A partial discharge diagnosis method and system based on statistical indicators, storage medium and computing device

Abstract

The application discloses a kind of based on statistical index's partial discharge diagnostic method, system, storage medium and computing device.The method is first to the partial discharge PRPS data sample is standardized and normalized processing, and corresponding fault type is converted into category label, generate high-quality sample set.Then, construct the deep neural network diagnostic model including the local statistical feature extraction network based on sliding window and fully connected classifier, complete the calculation of partial discharge category.Through calculating sample set category weight and using cross-entropy loss function with category weight, the deep neural network diagnostic model is trained.For the single partial discharge PRPS signal to be diagnosed, first normalized processing is carried out, then input into the model trained, and finally the corresponding partial discharge category label and its probability are output.

Description

Technical Field

[0001] This invention belongs to the field of power equipment fault diagnosis technology, specifically relating to a partial discharge diagnosis method, system, storage medium, and computing device based on statistical indicators. Background Technology

[0002] Partial discharge pulse sequence (PRPS) data, as an important signal carrier reflecting the characteristics of partial discharge, contains rich physical and statistical information. Traditional partial discharge diagnostic methods often rely on single or limited statistical features, neglecting the comprehensive utilization of the multi-dimensional and multi-level statistical characteristics of the signal. This results in poor adaptability of diagnostic models to complex operating conditions, and limited accuracy and robustness. In addition, partial discharge signals exhibit significant non-stationarity and time-varying characteristics due to factors such as the acquisition environment and equipment status. The statistical features of a single window are insufficient to fully characterize the signal features, thus limiting diagnostic performance.

[0003] With the rapid development of deep learning technology, partial discharge diagnosis methods based on deep neural networks have gradually become a research hotspot. Deep networks can automatically learn multi-level signal features, improving the automation and accuracy of diagnosis. However, most current deep learning methods lack parallel mining of multi-angle statistical indicators of signals during the feature extraction stage, failing to fully utilize the synergistic effect of multi-class statistical features, resulting in insufficient model representation of the complex internal structure and temporal changes of signals. Furthermore, the distribution of partial discharge fault category samples is often significantly imbalanced, and conventional training strategies have failed to effectively address this imbalance, leading to lower diagnostic accuracy for a few categories and affecting the overall diagnostic performance.

[0004] Therefore, in response to the aforementioned technical bottlenecks, there is an urgent need for a high-precision partial discharge fault diagnosis method based on multidimensional statistical indicators. Summary of the Invention

[0005] Purpose of the invention: To improve the accuracy of partial discharge diagnosis of power equipment, this invention provides a partial discharge diagnosis method, system, storage medium, and computing device based on statistical indicators.

[0006] Technical solution:

[0007] In a first aspect, the present invention proposes a partial discharge diagnosis method based on statistical indicators, comprising:

[0008] Partial discharge data samples are preprocessed, and sample labels are set according to the partial discharge fault type to form a sample set;

[0009] A partial discharge diagnostic model based on a deep neural network is constructed, including a feature extraction network, a fully connected classifier, and a linear transformation network, and the sample set is used as the input data of the partial discharge diagnostic model.

[0010] The feature extraction network introduces a sliding window mechanism to extract statistical indicators of the input data in several preset windows; the fully connected classifier processes the statistical indicators under all windows to form a set of statistical features; the linear transformation network receives the output of the fully connected classifier, completes the calculation of the partial discharge fault category, and outputs the score of each sample label.

[0011] The category weights are calculated based on the scores of the sample labels, and the partial discharge diagnostic model is trained using the cross-entropy loss function with category weights. The parameters are optimized through the backpropagation algorithm.

[0012] The partial discharge signal to be diagnosed is extracted and input into the trained partial discharge diagnosis model to obtain the diagnosis result of the partial discharge signal to be diagnosed.

[0013] Furthermore, the preprocessing includes standardizing and normalizing the partial discharge data samples to obtain a partial discharge data sample matrix; the sample labels include normal, spike, suspended, air gap, along surface, and particle, represented by 0, 1, 2, 3, 4, and 5 respectively.

[0014] Furthermore, the sample set includes a partial discharge data sample matrix and a sample label matrix, wherein the elements of the sample label matrix are the sample labels corresponding to the elements at the same position in the partial discharge data sample matrix.

[0015] Furthermore, the statistical indicators include: absolute maximum value indicator, peak-to-peak value indicator, root mean square value indicator, L1 energy indicator, activity ratio indicator, peak count indicator, burst rate indicator, active energy ratio indicator, energy entropy indicator, difference average absolute value indicator, difference maximum absolute value indicator, window standard deviation indicator, phase dispersion indicator, spacing variance indicator, and phase deviation indicator.

[0016] Furthermore, the fully connected classifier performs layer normalization on the statistical indicators to form a statistical feature set; the features in the statistical feature set are then sequentially fed into a dimensionality reduction network and a mapping network to obtain an optimized statistical feature set.

[0017] Furthermore, the linear transformation network includes an activation layer and a forgetting layer. The preferred statistical feature set is calculated using the ReLU activation function of the activation layer to obtain the first layer calculation result. The forgetting layer processes the first layer calculation result to obtain the second layer calculation result, which includes sample labels and predicted probabilities.

[0018] Furthermore, the partial discharge signal to be diagnosed is generated based on a single partial discharge data point collected by the sensor, and is preprocessed in the same manner as the sample set.

[0019] Secondly, the present invention proposes a partial discharge diagnostic system based on statistical indicators, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of any of the aforementioned methods.

[0020] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the aforementioned methods.

[0021] Fourthly, the present invention also provides a computing device including computer program instructions that, when executed by a processor, implement the steps of any of the aforementioned methods.

[0022] Beneficial effects:

[0023] This invention proposes a partial discharge diagnosis method based on statistical indicators. By standardizing and normalizing partial discharge PRPS data samples and converting them into category labels according to corresponding fault types, a high-quality sample set is constructed, ensuring the standardization and consistency of data input. At the same time, a parallel extraction technique of local statistical features based on sliding windows is adopted to comprehensively cover statistical indicators of multiple dimensions such as amplitude / energy, activity / density, energy concentration, local variation, phase dispersion, intermittency, and phase deviation. This fully explores the multi-level and multi-angle features of partial discharge signals, effectively improving the richness and identification ability of signal representation.

[0024] The fully connected classifier designed in this invention enhances the model's nonlinear expressiveness and generalization performance by deeply fusing layer normalization, dimensionality reduction, and mapping networks, combined with ReLU activation and Dropout regularization techniques, effectively preventing overfitting. By introducing a cross-entropy loss function with class weights, the problem of imbalanced sample classes is effectively alleviated, improving the recognition accuracy of minority classes and enhancing the model's robustness in practical applications. The backpropagation algorithm is used to optimize model parameters, achieving efficient training of the diagnostic model.

[0025] Ultimately, the normalization processing and input prediction of a single partial discharge PRPS signal can accurately output the corresponding partial discharge category label and its probability, providing a scientific basis for equipment insulation status assessment. Attached Figure Description

[0026] Figure 1 This is a flowchart of the method of the present invention;

[0027] Figure 2This is a structural diagram of the partial discharge diagnostic model of the present invention. Detailed Implementation

[0028] The present invention will be further explained below with reference to the accompanying drawings and specific embodiments. The present invention proposes a partial discharge diagnosis method based on region clustering, the specific process of which is as follows: Figure 1 As shown, it includes the following steps:

[0029] Step 1: Standardize and normalize the partial discharge (PRPS) data sample data, convert the corresponding fault types into category labels, and generate a sample set.

[0030] Extract PRPS data for several integer periods, remove redundant data, and normalize the extracted PRPS data using a standard data normalization method. Partial discharge fault types include six categories: normal, spike, suspended, air gap, surface, and particulate. These are labeled sequentially with integers from 0 to 5, resulting in the following sample set:

[0031]

[0032] in, For the corresponding sample set, The total number of samples; For the standardized and normalized PRPS data matrix, For the corresponding element, For data length; This is the label matrix corresponding to the sample. For the corresponding element.

[0033] Step 2: Construct a deep neural network-based partial discharge diagnostic model based on statistical indicators, including a sliding window-based local statistical feature extraction network and a fully connected classifier. The structure of the partial discharge diagnostic model is as follows: Figure 2 As shown.

[0034] The input to the local statistical feature extraction network is partial discharge PRPS data of a specified window length. The calculation method for the number of sliding windows can be expressed as follows:

[0035]

[0036] in, This represents the number of sliding windows; The length of the window; This represents the window sliding step size. This allows us to obtain the partial discharge PRPS data for each sliding window. Represented as:

[0037]

[0038] in, Index the data within each sliding window. Number the windows. Within each window, extract statistical features of the partial discharge PRPS data in parallel, and record the batch number of the statistical feature calculation as... And record Number the elements in the current batch, window The elements in the partial discharge PRPS data are represented as The statistical features described correspond to 15 indicators, which are introduced below according to indicator type.

[0039] (1) Amplitude / Energy:

[0040]

[0041] Absolute maximum value index The maximum value of all data points within the window is taken. The peak amplitude of the partial discharge pulse directly reflects the instantaneous intensity of the discharge. Different fault types (such as spike discharge, which often produces high-amplitude pulses, while air gap discharge may have lower amplitudes) differ in their peak values. This indicator can highlight the strongest discharge event within the window, helping to identify the severity and type of the discharge.

[0042] Peak-to-peak value index The peak-to-peak value is the difference between the maximum and minimum values ​​within the window. It reflects the signal's fluctuation range within that window. For partial discharges, amplitude fluctuations can vary depending on the discharge mechanism; for example, floating discharges may produce stable pulse amplitudes, while surface discharges may exhibit larger fluctuations. This indicator captures the dynamic range of the signal, enhancing the characterization of discharge stability.

[0043] Root mean square index The root mean square (RMS) is the square root of the average of the squares of all data points within the window, representing the effective value of the signal. It measures the average energy level of the signal and is sensitive to amplitude variations. The energy distribution of partial discharge is an important basis for distinguishing fault types; for example, particulate discharge may have high energy accumulation, while normal discharge has lower energy. This index integrates the contributions of all points within the window and effectively characterizes the overall discharge intensity.

[0044] L1 energy index L1 energy is the sum of the absolute values ​​of all data points within the window, i.e., the L1 norm. The L1 energy index is the accumulation of absolute signal values. It is less sensitive to abnormally large values ​​(such as spikes) than the root mean square (RMS) value, and can more robustly reflect the overall discharge activity within the window. In partial discharge, different faults have different pulse densities and amplitude distributions. L1 energy can capture the cumulative effect; for example, air gap discharge may produce dense but small-amplitude pulses, with potentially higher L1 energy and relatively lower RMS values, thus providing complementary information.

[0045] (2) Activity / Density Category:

[0046]

[0047] Active ratio indicator , which is the proportion of data points within the window that exceed a preset threshold, i.e., the ratio of active points to window length. For active mask, The activity threshold; This is a mask function, where 1 indicates presence and 0 indicates absence. The activity ratio reflects the frequency of discharges. Since different fault types (e.g., air gap discharges may produce sparse but regular pulses, while particle discharges may produce continuous and dense pulses) differ in discharge density, this metric effectively distinguishes between sparse and dense discharge patterns. The threshold can be set based on the background noise level or adaptively to ensure that only true discharge events are counted.

[0048] Peak count index , which represents the number of local maxima (peaks) within the window. Peak count directly counts the number of discharges occurring within the window, reflecting the frequency of discharges. Different fault types may have different discharge repetition rates; for example, spike discharges may manifest as high-frequency pulses, while floating discharges may occur less frequently but with stable amplitudes. This indicator helps quantify the frequency of discharge activity and complements the activity ratio: the activity ratio focuses on the total active duration, while the peak count focuses on the number of events.

[0049] Outbreak rate index This is the ratio of peak count to the total number of active points within the window (i.e., window length multiplied by the active percentage). If discharge activity is continuous and each active point is almost a peak, the burst rate is close to 1; if discharge activity is sparse and the intervals between peaks are long, the burst rate is low. For example, surface discharge may exhibit continuous small fluctuations, resulting in fewer peaks and a larger number of active points, leading to a low burst rate; while particulate discharge may produce rapid, continuous multi-peak pulses, resulting in a high burst rate. This indicator helps identify the clustering and suddenness of discharges, enhancing the ability to distinguish between different fault modes. It is a very small non-zero constant used to prevent division by zero errors.

[0050] (3) Energy Concentration Category:

[0051]

[0052] Active energy percentage index , is the ratio of the sum of instantaneous energies with amplitudes exceeding the threshold within the window to the total energy of the window. Where, This represents instantaneous energy. The active energy percentage measures the proportion of energy carried by an effective discharge event in the total energy. If this ratio is close to 1, it indicates that almost all the signal energy is concentrated on the discharge pulse, with minimal contribution from background noise or non-discharge components; if the ratio is small, it indicates that the energy is dispersed in inactive regions, potentially containing more noise or continuous small fluctuations.

[0053] Energy entropy index Defined as a concept based on information theory, it treats the proportion of instantaneous energy of each data point to the total energy as a probability and calculates its Shannon entropy. In the formula... This represents the energy probability distribution. The energy entropy value reflects the uniformity of energy distribution within the window: a larger entropy indicates a more uniform energy distribution (i.e., energy contributions at each point are similar); a smaller entropy indicates that energy is concentrated at a few points. Energy entropy quantifies the disorder or concentration of energy distribution and is an important characteristic for distinguishing discharge types. It is unaffected by the absolute amplitude of the signal, focusing only on the relative distribution, and therefore possesses scale invariance. Combined with the proportion of active energy, it can more comprehensively describe energy characteristics.

[0054] (4) Local changes:

[0055]

[0056] Difference Mean Absolute Value Index , is defined as the absolute value of the amplitude difference between two adjacent points within the calculation window, and then the average value is taken. This is local differential. This index quantifies the average amplitude of local fluctuations in a signal, that is, the average change of the signal within a unit of time (sampling interval). The rising and falling edges of a partial discharge pulse can cause a large difference between adjacent points, while the difference is smaller in the steady-state segment.

[0057] Maximum absolute value of difference index , defined as the maximum absolute value of the differences between all adjacent points within a window. It is used to capture the most dramatic single change within the window, identify strong abrupt events, and distinguish pulse shapes.

[0058] Window standard deviation index The standard deviation is defined as the dispersion of all data points within a window relative to the mean. This is the mean value within the window. It is used to characterize overall volatility and supplement energy distribution, reflecting the amplitude of fluctuations more purely.

[0059] (5) Phase dispersion type, intermittent type, phase deviation type:

[0060]

[0061] Phase dispersion index This index typically measures the concentration of discharge energy in the phase using weighted variance. This is the weighted average of the amplitude values. The larger the value, the more dispersed the discharge energy is in phase; the smaller the value, the more concentrated the energy is in a certain phase region. For example, air gap discharges are usually symmetrically distributed in the positive and negative half-cycles, with a large phase dispersion; while spike discharges may be concentrated near the voltage peak, with a smaller dispersion.

[0062] Spacing variance index This is used to describe the discontinuity of the discharge pulse, that is, the irregularity of the switching frequency or interval between active and inactive states. To enable active mask difference, This represents the average value of the active mask. The larger this value, the more uneven the discharge intervals and the stronger the intermittency. For example, suspended discharges may exhibit stable periodic pulses with uniform intervals and small intermittency; while particle discharges may occur randomly with large interval variations and strong intermittency.

[0063] Phase deviation index This value measures the asymmetry of discharge energy between the positive and negative half-cycles, that is, the degree of shift in the discharge phase relative to the voltage peak phase. A value close to 0.5 indicates a symmetrical distribution, while a deviation from 0.5 indicates that the discharge is biased towards a certain half-cycle. For example, surface discharge may mainly occur in one polarity half-cycle, while discharge in normal insulation may be symmetrical.

[0064] The aforementioned 15 indicators are innovative indicators proposed for five specific discharge fault mechanisms: "spikes, suspension, air gaps, surface discharges, and particles," and are combined with a sliding window data partitioning method. This invention utilizes these 15 indicators to comprehensively characterize partial discharge signals from different physical dimensions (intensity, frequency, distribution, rate of change, etc.), aiming to capture all key information regarding the global and local, time and phase domains, intensity, and volatility of partial discharge signals. Compared to traditional methods, the indicator system established in this invention can obtain richer and more comprehensive information. For example, the burst rate indicator (S7) is specifically used to characterize the random and sudden characteristics of particle discharge, while the phase deviation indicator (S... 15 This is used to capture the polarity asymmetry of surface discharge.

[0065] The fully connected classifier first integrates the metrics from all windows, performs layer normalization, and forms a set of statistical features, represented as: ,in It is a set of statistical features normalized to the layer, with dimension . , The total number of windows, For batch size, This is the layer normalization operator.

[0066] The features in the statistical feature set are sequentially fed into a dimensionality reduction network and a mapping network to achieve optimal selection of statistical features, as shown below: ,in Optimize results based on statistical characteristics; For dimensionality reduction networks, the dimension is... 8 is the preferred feature dimension after dimensionality reduction; For tensor flattening operators, the dimension is . ; For mapping networks, the dimension is 256 is the preferred feature dimension after flattening.

[0067] The reduced and flattened feature dimensions are both hyperparameters that can be tuned based on model complexity, data volume, and classification performance. This paper demonstrates through experiments that 8-dimensional and 256-dimensional features achieve a good balance between preserving key information and reducing redundancy.

[0068] Finally, a linear transformation network containing ReLU activation and Dropout layers is used to calculate the partial discharge category. The linear transformation network consists of two layers, the first layer of which is represented as follows: ,in This is the result of the first layer calculation; Calculate the operator for the ReLU activation function; For the corresponding weights, The hidden layer dimension is 512; For the corresponding bias, In the linear transformation network, the dimensions of each layer are hyperparameters selected in the design and are not absolutely immutable. However, in practical applications, researchers can flexibly adjust the dimension size according to task requirements (such as changes in the number of categories or different data scales), as long as the network structure remains reasonable.

[0069] To prevent overfitting, a Dropout layer is added after the ReLU output as the second layer of the linear transformation network. The final output is represented as follows: ,in For the final output, ; This represents the total number of categories, with a value of 6, corresponding to 6 types of partial discharge fault states. For network forgetting operator; For the corresponding weights, ; For the corresponding bias, .

[0070] The final output result These are the unnormalized raw scores (logits) of the network. To obtain the probability that the signal to be diagnosed belongs to each category, it is necessary to... Applying the Softmax function yields a probability distribution vector, with each element ranging from [0,1] and summing to 1. The diagnostic result typically selects the category with the highest probability as the predicted fault type. It can also output the probability values ​​for all categories for further analysis (such as confidence assessment).

[0071] Step 3: Calculate the class weights of the sample set, train the partial discharge diagnostic model using the cross-entropy loss function with class weights, and optimize the neural network parameters based on the loss function using the backpropagation algorithm.

[0072] The method for calculating the class weights of the sample set is expressed as follows:

[0073]

[0074] in, For the class weights of the sample set, Assign category number; This is the base-e logarithmic operator; For category The number of times it appears in the sample category matrix Y; This is the maximum value operator.

[0075] The cross-entropy loss with class weights is expressed as:

[0076]

[0077] in, This represents the cross-entropy loss value. For sample numbers within a batch; For the first The weight of the category to which each sample belongs, the specific value of which is given above. The calculation formula is determined; For exponentiation operator; This represents the final calculation result of the batch of samples fed into the partial discharge diagnostic model. The subscript y(b) indicates the first... A function representing the category to which a sample belongs, such as y(1)=0, which means that the first sample belongs to category number 0 (normal), and the subscript c is the category number index.

[0078] The deep neural network based on statistical indicators is trained by calculating the class weights of the sample set and using a cross-entropy loss function with class weights. The neural network parameters are then optimized using the backpropagation algorithm based on the loss function, thus completing the training. Training is complete when the set number of iterations is reached.

[0079] Step 4: Extract a single partial discharge PRPS signal, perform normalization processing to generate a signal to be diagnosed, and input it into a trained deep neural network based on statistical indicators to finally obtain the partial discharge category label and corresponding probability of the signal to be diagnosed.

[0080] The normalization method for the signal to be diagnosed is the same as the sample set construction method in step 1. In the deep neural network partial discharge diagnostic model based on statistical indicators, the calculation method for the final predicted probability of each category is expressed as follows:

[0081]

[0082] in, Predict the probability for the corresponding category; For category numbering, , It is the total number of categories; The output results are for the corresponding category model. The calculation method for the final partial discharge category label of the signal to be diagnosed is expressed as follows: ,in For the final category label; To obtain the category index function that maximizes the probability.

[0083] On the other hand, the present invention also proposes a partial discharge diagnostic system based on statistical indicators, comprising:

[0084] Sample set generation module: Standardizes and normalizes partial discharge PRPS data samples, converts them to corresponding category labels, and generates a sample set.

[0085] The partial discharge diagnostic model construction module employs a sliding window-based local statistical feature extraction network to extract statistical features in parallel. All statistical indicators are then merged and normalized layer by layer before being fed into a dimensionality reduction network and a mapping network. Finally, the model passes through a two-layer linear transformation network containing ReLU activation and Dropout layers to calculate the partial discharge category.

[0086] Partial Discharge Diagnostic Model Training Module: By calculating the class weights of the sample set, the deep neural network based on statistical indicators is trained using a cross-entropy loss function with class weights. The neural network parameters are optimized based on the loss function using a backpropagation algorithm to complete the training.

[0087] Partial Discharge Diagnostic Model Inference Module: Extracts a single partial discharge PRPS signal, performs normalization processing, generates a signal to be diagnosed, and inputs it into a trained deep neural network based on statistical indicators to finally obtain the partial discharge category label and corresponding probability of the signal to be diagnosed.

Claims

1. A partial discharge diagnosis method based on statistical indicators, characterized in that, include: Partial discharge data samples are preprocessed, and sample labels are set according to the partial discharge fault type to form a sample set; A partial discharge diagnostic model based on a deep neural network is constructed, including a feature extraction network, a fully connected classifier, and a linear transformation network, and the sample set is used as the input data of the partial discharge diagnostic model. The feature extraction network introduces a sliding window mechanism to extract statistical indicators of the input data in several preset windows; the fully connected classifier processes the statistical indicators under all windows to form a set of statistical features. The linear transformation network receives the output of the fully connected classifier, completes the partial discharge fault category calculation, and outputs the score of each sample label; The category weights are calculated based on the scores of the sample labels, and the partial discharge diagnostic model is trained using the cross-entropy loss function with category weights. The parameters are optimized through the backpropagation algorithm. The partial discharge signal to be diagnosed is extracted and input into the trained partial discharge diagnosis model to obtain the diagnosis result of the partial discharge signal to be diagnosed.

2. The partial discharge diagnostic method according to claim 1, characterized in that, The preprocessing includes standardizing and normalizing the partial discharge data samples to obtain a partial discharge data sample matrix; the sample labels include normal, spike, suspended, air gap, along surface, and particle, which are represented by 0, 1, 2, 3, 4, and 5 respectively.

3. The partial discharge diagnostic method according to claim 2, characterized in that, The sample set includes a partial discharge data sample matrix and a sample label matrix, wherein the elements of the sample label matrix are the sample labels corresponding to the elements at the same position in the partial discharge data sample matrix.

4. The partial discharge diagnostic method according to claim 3, characterized in that, The statistical indicators include: absolute maximum value, peak-to-peak value, root mean square value, L1 energy, activity ratio, peak count, burst rate, active energy percentage, energy entropy, difference mean absolute value, difference maximum absolute value, window standard deviation, phase dispersion, spacing variance, and phase deviation.

5. The partial discharge diagnostic method according to claim 4, characterized in that, The fully connected classifier performs layer normalization on the statistical indicators to form a set of statistical features; the features in the set of statistical features are then fed into a dimensionality reduction network and a mapping network in sequence to obtain an optimized set of statistical features.

6. The partial discharge diagnostic method according to claim 5, characterized in that, The linear transformation network includes an activation layer and a forgetting layer. The preferred statistical feature set is calculated using the ReLU activation function of the activation layer to obtain the first layer calculation result. The forgetting layer processes the first layer calculation result to obtain the second layer calculation result, which includes sample labels and predicted probabilities.

7. The partial discharge diagnostic method according to claim 6, characterized in that, The partial discharge signal to be diagnosed is generated based on a single partial discharge data point collected by the sensor and is preprocessed in the same manner as the sample set.

8. A partial discharge diagnostic system based on statistical indicators, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of any of the methods of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of any of the methods described in claims 1 to 7.

10. A computing device comprising computer program instructions, characterized in that, When executed by a processor, the computer program instructions implement the steps of any of the methods described in claims 1 to 7.

Images