Method and system for evaluating credibility of artificial intelligence model in electric power field

Abstract

According to the credibility evaluation method and system for the artificial intelligence model in the electric power field, the uncertainty of output of the middle layer of the AI model is quantified through the joint information entropy, a basis is provided for understanding the working mechanism and the information transmission process in the AI model, meanwhile, the credibility of the model can be evaluated from the perspective of the information theory, and the reliability of the AI model is improved. The change trend of the joint information entropy of each layer of the AI model can explain the credibility transfer mode of each layer of the AI model, uncertainty sources and weak links of information transfer in the model can be effectively identified, and reference is provided for optimization and improvement of the model.

Description

Technical Field

[0001] The present invention belongs to the technical field related to artificial intelligence, and in particular relates to a credibility assessment method and system for an artificial intelligence model in the electric power field. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] With the gradual expansion of the scale of power grids and the widespread access to new energy sources, traditional stability assessment methods such as time domain simulation face problems such as inaccurate modeling, mismatched operation modes, and long calculation time in practical applications, and power grid stability assessment faces new challenges. Power grid stability judgment methods based on artificial intelligence have received widespread attention due to their powerful data processing and learning capabilities. With the advancement of artificial intelligence (AI) technology, its application in stability assessment is no longer limited to traditional machine learning models such as support vector machines. Deep learning models represented by neural networks have become a research hotspot and a potential feasible way to efficiently and accurately assess the stability of new power systems.

[0004] At present, artificial intelligence has been widely used in image recognition, medical and other fields, as well as various aspects of the power system, including load forecasting, fault diagnosis, equipment defect identification, electricity price forecasting, etc. However, there are still many risks and shortcomings in actual scenarios with high security requirements such as power grid stability judgment. For example, the black box characteristics of artificial intelligence models make it impossible for power grid operators to trust the stability judgment results of artificial intelligence models. Therefore, analyzing the credibility transfer mode of artificial intelligence models and analyzing the credibility of the output of each layer of the model are of great significance to understanding the internal working mechanism of artificial intelligence models and realizing the application of artificial intelligence models in power systems. Therefore, how to conduct in-depth and effective analysis of the information transmission process between internal levels of artificial intelligence is of great significance to understanding the behavior of artificial intelligence models and subsequent targeted iterative improvements to the weak links of artificial intelligence models. Summary of the invention

[0005] In order to overcome the shortcomings of the above-mentioned prior art, the present invention provides a credibility assessment method and system for an artificial intelligence model in the power field. By quantifying the uncertainty of the output of the intermediate layer of the AI ​​model by combining information entropy, a basis is provided for understanding the internal working mechanism and information transmission process of the AI ​​model. At the same time, the credibility of the model can be evaluated from the perspective of information theory, and the sources of uncertainty in the AI ​​model and the weak links in information transmission can be effectively identified, thereby providing a reference for the optimization and improvement of the AI ​​model.

[0006] In order to achieve the above object, the present invention adopts the following technical solution:

[0007] In a first aspect, the present invention provides a credibility assessment method for an artificial intelligence model in the electric power field, comprising:

[0008] Obtain the intermediate layer output features of the AI ​​model to be evaluated and perform preprocessing;

[0009] Perform dimension reduction processing on the preprocessed intermediate layer output features to obtain dimension reduction features;

[0010] Estimate the probability distribution of reduced-dimensional features based on kernel density estimation;

[0011] Calculate the joint information entropy according to the probability distribution corresponding to the calculated dimensionality reduction features;

[0012] According to the joint information entropy corresponding to each intermediate layer of the artificial intelligence model to be evaluated, the credibility assessment result of the artificial intelligence model to be evaluated is obtained.

[0013] In a second aspect, the present invention provides a credibility assessment system for an artificial intelligence model in the electric power field, comprising:

[0014] An acquisition module, which is configured to: acquire the intermediate layer output features of the artificial intelligence model to be evaluated and perform preprocessing;

[0015] A dimension reduction module is configured to: perform dimension reduction processing on the preprocessed intermediate layer output features to obtain dimension reduction features;

[0016] An estimation module, configured to: estimate the probability distribution of the dimension reduction feature based on a kernel density estimation method;

[0017] A calculation module, configured to: calculate the joint information entropy according to the probability distribution corresponding to the calculated dimensionality reduction features;

[0018] The evaluation module is configured to obtain the credibility evaluation result of the artificial intelligence model to be evaluated based on the joint information entropy corresponding to each intermediate layer of the artificial intelligence model to be evaluated.

[0019] In a third aspect, the present invention provides an electronic device comprising a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the method described in the first aspect is performed.

[0020] In a fourth aspect, the present invention provides a computer-readable storage medium for storing computer instructions, wherein when the computer instructions are executed by a processor, the method described in the first aspect is performed.

[0021] In a fifth aspect, the present invention provides a computer program product, comprising a computer program, which, when executed by a processor, implements the method described in the first aspect.

[0022] One or more of the above technical solutions have the following beneficial effects:

[0023] In the present invention, the uncertainty of the output of the middle layer of the AI ​​model is quantified by the joint information entropy, which provides a basis for understanding the internal working mechanism and information transmission process of the AI ​​model, and can evaluate the credibility of the model from the perspective of information theory. The changing trend of the joint information entropy of each layer of the AI ​​model can explain the credibility transfer pattern of each layer of the AI ​​model, and can effectively identify the sources of uncertainty in the model and the weak links in information transmission, and provide a reference for the optimization and improvement of the model.

[0024] In the present invention, the output features of the middle layer of the AI ​​model are subjected to dimensionality reduction processing, which can improve the accuracy of probability distribution estimation and reduce computational complexity.

[0025] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0027] Figure 1 This is a flow chart of a credibility assessment method for an artificial intelligence model in the electric power field in Embodiment 1 of the present invention;

[0028] Figure 2 This is the change trend of the frequency stability discrimination model combined with information entropy in the first embodiment of the present invention. DETAILED DESCRIPTION

[0029] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0030] It should be noted that the terms used herein are for describing specific embodiments only and are not intended to be limiting of exemplary embodiments according to the present invention.

[0031] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0032] The artificial intelligence model to be evaluated in this embodiment is an artificial intelligence model in the power field, such as: a model for identifying defects or faults of power equipment based on images, a model for identifying or classifying power equipment based on images, etc.

[0033] Embodiment 1

[0034] In order to deeply analyze the changing process of the internal state of the artificial intelligence model, this embodiment uses the joint information entropy to quantify the complexity and uncertainty of the output of the intermediate layer of the AI ​​model. The changing trend of the joint information entropy can effectively reflect the model credibility transfer pattern.

[0035] like Figure 1 As shown, this embodiment discloses a credibility assessment method for an artificial intelligence model in the electric power field, including:

[0036] Step 1: Obtain the intermediate layer output features of the AI ​​model to be evaluated and perform preprocessing.

[0037] Specifically, preprocessing includes: converting the high-dimensional features output by the middle layer of the artificial intelligence model to be evaluated into a one-dimensional vector to form a two-dimensional matrix, that is, the number of samples * the number of features.

[0038] For example, if the input of the artificial intelligence model to be evaluated has m samples, each sample is flattened into a one-dimensional vector of length n, and the one-dimensional vectors of all samples are combined into an m*n two-dimensional matrix, where the number of rows of the matrix is ​​equal to the number of samples, and the number of columns is equal to the number of features.

[0039] Step 2: Perform dimensionality reduction processing on the preprocessed intermediate layer output features to obtain dimensionality reduction features.

[0040] Specifically, since the output of the middle layer of the AI ​​model usually has a high dimension and a sparse sample density, this embodiment uses the principal component analysis method to process the preprocessed middle layer output features to reduce the feature dimension, thereby improving the accuracy of the probability distribution estimation and reducing the computational complexity.

[0041] Since the output of the middle layer of the AI ​​model is usually of high dimensionality, the number of features in the two-dimensional matrix obtained after preprocessing is large, even up to tens of thousands. This embodiment uses PCA principal component analysis to reduce the number of features to k while retaining the main information in the data, forming an m*k matrix after dimensionality reduction, where k is the number of features after dimensionality reduction.

[0042] Assume that the output of a certain layer in the middle of the model contains 2000 samples and 128 features, that is, a 2000*128 two-dimensional matrix. The matrix is ​​subjected to PCA dimensionality reduction. The main processing process is: standardize the two-dimensional matrix so that the mean of each feature is 0 and the variance is 1; then calculate the corresponding covariance matrix, and perform eigenvalue decomposition on the covariance matrix to obtain eigenvalues ​​and eigenvectors. The size of the eigenvalue represents the variance of the original data in the direction of the eigenvector. Select k eigenvectors with larger eigenvalues ​​as principal components and form a new feature space; project the original data into the new feature space to obtain the reduced-dimensional data, that is, an m*k matrix.

[0043] Step 3: Estimate the probability distribution of the reduced dimension features based on the kernel density estimation method.

[0044] Specifically, kernel density estimation is a non-parametric statistical method used to estimate the probability density function of a random variable. Appropriate kernel function and bandwidth should be selected before calculation. The choice of kernel function has an important influence on the result of kernel density estimation. Commonly used kernel functions include Gaussian kernel, uniform kernel, triangular kernel, etc. Gaussian kernel can usually be selected because of its good mathematical properties and smoothness; bandwidth is a key parameter in kernel density estimation, which determines the width and smoothness of the kernel function. The selection of bandwidth can be carried out through cross-validation, maximum likelihood estimation and other methods.

[0045] After selecting the appropriate kernel function and bandwidth, the joint probability density of each feature in the sample can be calculated. Assume that there are n samples, each sample contains k features. Take one sample z as an example, and calculate the high-dimensional vector corresponding to the estimated sample z and each sample z. i The distance u between the corresponding high-dimensional vectors i =(zz i ) / h, where h is the bandwidth; the calculated distance u i Substitute into the selected kernel function K(u) to get each sample z i The probability density contribution of the sample z to be estimated; add the probability density contributions of all samples and divide them by the total number of samples n and the bandwidth h to obtain the joint probability density estimate of the k features of the sample z to be estimated, that is:

[0046]

[0047] in, It is the joint probability density estimate of the k features of the sample z to be estimated, and the bandwidth h is a parameter in the kernel density estimation.

[0048] Step 4: Calculate the joint information entropy based on the probability distribution corresponding to the calculated dimensionality reduction features.

[0049] For two random variables X and Y, their joint information entropy H(X,Y) is defined as:

[0050]

[0051] More generally, for n random variables X1, X2, …, X n , and its joint information entropy is calculated as:

[0052]

[0053] Among them, p(x1,x 2, ...,xn ) are n random variables X1, X2, …, X n The joint probability density of can be calculated by the kernel density KDE.

[0054] In this embodiment, the k features after dimensionality reduction are regarded as k random variables, namely X1, X2, ..., X k , and the calculation result of the joint probability of KDE is substituted into the formula to obtain the calculation result of the joint information entropy. That is, the calculation formula of the joint information entropy is:

[0055]

[0056] Among them, X1, X2, …, X k are the k features after dimensionality reduction, p(x1,x2,...,x k ) is the joint probability density of k features corresponding to a sample.

[0057] The value of the joint information entropy corresponding to the middle layer of the AI ​​model to be evaluated is proportional to the uncertainty of the data distribution, that is, the greater the uncertainty of the data distribution, the greater the value of the joint information entropy, and vice versa, the smaller the uncertainty of the data distribution, the smaller the joint information entropy. If the probability of each value of the random variable is equal, then its joint information entropy reaches the maximum value.

[0058] Step 5: Based on the joint information entropy corresponding to each intermediate layer of the artificial intelligence model to be evaluated, obtain the credibility assessment result of the artificial intelligence model to be evaluated.

[0059] The artificial intelligence model aims to explore the mapping relationship between input and output by gradually extracting information from the input data. It explores specific rules from complex and disordered input data, and gradually retains the high-level features that are most relevant to the output results by integrating or aggregating features and transforming features. In this process, the randomness and uncertainty of data distribution should gradually decrease and approach a specific distribution. Specifically, the artificial intelligence model uses a series of processing methods such as convolution and pooling to transform the input data from a disordered state to an ordered state, thereby extracting information closely related to the output data and further revealing the mapping relationship between input and output.

[0060] The uncertainty of the output of the artificial intelligence model can reveal the credibility of the model. The smaller the uncertainty, the higher the credibility of the model. Taking the output layer as an example, if the probability distribution of the prediction result is more concentrated, that is, the smaller the uncertainty, the higher its credibility; conversely, if the probability distribution of the prediction result is more uniform, for example, the probability of the model predicting "stable" and "instability" results is almost equal, then the greater the uncertainty, the lower its credibility.

[0061] The joint information entropy can quantify uncertainty and can be an effective credibility measurement tool. Joint information entropy is a concept in information theory and an effective measure of the uncertainty of random variables. It reveals the degree of chaos or randomness in the data distribution. Specifically, the joint information entropy represents the degree of uncertainty when two or more variables occur simultaneously.

[0062] Joint information entropy can be an effective credibility measurement tool. The smaller the joint information entropy, the greater the credibility, and vice versa. The joint information entropy of the output of each layer of the AI ​​model is calculated using the joint information entropy. The change trend of the joint information entropy can reveal the increase or decrease in credibility. If the joint information entropy of a certain layer is significantly reduced, it indicates that the layer has less uncertainty and randomness in the output by compressing and extracting features, and the credibility of the layer increases accordingly. If the joint information entropy of a certain layer increases significantly, it indicates that the layer has introduced more complexity and uncertainty, and the credibility decreases. By analyzing the change trend of the joint information entropy, we can gain a deep understanding of the credibility transfer mode of the AI ​​model. This process is crucial to revealing the change process of the internal state of the AI ​​model and its working mechanism. It is an effective means to understand the behavior of complex AI models and to make targeted iterative improvements to the weak links of the AI ​​model.

[0063] Taking the frequency stability discrimination model as an example, the AI ​​model consists of a convolutional layer, a pooling layer, and a fully connected layer structure. The joint information entropy output by each layer of the model is calculated using the method of this embodiment. The results are as follows: Figure 2 As shown in the figure, it can be seen that the joint information entropy gradually decreases with the deepening of the number of layers, indicating that the credibility gradually increases with the deepening of the number of layers. The convolution layer reduces redundant information by extracting features, reduces the complexity and uncertainty of the data, and makes the joint information entropy smaller; the pooling layer downsamples and filters the features, further reducing the dimension and redundancy of the data, so that the joint information entropy is reduced; the fully connected layer can better capture the laws in the data by integrating and reorganizing features and introducing nonlinear activation functions, thereby improving the model's ability to fit the data and the prediction accuracy, which is conducive to reducing the uncertainty of the data, and the joint information entropy is further reduced.

[0064] Embodiment 2

[0065] The purpose of this embodiment is to provide a credibility assessment system for an artificial intelligence model in the electric power field, including:

[0066] An acquisition module, which is configured to: acquire the intermediate layer output features of the artificial intelligence model to be evaluated and perform preprocessing;

[0067] A dimension reduction module is configured to: perform dimension reduction processing on the preprocessed intermediate layer output features to obtain dimension reduction features;

[0068] An estimation module, configured to: estimate the probability distribution of the dimension reduction feature based on a kernel density estimation method;

[0069] A calculation module, configured to: calculate the joint information entropy according to the probability distribution corresponding to the calculated dimensionality reduction features;

[0070] The evaluation module is configured to obtain the credibility evaluation result of the artificial intelligence model to be evaluated based on the joint information entropy corresponding to each intermediate layer of the artificial intelligence model to be evaluated.

[0071] In further embodiments, there is also provided:

[0072] An electronic device includes a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the method described in Embodiment 1 is performed. For the sake of brevity, no further description is given here.

[0073] It should be understood that in this embodiment, the processor may be a central processing unit CPU, and the processor may also be other general-purpose processors, digital signal processors DSP, application-specific integrated circuits ASIC, off-the-shelf programmable gate arrays FPGA or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0074] The memory may include a read-only memory and a random access memory, and provide instructions and data to the processor. A portion of the memory may also include a non-volatile random access memory. For example, the memory may also store information about the device type.

[0075] A computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by a processor, the method described in embodiment 1 is completed.

[0076] The method in the first embodiment can be directly embodied as a hardware processor, or a combination of hardware and software modules in the processor. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware. To avoid repetition, it will not be described in detail here.

[0077] A computer program product includes a computer program, and when the computer program is executed by a processor, the method described in the first embodiment is implemented.

[0078] The present invention also provides at least one computer program product tangibly stored on a non-transitory computer-readable storage medium. The computer program product includes computer executable instructions, such as instructions included in a program module, which are executed in a device on a real or virtual processor of the target to perform the process / method as described above. Typically, program modules include routines, programs, libraries, objects, classes, components, data structures, etc. that perform specific tasks or implement specific abstract data types. In various embodiments, the functions of program modules can be combined or divided between program modules as needed. Machine executable instructions for program modules can be executed in local or distributed devices. In distributed devices, program modules can be located in local and remote storage media.

[0079] The computer program code for implementing the method of the present invention can be written in one or more programming languages. These computer program codes can be provided to the processor of a general-purpose computer, a special-purpose computer or other programmable data processing device, so that the program code, when executed by the computer or other programmable data processing device, causes the function / operation specified in the flow chart and / or block diagram to be implemented. The program code can be executed completely on a computer, partially on a computer, as an independent software package, partially on a computer and partially on a remote computer or completely on a remote computer or server.

[0080] In the context of the present invention, computer program codes or related data may be carried by any appropriate carrier to enable a device, apparatus or processor to perform the various processes and operations described above. Examples of carriers include signals, computer readable media, etc. Examples of signals may include electrical, optical, radio, acoustic or other forms of propagation signals, such as carrier waves, infrared signals, etc.

[0081] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with this embodiment can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0082] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A credibility assessment method for an artificial intelligence model in the power field, characterized in that: include: Obtain the intermediate layer output features of the AI ​​model to be evaluated and perform preprocessing; Perform dimension reduction processing on the preprocessed intermediate layer output features to obtain dimension reduction features; Estimate the probability distribution of reduced-dimensional features based on kernel density estimation; Calculate the joint information entropy according to the probability distribution corresponding to the calculated dimensionality reduction features; According to the joint information entropy corresponding to each intermediate layer of the artificial intelligence model to be evaluated, the credibility assessment result of the artificial intelligence model to be evaluated is obtained.

2. The credibility assessment method of an artificial intelligence model in the electric power field according to claim 1, characterized in that: The obtained intermediate layer output features of the artificial intelligence model to be evaluated are preprocessed, specifically: the intermediate layer output features of the artificial intelligence model to be evaluated are converted into a one-dimensional vector to form a two-dimensional matrix.

3. The credibility assessment method of an artificial intelligence model in the electric power field according to claim 1, characterized in that: The principal component analysis method is used to reduce the dimension of the preprocessed intermediate layer output features.

4. The credibility assessment method of an artificial intelligence model in the electric power field according to claim 1, characterized in that: The probability distribution of dimensionality reduction features is estimated based on the kernel density estimation method, specifically: Calculate the distance between the dummy sample to be estimated and each sample; The calculated distance is combined with the kernel function formula to obtain the probability density contribution of each sample to the sample to be estimated; The contribution of all samples to the probability density of the sample to be estimated is added up, and combined with the total number of samples and the bandwidth, the estimated value of the joint probability density at the sample to be estimated is obtained.

5. The credibility assessment method of an artificial intelligence model in the electric power field according to claim 1, characterized in that: The value of the joint information entropy corresponding to the intermediate layer of the artificial intelligence model to be evaluated is proportional to the uncertainty of the data distribution.

6. The credibility assessment method of an artificial intelligence model in the electric power field according to claim 1, characterized in that: According to the changing trend of the joint information entropy corresponding to each intermediate layer sequentially included in the artificial intelligence model to be evaluated, the credibility of each intermediate layer and the credibility evaluation result of the artificial intelligence model to be evaluated are obtained.

7. A credibility assessment system for artificial intelligence models in the power field, characterized in that: include: An acquisition module, which is configured to: acquire the intermediate layer output features of the artificial intelligence model to be evaluated and perform preprocessing; A dimension reduction module is configured to: perform dimension reduction processing on the preprocessed intermediate layer output features to obtain dimension reduction features; An estimation module, configured to: estimate the probability distribution of the dimension reduction feature based on a kernel density estimation method; A calculation module, configured to: calculate the joint information entropy according to the probability distribution corresponding to the calculated dimensionality reduction features; The evaluation module is configured to obtain the credibility evaluation result of the artificial intelligence model to be evaluated based on the joint information entropy corresponding to each intermediate layer of the artificial intelligence model to be evaluated.

8. An electronic device, characterized in that: The method comprises a memory and a processor and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the method according to any one of claims 1 to 6 is completed.

9. A computer-readable storage medium, characterized in that: Used to store computer instructions, which, when executed by a processor, complete the method described in any one of claims 1 to 6.

10. A computer program product, characterized in that The invention comprises a computer program, which, when executed by a processor, implements the method according to any one of claims 1 to 6.

Images