors; and the target calibration table is processed sequentially through a recurrent neural network module, a fully connected layer, and a Transformer module to obtain the completed target calibration table. This automatically completes the calibration table through a preset neural network model, reducing the workload required for calibration and lowering the calibration cost. This shortens the calibration cycle, improves calibration efficiency, and helps increase the speed of vehicle model upgrades.

[0085] based on Figure 1 The method in this specification also provides some specific implementation plans of the method, which are described below.

[0086] Optionally, the preset neural network model includes a recurrent neural network module, a fully connected layer, and a Transformer module in sequence; inputting the target input tensor into the preset neural network model to obtain an output matrix for the target calibration table includes:

[0087] Inputting the target input tensor into the recurrent neural network module to obtain an intermediate input tensor corresponding to the target input tensor;

[0088] Input the intermediate input tensor into the fully connected layer to obtain a query matrix, a key matrix, and a value matrix corresponding to the target input tensor;

[0089] The query matrix, the key matrix and the value matrix are input into the Transformer module to obtain an output matrix for the target calibration table.

[0090] Figure 2 The figure shows the structure of the preset neural network model provided in the embodiment of the present application. Figure 2 The preset neural network model may be a hybrid neural network model that sequentially includes a recurrent neural network module, a fully connected layer, and a Transformer module. The sample feature vector of the preset neural network model is generated based on a sample calibration table of the same type as the target calibration table.

[0091] In the embodiment of the present application, the recurrent neural network module is used to capture the temporal dependency in the sequence data and perform preliminary feature extraction on the target input vector. The recurrent neural network module can be a basic RNN model, LSTM, GRU, Bi-RNN, or deep RNN.

[0092] In an embodiment of the present application, the recurrent neural network module is used as a backbone network for feature extraction, which is used to extract the context dependency between target input vectors in the target input tensor and capture the dynamic change pattern of the calibration parameter sequence data.

[0093] In an embodiment of the present application, the intermediate input tensor may include intermediate input vectors corresponding to each target input vector in the target input tensor, which is obtained by processing the target input vector by the recurrent neural network module.

[0094] In the embodiment of the present application, the fully connected layer may include a query weight matrix W q , key weight matrix W k Sum value weight matrix W v Three fully connected layers are used to convert the intermediate input vector into the key vector, value vector and query vector corresponding to the target input vector, linearly transform the output features of the recurrent neural network module and align the dimensions, and generate the key matrix, value matrix and query matrix adapted to the input format of the Transformer module so that their dimensions meet the input requirements of the Transformer module.

[0095] In an embodiment of the present application, the Transformer module can be used to capture the correlation between different positions in the input data using a self-attention mechanism, and finally output an output matrix for representing the completed target calibration table.

[0096] In an embodiment of the present application, the preset neural network model cascades a recurrent neural network module, a fully connected layer and a Transformer module, which sequentially implements local temporal feature extraction, feature space transformation and global dependency modeling to form an end-to-end multidimensional calibration table completion architecture.

[0097] In an embodiment of the present application, the sample feature vector is generated based on a sample calibration table of the same type as the target calibration table, and is processed through segmentation, vectorization, and tensor stacking processes to ensure consistency in parameter dimensions and distribution characteristics between the input data and the target calibration table.

[0098] In an embodiment of the present application, the Transformer module is used to process the query matrix, the key matrix and the value matrix based on the attention mechanism to obtain an output matrix for representing the completed target calibration table.

[0099] In the embodiment of the present application, through the collaborative processing of the recurrent neural network module, the fully connected layer and the Transformer module, the advantages of each network structure can be fully utilized to accurately capture the characteristics and correlations of the target calibration table data. The deep mining of sequence features by the recurrent neural network module, combined with the dimensional adaptation of the fully connected layer and the efficient attention mechanism of the Transformer module, can quickly and accurately complete the missing data in the target calibration table. Compared with traditional manual calibration or single model processing methods, it greatly reduces the manpower investment and time cost required for calibration, effectively improving calibration efficiency and accuracy.

[0100] Optionally, before inputting the target input tensor into a preset neural network model, the method further includes:

[0101] For each sample calibration table in the training set, the sample calibration table is segmented to obtain a sample table segment corresponding to the sample calibration table; the sample calibration table is of the same type as the target calibration table;

[0102] Performing the vectorization process on the sample table segments to obtain an initial sample vector corresponding to each sample table segment in the sample calibration table;

[0103] Construct the target sample matrix based on the initial sample vectors corresponding to all sample table fragments;

[0104] The initial model is trained based on the target sample matrix, with the goal of minimizing the value of the loss function, and the model parameters of the initial model are adjusted to obtain a preset neural network model; the initial model includes a recurrent neural network module, a fully connected layer and a Transformer module in sequence; the loss function is used to represent the difference between the actual value and the predicted value of the calibration parameter.

[0105] In the embodiment of the present application, the sample calibration table is a calibration table of the same type as the target calibration table, and can be obtained by collecting historical calibration tables.

[0106] In some cases, to improve model accuracy and reduce processing difficulty, the initial sample calibration table can be scaled so that the sample calibration table and the target calibration table have not only the same dimensions but also the same number of dimensions. That is, if the sample calibration table and the target calibration table are viewed as tensors, the two tensors have the same shape, not only do the coordinate axes correspond one-to-one, but the number of parameters on the corresponding coordinate axes (i.e., the number of dimensions) is also the same. Taking a motor calibration table as an example, although different motor models have different output torque ranges and speed ranges, by adjusting the parameter interval, the number of parameters on the torque axis and the number of parameters on the speed axis are the same.

[0107] In an embodiment of the present application, before the sample calibration table is segmented, data cleaning may be performed on the sample calibration table to check and process missing values ​​to ensure data consistency and integrity.

[0108] In the embodiment of the present application, the steps of segmenting, vectorizing, etc. for processing the sample calibration table are the same as the processing process of the target calibration table.

[0109] In the embodiment of the present application, the initial sample vector is sequence data composed of calibration parameters in the sample table segment.

[0110] In an embodiment of the present application, the target sample matrix includes a sample feature vector corresponding to each sample table segment in the sample calibration table. The sample feature vector is obtained by processing the initial sample vector.

[0111] In this embodiment of the present application, the loss function is used to represent the difference between the actual value and the predicted value of the calibration parameter. During the training process, some calibration parameters in the sample calibration table are masked, and the target sample matrix corresponding to the sample calibration table is completed using the initial model to obtain the predicted value of the masked portion. The actual value refers to the actual value of the calibration parameter in the sample calibration table; the predicted value refers to the corresponding calibration parameter obtained after the initial model completes the masked portion.

[0112] In the embodiment of the present application, by using a sample calibration table of the same type as the target calibration table as the training set and iteratively optimizing the initial model parameters under the constraints of a loss function, the preset neural network model is driven to automatically mine the physical mapping relationships and engineering experience knowledge implicit in the historical calibration data during the training process, effectively learning the inherent correlation mechanism between the operating parameters and the calibration parameters. This training mechanism based on homologous data fully utilizes the prior knowledge contained in the historical calibration data, avoids the inefficient use of historical data in traditional methods, and enables the model to accurately capture the structural characteristics and parameter variation patterns of similar calibration tables, thereby significantly improving the accuracy and robustness of the calibration table completion task.

[0113] Optionally, constructing a target sample matrix according to the initial sample vectors corresponding to all sample table fragments includes:

[0114] Performing stacking processing on the initial sample vectors to form a first sample matrix;

[0115] performing a normalization process on the first sample matrix based on the normalization parameter of the first sample matrix to obtain a second sample matrix;

[0116] Based on the projection matrix, a dimensionality reduction process is performed on the second sample matrix to obtain a target sample matrix; the projection matrix is ​​obtained by performing principal component analysis on the second sample matrix.

[0117] In the embodiment of the present application, the normalization process may include Min-Max normalization and Z-Score normalization.

[0118] In the embodiment of the present application, the normalization parameter refers to a statistic calculated from the first sample matrix, such as the mean μ and standard deviation σ of each feature dimension.

[0119] In an embodiment of the present application, taking Z-Score standardization as an example, the standardization parameters based on the first sample matrix and the standardization of the first sample matrix can refer to forming a standardization parameter by calculating the mean μ and standard deviation σ of each feature dimension (column vector) in the first sample matrix; then performing an operation on each element in the matrix: subtracting the mean μ of the corresponding feature dimension from the original value, and dividing it by the standard deviation σ, that is, x′=(x-μ) / σ. By standardizing the first sample matrix, the dimensional differences of different parameters are eliminated. This standardization ensures that the weights of each parameter are balanced when the subsequent covariance matrix is ​​calculated, avoids the problem of feature dominance caused by differences in numerical scales, and improves the accuracy and efficiency of subsequent processing.

[0120] In the embodiment of the present application, the dimensionality reduction process may refer to principal component analysis (PCA) processing, which may include the following steps:

[0121] By formula Calculate the covariance matrix C of the second sample matrix X, where n is the number of sample eigenvectors. Perform eigenvalue decomposition on the covariance matrix C, obtaining eigenvectors sorted by eigenvalue from largest to smallest, forming a set of orthogonal basis vectors. Select the eigenvectors corresponding to the first k largest eigenvalues ​​as the principal component directions as needed to construct the projection matrix P. The eigenvalues ​​represent the variance contributions of the principal components, and the eigenvectors indicate the directions of the principal components. Multiply the second sample matrix X by the projection matrix P (i.e., Y = PX) to map the original high-dimensional data into a low-dimensional space, generating the target sample matrix Y.

[0122] In an embodiment of the present application, by stacking, standardizing and reducing the dimensions of the initial sample vectors, the multidimensional features of the sample data can be effectively integrated, redundant information can be eliminated and the key parameter associations can be retained. Standardization eliminates the dimensional differences of different parameter dimensions to ensure the stability of model training; the dimensionality reduction processing based on the projection matrix mines the principal component features of the data, while reducing the computational complexity and retaining the core variation rules of the sample data, so that the target sample matrix can more efficiently characterize the key features of the sample calibration table. This process significantly compresses the feature dimensions and improves the data characterization efficiency while retaining the nonlinear correlation characteristics between the calibration table parameters, so that the target sample matrix can carry the physical laws and empirical knowledge of the multidimensional calibration table in the form of a low-dimensional dense vector, providing high signal-to-noise ratio input features for the preset neural network model, and reducing the computational complexity of model training.

[0123] Optionally, the vectorizing the target table fragment to obtain a target input vector corresponding to the target table fragment and a target input tensor composed of the target input vector includes:

[0124] Performing vectorization processing on the target table fragment to obtain an initial input vector and an initial input tensor composed of the initial input vector; the initial input vector is sequence data composed of calibration parameters contained in the target table fragment;

[0125] Based on a standardization parameter, the initial input tensor is standardized to obtain a second input tensor; the standardization parameter is obtained by statistically analyzing a sample calibration table in a training set of the preset neural network model;

[0126] Based on the projection matrix, the second input tensor is subjected to dimensionality reduction processing to obtain the target input tensor; the projection matrix is ​​obtained by performing principal component analysis on the sample calibration table in the training set of the preset neural network model.

[0127] In this embodiment of the present application, the initial input vector is parameter sequence data composed of calibration parameters contained in the target table segment of the target calibration table arranged in a predetermined order, and is strictly consistent with the construction rules of the initial sample vector described in the sample calibration table training phase. The initial input vector has the same dimension as the initial sample vector.

[0128] In this embodiment, the initial input tensor is obtained by stacking the initial input vectors; for example, the initial input vectors corresponding to the target table segments are stacked along the feature dimension. Its tensor structure (order and dimension) is aligned with the construction of the first sample matrix, ensuring the effective reuse of the normalization parameters during the normalization phase.

[0129] In the embodiment of the present application, the normalization parameter is obtained by performing statistics on the initial sample vector; and the initial sample vector is obtained by performing the vectorization processing on the sample table segment.

[0130] In an embodiment of the present application, the standardized parameters of each feature dimension calculated by the first sample matrix in the training phase are used to perform a linear transformation on the parameter values ​​of the corresponding dimensions in the initial input tensor, so that the parameter distributions of the target calibration table and the sample calibration table are aligned.

[0131] In an embodiment of the present application, the projection matrix is ​​obtained by performing principal component analysis on the second sample matrix; the second sample matrix includes a standardized initial sample vector.

[0132] In an embodiment of the present application, the projection matrix obtained by decomposing the projection matrix of the second sample matrix in the training phase is used to perform matrix multiplication on the standardized second input tensor and the projection matrix to obtain the target input tensor, thereby realizing the mapping of the high-dimensional parameter space to the low-dimensional principal component space.

[0133] In an embodiment of the present application, the target calibration table and the sample calibration table are strictly consistent in the feature processing process by reusing the standardized parameters and dimensionality reduction projection matrix established in the training phase. The standardization process eliminates the dimensional differences and distribution offsets of the parameters under different working conditions, aligns the parameter distribution of the target input tensor with the training set, and avoids the degradation of model performance caused by data offset. The dimensionality reduction process based on the projection matrix retains the characteristic direction of the principal component through the projection matrix, removes the linear redundant information between the parameters while compressing the feature dimension, and enables the target input tensor to carry the key parameter distribution law in a low-dimensional dense form. This processing mechanism not only maintains the original statistical correlation characteristics between the calibration table parameters, but also makes the input data isomorphic with the training feature space of the preset neural network model, effectively improving the model's completion accuracy and cross-working condition generalization ability for the missing areas of the target calibration table parameters, and improving the quality and efficiency of calibration table completion.

[0134] Optionally, obtaining a completed target calibration table according to the output matrix includes:

[0135] Based on the projection matrix, performing dimension recovery processing on the first output matrix output by the preset neural network model to obtain a second output matrix;

[0136] Based on the normalization parameters, the second output matrix is ​​subjected to inverse normalization processing to obtain a completed target calibration table; the inverse normalization processing is an inverse process of the normalization processing.

[0137] In an embodiment of the present application, the first output matrix output by the Transformer module is inversely transformed based on the projection matrix of the second sample matrix. This process is the reverse operation corresponding to the dimensionality reduction process in the training phase. In the training phase, the input data is reduced in dimensionality by the projection matrix of the second sample matrix, and the high-dimensional parameter space is mapped to the low-dimensional principal component space to reduce the complexity of the data and retain key information. Here, in order to restore the results of the model output to the dimension and feature space of the original data, it is necessary to use the projection matrix for inverse transformation. That is, the first output matrix is ​​multiplied by the inverse matrix of the projection matrix to obtain the second output matrix, thereby restoring the first output matrix output by the Transformer module from the low-dimensional principal component space to the high-dimensional parameter space.

[0138] In an embodiment of the present application, the data in the second output matrix is ​​normalized data with dimension differences removed. In order to restore the normalized data to the dimensions of the original data, the second output matrix is ​​denormalized to convert the normalized data back to the dimensions of the original data.

[0139] In an embodiment of the present application, the output of the Transformer module is inversely transformed using the projection matrix of the second sample matrix, and the output matrix of the low-dimensional space is mapped back to the high-dimensional original space to restore the complete dimensional information of the data. Then, based on the standardized parameters of the first sample matrix, the output matrix after the inverse transformation is denormalized, and the data is converted from the standardized space back to the original data space to ensure that the output matrix has the same dimension and physical meaning as the original calibration table. This ensures that the completed target calibration table can be directly applied to the parameter configuration of the actual control system, achieving end-to-end consistency from model output to engineering application.

[0140] Optionally, inputting the target input tensor into the recurrent neural network module to obtain an intermediate input tensor corresponding to the target input tensor includes:

[0141] For each target input vector in the target input tensor, input the target input vector into the recurrent neural network module to obtain an intermediate vector corresponding to the target input vector; the intermediate vector is the output vector of the last time step of the recurrent neural network module;

[0142] According to the intermediate vector corresponding to each target input vector, the intermediate input tensor corresponding to the target input vector is obtained.

[0143] In an embodiment of the present application, the target input vector is expanded into a parameter sequence according to the time step, and input into the recurrent neural network module for iterative calculation by time step. The temporal context information is transmitted through the hidden state, thereby capturing the nonlinear evolution law of the calibration parameters along the coordinate axis and the long-range dependency characteristics across time steps. Specifically, the target input vector is sequentially input into the hidden layer of the recurrent neural network as sequence data. The hidden state of each time step receives the information of the current input and the hidden state of the previous time step, and realizes the memory and transmission of historical information through the recurrent connection. The calculation formula is:

[0144] y'=h t (y)

[0145] h t =f tanh (W ih y t +b ih +W hh h (t―1) +b hh )

[0146] Among them, y is the target input vector, y' is the intermediate vector; h t represents the hidden state at time step t. tanh W represents the activation function tanh(). ih Represents the weight matrix input to the hidden layer. t is the input data at time step t. b ih W represents the bias term input to the hidden layer. hh Represents the weight matrix from hidden layer to hidden layer, which is used to transfer the hidden state information of the previous time step.

[0147] After completing the calculations for all time steps, the hidden layer output vector of the last time step is selected as the intermediate vector corresponding to the target input vector. This intermediate vector integrates the global context information of the target input vector sequence, effectively capturing the long-range feature correlations of the calibration parameter sequence data, and providing a high-level abstract representation containing sequence features for subsequent processing in the fully connected layer and Transformer module.

[0148] In an embodiment of the present application, the intermediate vector is the hidden state vector output by the recurrent neural network module at the last time step. This vector integrates the timing characteristics and dynamic patterns of all time steps of the target input vector and can be used to represent the complete sequence information in the target input vector.

[0149] In an embodiment of the present application, the intermediate vectors are stacked to obtain the intermediate input tensor corresponding to the target input vector, ensuring strict alignment with the input dimensions of the three fully connected layers: the key weight matrix, the value weight matrix, and the query weight matrix.

[0150] In the embodiment of the present application, the processing mechanism of the target input vector by the above-mentioned recurrent neural network module can deeply explore the complex variation patterns of the calibration parameters in the multi-dimensional working condition parameter space, effectively overcoming the limitation of traditional methods in capturing nonlinear dynamic characteristics. The integration of global context information by the intermediate vector enables the subsequent fully connected layer and Transformer module to perform efficient calculations based on highly abstract feature representations, significantly reducing the computational overhead caused by data redundancy, thereby enabling the preset neural network model to more accurately learn the mapping relationship between working condition parameters and calibration parameters, and improving the reliability and stability of the target calibration table completion results.

[0151] Optionally, the target calibration table is a motor calibration table of a target motor, the operating condition parameters include a speed value and a torque value of the target motor, and the calibration parameters include a control current value of the target motor;

[0152] The target calibration table is segmented to obtain target table segments of the target calibration table, including:

[0153] Slicing the motor calibration table according to a preset speed interval to obtain target table segments of the motor calibration table;

[0154] The vectorizing the target table fragment includes:

[0155] For each target table segment, the control current values ​​are arranged in order of rotational speed priority.

[0156] In an embodiment of the present application, the coordinate axes in the motor calibration table include a speed axis and a torque axis; the segmentation of the target calibration table may refer to dividing the motor calibration table into several table segments along a vertical plane along the speed axis; each table segment covers a specific speed range and completely includes the control current values ​​corresponding to all torques within the speed range.

[0157] In some cases, the target calibration table can be segmented using a sliding window method. Specifically, a sliding window of fixed width (e.g., 500 rpm) is set along the speed axis and moved along the speed scale at a preset step size (which can be the same as the width of the sliding window). Each window captures the entire torque parameter plane within a continuous speed range. When the window reaches the end of the speed axis, the window size is automatically adjusted to cover the remaining parameter area, thereby generating multiple target table segments covering the entire speed range and with complete torque parameters.

[0158] In this embodiment, the control current values ​​corresponding to different torques at the same speed are first arranged in ascending order, with all control current values ​​corresponding to a speed value being a subsequence. These subsequences are then arranged in ascending order according to the speed value to obtain an initial input vector. This initial input vector thus retains the correspondence between the speed, torque, and control current values ​​in the motor calibration table.

[0159] Based on the same technical concept, the present application provides an example of a method for completing a motor calibration table. The horizontal axis of the motor calibration table is the speed axis, and the vertical axis is the torque axis. The values ​​in the table are the control current values ​​corresponding to different motor speeds and torques.

[0160] First, the motor calibration table is segmented according to preset speed intervals to obtain target table segments of the motor calibration table. The segmentation method can use a sliding window method, and the step size and window size (i.e., speed interval) can be the same, for example, both set to 500 rpm. Each segment represents the corresponding relationship between torque and current within that speed range.

[0161] Secondly, each segment is vectorized in column-first (speed-first) order to obtain the initial input vector; for the area to be completed, zero-padding is used for masking, and the filling position is distinguished from the valid data by the marker bit. The mean and standard deviation calculated in the training phase are reused to perform Z-score normalization on the initial input vector in sequence; and PCA dimensionality reduction is performed on the projection matrix of the training phase to obtain the target input vector after dimensional compression. Multiple target input vectors are stacked along the batch dimension to generate the target input tensor. In order to ensure calibration accuracy, it is best to include some continuous calibration parameters in the motor calibration table to be completed, and at least one complete target input tensor is obtained.

[0162] The target input tensors corresponding to each fragment are stacked to obtain the input matrix corresponding to the motor calibration table to be completed.

[0163] The input matrix is ​​input into the preset neural network model, and is processed in sequence by the recurrent neural network module, the fully connected layer and the Transformer module. The specific process is as follows: the target input tensor is input into the recurrent neural network module to obtain the output vector of the last time step, and the intermediate input tensor is stacked to generate. Using the three fully connected layers of the key weight matrix, the value weight matrix and the query weight matrix, each intermediate input tensor is linearly transformed to obtain the key vector, value vector and query vector corresponding to each intermediate input tensor, that is, the key matrix, value matrix and query matrix. The key matrix, value matrix and query matrix are input into the Transformer module to obtain the output matrix output by the Transformer module;

[0164] Perform PCA inverse processing and denormalization on the output matrix to obtain the completed motor calibration table.

[0165] Exemplary devices

[0166] The device embodiments of this application can be used to execute the method embodiments of this application. For details not disclosed in the device embodiments of this application, please refer to the method embodiments of this application.

[0167] Figure 3 FIG. 1 is a block diagram of a calibration parameter completion device provided by an embodiment of the present application. Figure 3 As shown, the device 300 includes:

[0168] The acquisition module 310 is used to obtain a target calibration table to be completed; the target calibration table is used to determine the calibration parameters corresponding to the operating condition parameters by a table lookup method;

[0169] a segmentation module 320 for segmenting the target calibration table to obtain target table segments of the target calibration table;

[0170] a vectorization module 330 configured to perform vectorization processing on the target table segment to obtain a target input vector corresponding to the target table segment and a target input tensor composed of the target input vector; the target input vector is used to represent sequence data composed of calibration parameters in the target table segment;

[0171] A completion module 340 is configured to input the target input tensor into a preset neural network model to obtain an output matrix for the target calibration table;

[0172] The output module 350 is configured to obtain a completed target calibration table according to the output matrix.

[0173] Optionally, the preset neural network model includes a recurrent neural network module, a fully connected layer and a Transformer module in sequence;

[0174] The completion module 340 includes:

[0175] An RNN unit, configured to input the target input tensor into the recurrent neural network module to obtain an intermediate input tensor corresponding to the target input tensor;

[0176] a conversion unit, configured to input the intermediate input tensor into the fully connected layer to obtain a query matrix, a key matrix, and a value matrix corresponding to the target input tensor;

[0177] A Transformer unit is configured to input the query matrix, the key matrix, and the value matrix into the Transformer module to obtain an output matrix for the target calibration table.

[0178] Optionally, the segmentation module 320 is configured to segment each sample calibration table in the training set to obtain a sample table segment corresponding to the sample calibration table; the sample calibration table is of the same type as the target calibration table;

[0179] The vectorization module 330 is configured to perform the vectorization process on the sample table segments to obtain an initial sample vector corresponding to each sample table segment in the sample calibration table;

[0180] The apparatus 300 further includes:

[0181] A sample construction module is used to construct a target sample matrix based on the initial sample vectors corresponding to all sample table fragments;

[0182] A training module is used to train the initial model based on the target sample matrix, with the goal of minimizing the value of the loss function, adjust the model parameters of the initial model, and obtain a preset neural network model; the initial model includes a recurrent neural network module, a fully connected layer and a Transformer module in sequence; the loss function is used to represent the difference between the actual value and the predicted value of the calibration parameter.

[0183] Optionally, the sample construction module includes:

[0184] a stacking unit, configured to stack the initial sample vectors to form a first sample matrix;

[0185] a normalization unit, configured to perform normalization processing on the first sample matrix based on a normalization parameter of the first sample matrix to obtain a second sample matrix;

[0186] The dimensionality reduction unit is used to perform dimensionality reduction processing on the second sample matrix based on the projection matrix to obtain a target sample matrix; the projection matrix is ​​obtained by performing principal component analysis on the second sample matrix.

[0187] Optionally, the vectorization module 330 is further configured to:

[0188] Performing vectorization processing on the target table fragment to obtain an initial input vector and an initial input tensor composed of the initial input vector; the initial input vector is sequence data composed of calibration parameters contained in the target table fragment;

[0189] The normalization unit is configured to perform normalization processing on the initial input tensor based on a normalization parameter to obtain a second input tensor; the normalization parameter is obtained by statistically analyzing a sample calibration table in a training set of the preset neural network model;

[0190] The dimensionality reduction unit is used to perform dimensionality reduction processing on the second input tensor based on the projection matrix to obtain the target input tensor; the projection matrix is ​​obtained by performing principal component analysis on the sample calibration table in the training set of the preset neural network model.

[0191] Optionally, the output module 350 is configured to:

[0192] Based on the projection matrix, performing dimension recovery processing on the first output matrix output by the preset neural network model to obtain a second output matrix;

[0193] Based on the normalization parameters, the second output matrix is ​​subjected to inverse normalization processing to obtain a completed target calibration table; the inverse normalization processing is an inverse process of the normalization processing.

[0194] Optionally, the RNN unit is used to:

[0195] For each target input vector in the target input tensor, input the target input vector into the recurrent neural network module to obtain an intermediate vector corresponding to the target input vector; the intermediate vector is the output vector of the last time step of the recurrent neural network module;

[0196] According to the intermediate vector corresponding to each target input vector, the intermediate input tensor corresponding to the target input vector is obtained.

[0197] Optionally, the target calibration table is a motor calibration table of a target motor, the operating condition parameters include a speed value and a torque value of the target motor, and the calibration parameters include a control current value of the target motor;

[0198] The segmentation module 320 is used to:

[0199] Slicing the motor calibration table according to a preset speed interval to obtain target table segments of the motor calibration table;

[0200] The vectorization module 330 is further configured to:

[0201] For each target table segment, the control current values ​​are arranged in order of rotational speed priority.

[0202] Exemplary electronic devices

[0203] Below, reference Figure 4 To describe the electronic device according to the embodiment of the present application. Figure 4 The figure shows a block diagram of an electronic device according to an embodiment of the present application.

[0204] like Figure 4 As shown, electronic device 400 includes one or more processors 410 and memory 420 .

[0205] The processor 410 may be a central processing unit (CPU) or other forms of processing units having data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device 400 to perform desired functions.

[0206] The memory 420 may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, random access memory (RAM) and / or cache memory (cache), etc. The non-volatile memory may include, for example, read-only memory (ROM), a hard disk, a flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and the processor 410 may execute the program instructions to implement the calibration parameter completion method and / or other desired functions of the various embodiments of the present application described above. Various contents such as category correspondences may also be stored in the computer-readable storage medium.

[0207] In one example, the electronic device 400 may further include an input device 430 and an output device 440 , and these components are interconnected via a bus system and / or other forms of connection mechanisms (not shown).

[0208] In addition, the input device 430 may also include, for example, a keyboard, a mouse, etc. The output device 440 may output various information to the outside. The output device 440 may include, for example, a display, a speaker, a printer, a communication network and its connected remote output device, etc.

[0209] Of course, to simplify, Figure 4Only some of the components related to the present application in the electronic device 400 are shown, and components such as buses, input / output interfaces, etc. are omitted. In addition, the electronic device 400 may further include any other appropriate components according to specific application scenarios.

[0210] Exemplary computer program products and computer-readable storage media

[0211] In addition to the above-mentioned methods and devices, an embodiment of the present application may also be a computer program product, which includes computer program instructions, which, when executed by a processor, enable the processor to execute the steps of the calibration parameter completion method according to various embodiments of the present application described in the above-mentioned "Exemplary Method" section of this specification.

[0212] The computer program product may be written in any combination of one or more programming languages ​​to implement the program code for performing the operations of the embodiments of the present application, including object-oriented programming languages ​​such as Java, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computing device, partially on the user's computing device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0213] In addition, an embodiment of the present application may also be a computer-readable storage medium having computer program instructions stored thereon, which, when executed by a processor, causes the processor to execute the steps of the method for completing calibration parameters according to various embodiments of the present application described in the above "Exemplary Method" section of this specification.

[0214] The computer-readable storage medium can adopt any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium can, for example, include but is not limited to a system, device or component of electricity, magnetism, light, electromagnetic, infrared, or semiconductor, or any combination thereof. More specific examples (non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof.

[0215] The basic principles of the present application have been described above in conjunction with specific embodiments. However, it should be noted that the advantages, strengths, and effects mentioned in this application are merely illustrative and not restrictive, and it should not be assumed that these advantages, strengths, and effects are required of each embodiment of this application. In addition, the specific details disclosed above are merely illustrative and facilitating understanding, and are not restrictive. The above details do not limit this application to necessarily being implemented using the above specific details.

[0216] The block diagrams of the devices, devices, equipment, and systems involved in this application are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As will be appreciated by those skilled in the art, these devices, devices, equipment, and systems can be connected, arranged, or configured in any manner. Words such as "include," "comprise," "have," and the like are open-ended words, meaning "including but not limited to," and can be used interchangeably therewith. The words "or" and "and" used herein refer to the words "and / or" and can be used interchangeably therewith, unless the context clearly indicates otherwise. The word "such as" used herein refers to the phrase "such as but not limited to," and can be used interchangeably therewith.

[0217] It should also be noted that in the apparatus, device, and method of the present application, each component or each step can be decomposed and / or recombined, and such decomposition and / or recombination should be regarded as equivalent solutions of the present application.

[0218] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects without departing from the scope of the present application. Therefore, the present application is not intended to be limited to the aspects shown herein, but rather to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0219] It should be understood that the qualifiers "first", "second", "third", "fourth", "fifth" and "sixth" used in the description of the embodiments of the present application are only used to more clearly illustrate the technical solutions and cannot be used to limit the scope of protection of the present application.

[0220] The above description has been provided for the purpose of illustration and description. Furthermore, this description is not intended to limit the embodiments of the present application to the forms disclosed herein. Although a number of example aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations thereof.

Claims

1. A method for completing calibration parameters, characterized in that: include: Get the target calibration table to be completed; The target calibration table is used to determine the calibration parameters corresponding to the operating condition parameters by a table lookup method; Slicing the target calibration table to obtain target table segments of the target calibration table; Performing vectorization processing on the target table fragment to obtain a target input vector corresponding to the target table fragment and a target input tensor composed of the target input vector; The target input vector is used to represent sequence data consisting of calibration parameters in the target table segment; Inputting the target input tensor into a preset neural network model to obtain an output matrix for the target calibration table; According to the output matrix, a completed target calibration table is obtained.

2. The method according to claim 1, characterized in that The preset neural network model includes a recurrent neural network module, a fully connected layer and a Transformer module in sequence; The step of inputting the target input tensor into a preset neural network model to obtain an output matrix for the target calibration table includes: Inputting the target input tensor into the recurrent neural network module to obtain an intermediate input tensor corresponding to the target input tensor; Input the intermediate input tensor into the fully connected layer to obtain a query matrix, a key matrix, and a value matrix corresponding to the target input tensor; The query matrix, the key matrix and the value matrix are input into the Transformer module to obtain an output matrix for the target calibration table.

3. The method according to claim 1, characterized in that Before inputting the target input tensor into the preset neural network model, the method further includes: For each sample calibration table in the training set, the sample calibration table is segmented to obtain a sample table segment corresponding to the sample calibration table; the sample calibration table is of the same type as the target calibration table; Performing the vectorization process on the sample table segments to obtain an initial sample vector corresponding to each sample table segment in the sample calibration table; Construct the target sample matrix based on the initial sample vectors corresponding to all sample table fragments; The initial model is trained based on the target sample matrix, with the goal of minimizing the value of the loss function, and the model parameters of the initial model are adjusted to obtain a preset neural network model; the initial model includes a recurrent neural network module, a fully connected layer and a Transformer module in sequence; the loss function is used to represent the difference between the actual value and the predicted value of the calibration parameter.

4. The method according to claim 3, characterized in that The target sample matrix is ​​constructed based on the initial sample vectors corresponding to all sample table fragments, including: Performing stacking processing on the initial sample vectors to form a first sample matrix; performing a normalization process on the first sample matrix based on the normalization parameter of the first sample matrix to obtain a second sample matrix; Based on the projection matrix, a dimensionality reduction process is performed on the second sample matrix to obtain a target sample matrix; the projection matrix is ​​obtained by performing principal component analysis on the second sample matrix.

5. The method according to claim 1, wherein The vectorizing process of the target table fragment to obtain a target input vector corresponding to the target table fragment and a target input tensor composed of the target input vector includes: Performing vectorization processing on the target table fragment to obtain an initial input vector and an initial input tensor composed of the initial input vector; the initial input vector is sequence data composed of calibration parameters contained in the target table fragment; Based on a standardization parameter, the initial input tensor is standardized to obtain a second input tensor; the standardization parameter is obtained by statistically analyzing a sample calibration table in a training set of the preset neural network model; Based on the projection matrix, the second input tensor is subjected to dimensionality reduction processing to obtain the target input tensor; the projection matrix is ​​obtained by performing principal component analysis on the sample calibration table in the training set of the preset neural network model.

6. The method according to claim 5, characterized in that The method of obtaining a completed target calibration table according to the output matrix includes: Based on the projection matrix, performing dimension recovery processing on the first output matrix output by the preset neural network model to obtain a second output matrix; Based on the normalization parameters, the second output matrix is ​​subjected to inverse normalization processing to obtain a completed target calibration table; the inverse normalization processing is an inverse process of the normalization processing.

7. The method according to claim 2, characterized in that Inputting the target input tensor into the recurrent neural network module to obtain an intermediate input tensor corresponding to the target input tensor includes: For each target input vector in the target input tensor, input the target input vector into the recurrent neural network module to obtain an intermediate vector corresponding to the target input vector; the intermediate vector is the output vector of the last time step of the recurrent neural network module; According to the intermediate vector corresponding to each target input vector, the intermediate input tensor corresponding to the target input vector is obtained.

8. The method according to claim 1 or 5, characterized in that The target calibration table is a motor calibration table of a target motor, the operating condition parameters include a speed value and a torque value of the target motor, and the calibration parameters include a control current value of the target motor; The target calibration table is segmented to obtain target table segments of the target calibration table, including: Slicing the motor calibration table according to a preset speed interval to obtain target table segments of the motor calibration table; The vectorizing the target table fragment includes: For each target table segment, the control current values ​​are arranged in order of rotational speed priority.

9. A calibration parameter completion device, characterized in that: include: An acquisition module is used to obtain the target calibration table to be completed; The target calibration table is used to determine the calibration parameters corresponding to the operating condition parameters by a table lookup method; a segmentation module, configured to segment the target calibration table to obtain target table segments of the target calibration table; a vectorization module, configured to perform vectorization processing on the target table fragment to obtain a target input vector corresponding to the target table fragment and a target input tensor composed of the target input vector; The target input vector is used to represent sequence data consisting of calibration parameters in the target table segment; A completion module, configured to input the target input tensor into a preset neural network model to obtain an output matrix for the target calibration table; The output module is used to obtain a completed target calibration table according to the output matrix.

10. An electronic device, characterized in that: include: processor; a memory for storing instructions executable by the processor; The processor is configured to execute the method according to any one of claims 1 to 8.

11. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer program instructions, which, when executed by a processor, cause the processor to perform the method according to any one of claims 1 to 8.

Images