Battery health state assessment method, assessment system and related components

Abstract

The invention provides a battery health state evaluation method. The method comprises the following steps: collecting discharge measurement data and recovery voltage data of a battery through a BMS (Battery Management System); converting the recovery voltage data into a query matrix by using cosine position coding, and extracting time sequence characteristics of the discharge measurement data to form a key matrix and a value matrix; fusing the query matrix, the key matrix and the value matrix to obtain a measurement embedded vector; performing semantic fusion according to the measurement embedding vector and a preset battery degradation prompt to obtain a joint embedding representation; inputting the joint embedded representation into a preset battery analysis model to obtain a battery health prediction value; comparing and evaluating according to the battery health prediction value and a preset threshold value, and outputting a health state report of the battery; according to the method, an industrial large model framework integrating sequence reconstruction embedding, domain knowledge guidance and sparse model adaptation is designed, deep mining and self-adaptive learning of battery degradation characteristics are achieved, and finally a high-reliability health state report and early warning information are output.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of battery state of health evaluation, and in particular to a battery state of health evaluation method and system based on a battery management system (BMS), a device, and a storage medium. BACKGROUND

[0002] In the field of data-driven battery state of health prediction, raw time series data such as voltage and current collected by a battery management system (BMS) are usually directly used for modeling. However, the discharge process of a battery and the recovery voltage process after static state have different variation rates and sampling rules. This heterogeneity in time sequence makes it difficult for traditional embedding methods to effectively align and fuse multi-source data, resulting in the loss or redundancy of key degradation features.

[0003] To this end, we propose a battery state of health evaluation method, system, and related components to address the key issues of data heterogeneity fusion difficulty, shallow physical mechanism embedding, and weak model generalization capability when accurately and reliably predicting the state of health of power batteries under complex and variable working conditions. SUMMARY

[0004] The purpose of the present application is to solve the problems existing in the prior art. To achieve the above purpose, the present application adopts the following technical solutions: A battery state of health evaluation method, the evaluation method comprising: S101: collecting discharge measurement data and recovery voltage data of a battery through a BMS; By way of example, the discharge measurement data includes a discharge voltage value sequence, a discharge current value sequence, and a discharge temperature value sequence recorded at a fixed sampling frequency under a battery load working state. These time series data reflect the dynamic characteristics and thermal behavior of the battery during energy output. The recovery voltage data are an end voltage recovery curve collected at a lower frequency after the battery ends discharge and enters a static state. This curve can reveal the polarization characteristics and charge redistribution process of the battery internally. Due to the limitations of sensor accuracy and electromagnetic interference in actual application environments, the original collected signals need to be preprocessed, including using a sliding average filter with a window length of 5 to eliminate high-frequency noise, and using a linear interpolation algorithm to align the non-uniform sampling points of the recovery voltage data to standard timestamps.

[0005] S102: converting the recovery voltage data into a query matrix using cosine position encoding, and extracting the time series features of the discharge measurement data to form a key matrix and a value matrix; For example, after data acquisition and preprocessing, cosine position encoding is used to transform the recovered voltage data into a query matrix. Simultaneously, a multi-layer neural network architecture is used to extract deep temporal features of the discharge measurement data to form a key matrix and a value matrix. For the recovered voltage data, its normalized numerical sequence is input into a cosine position encoder with a 256-dimensional output space. This encoder calculates cosine function values ​​at different frequencies to generate feature vectors containing absolute and relative position information for the voltage value at each time step. All these feature vectors are combined to form the query matrix. For multi-dimensional discharge measurement data containing voltage, current, and temperature, local feature extraction is first performed using a one-dimensional convolutional neural network. This network uses 32 convolutional kernels of length 3 to scan the input sequence to capture short-term dependency patterns. Two fully connected layers then map the features to the same vector space as the query matrix, generating a key matrix and a value matrix. The key matrix represents contextual information, and the value matrix represents the substantive feature content.

[0006] S103: The query matrix, the key matrix, and the value matrix are fused to obtain a measurement embedding vector; For example, during the fusion process, the dot product between the query matrix and the transpose of the key matrix is ​​calculated. The result is then scaled by dividing by the square root of the vector dimension. Next, the scaled attention score is normalized using the Softmax function, generating probability distribution weights representing the correlation strength between different time steps. These weights are used to perform a weighted summation operation on the value matrix, thereby constructing a context-aware feature representation that simultaneously reflects the recovery voltage characteristics and the discharge process state. This feature representation is then subjected to nonlinear transformation and dimensionality compression through a feedforward neural network with 512 neurons, outputting a fixed-length 128-dimensional measurement embedding vector. This vector comprehensively encodes health-related information of the battery during both dynamic operation and static recovery phases.

[0007] S104: Perform semantic fusion based on the measured embedding vector and the preset battery degradation prompt to obtain a joint embedding representation; For example, battery degradation tips are a set of structured text descriptions containing knowledge about the impact of temperature on capacity decay, the differences in degradation rates at different discharge rates, and typical characteristics of battery cycle life. These text tips are first converted into 256-dimensional semantic vectors through a pre-trained language model embedding layer, forming a knowledge embedding representation. This knowledge embedding representation is then concatenated with the measurement embedding vector and input into a feature fusion module with a multi-head attention mechanism. This module uses eight attention heads to compute in parallel, mining potential correlations between data features and knowledge semantics from different representation subspaces. The fused output is then processed through layer normalization and residual connections to generate a joint embedding representation that incorporates both measured data features and domain knowledge.

[0008] S105: Input the joint embedded representation into the preset battery analysis model to obtain the battery health prediction value; For example, the joint embedding representation is input to a pre-defined battery analysis model, and a battery health prediction value is obtained through multi-level transformation and computation. The battery analysis model consists of a backbone network, a sparse attention module, and a regression output layer. The backbone network contains a 12-layer Transformer decoder structure, with each layer having 768-dimensional hidden states and 12 attention heads. The joint embedding representation first undergoes forward propagation through all layers of the backbone network, extracting high-level abstract features through a self-attention mechanism and a feedforward neural network to generate a sequence of hidden states rich in semantic information. This hidden state sequence then enters the sparse attention module, where the 30% of attention connections with the highest importance are selected using a relevance scoring function based on the L1 norm. Weighted calculations are performed only on these connections to focus on key information. The processed feature vector is then passed through a regression output layer consisting of two fully connected layers, which maps the 1024-dimensional input features to a single numerical output representing the battery health state.

[0009] S106: Based on the comparison and evaluation of the battery health prediction value and the preset threshold, output the battery health status report.

[0010] For example, the battery health prediction value is a numerical value between 0 and 1, representing the percentage of current battery capacity to initial capacity. This value is compared and analyzed against a preset set of thresholds, including a 0.8 threshold to distinguish between normal aging and abnormal degradation, and a 0.7 warning value to indicate severe performance decline. Based on the comparison results, an exponential smoothing model is used to extrapolate the trend using historical health prediction value sequences, calculating an estimate of remaining lifespan in units of charge-discharge cycles. All this information is integrated into a structured health status report, which includes key information such as the current health status level, remaining lifespan prediction, performance degradation warning signs, and maintenance recommendations, and is visualized through the battery management system's user interface or remote monitoring platform.

[0011] A battery health status assessment system, based on the assessment method described above, includes: The data acquisition module is used to acquire battery discharge measurement data and recovery voltage data through the BMS; The matrix generation module is used to convert the recovery voltage data into a query matrix using cosine position encoding, and to extract the time-series features of the discharge measurement data to form a key matrix and a value matrix. A matrix fusion module is used to fuse the query matrix, the key matrix, and the value matrix to obtain a measurement embedding vector; The semantic fusion module is used to perform semantic fusion based on the measurement embedding vector and the preset battery degradation prompt to obtain a joint embedding representation; The health analysis module is used to input the joint embedded representation into a preset battery analysis model to obtain a battery health prediction value; The health comparison module is used to compare and evaluate the battery health prediction value with a preset threshold, and output a health status report of the battery.

[0012] An electronic device includes a memory and a processor; the memory is used to store computer programs; The processor is configured to execute the computer program and, in executing the computer program, implement the steps of the evaluation method as described above.

[0013] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the evaluation method as described above.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention uses cosine position encoding to process and recover voltage data to generate a query matrix. It then constructs a measurement embedding vector by combining a key-value matrix built from discharge measurement data. By semantically fusing the measurement embedding vector with preset battery degradation prompts, a joint embedding representation is constructed, which combines data-driven and knowledge-guided approaches. The joint embedding representation is analyzed by a battery analysis model to obtain a battery health prediction value. A health status report is generated based on the battery health prediction value, improving adaptability to different battery application scenarios and thus enhancing the accuracy of battery health status assessment. Attached Figure Description

[0015] Figure 1 A schematic flowchart illustrating the battery health status assessment method; Figure 2 This is a schematic block diagram of a battery health status assessment system. Detailed Implementation

[0016] To more clearly illustrate the technical solution of this application, the technical solution of this application will be described below through specific embodiments. It should be noted that the specific embodiments are used to expand the description of the technical solution of this application, and are not intended to limit this application.

[0017] In some embodiments, refer to Figure 1A battery health status assessment method, based on the collection of battery discharge measurement data and recovery voltage data by a battery management system, includes: collecting discharge voltage sequence, discharge current sequence, discharge temperature sequence and recovery voltage sequence by the BMS; filtering the discharge voltage sequence, discharge current sequence and discharge temperature sequence and arranging them in chronological order as discharge measurement data; and performing linear interpolation on the recovery voltage sequence to obtain recovery voltage data.

[0018] For example, the discharge voltage sequence records the battery's terminal voltage changes under load conditions at a sampling frequency of 10 points per second; the discharge current sequence synchronously monitors current intensity fluctuations in the circuit; the discharge temperature sequence acquires temperature readings through sensors placed on the battery casing surface; and the recovery voltage sequence collects the battery voltage recovery process at a frequency of once per minute during the resting phase after discharge. The discharge voltage, discharge current, and discharge temperature sequences are filtered using a sliding window. A mean filter with a window width of 5 sampling points is used to smooth the original signal to eliminate high-frequency noise interference and ensure the stability of subsequent feature extraction. The filtered discharge voltage, discharge current, and discharge temperature sequences are arranged and combined according to their strict time correspondence to form discharge measurement data, maintaining the synchronization of the three physical quantities in the time dimension. Linear interpolation is performed on the recovery voltage sequence, constructing a linear function between adjacent sampling points and calculating the voltage value at intermediate moments on a standardized time grid. This ensures that the sampling interval of the recovery voltage data is consistent with the discharge measurement data, resolving the time alignment problem caused by sampling frequency differences.

[0019] In some embodiments, cosine position encoding is used to transform the recovery voltage data into a query matrix, and the temporal features of the discharge measurement data are extracted to form a key matrix and a value matrix. This includes: performing one-dimensional convolution and linear layer mapping on the discharge measurement data to obtain a discharge feature matrix; performing cosine position encoding on the recovery voltage data to obtain recovery voltage features; performing linear mapping based on the recovery voltage features to obtain a query matrix; and performing linear mapping based on the discharge feature matrix to obtain a key matrix and a value matrix.

[0020] For example, a one-dimensional convolution operation is performed on the discharge measurement data. 32 convolutional kernels of length 3 slide along the time dimension to extract local temporal patterns. The kernels move on the input sequence with a stride of 1 and generate a feature map by calculating the dot product. This feature map is processed by the ReLU activation function and then fed into a linear layer with 256 output units for high-dimensional spatial projection, resulting in a discharge feature matrix containing rich contextual semantics. Cosine position encoding is performed on the recovery voltage data. Cosine function values ​​at different frequencies are calculated to generate a feature vector containing absolute position information for the voltage value at each time step. The wavelength of the cosine function varies geometrically from 2π to 10000π to ensure that each position receives a unique encoded representation. The cosine-encoded recovery voltage features are then input into a linear projection layer with a 256-dimensional output space. This projection layer converts the position-encoded features into a query matrix suitable for attention calculation using a learnable weight matrix. Simultaneously, the discharge feature matrix is ​​input into two independent linear mapping layers. One mapping layer generates a key matrix for calculating similarity, and the other mapping layer generates a value matrix containing substantial feature content. These two matrices have the same vector dimension as the query matrix to ensure the normal calculation of the attention mechanism.

[0021] In some embodiments, fusing the query matrix, key matrix, and value matrix to obtain a measurement embedding vector includes: performing scaled dot product attention calculation based on the query matrix and key matrix to obtain a weight matrix; performing weighted summation on the value matrix based on the weight matrix to obtain cross-attention fusion features; performing layer normalization processing on the cross-attention fusion features to obtain normalized fusion features; and performing feedforward neural network mapping processing on the normalized fusion features to obtain a measurement embedding vector.

[0022] For example, scaling dot product attention is performed based on the query matrix and key matrix. The two matrices are multiplied and then divided by the square root of the vector dimension (8) for numerical scaling to prevent the vanishing gradient problem. Then, the Softmax function is applied to normalize each row so that the sum of all attention weights is 1, resulting in a weight matrix representing the correlation strength between different time steps. A weighted summation operation is then performed on the value matrix based on the weight matrix, where each row of the weight matrix is ​​linearly combined with the column vectors of the value matrix to generate cross-attention fusion features. These features fuse battery behavior information from two different time scales to simultaneously reflect recovery voltage characteristics and discharge process states. Layer normalization is then applied to the cross-attention fusion features, calculating the mean and variance of the batch of data features and standardizing each feature dimension, making the network training process more stable and accelerating model convergence. The normalized fused features are input into a feedforward neural network for processing. This network contains two linear transformation layers and uses the ReLU activation function in the middle to introduce non-linear transformation capability. The first linear layer expands the 128-dimensional input to 512 dimensions, and the second linear layer compresses the 512-dimensional features back to 128 dimensions. The output is a 128-dimensional measurement embedding vector, which serves as the input for subsequent knowledge fusion.

[0023] In some embodiments, semantic fusion is performed based on the measurement embedding vector and a preset battery degradation prompt to obtain a joint embedding representation, including: segmenting the text prompt to obtain a prompt token sequence, wherein the text prompt includes: temperature influence, discharge mode influence and task instruction; vectorizing the prompt token sequence using a preset embedding layer to obtain a knowledge embedding vector; concatenating the measurement embedding vector and the knowledge embedding vector to obtain a joint vector; and inputting the joint vector into a preset multi-head self-attention layer to output the joint embedding representation.

[0024] For example, the text prompts are segmented into word units, dividing the natural language text describing the effects of temperature, discharge mode, and task instructions into independent lexical units. The description of temperature effects includes technical terms such as "high temperature accelerates electrolyte decomposition," the description of discharge mode effects includes mechanistic explanations such as "high-rate discharge leads to lithium deposition," and the task instructions explicitly provide operational guidance such as "predict battery health status." All these lexical units form a prompt token sequence that is easy for the model to process. A pre-defined embedding layer is used to vectorize the prompt token sequence, mapping each lexical unit to a 256-dimensional dense vector representation. The sequence of all lexical vectors constitutes a knowledge embedding vector that preserves semantic relationships. The measurement embedding vector and the knowledge embedding vector are concatenated, connecting them along the feature dimension to form a 512-dimensional joint vector. This joint vector simultaneously contains feature information extracted from the measured data and semantic information extracted from domain knowledge. The joint vector is input into a pre-defined multi-head self-attention layer for processing. This layer contains eight independent attention heads that compute feature interactions in different representation subspaces in parallel. Each attention head learns different attention patterns and mines deep associations between data features and domain knowledge through a self-attention mechanism. The output is a joint embedding representation that includes both measured data characteristics and domain knowledge.

[0025] In some embodiments, the pre-defined battery analysis model includes a backbone network, a sparse attention module, a bidirectional gated recurrent unit, and a regression output layer. The joint embedding representation is input into the pre-defined battery analysis model to obtain a battery health prediction value. This includes: inputting the joint embedding representation into the backbone network to obtain a hidden state sequence; performing sparse correlation scoring on the hidden state sequence based on the sparse attention module to obtain attention weights; selectively aggregating the hidden state sequence according to the attention weights to obtain a feature representation; inputting the feature representation into the bidirectional gated recurrent unit for temporal dependency modeling to obtain a temporal feature vector; and inputting the temporal feature vector into the regression output layer for linear transformation to obtain the battery health prediction value.

[0026] For example, the backbone network in the pre-built battery analysis model receives a joint embedding representation as input. This backbone network consists of 12 stacked Transformer decoders, each containing 768-dimensional hidden states and 12 attention heads. High-level semantic features are extracted through multi-layer nonlinear transformations. The output of each layer serves as the input to the next layer, progressively abstracting the feature representation, resulting in a hidden state sequence rich in semantic information. A sparse attention module performs sparse relevance scoring on the hidden state sequence, calculating the correlation between each query vector and key vector and employing a Top-k selection strategy to retain only the top 30% of connections. This sparsity processing effectively reduces computational redundancy and allows the model to focus on the most relevant feature information. Based on the sparsified attention weights, the hidden state sequence is selectively aggregated, and the feature vectors corresponding to high weights are weighted and combined to obtain a condensed feature representation. This representation retains the most informative parts of the original sequence while filtering out secondary information. The feature representation is input into a bidirectional gated recurrent unit for temporal dependency modeling. This unit captures long-term dependencies in the sequence through recurrent neural networks in both forward and backward directions. Each direction uses 128 hidden units and processes the input sequence sequentially according to time steps. The last hidden states of the two directions are concatenated to form a 256-dimensional temporal feature vector. The temporal feature vector is then input into a regression output layer for linear transformation. This layer maps the 256-dimensional input features to a single numerical prediction of the battery health state through a weight matrix. The output value is limited to the range of 0 to 1 to represent the percentage ratio of battery capacity to initial capacity.

[0027] In some embodiments, a battery health status report is output based on a comparison and evaluation of the battery health prediction value and a preset threshold, including: reading historical data packets and determining a preset health status threshold based on the historical data packets; comparing the battery health prediction value with the preset health status threshold to obtain a performance degradation warning sign; generating an audible and visual alarm signal or a cloud notification message based on the performance degradation warning sign; and synthesizing a battery health status report based on the battery health prediction value and the performance degradation warning sign.

[0028] For example, historical data packets stored in the database are read. These packets contain historical health status records of batteries of the same type under the same operating conditions, including capacity decay curves, internal resistance growth trends, and cycle life statistics. Based on these records, a series of preset health status thresholds are determined to assess the current battery status. The real-time calculated battery health prediction values ​​are compared and analyzed with the preset health status thresholds. When the predicted value is below a critical threshold of 0.8, a performance degradation warning flag is generated; when the predicted value is below a warning threshold of 0.7, a severe warning flag is generated. These flags are stored in a register in binary code for subsequent processing. Based on the type and level of the performance degradation warning flag, an audible and visual alarm signal of corresponding intensity or a formatted cloud notification message is generated. The audible and visual alarm signal distinguishes the warning level by modulating pulse waveforms of different frequencies, while the cloud notification message encapsulates information such as the battery identifier, health status value, and timestamp according to a predefined JSON format. Battery health prediction values, performance degradation warning signs, and related operating parameters are integrated into a structured data set. A battery health status report containing current status assessment, warning information, and maintenance suggestions is generated according to a preset report template. This report is transmitted to a display device or remote monitoring platform through the communication interface of the battery management system for users to view and make decisions.

[0029] This invention uses cosine position encoding to process and recover voltage data to generate a query matrix. It then constructs a measurement embedding vector by combining a key-value matrix built from discharge measurement data. By semantically fusing the measurement embedding vector with preset battery degradation prompts, a joint embedding representation is constructed, which combines data-driven and knowledge-guided approaches. The joint embedding representation is analyzed by a battery analysis model to obtain a battery health prediction value. A health status report is generated based on the battery health prediction value, improving adaptability to different battery application scenarios and thus enhancing the accuracy of battery health status assessment.

[0030] Please see Figure 2 , Figure 2 This is a schematic block diagram of a BMS-based battery health status assessment system provided in an embodiment of this application. The BMS-based battery health status assessment device 200 is used to execute the aforementioned BMS-based battery health status assessment method. The BMS-based battery health status assessment device 200 can be configured in a server.

[0031] The server can be a standalone server, a server cluster, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms.

[0032] like Figure 2 As shown, the battery health status assessment device 200 based on BMS includes: a data acquisition module 201, a matrix generation module 202, a matrix fusion module 203, a semantic fusion module 204, a health analysis module 205, and a health comparison module 206.

[0033] The data acquisition module 201 is used to acquire battery discharge measurement data and recovery voltage data through the BMS.

[0034] The matrix generation module 202 is used to convert the recovered voltage data into a query matrix using cosine position encoding, and to extract the time-series features of the discharge measurement data to form a key matrix and a value matrix.

[0035] The matrix fusion module 203 is used to fuse the query matrix, key matrix and value matrix to obtain the measurement embedding vector.

[0036] The semantic fusion module 204 is used to perform semantic fusion based on the measured embedding vector and the preset battery degradation cues to obtain a joint embedding representation.

[0037] The health analysis module 205 is used to input the joint embedded representation into a preset battery analysis model to obtain a battery health prediction value.

[0038] The health comparison module 206 is used to compare and evaluate the battery health prediction value with a preset threshold and output a battery health status report.

[0039] This application provides an electronic device, which includes a memory and a processor; the memory is used to store a computer program; the processor is used to execute the computer program and, when executing the computer program, implement a battery health status assessment method based on a BMS as described in any of the embodiments of this application.

[0040] This application provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it enables the processor to implement a battery health status assessment method based on a BMS as described in any of the embodiments of this application.

[0041] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method of battery state of health estimation, characterized by, The evaluation method comprises: S101: collecting discharge measurement data and recovery voltage data of the battery by the BMS; S102: converting the recovery voltage data into a query matrix using cosine position encoding, and extracting time sequence features of the discharge measurement data to form a key matrix and a value matrix; S103: fusing the query matrix, the key matrix, and the value matrix to obtain a measurement embedding vector; S104: performing semantic fusion according to the measurement embedding vector and a preset battery degradation prompt to obtain a joint embedding representation; S105: inputting the joint embedding representation into a preset battery analysis model to obtain a battery health prediction value; S106: outputting a health status report of the battery according to a comparison evaluation of the battery health prediction value and a preset threshold value.

2. The method of claim 1, wherein, The discharge measurement data comprises a discharge voltage value sequence, a discharge current value sequence, and a discharge temperature value sequence.

3. The method of claim 1, wherein, The recovery voltage data is an end voltage recovery curve collected after the battery ends discharging and enters a resting state.

4. The method of claim 1, wherein, The time sequence features of the discharge measurement data are extracted by a multi-layer neural network architecture.

5. The method of claim 1, wherein, The battery analysis model is composed of a backbone network, a sparse attention module, and a regression output layer.

6. The method of claim 1, wherein, The battery health prediction value is a value between 0 and 1, which is a percentage ratio of the current battery capacity to the initial capacity.

7. The method of claim 1, wherein, The health status report of the battery includes a current health status level, a remaining service life prediction, and a performance degradation warning sign.

8. A battery state-of-health evaluation system based on the evaluation method of claim 1, characterized by, Comprise: a data collection module for collecting discharge measurement data and recovery voltage data of the battery by the BMS; a matrix generation module for converting the recovery voltage data into a query matrix using cosine position encoding, and extracting time sequence features of the discharge measurement data to form a key matrix and a value matrix; a matrix fusion module for fusing the query matrix, the key matrix, and the value matrix to obtain a measurement embedding vector; a semantic fusion module for performing semantic fusion according to the measurement embedding vector and a preset battery degradation prompt to obtain a joint embedding representation; a health analysis module for inputting the joint embedding representation into a preset battery analysis model to obtain a battery health prediction value; a health comparison module for outputting a health status report of the battery according to a comparison evaluation of the battery health prediction value and a preset threshold value.

9. An electronic device, comprising: comprise a memory and a processor; the memory is used to store a computer program; the processor is used to execute the computer program and implement the steps of the evaluation method according to any one of claims 1-7 when executing the computer program.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, the computer program is executed by the processor to implement the steps of the evaluation method according to any one of claims 1-7.

Images