AI-assisted full-automatic weather radar spare part performance detection and evaluation method and device

Abstract

The invention provides an AI-assisted full-automatic weather radar spare part performance detection and evaluation method and device, and the method employs an integrated machine learning algorithm to integrate the spare part attribute information and test response data of a weather radar spare part into the same evaluation detection learning task, so as to carry out the evaluation of the performance of the weather radar spare part. Compared with weather radar spare part performance evaluation by adopting single weather radar spare part data, the weather radar spare part performance evaluation method based on the weather radar spare part data from multiple angles can obtain a performance evaluation result with higher confidence. In addition, compared with a mode that weather radar spare part performance detection and evaluation are performed on different types of weather radar spare part data through different algorithms and then results are combined, the method can prevent disordered extension of algorithm loss, so that the precision of weather radar spare part performance detection and evaluation can be improved.

Description

Technical Field

[0001] The present application relates to the field of data processing, and specifically, to an AI-assisted fully automatic weather radar spare parts performance detection and evaluation method and device. Background Art

[0002] With the development of weather radar technology and its increasing application in weather forecasting, disaster warning, etc., the performance evaluation of weather radar spare parts has become particularly important. Traditional weather radar spare parts performance evaluation methods usually rely on a single data source, such as spare parts attribute information or test response data. Although these methods can provide certain performance evaluation results, they have limitations in accuracy and reliability. In addition, existing evaluation methods often use different algorithms to process different types of data separately, and then merge the results. This method may lead to disordered extension of algorithm loss, thereby affecting the final evaluation accuracy. In recent years, with the advancement of artificial intelligence technology, especially the widespread application of machine learning algorithms in various fields, researchers have begun to explore how to use these advanced technologies to improve the accuracy and reliability of weather radar spare parts performance evaluation. However, how to effectively integrate data from different perspectives and evaluate them within a unified framework remains a challenge. Summary of the invention

[0003] The purpose of the present invention is to provide an AI-assisted fully automatic weather radar spare parts performance detection and evaluation method and device. This application is implemented as follows:

[0004] In a first aspect, the present application provides an AI-assisted fully automatic weather radar spare parts performance detection and evaluation method, the method comprising: obtaining spare parts attribute information of the weather radar spare parts to be evaluated, test response data of the weather radar spare parts to be evaluated, and a performance priori mark set corresponding to multiple radar spare parts performance types; performing implicit representation mining on the spare parts attribute information according to a first machine learning algorithm to obtain an implicit representation of the spare parts attribute information; performing implicit representation mining on the test response data according to a second machine learning algorithm to obtain an implicit representation of the test response; performing implicit representation mining on the performance prior mark set according to a third machine learning algorithm to obtain an implicit representation of the performance prior mark, the first machine learning algorithm, the second machine learning algorithm and the third machine learning algorithm are debugged together according to the same batch of debugging example templates; determining the tendency support coefficient corresponding to the weather radar spare parts to be evaluated and each radar spare parts performance type according to the implicit representation of the spare parts attribute information, the implicit representation of the test response and the implicit representation of the performance prior mark; determining the spare parts performance evaluation result of the weather radar spare parts to be evaluated according to the tendency support coefficient.

[0005] In a second aspect, the present application provides an AI-assisted fully automatic weather radar spare parts performance detection and evaluation device, the device comprising: a data acquisition module, used to obtain spare parts attribute information of the weather radar spare parts to be evaluated, test response data of the weather radar spare parts to be evaluated, and a performance priori label set corresponding to multiple radar spare parts performance types; a first implicit representation module, used to perform implicit representation mining on the spare parts attribute information according to a first machine learning algorithm to obtain an implicit representation of the spare parts attribute information; a second implicit representation module, used to perform implicit representation mining on the test response data according to a second machine learning algorithm to obtain an implicit representation of the test response; a third implicit representation module A representation module is used to perform implicit representation mining on a set of performance prior markers according to a third machine learning algorithm to obtain an implicit representation of the performance prior markers, wherein the first machine learning algorithm, the second machine learning algorithm and the third machine learning algorithm are obtained by debugging together according to the same batch of debugging example templates; a support coefficient determination module is used to determine the tendency support coefficient corresponding to the weather radar spare parts to be evaluated and each radar spare parts performance type according to the implicit representation of spare parts attribute information, the implicit representation of test response and the implicit representation of performance prior markers; a spare parts performance evaluation module is used to determine the spare parts performance evaluation results of the weather radar spare parts to be evaluated according to the tendency support coefficient.

[0006] Beneficial effects of the present application: The embodiment of the present application uses an integrated machine learning algorithm to integrate the spare parts attribute information and test response data of weather radar spare parts into the same evaluation detection learning task to evaluate the performance of weather radar spare parts. Compared with the use of a single type of weather radar spare parts data to evaluate the performance of weather radar spare parts, the present application performs weather radar spare parts performance evaluation based on weather radar spare parts data from multiple angles, and can obtain performance evaluation results with higher confidence. In addition, by comparing the method of performing weather radar spare parts performance detection and evaluation on weather radar spare parts data of different categories through different algorithms and then merging the results, the present application can prevent the disordered extension of algorithm losses, thereby increasing the accuracy of weather radar spare parts performance detection and evaluation. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in describing the embodiments of the present application are briefly introduced below.

[0008] Figure 1 This is a flow chart of an AI-assisted fully automatic weather radar spare parts performance detection and evaluation method provided in an embodiment of the present application.

[0009] Figure 2 It is a schematic diagram of the functional module architecture of the AI-assisted fully automatic weather radar spare parts performance detection and evaluation device provided in an embodiment of the present application.

[0010] Figure 3It is a schematic diagram of the composition of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0011] The following describes the embodiments of the present application in conjunction with the drawings in the embodiments of the present application. The terms used in the implementation method part of the embodiments of the present application are only used to explain the specific embodiments of the present application, and are not intended to limit the present application.

[0012] The execution subject of the AI-assisted fully automatic weather radar spare parts performance detection and evaluation method in the embodiment of the present application is a computer device, including but not limited to a server, a personal computer, a laptop computer, a tablet computer, etc. Figure 1 As shown, the method includes:

[0013] Step S100: obtaining spare part attribute information of a weather radar spare part to be evaluated, test response data of the weather radar spare part to be evaluated, and a performance priori mark set corresponding to multiple radar spare part performance types.

[0014] In step S100, the computer system first obtains the spare parts attribute information, test response data and a set of performance priori marks corresponding to the performance types of multiple radar spare parts of the weather radar spare parts to be evaluated. At this stage, the primary task of the system is to retrieve all attribute information related to the weather radar spare parts to be evaluated from the database or external interface. These attribute information usually include but are not limited to the model, manufacturer, production date, service life expectation, etc. of the spare parts. For example, for a radar transmitter module to be evaluated, its attribute information may include detailed information such as the module model is RTM-1234, the manufacturer is XYZ company, the production date is January 2022, and the design life is 10 years. At the same time, the computer system also needs to collect the test response data of the spare parts. Test response data refers to a series of measurements or feedback information recorded when the spare parts are running in the actual operating environment or under simulated conditions. These data cover the performance of spare parts under different working conditions, such as key indicators such as operating temperature range, signal strength change, and frequency stability. Continuing with the above radar transmitter module as an example, the test response data may include power output stability at different ambient temperatures (-20°C to +50°C), and signal transmission efficiency under various humidity conditions.

[0015] In addition, the system also needs to obtain a set of performance priori markers corresponding to multiple radar spare parts performance types. This set contains historical standards and empirical data for performance evaluations of similar spare parts, which form the basic reference for evaluating the performance of new spare parts. For example, the performance priori marker set may include indicators such as the power output level, signal distortion rate, and noise level that radar spare parts should have under normal working conditions. Each performance type has a corresponding priori marker. For example, the priori marker of the power output level may be set to within the range of -20°C to +50°C, and the output power fluctuation shall not exceed ±5%.

[0016] In the process of obtaining spare parts attribute information, the computer system converts each attribute into a structured form for subsequent processing. For example, the model RTM-1234 is converted into a model number, represented by an integer or string; the production date January 2022 is converted into a timestamp or date format; the design life of 10 years may be represented as 3650 days or the corresponding number of seconds. This information will be organized into a feature vector or array as input for subsequent machine learning models. An example of a feature vector may be [1234,1640995200,3650], where 1234 represents the model number, 1640995200 is the Unix timestamp of January 1, 2022, and 3650 is the number of days of design life. The test response data is also structured in numerical form. For example, the operating temperature range of -20℃ to +50℃ may be converted into the interval [-20,50]. The performance of power output stability at different temperatures may constitute a multidimensional array, such as [[20,-3.5],[25,-4.0],[30,-4.5]], where the first element of each row represents the ambient temperature and the second element represents the percentage change in power output at that temperature.

[0017] The performance priori marker set is a set of benchmark values ​​composed of historical experience and standards for comparison and evaluation. For example, the priori marker of the power output level may be a threshold range, such as [-5,5], which means that the difference between the actual power output and the nominal value should be controlled within this range.

[0018] Through the above steps, the computer system prepares a comprehensive and structured data set for subsequent performance evaluation, ensuring the accuracy and effectiveness of the evaluation process. This stage of work provides a solid foundation for subsequent analysis, enabling subsequent machine learning models to learn and predict based on sufficient information.

[0019] Step S200: performing implicit representation mining on spare part attribute information according to a first machine learning algorithm to obtain implicit representation of spare part attribute information.

[0020] In step S200, the computer system mines the implicit representation of the spare parts attribute information according to the first machine learning algorithm to obtain the implicit representation of the spare parts attribute information. In this process, the system first extracts data from the spare parts attribute information according to the preset data capacity to obtain the target spare parts attribute data. If the data capacity of the spare parts attribute information is less than the preset data capacity, the system will fill the spare parts attribute information with data items according to the preset data capacity to obtain the target spare parts attribute data; if the data capacity of the spare parts attribute information is greater than or equal to the preset data capacity, the system will delete the data items of the spare parts attribute information according to the preset data capacity, and also obtain the target spare parts attribute data. Taking the radar transmitter module to be evaluated as an example, it is assumed that its spare parts attribute information includes model, manufacturer, production date, design life, etc. Assuming that the preset data capacity is five attributes, but the actual spare parts attribute information has only four attributes, the system will add a virtual attribute, such as "no special remarks", to the data to meet the requirements of five attributes. On the contrary, if the spare parts attribute information exceeds five attributes, the system will select the five most important attributes to retain, and the other attributes will be ignored.

[0021] Once the target spare part attribute data is obtained, the computer system will use the first machine learning algorithm to perform implicit representation mining on this data. The goal of implicit representation mining is to convert the original spare part attribute information into a set of vectors or arrays that can reflect the intrinsic characteristics of the spare parts. For example, the model RTM-1234 can be represented as an integer or string, the production date January 2022 is converted into a timestamp or date format, and the design life of 10 years may be represented as 3650 days. Assume that the processed spare part attribute information is as follows: [Model: "RTM-1234", Manufacturer: "XYZ", Production Date: "2022-01-01", Design Life: "3650 Days"].

[0022] To further explore the potential associations of this information, the system uses a machine learning algorithm, such as a deep neural network (DNN). In a DNN, the input layer receives spare part attribute information, the hidden layer captures the complex relationship between spare part attributes through nonlinear transformation, and the output layer generates a fixed-length vector, which is the implicit representation of spare part attribute information. For example, the input layer receives the above spare part attribute information, and the hidden layer may generate a vector of length n through a series of matrix multiplications and activation functions (such as ReLU), which represents the implicit representation of spare part attribute information.

[0023] Specifically, assume that the input layer receives a four-dimensional vector representing four attributes: model, manufacturer, production date, and design life. After being processed by multiple hidden layers, the output layer generates a new vector, such as [0.5, -0.3, 0.2, ..., 0.1], where each element is calculated through a series of weight matrices W and bias vectors b. For example, the calculation formula for the first hidden layer can be expressed as: h1 = σ (W1x + b1), where x is the input vector, W1 is the weight matrix of the first hidden layer, b1 is the bias vector, and σ is the activation function.

[0024] In this way, the computer system not only transforms the spare parts attribute information into a compact form, but also captures the potential connections between spare parts attributes, allowing subsequent steps to more effectively use this information for performance evaluation. This method improves the accuracy and reliability of the evaluation and helps to make more scientific and reasonable judgments in the process of fully automatic weather radar spare parts performance detection and evaluation.

[0025] Step S300: performing implicit representation mining on the test response data according to the second machine learning algorithm to obtain the test response implicit representation.

[0026] In step S300, the computer system performs implicit representation mining on the test response data according to the second machine learning algorithm to obtain the test response implicit representation. In this process, the system first standardizes or normalizes the test response data to eliminate the influence of different dimensions and ensure that all features are compared at the same level. Assuming that the test response data contains indicators such as voltage, current, and signal strength of radar spare parts under different working conditions, the system will convert these raw data into a standard normal distribution form with a mean of 0 and a standard deviation of 1, or normalize them to between 0 and 1.

[0027] After completing the data preprocessing, the computer system will apply a second machine learning algorithm to extract implicit representations from the test response data. This algorithm may be a deep learning model, such as a convolutional neural network (CNN) or a recurrent neural network (RNN), which is used to capture complex patterns in the data. Taking CNN as an example, in the scenario of performance detection and evaluation of weather radar spare parts, the system can identify specific patterns from the test response data, such as the signal attenuation or noise level of the spare parts at a specific frequency. Assuming that the test response data is a time series signal composed of multiple sampling points, the system will use a one-dimensional convolution layer to extract time domain features, and then use a fully connected layer to further process these features to generate a fixed-length vector as the implicit representation of the test response. Specifically, if the test response data is a one-dimensional time series signal containing N sampling points, each sampling point records information in three dimensions such as voltage, current, and signal strength, the system will first convert it into a two-dimensional array with a shape of (N,3). Then, a one-dimensional convolution layer with a specific filter size and step size is used to convolve this array to capture local patterns in the time series. Assuming a filter of size 3 is used, the convolution operation can be expressed as: t =f(w1·x t-1 +w2·x t +w3·x t+1 +b), where x t represents the input vector at time t, w1, w2, w3 are filter weights, b is the bias term, and f is the activation function, such as ReLU.

[0028] Next, the system reduces the spatial dimension through the pooling layer, extracts the most important features, and maps the extracted features to a low-dimensional space through the fully connected layer to generate an implicit representation of the test response. For example, after convolution and pooling operations, the system may obtain an array of shape (M, k), where M represents the length of the sequence after processing and k represents the number of features. The fully connected layer compresses these features into a vector of length L as the implicit representation of the test response. This vector can effectively reflect the key characteristics of the test response data, thereby helping the system to perform more accurate performance evaluation in subsequent steps.

[0029] Throughout the process, the computer system used the second machine learning algorithm to effectively mine the implicit representation of the test response data, which provided a solid foundation for subsequent analysis. In this way, the system can better understand the deep meaning behind the test response data, thereby improving the accuracy and reliability of the performance evaluation of fully automatic weather radar spare parts.

[0030] Step S400: Perform implicit representation mining on the performance prior mark set according to the third machine learning algorithm to obtain implicit representation of the performance prior mark, wherein the first machine learning algorithm, the second machine learning algorithm and the third machine learning algorithm are obtained by debugging together according to the debugging example templates of the same batch.

[0031] In step S400, the computer system mines the implicit representation of the performance prior tag set according to the third machine learning algorithm to obtain the implicit representation of the performance prior tag. At this stage, the system uses a machine learning model specially designed to mine the potential features of the tag data. The model is trained and optimized based on the same batch of debugging example templates together with the first and second machine learning algorithms previously used to process the test response data, ensuring the synergy and consistency between the models.

[0032] For the scenario of performance detection and evaluation of weather radar spare parts, the performance prior tag set includes historical maintenance records, fault reports, and various statistical data under normal operating conditions. These tag data may include information such as the service life of spare parts, maintenance frequency, fault type and its probability of occurrence. For ease of processing, the computer system first converts these unstructured text data into structured numerical data. For example, for each type of spare part, the system will build a data set containing features such as average service life, failure interval time, and failure rate based on its historical records.

[0033] The computer system processes the labeled data using a third machine learning algorithm, typically a deep neural network such as an autoencoder or a long short-term memory network (LSTM). In this case, the LSTM model is a suitable choice given the time series characteristics of the performance prior label set. The LSTM model can effectively capture long-term dependencies, which is critical to understanding the performance of spare parts over time. The system first inputs the above feature data set into the LSTM model, which stores and processes sequence information through its internal memory units. The input at each time step is a feature vector. For example, the vector can be expressed as (Xt = [lifespan, repair_interval, failure_rate]), where lifespan represents the service life, repair_interval represents the average time interval between two repairs, and failure_rate represents the probability of failure.

[0034] The LSTM model processes these inputs through its internal state transition equation. Let h t is the hidden state, c t is the cell state, then the state update formula can be expressed as:

[0035] it =σ(W i ·[h t-1 ,X t ]+b i );

[0036] f t =σ(W f ·[h t-1 ,X t ]+b f );

[0037] c t =f t ⊙c t-1 +i t ⊙tanh(W c ·[h t-1 ,X t ]+b c );

[0038] o t =σ(W o ·[ t-1 ,X t ]+b o );

[0039] h t =o t ⊙tanh(c t );

[0040] Where σ is the sigmoid function, W i ,W f ,W c ,W o is the weight matrix, b i ,b f ,b c ,b o is the bias term, and ⊙ represents element-by-element multiplication. In this way, the LSTM model can capture the dynamic change patterns in time series data.

[0041] Finally, the LSTM model compresses the entire time series data into a fixed-length vector, namely the performance prior labeled implicit representation. This vector not only contains the main information of the original labeled data, but also implies the inherent rules in the time series. For example, for a specific model of radar spare parts, the performance prior labeled implicit representation may be a vector V = [0.12, -0.45, 0.89, ...], which condenses the various performance change trends that may occur in this model of spare parts during its life cycle.

[0042] Through the operation of step S400, the computer system transforms the complex performance priori marker set into a concise and informative implicit representation, providing strong support for subsequent performance evaluation. This process demonstrates the powerful ability of AI-assisted technology in processing large-scale, multi-dimensional data sets, and helps to improve the accuracy and efficiency of weather radar spare parts performance detection and evaluation.

[0043] Step S500: determining the tendency support coefficient corresponding to each radar spare part performance type of the weather radar spare part to be evaluated according to the spare part attribute information implicit representation, the test response implicit representation and the performance priori mark implicit representation.

[0044] In step S500, the computer system determines the tendency support coefficient of the weather radar spare part to be evaluated corresponding to each radar spare part performance type according to the implicit representation of spare part attribute information, the implicit representation of test response and the implicit representation of performance prior mark. This process is implemented through implicit interaction and quantitative product calculation to quantify the tendency of spare parts under different performance types.

[0045] First, the computer system implicitly interacts the implicit representation of spare part attribute information with the implicit representation of test response to obtain an interactive implicit representation. This interactive process takes into account the importance parameters of the two representations. For example, assume that the implicit representation of spare part attribute information is a vector of length 10, and the implicit representation of test response is also a vector of length 10. The system performs a dot product operation on these two vectors to obtain a new vector as the interactive implicit representation. Specifically, the calculation of the interactive implicit representation can be expressed as: Where A is the implicit representation of spare part attribute information, T is the implicit representation of test response, Represents bitwise multiplication (Hadamard product).

[0046] Next, the system determines the dot product calculation value of the interaction implicit representation and the performance prior label implicit representation. Assuming that the performance prior label implicit representation is also a vector of length 10, the system will calculate the dot product between the interaction implicit representation and the performance prior label implicit representation. The dot product calculation formula is: S = I·P, where I is the interaction implicit representation, P is the performance prior label implicit representation, and · represents the dot product of the vector.

[0047] Finally, the system determines the trend support coefficient corresponding to the weather radar spare parts to be evaluated and each radar spare parts performance type based on the calculated value of the quantity product. Assume that there are three types of radar spare parts performance: high stability, high sensitivity and long life. For each performance type, the system will repeat the above process and calculate the corresponding trend support coefficient. For example, the trend support coefficient calculation for high stability can be expressed as: C stable =(I stable ·P stable ), where I stableFor interactive implicit representation with high stability, P stable A priori labeling of implicit representations for high stability performance.

[0048] In actual operation, the system first obtains the importance parameters corresponding to the implicit representation of the spare part attribute information and the implicit representation of the test response. This process includes evaluating the basic performance conditions of the spare part attribute information, obtaining the evaluation results of the basic performance conditions of the spare part, and determining the importance parameters based on the evaluation results. For example, the system may find that the model of a spare part is RTM-1234, the manufacturer is XYZ Company, the production date is January 2022, and the design life is 10 years. By evaluating these attributes, the system determines that the model and design life are of higher importance, so it assigns them higher weights.

[0049] Then, the system implicitly interacts the implicit representation of spare part attribute information and the implicit representation of test response according to the importance parameter to obtain the interactive implicit representation. Assuming that the implicit representation of spare part attribute information is [0.5, 0.3, 0.2, 0.1] and the implicit representation of test response is [0.4, 0.6, 0.3, 0.7], the system weights these two vectors according to the importance parameter and then performs a dot product operation to obtain the interactive implicit representation. For example, the weighted implicit representation of spare part attribute information is [0.7, 0.4, 0.2, 0.1], the implicit representation of test response is [0.6, 0.7, 0.3, 0.5], and the interactive implicit representation is [0.42, 0.28, 0.06, 0.05].

[0050] Finally, the system calculates the product of the interaction implicit representation and the performance priori mark implicit representation to obtain the tendency support coefficient. Assuming that the performance priori mark implicit representation is [0.3, 0.2, 0.1, 0.4], the product calculation value is (S = 0.42 × 0.3 + 0.28 × 0.2 + 0.06 × 0.1 + 0.05 × 0.4 = 0.216). This value reflects the tendency of spare parts in terms of high stability.

[0051] Step S600: determining the spare parts performance evaluation result of the weather radar spare parts to be evaluated according to the tendency support coefficient.

[0052] In step S600, the computer system determines the spare parts performance evaluation results of the weather radar spare parts to be evaluated according to the tendency support coefficient. In this process, the system comprehensively considers the tendency support coefficients obtained by implicitly representing the spare parts attribute information, implicitly representing the test response, and implicitly representing the performance priori mark to obtain the final performance evaluation conclusion. First, the computer system further analyzes the tendency of each spare parts performance type according to the tendency support coefficient obtained in step S500. Assume that the system has calculated the tendency support coefficients of three radar spare parts performance types, namely high stability, high sensitivity, and long life. For example, for a radar spare part to be evaluated, the system may obtain the following tendency support coefficients: high stability is 0.8, high sensitivity is 0.6, and long life is 0.7. Next, the system uses a certain strategy or algorithm to determine the performance evaluation results of the spare parts based on these tendency support coefficients. This strategy may be based on a simple threshold judgment, or it may use a more complex machine learning model for prediction. For example, the system can set a threshold, and when the tendency support coefficient of a certain performance type exceeds the threshold, it is considered that the spare part performs well in this performance type. Assuming that the threshold is set to 0.5, in the above example, the system can determine that the spare part performs well in terms of high stability, high sensitivity and long life.

[0053] To further refine the evaluation results, the system can use classification models, such as logistic regression, support vector machine (SVM), or decision tree, to predict the specific performance level of the spare part. For example, the system may use a logistic regression model to predict whether a spare part meets the high stability standard. The output of the logistic regression model is a probability value between 0 and 1, indicating the probability that the spare part meets the standard. Assuming that the output of the logistic regression model is 0.85, the system can judge that the spare part meets the standard in terms of high stability based on this probability value.

[0054] In the specific implementation process, the system passes the propensity support coefficient as an input feature to the classification model. For example, the system can build a logistic regression model whose input is the propensity support coefficient vector C = C stable ,C sensitive ,C longevity ], and the output is the probability of whether the spare part meets the high stability standard. The calculation formula of the logistic regression model can be expressed as:

[0055]

[0056] Among them, w is the weight vector, b is the bias term, and w T C represents the dot product of the weight vector and the tendency support coefficient vector.

[0057] Based on the above calculation results, the system determines the evaluation results of the spare parts in each performance type. For example, assuming that the weight vector of the logistic regression model is w = [0.5, 0.3, 0.2] and the bias term is b = -0.1, then for the spare parts in the above example, the calculation process is:

[0058]

[0059] Since 0.68 is greater than the set threshold of 0.5, the system can determine that the spare part performs well in terms of high stability.

[0060] The same process applies to other performance types, and the system will calculate the performance evaluation results of the spare parts in terms of high sensitivity and long life respectively. Finally, the system will combine the evaluation results of all performance types to draw an overall performance evaluation conclusion for the weather radar spare parts to be evaluated.

[0061] For example, suppose the system evaluates the above process as follows: high stability is good, high sensitivity is fair, and long life is excellent. Based on this, the system can generate a detailed evaluation report, indicating where the spare part performs well and where it needs improvement, and provide specific improvement suggestions.

[0062] Through the operation of step S600, the computer system determines the spare parts performance evaluation results of the weather radar spare parts to be evaluated according to the tendency support coefficient. This process demonstrates the powerful function of AI technology in integrating multi-source information and improving decision-making accuracy and reliability, which helps to improve the accuracy and efficiency of weather radar spare parts performance detection and evaluation.

[0063] As an implementation method, the second machine learning algorithm includes a first branch algorithm and a second branch algorithm. Then, step S300, performing implicit representation mining on the test response data according to the second machine learning algorithm to obtain the test response implicit representation, may specifically include:

[0064] Step S310: performing a feature expansion operation on the implicit representation to be executed corresponding to the test response data according to the first branch algorithm to obtain an expanded implicit representation;

[0065] Step S320: perform implicit representation mapping on the extended implicit representation according to the second branch algorithm to obtain a test response implicit representation.

[0066] Steps S310 and S320 involve the process of the computer system using the first branch algorithm and the second branch algorithm to mine the implicit representation of the test response data. In this process, the computer system first performs feature expansion operations on the implicit representation to be executed according to the first branch algorithm, and then obtains the extended implicit representation. Subsequently, the system further processes the extended implicit representation according to the second branch algorithm, and finally obtains the test response implicit representation.

[0067] Specifically, in step S310, the computer system uses the first branch algorithm to process the original test response data. The task of this stage is to extract and expand the features of the original data, with the aim of enhancing the representativeness of the data so that the subsequent algorithm can more effectively identify the pattern. For example, if the test response data contains the working status records of radar spare parts under different environmental conditions, the first branch algorithm can be a convolutional neural network (CNN) for capturing spatial structural information, or a long short-term memory network (LSTM) for discovering dependencies in time series. Taking CNN as an example, the system uses a series of convolution kernels to slide the input data to extract local features. Assuming that the input data is a time series of a series of radar signal intensities, the system can set multiple 1D convolution kernels of different sizes, corresponding to signal segments of different lengths. For example, using convolution kernels with widths of 3, 5, and 7, feature maps in different time windows can be obtained. After the convolution layer, the system may obtain a new set of feature vectors. For example, for a specific radar spare part, a feature vector F containing 100 elements may be obtained. expanded =[f1,f2,...,f 100 ], each element represents an eigenvalue extracted from the original signal. This eigenvector is the extended implicit representation.

[0068] In step S320, the computer system continues to process the extended implicit representation using the second branch algorithm. This step is intended to map the extended implicit representation to a higher level of abstract space, thereby obtaining the test response implicit representation. The second branch algorithm may be another form of neural network or dimensionality reduction algorithm, such as an autoencoder. Taking the autoencoder as an example, the system first converts the extended implicit representation F into the test response implicit representation through the encoder part. expanded Compressed into a lower dimensional vector H = [h1,h2,...,h k ], where k<100. This process can be expressed as:

[0069] H=Encoder(F expanded );

[0070] Next, the system uses the decoder part to try to reconstruct the original feature vector. This reconstruction process can be achieved by minimizing the reconstruction error, for example using the mean square error (MSE) as the loss function.

[0071] Through the operations of steps S310 and S320, the computer system performs implicit representation mining on the test response data according to the second machine learning algorithm, thereby obtaining the test response implicit representation. This process not only improves the availability of data, but also provides a more accurate basis for subsequent performance evaluation. For example, the system can determine the performance evaluation result of the spare part based on the obtained test response implicit representation, further combining the implicit representation of spare part attribute information and the implicit representation of performance priori mark.

[0072] As an implementation manner, step S310, performing a feature extension operation on the implicit representation to be executed corresponding to the test response data according to the first branch algorithm to obtain an extended implicit representation, may specifically include:

[0073] Step S311: obtaining continuous data items in the test response data, and performing feature expansion operations on the continuous data items according to the first linear weighted array and the first deviation array to obtain a continuous array;

[0074] Step S312: Obtain discrete data items in the test response data, and perform feature expansion operations on implicit representations to be executed corresponding to the discrete data items according to the second linear weighted array and the second deviation array to obtain a discrete array;

[0075] Step S313: concatenate the continuous array and the discrete array to obtain an extended implicit representation.

[0076] In step S310, the computer system performs a feature expansion operation on the implicit representation to be executed corresponding to the test response data according to the first branch algorithm to obtain an extended implicit representation. This process is divided into three specific steps: obtaining continuous data items and performing feature expansion operations, obtaining discrete data items and performing feature expansion operations, and finally splicing the two together to form an extended implicit representation.

[0077] In step S311, the computer system obtains continuous data items in the test response data, and performs feature expansion operations on these data items according to the first linear weighted array and the first deviation array to obtain a continuous array. For example, assuming that the test response data contains continuous data such as power output and signal strength of radar spare parts at different temperatures, these data can be represented as an array x cont =[x1,x2,...,x m ], where m is the number of continuous data items. The system uses a linear weighted array w cont =[w1,w2,...,w m ] and the deviation array b cont =[b1,b2,...,b m ] to expand these data items. The feature expansion operation can be expressed as:

[0078] z cont =w cont ·x cont +b cont ;

[0079] where z cont It is a continuous array after expansion.

[0080] In step S312, the computer system obtains discrete data items in the test response data, and performs feature expansion operations on these data items according to the second linear weighted array and the second deviation array to obtain a discrete array. Assuming that the test response data contains discrete data such as fault codes, these data can be represented as an array x disc =x1,x2,...,x n ], where n is the number of discrete data items. The system first performs one-hot encoding on the discrete data items, converting each discrete data item into a vector with a length equal to the number of categories, where only one position is 1 and the rest are 0. For example, suppose x disc =[1,2,3] represents three different fault codes, and the number of categories is 3, so the result after one-hot encoding is x' disc =[[1,0,0],[0,1,0],[0,0,1]]. Then, the system uses the linear weighted array w disc =[w1,w2,...,w n ] and the deviation array b disc =[b1,b2,...,b n ] to expand these one-hot encoded data items. The feature expansion operation can be expressed as:

[0081] z disc =w disc ·x' disc +b disc ;

[0082] In step S313, the computer system performs array concatenation on the continuous array and the discrete array to obtain an extended implicit representation.

[0083] Through the operations of the above steps S311, S312 and S313, the computer system performs feature expansion operations on the test response data according to the first branch algorithm to obtain an extended implicit representation. This process not only enhances the characterization capability of the original data, but also provides richer information for subsequent implicit representation mapping. For example, the system can determine the performance evaluation results of the spare parts based on the obtained extended implicit representation, further combining the implicit representation of spare parts attribute information and the implicit representation of performance prior marking. This process demonstrates the powerful ability of AI technology in processing complex multidimensional data, and helps to improve the accuracy and efficiency of weather radar spare parts performance detection and evaluation.

[0084] As an implementation mode, step S312, obtaining discrete data items in the test response data, performing feature expansion operations on the implicit representations to be executed corresponding to the discrete data items according to the second linear weighted array and the second deviation array, to obtain a discrete array, may specifically include:

[0085] Step S3121: Obtain discrete data items in the test response data, and perform one-hot encoding on the discrete data items to obtain encoded continuous data items;

[0086] Step S3122: Perform a feature expansion operation on the encoded continuous data items according to the second linear weighted array and the second deviation array to obtain a discrete array.

[0087] In step S312, the computer system obtains discrete data items in the test response data, and performs feature expansion operations on the discrete data items according to the second linear weighted array and the second deviation array to obtain a discrete array. This process is divided into two specific steps: first, the discrete data items are uniquely encoded to obtain encoded continuous data items; then, the encoded continuous data items are feature expanded according to the linear weighted array and the deviation array to finally obtain a discrete array.

[0088] In step S3121, the computer system obtains discrete data items in the test response data and performs one-hot encoding on these discrete data items to obtain encoded continuous data items. Discrete data items usually include some classified or label data, such as fault codes and working status of radar spare parts. Assume that the discrete data items in the test response data are an array x disc =[1,2,3,1,2], indicating different fault codes or working states. To facilitate subsequent feature expansion operations, the system first performs one-hot encoding on these discrete data items. One-hot encoding is a method of converting discrete data items into continuous data items by converting each discrete data item into a vector with a length equal to the number of categories, where only one position is 1 and the rest are 0. Assuming that there are 3 different categories of discrete data items, namely fault codes 1, 2, and 3, the result after one-hot encoding is:

[0089]

[0090] Each row vector represents a one-hot encoding of a discrete data item. For example, the first row [1,0,0] represents fault code 1, the second row [0,1,0] represents fault code 2, and so on.

[0091] In step S3122, the computer system performs feature expansion operation on the encoded continuous data items according to the second linear weighted array and the second bias array to obtain a discrete array. This process further enriches the feature representation through linear combination and bias terms. Assume that the second linear weighted array is w disc =[0.3,0.4,0.3], the second deviation array is b disc =1,2,3], the feature expansion operation can be expressed as:

[0092] z disc =w disc ·x' disc +b disc ;

[0093] Specifically, for each one-hot encoded row vector, the feature expansion operation can be expressed as:

[0094] z disc,i =w disc ·x' disc,i +b disc ;

[0095] where x' disc,i represents the i-th one-hot encoded row vector, z disc,i Represents the i-th expanded discrete array. For example, for the first one-hot encoded row vector [1,0,0], the expanded result is [1.3,2,3].

[0096] Similarly, for the second one-hot encoded row vector [0,1,0], the expanded result is [1,2.4,3].

[0097] Finally, all expanded discrete arrays z disc for:

[0098]

[0099] Through the operations of steps S3121 and S3122, the computer system performs feature expansion operations on discrete data items according to the second linear weighted array and the second deviation array to obtain a discrete array. This process not only enhances the characterization capability of the original data, but also provides richer information for subsequent implicit representation mapping. For example, the system can further combine the continuous array and other implicit representations based on the obtained discrete array to determine the performance evaluation results of the spare parts.

[0100] As an implementation manner, step S500, determining the tendency support coefficient corresponding to the weather radar spare part to be evaluated and each radar spare part performance type according to the implicit representation of spare part attribute information, the implicit representation of test response and the implicit representation of performance priori mark, may specifically include:

[0101] Step S510: implicitly interacting the implicit representation of spare part attribute information with the implicit representation of test response to obtain an interactive implicit representation;

[0102] Step S520: determining a calculated value of a scalar product of an implicit representation of the interaction and an implicit representation of a priori performance mark;

[0103] Step S530: Determine the tendency support coefficient corresponding to the weather radar spare part to be evaluated and each radar spare part performance type according to the quantity product calculation value.

[0104] In step S500, the computer system determines the tendency support coefficients corresponding to the weather radar spare parts to be evaluated and each radar spare parts performance type according to the implicit representation of spare parts attribute information, the implicit representation of test response and the implicit representation of performance prior mark. This process is divided into three specific steps: implicitly interacting the implicit representation of spare parts attribute information and the implicit representation of test response to obtain the interactive implicit representation; determining the quantitative product calculation value of the interactive implicit representation and the implicit representation of performance prior mark; and determining the tendency support coefficients corresponding to the weather radar spare parts to be evaluated and each radar spare parts performance type according to the quantitative product calculation value.

[0105] In step S510, the computer system implicitly interacts the implicit representation of spare part attribute information with the implicit representation of test response to obtain an interactive implicit representation. This process enhances the feature representation by calculating the interaction between the implicit representation of spare part attribute information and the implicit representation of test response. For example, the implicit representation of the assumed equipment component attribute information is A = [0.5, 0.3, 0.2, 0.1], and the implicit representation of the test response is T = [0.4, 0.6, 0.3, 0.7]. The system first obtains the importance parameters corresponding to the two vectors, and then implicitly interacts based on these parameters. The importance parameter of the implicit representation of the assumed equipment component attribute information is α = [0.7, 0.6, 0.5, 0.4], and the importance parameter of the implicit representation of the test response is β = [0.6, 0.5, 0.4, 0.3].

[0106] The specific calculation process of implicit interaction can be expressed as:

[0107] I = α°A°β°T;

[0108] Where ° represents bitwise multiplication (Hadamard product).

[0109] In step S520, the computer system determines a scalar product calculation value of the interaction implicit representation and the performance priori mark implicit representation. Assuming that the performance priori mark implicit representation is P = [0.3, 0.2, 0.1, 0.4], the scalar product calculation value can be expressed as: S = I·P.

[0110] In step S530, the computer system determines the tendency support coefficient corresponding to the weather radar spare part to be evaluated and each radar spare part performance type according to the calculated value of the quantity product. Assume that there are three radar spare part performance types: high stability, high sensitivity and long life. The system will calculate the tendency support coefficient for each performance type respectively.

[0111] Specifically, it is assumed that the system uses a simple threshold to determine the tendency support coefficient. For example, if the quantity product calculation value is greater than 0.05, the tendency support coefficient is considered to be high. In this case, the system can determine that the tendency support coefficient of the weather radar spare part to be evaluated is high in terms of high stability. Similarly, for the high sensitivity and long life performance types, the system will calculate the corresponding tendency support coefficients respectively.

[0112] Through the operations of steps S510, S520 and S530, the computer system determines the tendency support coefficient corresponding to the weather radar spare parts to be evaluated and each radar spare parts performance type according to the implicit representation of spare parts attribute information, the implicit representation of test response and the implicit representation of performance prior mark. This process demonstrates the powerful function of AI technology in integrating multi-source information and improving decision-making accuracy and reliability, which helps to improve the accuracy and efficiency of weather radar spare parts performance detection and evaluation.

[0113] As an implementation manner, step S510, implicitly interacting the implicit representation of spare part attribute information with the implicit representation of the test response to obtain an interactive implicit representation, may specifically include:

[0114] Step S511: respectively obtaining importance parameters corresponding to the implicit representation of spare part attribute information and the implicit representation of test response;

[0115] Step S512: implicitly interact the implicit representation of spare part attribute information and the implicit representation of test response according to the importance parameter to obtain an interactive implicit representation.

[0116] In step S510, the computer system implicitly interacts the implicit representation of spare part attribute information with the implicit representation of test response to obtain an interactive implicit representation. This process is divided into two specific steps: obtaining importance parameters corresponding to the implicit representation of spare part attribute information and the implicit representation of test response respectively; implicitly interacting the implicit representation of spare part attribute information with the implicit representation of test response according to these importance parameters, and finally obtaining an interactive implicit representation.

[0117] In step S511, the computer system obtains the importance parameters corresponding to the implicit representation of the spare part attribute information and the implicit representation of the test response. This process aims to determine the importance of each feature in each implicit representation so that key information can be highlighted in subsequent implicit interactions. The importance parameters can be obtained in a variety of ways, for example, based on historical data statistical analysis, expert knowledge input, or through machine learning model training.

[0118] For example, the implicit representation of the device attribute information is A = [0.5, 0.3, 0.2, 0.1], and the implicit representation of the test response is T = [0.4, 0.6, 0.3, 0.7]. The computer system needs to assign importance parameters to these two vectors respectively. The importance parameter of the implicit representation of the device attribute information is α = [0.7, 0.6, 0.5, 0.4]\), and the importance parameter of the implicit representation of the test response is β = [0.6, 0.5, 0.4, 0.3]. These importance parameters can be obtained by analyzing historical data, for example, by determining the importance of each feature through principal component analysis (PCA) or other feature selection methods.

[0119] In step S512, the computer system performs implicit interaction based on the importance parameters of the implicit representation of the spare part attribute information and the implicit representation of the test response to obtain an interactive implicit representation. This process enhances the feature representation by calculating the interaction between the implicit representation of the spare part attribute information and the implicit representation of the test response. The specific calculation process of the implicit interaction can be expressed as: I = α°A°β°T.

[0120] Through the operations of steps S511 and S512, the computer system implicitly interacts with the importance parameters corresponding to the implicit representation of the spare part attribute information and the implicit representation of the test response to obtain an interactive implicit representation. This process enhances the representation capability of the original data and provides richer information for subsequent quantitative product calculations.

[0121] For example, the implicit representation of the attribute information of the hypothetical equipment A represents the basic attributes of the spare part, such as model, manufacturer, production date, etc., while the implicit representation of the test response T represents the test response data of the spare part in actual use, such as operating temperature, voltage range, etc. Through implicit interaction, the system can better capture the intrinsic connection between the attribute information of the spare part and the test response data, thereby providing more accurate support for subsequent performance evaluation.

[0122] As an implementation manner, step S511, respectively obtaining importance parameters corresponding to the implicit representation of the spare part attribute information and the implicit representation of the test response, may specifically include:

[0123] Step S5111: perform a basic performance condition evaluation on the spare part attribute information to obtain a basic performance condition evaluation result of the spare part;

[0124] Step S5112: Determine the importance parameters corresponding to the implicit representation of spare part attribute information and the implicit representation of test response according to the evaluation result of spare part basic performance condition.

[0125] In step S511, the computer system obtains the importance parameters corresponding to the implicit representation of the spare part attribute information and the implicit representation of the test response, respectively. This process is divided into two specific steps: performing a spare part basic performance condition evaluation on the spare part attribute information to obtain a spare part basic performance condition evaluation result; and determining the importance parameters corresponding to the implicit representation of the spare part attribute information and the implicit representation of the test response according to the spare part basic performance condition evaluation result.

[0126] In step S5111, the computer system performs a basic performance condition evaluation on the spare part attribute information to obtain a basic performance condition evaluation result of the spare part. This process is intended to determine the importance of each feature by evaluating the spare part attribute information, thereby providing guidance for subsequent implicit interactions. The basic performance condition evaluation of spare parts can be completed based on historical data, expert knowledge, or through a machine learning model.

[0127] For example, spare parts attribute information may include model, manufacturer, production date, design life, etc. The computer system can make a preliminary assessment of the performance of the spare parts based on these attributes. Assume that the attribute information of the spare parts is implicitly represented as A = [0.5, 0.3, 0.2, 0.1], where each element represents the implicit representation of a different attribute. The system first evaluates these attributes. For example, the model may have a greater impact on the performance of the spare parts, followed by the production date and design life, and the manufacturer is relatively small. The system can use a simple scoring system or a machine learning model to evaluate the importance of these attributes.

[0128] Specifically, assume that the system uses a simple scoring system to evaluate spare part attribute information. The system can assign a score to each attribute based on historical data or expert knowledge, and a higher score indicates that the attribute has a greater impact on spare part performance. For example, the model score is 8, the production date score is 6, the design life score is 5, and the manufacturer score is 4. The system can convert these scores into importance parameters, for example, through normalization.

[0129] In step S5112, the computer system determines the importance parameters corresponding to the implicit representation of the spare part attribute information and the implicit representation of the test response based on the evaluation results of the spare part basic performance conditions. This process determines the importance of each feature through the evaluation results, thereby highlighting key information in implicit interaction. Assuming that the implicit representation of the test response is T = [0.4, 0.6, 0.3, 0.7], the system needs to assign corresponding importance parameters to it.

[0130] Assume that the test response data includes operating temperature, voltage range, signal strength, etc. The system can use a similar method to evaluate these features. For example, the operating temperature score is 7, the voltage range score is 5, the signal strength score is 6, and the other factors score is 3. Through normalization, the importance parameter implicitly represented by the test response is obtained:

[0131]

[0132] β≈[0.3333,0.2381,0.2857,0.1429];

[0133] Through the operations of steps S5111 and S5112, the computer system determines the importance parameters corresponding to the implicit representation of spare part attribute information and the implicit representation of test response according to the evaluation results of spare part basic performance conditions. This process enhances the representation ability of the original data and provides richer information for subsequent implicit interaction.

[0134] For example, the implicit representation of the attribute information of the hypothetical equipment A represents the basic attributes of the spare part, such as model, manufacturer, production date, etc., while the implicit representation of the test response T represents the test response data of the spare part in actual use, such as operating temperature, voltage range, etc. By determining the importance parameters, the system can better capture the intrinsic connection between the attribute information of the spare part and the test response data, thereby providing more accurate support for subsequent performance evaluation.

[0135] As an implementation manner, step S520, determining the calculated value of the product of the interaction implicit representation and the performance priori mark implicit representation, may specifically include:

[0136] Step S521: performing a uniform dimensional transformation on the interactive implicit representation according to the implicit representation dimension of the performance priori labeled implicit representation to obtain a target interactive implicit representation;

[0137] Step S522: Determine the scalar product of the target interaction implicit representation and the performance priori mark implicit representation to obtain a scalar product calculation value.

[0138] In step S520, the computer system determines the scalar product calculation value of the interaction implicit representation and the performance priori label implicit representation. This process is divided into two specific steps: performing a uniform dimension transformation on the interaction implicit representation according to the implicit representation dimension of the performance priori label implicit representation to obtain the target interaction implicit representation; and performing a scalar product determination on the target interaction implicit representation and the performance priori label implicit representation to obtain the scalar product calculation value.

[0139] In step S521, the computer system performs a dimensionality uniform transformation on the interactive implicit representation according to the implicit representation dimension of the performance prior labeling implicit representation to obtain the target interactive implicit representation. This process is intended to ensure that the interactive implicit representation and the performance prior labeling implicit representation have the same dimension for subsequent scalar product calculation. For example, assuming that the dimension of the performance prior labeling implicit representation is 4, that is, P = [0.3, 0.2, 0.1, 0.4], and the dimension of the interactive implicit representation is also 4, that is, I = [0.21, 0.09, 0.04, 0.012]. If the dimensions of the two are consistent, there is no need to perform a dimensionality uniform transformation; if the dimensions are inconsistent, interpolation, zero padding or other methods are required to match the dimensions. For example, assuming that the dimension of the interactive implicit representation is 3, that is, I = [0.21, 0.09, 0.04], and the dimension of the performance prior labeling implicit representation is 4. The system can increase the dimension of the interaction implicit representation to 4 by zero padding: I' = [0.21, 0.09, 0.04, 0]. In this way, the target interaction implicit representation I' has the same dimension as the performance prior label implicit representation P.

[0140] In step S522, the computer system determines the scalar product of the target interaction implicit representation and the performance prior mark implicit representation to obtain a scalar product calculation value. The scalar product calculation value can be used to measure the similarity between the interaction implicit representation and the performance prior mark implicit representation. The scalar product calculation process can be expressed as:

[0141] S = I'·P;

[0142] where · represents the dot product operation of the vectors.

[0143] The specific calculation process is:

[0144] S=[0.21,0.09,0.04,0]·[0.3,0.2,0.1,0.4];

[0145] S=0.21·0.3+0.09·0.2+0.04·0.1+0·0.4;

[0146] S = 0.063 + 0.018 + 0.004 + 0;

[0147] S = 0.085;

[0148] Through the operations of steps S521 and S522, the computer system performs a uniform dimensional transformation on the interactive implicit representation according to the implicit representation dimension of the performance priori mark implicit representation, and determines the scalar product of the target interactive implicit representation and the performance priori mark implicit representation, and obtains the scalar product calculation value. This process provides a basis for the subsequent determination of the tendency support coefficient corresponding to the weather radar spare parts to be evaluated and each radar spare parts performance type.

[0149] For example, suppose the computer system has obtained implicit representation of spare part attribute information, implicit representation of test response and implicit representation of performance prior label. Through implicit interaction and quantile product calculation, the system can quantify the tendency of spare parts under different performance types.

[0150] As an implementation manner, step S200, performing implicit representation mining on spare part attribute information according to a first machine learning algorithm to obtain implicit representation of spare part attribute information may specifically include:

[0151] Step S210: extracting the spare part attribute information according to the preset data capacity to obtain the target spare part attribute data;

[0152] Step S220: performing implicit representation mining on the target spare part attribute data according to the first machine learning algorithm to obtain implicit representation of spare part attribute information.

[0153] In step S200, the computer system performs implicit representation mining on the spare part attribute information according to the first machine learning algorithm to obtain the implicit representation of the spare part attribute information. This process is divided into two specific steps: extracting data from the spare part attribute information according to the preset data capacity to obtain the target spare part attribute data; and performing implicit representation mining on the target spare part attribute data according to the first machine learning algorithm to obtain the implicit representation of the spare part attribute information.

[0154] In step S210, the computer system extracts data from the spare parts attribute information according to the preset data capacity to obtain the target spare parts attribute data. This process is intended to ensure the integrity and consistency of the spare parts attribute information for subsequent processing. The preset data capacity refers to the size of the data set pre-defined by the system to ensure that all spare parts attribute information conforms to a unified data format and size. For example, assume that the preset data capacity is 5 attributes, and the spare parts attribute information may include model, manufacturer, production date, design life, warranty period, etc. If the actual number of attributes of the spare parts attribute information is less than the preset data capacity, the system needs to fill in the data; if the actual number of attributes is more than the preset data capacity, data deletion is required.

[0155] Assume that the spare part attribute information includes six attributes: model, manufacturer, production date, design life, warranty period, and use environment, and the preset data capacity is 5. The system will delete the last attribute, i.e., use environment, to meet the preset data capacity. At this time, the target spare part attribute data is: A = [model, manufacturer, production date, design life, warranty period]. If the spare part attribute information only contains four attributes, i.e., model, manufacturer, production date, and design life, the system needs to fill in the data. For example, a virtual attribute, such as "no special remarks", can be added at the end to reach the preset data capacity. At this time, the target spare part attribute data is: A = [model, manufacturer, production date, design life, no special remarks].

[0156] In step S220, the computer system performs implicit representation mining on the target spare part attribute data according to the first machine learning algorithm to obtain the implicit representation of the spare part attribute information. This process converts the spare part attribute information into a vector of a fixed length through a machine learning model, so as to better capture the intrinsic relationship between the spare part attribute information. For example, the first machine learning algorithm can be a deep neural network (DNN) for extracting features from the spare part attribute information and performing implicit representation mining.

[0157] Assume that the target spare part attribute data is: A = [model, manufacturer, production date, design life, warranty period]. The system can convert these attributes into numerical form, for example: A = [1234, XYZ, 2022-01-01, 10, 5]. Among them, 1234 represents the model number, XYZ represents the manufacturer name code, 2022-01-01 represents the production date, 10 represents the design life (years), and 5 represents the warranty period (years).

[0158] Next, the system uses DNN to process these values. DNN contains an input layer, several hidden layers, and an output layer. The input layer receives spare parts attribute information, the hidden layer captures the complex relationship between spare parts attributes through nonlinear transformation, and the output layer generates a fixed-length vector, that is, the spare parts attribute information implicit representation. For example, assume that the hidden layer of DNN contains two fully connected layers, and each fully connected layer uses the ReLU activation function. The input layer receives the spare parts attribute information vector A. After processing by the hidden layer, the output layer generates a vector of length 10, that is, the spare parts attribute information implicit representation A'. Specifically, assume that the vector received by the input layer is A=[1234,XYZ,2022-01-01,10,5], the first fully connected layer uses the weight matrix W1 and the bias vector b1, and the second fully connected layer uses the weight matrix W2 and the bias vector b2. The calculation process can be expressed as:

[0159] H1=ReLU(W1·A+b1);

[0160] H2=ReLU(W2·H1+b2);

[0161] Among them, H1 represents the output of the first hidden layer, and H2 represents the output of the second hidden layer. Finally, the output layer generates the implicit representation A' of the spare part attribute information.

[0162] For example, suppose W1 and b1 transform the input vector A into an intermediate vector H1 of length 10, and then W2 and b2 transform H1 into a vector A' of length 10, which implicitly represents the spare part attribute information.

[0163] Through the operations of steps S210 and S220 above, the computer system performs implicit representation mining on the spare parts attribute information according to the first machine learning algorithm, and obtains the implicit representation of the spare parts attribute information. This process provides an important basis for subsequent steps and helps to improve the accuracy and efficiency of weather radar spare parts performance detection and evaluation. Through data extraction and implicit representation mining, the system can better understand and utilize the interaction between spare parts attribute information, thereby playing an important role in fully automatic weather radar spare parts performance detection and evaluation.

[0164] As an implementation manner, step S210, extracting the spare part attribute information according to the preset data capacity to obtain the target spare part attribute data, may specifically include:

[0165] Step S211: if the data capacity of the spare part attribute information is less than the preset data capacity, fill the data items of the spare part attribute information according to the preset data capacity to obtain the target spare part attribute data;

[0166] Step S212: If the data capacity of the spare part attribute information is greater than or equal to the preset data capacity, data items of the spare part attribute information are deleted according to the preset data capacity to obtain target spare part attribute data.

[0167] In step S210, the computer system extracts data from the spare part attribute information according to the preset data capacity to obtain target spare part attribute data. This process is divided into two cases: if the data capacity of the spare part attribute information is less than the preset data capacity, the system fills the spare part attribute information with data items according to the preset data capacity to obtain target spare part attribute data; if the data capacity of the spare part attribute information is greater than or equal to the preset data capacity, the system deletes the spare part attribute information with data items according to the preset data capacity to obtain target spare part attribute data.

[0168] In step S211, when the data capacity of the spare part attribute information is less than the preset data capacity, the computer system fills the spare part attribute information with data items according to the preset data capacity to obtain the target spare part attribute data. For example, assuming that the preset data capacity is five attributes, and the spare part attribute information only includes three attributes: model, manufacturer, and production date. The system will add two virtual attributes on the basis of the original attributes to meet the preset data capacity. These two virtual attributes can be default values ​​or attributes generated based on rules, such as "standard warranty period" and "no special use environment". The purpose is to keep all spare part attribute information in a consistent data format for subsequent processing. For example, the spare part model "RDA-1234" has the original attribute information of model = RDA-1234, manufacturer = XYZ company, production date = 2022-01-01, and the target spare part attribute data after data item filling is model = RDA-1234, manufacturer = XYZ company, production date = 2022-01-01, standard warranty period = 3 years, no special use environment.

[0169] In step S212, when the data capacity of the spare part attribute information is greater than or equal to the preset data capacity, the computer system deletes data items of the spare part attribute information according to the preset data capacity to obtain the target spare part attribute data. For example, if the preset data capacity is still five attributes, and the spare part attribute information includes six attributes: model, manufacturer, production date, design life, warranty period, and use environment. The system will select five attributes from the six attributes to retain in order to meet the preset data capacity. This selection process can be based on priority or importance, and usually retains information that is more critical to the spare part performance detection and evaluation. For example, the system can choose to retain the five attributes of model, manufacturer, production date, design life, and warranty period, and ignore the use environment. Therefore, the original attribute information of the spare part with model "RDA-1234" is model = RDA-1234, manufacturer = XYZ company, production date = 2022-01-01, design life = 10, warranty period = 5 years, use environment = outdoor. After data item deletion, the target spare part attribute data is model = RDA-1234, manufacturer = XYZ company, production date = 2022-01-01, design life = 10 years, warranty period = 5 years.

[0170] When processing spare part attribute information, the computer system can apply some simple machine learning models to assist decision-making, especially when determining which attributes are more important or which attributes should be retained. For example, the system can use feature selection algorithms to evaluate the importance of different attributes for predicting spare part performance. A commonly used feature selection method is based on mutual information scoring, which measures the degree of correlation between spare part attributes and spare part performance. By calculating the mutual information score between each attribute and spare part performance, the system can select the top N attributes with the highest scores as the target spare part attribute data, where N is equal to the preset data capacity. The mutual information calculation formula is as follows:

[0171]

[0172] Where I(X;Y) represents the mutual information between attribute X and spare part performance Y, p(x,y) represents the probability of attribute X and spare part performance Y appearing at the same time, and p(x) and p(y) represent the probability of attribute X and spare part performance Y appearing respectively.

[0173] Through the operations of steps S211 and S212 above, the computer system extracts the spare parts attribute information according to the preset data capacity to obtain the target spare parts attribute data. This process provides an important basis for subsequent steps and helps to improve the accuracy and efficiency of weather radar spare parts performance detection and evaluation. By filling or deleting data items, the system can ensure that all spare parts attribute information has the same data capacity.

[0174] As an implementation manner, in step S100, before obtaining spare part attribute information of the weather radar spare part to be evaluated, test response data of the weather radar spare part to be evaluated, and a performance priori label set corresponding to multiple radar spare part performance types, the training process of each machine learning algorithm may also be performed, specifically including:

[0175] Step S10: Obtain a debugging example template, where the debugging example template includes template spare part attribute information of a template weather radar spare part, template test response data of a template weather radar spare part, a template radar spare part performance type priori mark of a template weather radar spare part, and a performance priori mark set corresponding to multiple radar spare part performance types.

[0176] In step S10, the computer system obtains a debugging example template, which includes template spare parts attribute information of a template weather radar spare part, template test response data, a priori mark of a template radar spare part performance type, and a set of performance priori marks corresponding to multiple radar spare parts performance types. This process is a prerequisite for training the machine learning algorithm, ensuring that the algorithm can learn based on high-quality data, thereby improving the accuracy of subsequent performance evaluation. The computer system first obtains the template spare parts attribute information of the template weather radar spare part. The template spare parts attribute information includes but is not limited to basic information such as model, manufacturer, production date, and design life. For example, the attribute information of a template spare part can be [model = RDA-1234}, manufacturer = XYZ company, production date = 2022-01-01, design life = 10 years]. These attribute information are basic descriptions of spare parts, which are essential for understanding the working characteristics and historical background of spare parts.

[0177] Next, the computer system obtains the template test response data of the template weather radar spare part. The template test response data reflects the performance of the spare part in actual operation or testing, and may include signal strength, operating temperature range, voltage fluctuation, etc. For example, assuming that the test response data of a template spare part is [signal strength = [85, 90, 87, 88, 92], operating temperature = [25, 30, 28, 27, 26], voltage fluctuation = [0.5, 0.6, 0.4, 0.3, 0.7]]. These data provide the performance of the spare part under different conditions and are the key basis for evaluating the performance of the spare part.

[0178] In addition, the computer system also needs to obtain the template radar spare parts performance type prior tags of the template weather radar spare parts. These prior tags are the predictions or expected values ​​of the spare parts performance types, such as "high stability", "high sensitivity" or "long life". The template radar spare parts performance type prior tags can be based on expert experience or historical data analysis. For example, assuming that a template spare part has a prior tag of [high stability, general sensitivity, long life], these tags help the system understand the expected performance of the spare parts in certain performance types.

[0179] Finally, the computer system obtains a priori performance tag set corresponding to multiple radar spare parts performance types. The prior performance tag set contains tag information of different performance types and is used for supervised learning when training the algorithm. For example, the prior performance tag set can be ["high stability": 1,"normal sensitivity": 0,"long life": 1], where 1 indicates good performance and 0 indicates normal performance. These tag sets are used to guide the algorithm to learn how to correctly evaluate the performance of spare parts.

[0180] Through the operation of step S10, the computer system obtains the debugging example template, which provides the necessary data support for the subsequent algorithm training. This process ensures that the algorithm can be trained on a rich data set, thereby improving its accuracy and robustness in the performance detection and evaluation of fully automatic weather radar spare parts. For example, through these template data, the system can train a neural network model to learn the association between spare parts attribute information and test response data, and evaluate the performance type of spare parts accordingly. This process demonstrates the powerful ability of AI technology in processing complex data, which helps to improve the accuracy and efficiency of weather radar spare parts performance detection and evaluation.

[0181] Step S20: performing implicit representation mining on the template spare part attribute information according to the first machine learning algorithm to obtain implicit representation of the template spare part attribute information.

[0182] In step S20, the computer system mines the template spare part attribute information for implicit representation according to the first machine learning algorithm to obtain the template spare part attribute information implicit representation. This process aims to convert the spare part attribute information into a compact and information-rich representation form through the machine learning algorithm, so as to better capture the intrinsic relationship between the spare part attributes.

[0183] The computer system processes the template spare part attribute information using a first machine learning algorithm. The first machine learning algorithm can be a deep neural network (DNN), such as a multi-layer perceptron (MLP), which is used to extract features from the spare part attribute information and perform implicit representation mining. For example, the template spare part attribute information may include information such as model, manufacturer, production date, and design life. In order to perform implicit representation mining, the computer system first converts these text information into numerical form. For example, the model can be converted into a model number 1234, the manufacturer can be converted into a manufacturer code, the production date can be converted into the number of days from a certain reference date, and the design life can be directly expressed in years. Next, the computer system uses the first machine learning algorithm to process these numerical spare part attribute information. Assume that a deep neural network with multiple hidden layers is used, which can capture the complex relationship between spare part attributes and convert it into a fixed-length vector. For example, assume that the first machine learning algorithm is a DNN with two hidden layers, each hidden layer has 10 neurons, and uses a ReLU activation function. The input layer receives the spare part attribute information vector, the hidden layer captures the complex relationship between spare part attributes through nonlinear transformation, and the output layer generates a fixed-length vector, which is the implicit representation of the template spare part attribute information.

[0184] Specifically, assuming that the vector received by the input layer is [1234, 1, 8212, 10], the first hidden layer uses the weight matrix W1 and the bias vector b1, and the second hidden layer uses the weight matrix W2 and the bias vector b2. The calculation process can be expressed as:

[0185] H1=ReLU(W1·A+b1);

[0186] H2=ReLU(W2·H1+b2);

[0187] Among them, A represents the input spare part attribute information vector, H1 represents the output of the first hidden layer, and H2 represents the output of the second hidden layer. Finally, the output layer generates the template spare part attribute information implicit representation A'.

[0188] Through the operation of step S20 above, the computer system performs implicit representation mining on the template spare part attribute information according to the first machine learning algorithm, and obtains the implicit representation of the template spare part attribute information. This process enhances the characterization capability of the original data and provides richer information for subsequent performance evaluation. For example, through implicit representation mining, the system can better understand and utilize the interaction between spare part attribute information, thereby playing an important role in the performance detection and evaluation of fully automatic weather radar spare parts.

[0189] Step S30: Perform implicit representation mining on the template test response data according to the second machine learning algorithm to obtain an implicit representation of the template test response.

[0190] In step S30, the computer system performs implicit representation mining on the template test response data according to the second machine learning algorithm to obtain the implicit representation of the template test response. The core of this step is to apply a specific machine learning model to analyze and transform the data generated by the spare part performance test, thereby revealing deeper information that helps to comprehensively evaluate the performance status of the spare part.

[0191] The second machine learning algorithm used by the computer system at this stage can be a deep convolutional neural network (CNN) that is specifically designed to handle pattern recognition tasks in template test response data. The template test response data may include operating data of radar spare parts under different environmental conditions, such as signal strength, noise level, echo delay, etc. For example, for a specific radar spare part, its test response data may be [signal strength = 80dBm, noise level = 10dB, echo delay = 2ms, temperature = 25°C], etc.

[0192] In order to perform implicit representation mining, the computer system first needs to convert these test response data into a form suitable for processing by the machine learning model. For example, each measurement value can be converted into a standardized value so that all data are on the same scale. Then, these values ​​can be formed into a multidimensional array as the input of the CNN. Assume that after preprocessing, the test response data array is [0.8, 0.1, 0.002, 0.25], where it is assumed that the temperature has been converted into a proportional value relative to a certain reference temperature.

[0193] Next, the computer system processes the test response data array using a second machine learning algorithm (CNN in this case). Assume that the CNN used contains an input layer, multiple convolutional layers, pooling layers, and fully connected layers. The input layer receives the test response data array, the convolutional layer detects local features in the input data through a series of filters, and the pooling layer further reduces the data dimension to reduce computational complexity. Finally, the fully connected layer is responsible for combining these features to form the final implicit representation.

[0194] Specifically, suppose the input layer receives an array of [0.8, 0.1, 0.002, 0.25] and the first convolutional layer uses the weight matrix W conv1 and the bias vector b conv1 , the subsequent pooling layer reduces the data dimension through the maximum pooling operation, and the second convolutional layer continues to use the weight matrix W conv2 and the bias vector b conv2 Further extract features, and finally the fully connected layer uses the weight matrix W fc and the bias vector b fc Generate a fixed-length vector, which is the implicit representation of the template test response. For example, suppose the array received by the input layer is [0.8, 0.1, 0.002, 0.25]. After a series of convolution and pooling operations, the template test response generated by the fully connected layer is implicitly represented as [0.7, -0.2, 0.4, 0.1, -0.3, 0.5, -0.1, 0.6, 0.2, -0.4].

[0195] It is worth noting that during the training phase, the computer system uses a loss function to measure the gap between the implicit representation predicted by the model and the actual target, and adjusts the model parameters through the back-propagation algorithm to minimize the loss. The loss function can be the mean square error (MSE), and the formula is:

[0196]

[0197] where y i represents the true value, Represents the predicted value, and n is the number of samples. By continuously iteratively updating the model parameters, the loss function value gradually decreases until the preset convergence standard is reached or the predetermined training rounds are completed.

[0198] In summary, through the operation of step S30, the computer system mines the template test response data for implicit representation according to the second machine learning algorithm, and obtains the template test response implicit representation. This step improves the depth of understanding of the test response data and provides a more accurate basis for the subsequent automatic weather radar spare parts performance detection and evaluation.

[0199] Step S40: implicitly interact the template spare part attribute information implicit representation with the template test response implicit representation to obtain a template interaction implicit representation.

[0200] In step S40, the computer system implicitly expresses the interaction between the implicit expression of the template spare part attribute information and the implicit expression of the template test response to obtain the template interactive implicit expression. The key to this step is to use advanced machine learning methods to combine the features extracted from the spare part attribute information with the performance of the spare part in the actual test, in order to discover the potential relationship between the two, thereby providing a more comprehensive data basis for the comprehensive evaluation of spare part performance.

[0201] The method adopted by the computer system at this stage may be a multi-layer perceptron (MLP) or a more complex model, such as the Transformer architecture, to capture the interaction between the implicit representations of the two different sources. The implicit representation of the template spare part attribute information may include attribute information about the spare part design specifications, material properties, manufacturing processes, etc., while the implicit representation of the template test response covers the functional performance indicators of the spare part under specific test conditions. For example, the implicit representation of the template spare part attribute information may be a vector composed of the following features: [design specification = 1, material hardness = 0.8, manufacturing accuracy = 0.9, service life = 0.7]; the implicit representation of the template test response is a vector formed based on the actual test data of the spare part: [signal strength = 0.8, noise level = 0.1, echo delay = 0.002, temperature effect = 0.25].

[0202] In order to achieve implicit representation interaction, the computer system first passes the two vectors as input to the interaction model. Taking Transformer as an example, the model is able to capture the dependencies between different features through the self-attention mechanism. After the input layer, the system maps each vector to a high-dimensional space, in which the relationship between different features can be quantified by dot products or other forms of similarity calculations. For example, for the two vectors above, the system can calculate the cosine similarity between them, as follows:

[0203]

[0204] Where a and b are two vectors, · represents the dot product operation, and ∥a∥ and ∥b∥ represent the norms of the vectors respectively.

[0205] At the interaction layer, the system uses the multi-head self-attention mechanism in Transformer to enhance the interaction between different features. For example, for the two features of design specifications and signal strength, the system can calculate the attention weight between them, the formula is:

[0206]

[0207] Where Q represents the query vector, K represents the key vector, V represents the value vector, and dk is the dimension of the key vector. In this way, the system can not only evaluate the importance of a single feature, but also understand the interaction between different features.

[0208] After calculating the interactions between all features, the system integrates these interaction results into a new vector, namely the template interaction implicit representation. For example, suppose that after the interaction layer, the system obtains a new vector [0.65, -0.12, 0.48, 0.19, -0.27, 0.52, -0.11, 0.61, 0.23, -0.42], which contains information about the interaction between spare part attributes and test responses.

[0209] During the training process, in order to optimize the model, the system defines a loss function to measure the difference between the model output and the expected output. Common loss functions include cross-entropy loss, which is suitable for classification tasks. By minimizing the loss function, the system can gradually adjust the model parameters to improve the accuracy and reliability of the model in predicting spare parts performance.

[0210] In summary, through the operation of step S40, the computer system effectively realizes the implicit representation interaction between the implicit representation of template spare part attribute information and the implicit representation of template test response, forms a template interaction implicit representation, and provides more in-depth data support for subsequent spare part performance evaluation.

[0211] Step S50: performing implicit representation mining on the performance priori label set according to the third machine learning algorithm to obtain implicit representation of the template performance priori label.

[0212] In step S50, the computer system mines the performance priori marker set for implicit representation according to the third machine learning algorithm to obtain the template performance priori marker implicit representation. This process aims to convert the performance priori marker set into a compact and informative vector form so as to be effectively combined and analyzed with other implicit representations later.

[0213] The computer system first obtains a performance priori tag set, which contains performance priori tags corresponding to multiple radar spare parts performance types. For example, the performance priori tag set may include labels such as "high stability", "high sensitivity", and "long life". These tags are based on historical data or expert experience and are used to describe the expected performance of spare parts under specific performance types. In order to facilitate processing by the machine learning algorithm, the computer system needs to convert these tags into numerical form. Assume that the performance priori tag set is ["high stability": 1,"normal sensitivity": 0,"long life": 1], where the numbers 1 and 0 represent good and normal performance, respectively.

[0214] Next, the computer system uses a third machine learning algorithm to mine implicit representations of the performance prior tag set. The third machine learning algorithm can be a multi-layer perceptron (MLP) or other types of neural networks, such as recurrent neural networks (RNN) or long short-term memory networks (LSTM), which are particularly good at processing sequence data and classification tasks. In this example, the system may use a simple MLP to convert the tag set into implicit representations.

[0215] Assume that the input layer receives a set of tags [1, 0, 1], where 1 indicates good performance and 0 indicates average performance. The system can pass this vector as input to the MLP. The MLP consists of an input layer, one or more hidden layers, and an output layer. The input layer receives the performance prior tag set vector, the hidden layer captures the complex relationship in the tag set through nonlinear transformation, and the output layer generates a fixed-length vector, which is the implicit representation of the template performance prior tag.

[0216] During the training process, in order to optimize the model, the system defines a loss function to measure the difference between the model output and the expected output. Common loss functions include mean square error (MSE) or cross-entropy loss (Cross-EntropyLoss), which are suitable for regression or classification tasks. By minimizing the loss function, the system can gradually adjust the model parameters to improve the accuracy and reliability of the model in predicting spare parts performance.

[0217] Through the operation of step S50 above, the computer system performs implicit representation mining on the performance prior marker set according to the third machine learning algorithm, and obtains the implicit representation of the template performance prior marker. This process enhances the characterization ability of the original marker data and provides richer information for subsequent performance evaluation. For example, through implicit representation mining, the system can better understand and utilize the information in the performance prior marker set, thereby playing an important role in the performance detection and evaluation of fully automatic weather radar spare parts.

[0218] Step S60: Determine a debugging error value according to the template interaction implicit representation, the template performance prior marking implicit representation and the template radar spare part performance type prior marking, and optimize the algorithm configuration variables of the first machine learning algorithm, the second machine learning algorithm and the third machine learning algorithm according to the debugging error value.

[0219] In step S60, the computer system determines the debugging error value according to the template interaction implicit representation, the template performance prior mark implicit representation and the template radar spare part performance type prior mark, and optimizes the algorithm configuration variables of the first machine learning algorithm, the second machine learning algorithm and the third machine learning algorithm according to the debugging error value. This process aims to improve the algorithm through the feedback mechanism so that it can more accurately meet the needs of weather radar spare part performance detection and evaluation.

[0220] The computer system first obtains the implicit representation of template interaction, the implicit representation of template performance prior labeling, and the prior labeling of template radar spare parts performance type. These implicit representations are generated by the machine learning model in the previous step. For example, the implicit representation of template interaction may be a feature vector extracted from the radar spare parts usage environment data, assuming its form is I = [0.5, 0.2, -0.3, 0.4]; the implicit representation of template performance prior labeling is the information mined from the spare parts performance prior labeling set, in the form of P' = [0.5, -0.2, 0.3, 0.1, 0.4, -0.1, 0.6, -0.3, 0.2, -0.4]; the prior labeling of template radar spare parts performance type represents the expected performance of spare parts under different performance types, in the form of T = [0.7, 0.3, 0.9, 0.2]. Next, the computer system calculates the debugging error value based on these implicit representations. The debugging error value reflects the gap between the algorithm prediction result and the actual performance indicator. The system may use Mean Squared Error (MSE) as the loss function to calculate the debugging error value.

[0221] Once the debugging error value is determined, the computer system begins to optimize the algorithm configuration variables of the first machine learning algorithm, the second machine learning algorithm, and the third machine learning algorithm. The optimization process typically involves gradient descent or other optimization algorithms to minimize the loss function. In each iteration, the system updates the weights and biases of the algorithms to reduce the debugging error value.

[0222] To specifically demonstrate how to optimize algorithm configuration variables, assume that the first machine learning algorithm is a convolutional neural network (CNN) used to identify defect features in radar spare parts images. The second machine learning algorithm is a recurrent neural network (RNN) used to process time series data, such as changes in the operating status of spare parts. The third machine learning algorithm is a multilayer perceptron (MLP) used to mine implicit representations of a set of performance prior labels. During the optimization process, the system adjusts the parameters of these models, such as the convolution kernel size in the CNN, the number of hidden units in the RNN, and the number of layers and nodes per layer in the MLP.

[0223] Through the continuous iterative optimization process, the computer system is able to gradually improve the performance of the three machine learning algorithms, enabling them to more accurately predict the performance of radar spare parts and reduce the debugging error value. This closed-loop feedback optimization mechanism ensures the continuous improvement of the system, and ultimately achieves the efficiency and accuracy of AI-assisted fully automatic weather radar spare parts performance detection and evaluation. For example, after several rounds of optimization, the system can more accurately distinguish the performance differences of different radar spare parts under various environmental conditions, thereby providing more reliable performance evaluation results.

[0224] Step S70: Iterate the steps of obtaining a debugging example template and optimizing the algorithm configuration variables of the first machine learning algorithm, the second machine learning algorithm, and the third machine learning algorithm according to the obtained debugging example template, and stop when the first machine learning algorithm, the second machine learning algorithm, and the third machine learning algorithm reach a preset debugging stop condition.

[0225] In step S70, the computer system iteratively executes the process of obtaining the debugging example template and optimizing the algorithm configuration variables of the first machine learning algorithm, the second machine learning algorithm, and the third machine learning algorithm until these algorithms reach the preset debugging stop condition. This process is designed to improve the performance of the machine learning model in the performance detection and evaluation of weather radar spare parts through continuous learning and adjustment.

[0226] The computer system first obtains debugging example templates from historical data. These templates include but are not limited to performance data of radar spare parts under specific environmental conditions, maintenance records during the use of spare parts, and performance evaluations annotated by experts. Template data may contain multiple types of features, such as the change pattern of radar signal strength, the change curve of spare part temperature over time, etc. For example, a specific debugging example template can be a time series of a set of radar echo data in the form of X = [x1, x2, ..., x t ], where each x i represents the radar echo intensity at time i. In addition, the template may also contain performance labels Y = [y1, y2, ..., y n ], each y i Corresponding to the performance score of the radar spare parts in a certain aspect. After obtaining the debugging example template, the computer system uses these template data to optimize the algorithm configuration variables of the first machine learning algorithm, the second machine learning algorithm and the third machine learning algorithm. The first machine learning algorithm may be a convolutional neural network (CNN) for analyzing the image features of radar spare parts; the second machine learning algorithm may be a long short-term memory network (LSTM) for processing the historical data sequence of radar spare parts; the third machine learning algorithm may be an integrated learning framework, such as a random forest or a gradient boosted tree (GBT), for comprehensively evaluating the overall performance of radar spare parts. The system adjusts the algorithm configuration variables through the back-propagation algorithm, that is, the parameters are updated by calculating the gradient of the loss function relative to each parameter. The loss function may be the mean square error (MSE).

[0227] In each iteration, the computer system uses the newly acquired debugging example template to evaluate the performance of the current model and updates the algorithm configuration variables accordingly. For example, for the CNN model, the system can adjust the size, step size, or activation function of the convolution kernel; for the LSTM model, the number of units or the dropout rate of the hidden layer may be changed; and for the random forest model, the number or maximum depth of the decision trees may be changed. The system decides when to end the iteration process based on the preset debugging stop conditions. These conditions may include, but are not limited to, reaching a predetermined number of iterations, the model performance no longer significantly improving, or the value of the loss function is below a certain threshold.

[0228] The computer system continuously monitors the performance indicators of the first machine learning algorithm, the second machine learning algorithm, and the third machine learning algorithm, such as precision, recall rate, or F1 score, throughout the iteration process, and adjusts the algorithm configuration variables based on these indicators. As the iteration proceeds, the model gradually learns more complex and useful feature representations, improving the accuracy of the performance detection and evaluation of weather radar spare parts. When the preset debugging stop condition is met, the system stops the iteration process and outputs the final optimized algorithm configuration. This process ensures that the machine learning model can perform optimally when processing weather radar spare parts performance detection tasks. For example, the optimized model can more accurately identify potential failure modes of radar spare parts and provide early warnings, thereby improving the reliability and availability of the radar system.

[0229] As an implementation manner, step S60, determining a debugging error value according to the template interaction implicit representation, the template performance priori mark implicit representation, and the template radar spare part performance type priori mark, and optimizing the algorithm configuration variables of the first machine learning algorithm, the second machine learning algorithm, and the third machine learning algorithm according to the debugging error value, may specifically include:

[0230] Step S61: Dimensionally adjust the template interaction implicit representation according to the fourth machine learning algorithm to obtain a target template interaction implicit representation;

[0231] Step S62: determining the quantitative product of the target template interactive implicit representation and the template performance priori marked implicit representation to obtain the template spare part performance evaluation result;

[0232] Step S63: determining a debugging error value according to the template spare part performance evaluation result and the template radar spare part performance type priori mark, and optimizing the algorithm configuration variables of the first machine learning algorithm, the second machine learning algorithm, the third machine learning algorithm, and the fourth machine learning algorithm according to the debugging error value;

[0233] Step S70, iteratively obtaining a debugging example template and optimizing the algorithm configuration variables of the first machine learning algorithm, the second machine learning algorithm, and the third machine learning algorithm according to the obtained debugging example template, and stopping when the first machine learning algorithm, the second machine learning algorithm, and the third machine learning algorithm reach a preset debugging stop condition, may specifically include: iteratively obtaining a debugging example template and optimizing the algorithm configuration variables of the first machine learning algorithm, the second machine learning algorithm, the third machine learning algorithm, and the fourth machine learning algorithm according to the obtained debugging example template, and stopping when the first machine learning algorithm, the second machine learning algorithm, the third machine learning algorithm, and the fourth machine learning algorithm reach a preset debugging stop condition.

[0234] In steps S60 and S70, the computer system determines the debugging error value according to the template interaction implicit representation, the template performance prior mark implicit representation and the template radar spare part performance type prior mark, and optimizes the algorithm configuration variables of the first machine learning algorithm, the second machine learning algorithm, the third machine learning algorithm and the fourth machine learning algorithm according to the debugging error value. This process continues in an iterative manner until all algorithms reach the preset debugging stop condition. This process aims to improve the performance of the machine learning model in the performance detection and evaluation of weather radar spare parts through continuous learning and adjustment.

[0235] In step S61, the computer system adjusts the dimension of the implicit representation of template interaction according to the fourth machine learning algorithm to obtain the implicit representation of target template interaction. This process is to ensure that the implicit representation of template interaction has the same dimension as the implicit representation of template performance prior labeling, so as to facilitate the subsequent scalar product calculation. The fourth machine learning algorithm can be a dimensionality reduction algorithm, such as principal component analysis (PCA) or an autoencoder. For example, assume that the implicit representation of template interaction is a vector I=[0.5, 0.2, -0.3, 0.4, 0.1, 0.3, -0.2, 0.6, 0.2, -0.4] with a length of 10, and the implicit representation of template performance prior labeling is a vector P'=[0.5, -0.2, 0.3, 0.1, 0.4] with a length of 5. In order to perform scalar product calculation, the system needs to adjust the implicit representation of template interaction to the same dimension as the implicit representation of template performance prior labeling. Assuming PCA dimensionality reduction is used, the target template interaction implicitly represented as I' = [0.4, 0.1, -0.2, 0.3, 0.2].

[0236] In step S62, the computer system performs a dot product determination on the target template interactive implicit representation and the template performance priori labeled implicit representation to obtain the template spare part performance evaluation result. Dot product calculation is a method of measuring the similarity between two vectors.

[0237] In step S63, the computer system determines the debugging error value according to the template spare part performance evaluation result and the template radar spare part performance type prior mark, and optimizes the algorithm configuration variables of the first machine learning algorithm, the second machine learning algorithm, the third machine learning algorithm and the fourth machine learning algorithm according to the debugging error value. The debugging error value reflects the gap between the model prediction result and the actual performance index. The system may use the mean square error (MSE) as the loss function to calculate the debugging error value. During the optimization process, the system adjusts the weights and biases of the algorithm to reduce the debugging error value. Through continuous iteration of the optimization process, the computer system can gradually improve the performance of the four machine learning algorithms so that they can more accurately predict the performance of radar spare parts and reduce the debugging error value.

[0238] In step S70, the computer system iteratively executes the process of obtaining debugging example templates and optimizing the algorithm configuration variables of the first machine learning algorithm, the second machine learning algorithm, the third machine learning algorithm, and the fourth machine learning algorithm until the algorithms reach the preset debugging stop condition. The system obtains debugging example templates from historical data, which include but are not limited to performance data of radar spare parts under specific environmental conditions, maintenance records during the use of spare parts, and performance evaluations annotated by experts. The template data may contain various types of features, such as the change pattern of radar signal strength, the change curve of spare part temperature over time, etc. In each iteration, the computer system uses the newly obtained debugging example template to evaluate the performance of the current model and updates the algorithm configuration variables accordingly. For example, for the first machine learning algorithm (such as CNN), the system can adjust the size, step size or activation function of the convolution kernel; for the second machine learning algorithm (such as LSTM), the number of units or the drop rate of the hidden layer may be changed; for the third machine learning algorithm (such as random forest), the number of decision trees or the maximum depth may be changed; for the fourth machine learning algorithm (such as PCA or Autoencoder), the number of dimensions after dimensionality reduction or the structure of the hidden layer may be adjusted. The system decides when to end the iteration process based on preset debugging stop conditions. These conditions may include but are not limited to reaching a predetermined number of iterations, model performance no longer significantly improving, or the value of the loss function falling below a certain threshold.

[0239] The computer system continuously monitors the performance indicators of the first machine learning algorithm, the second machine learning algorithm, the third machine learning algorithm, and the fourth machine learning algorithm, such as precision, recall rate, or F1 score, throughout the iteration process, and adjusts the algorithm configuration variables based on these indicators. As the iteration proceeds, the model gradually learns more complex and useful feature representations, which improves the accuracy of the performance detection and evaluation of weather radar spare parts. When the preset debugging stop condition is met, the system stops the iteration process and outputs the final optimized algorithm configuration. This process ensures that the machine learning model can perform optimally when processing weather radar spare parts performance detection tasks. For example, the optimized model can more accurately identify potential failure modes of radar spare parts and provide early warnings, thereby improving the reliability and availability of the radar system.

[0240] Based on Figure 1 Based on the same principle as the method shown in , the present application also provides an AI-assisted fully automatic weather radar spare parts performance detection and evaluation device 10, such as Figure 2 As shown, the device 10 comprises:

[0241] A data acquisition module 11 is used to acquire spare part attribute information of the weather radar spare part to be evaluated, test response data of the weather radar spare part to be evaluated, and a performance priori mark set corresponding to multiple radar spare part performance types;

[0242] A first implicit representation module 12, configured to perform implicit representation mining on spare part attribute information according to a first machine learning algorithm to obtain implicit representation of spare part attribute information;

[0243] A second implicit representation module 13, configured to perform implicit representation mining on the test response data according to a second machine learning algorithm to obtain an implicit representation of the test response;

[0244] A third implicit representation module 14 is used to perform implicit representation mining on the performance priori mark set according to a third machine learning algorithm to obtain an implicit representation of the performance priori mark, wherein the first machine learning algorithm, the second machine learning algorithm and the third machine learning algorithm are obtained by debugging together according to the same batch of debugging example templates;

[0245] A support coefficient determination module 15 is used to determine the tendency support coefficient corresponding to the weather radar spare part to be evaluated and each radar spare part performance type according to the implicit representation of spare part attribute information, the implicit representation of test response and the implicit representation of performance priori mark;

[0246] The spare parts performance evaluation module 16 is used to determine the spare parts performance evaluation result of the weather radar spare parts to be evaluated according to the tendency support coefficient.

[0247] The above embodiment introduces the AI-assisted fully automatic weather radar spare parts performance detection and evaluation device 10 from the perspective of a virtual module. The following introduces a computer device from the perspective of a physical module, as shown below:

[0248] The present application embodiment provides a computer device, such as Figure 3 As shown, the computer device 100 includes: a processor 101 and a memory 103. The processor 101 and the memory 103 are connected, such as through a bus 102. Optionally, the computer device 100 may also include a transceiver 104. It should be noted that in actual applications, the transceiver 104 is not limited to one, and the structure of the computer device 100 does not constitute a limitation on the embodiments of the present application.

[0249] An embodiment of the present application provides a computer device. The computer device in the embodiment of the present application includes: one or more processors; a memory; one or more computer programs, wherein the one or more computer programs are stored in the memory and configured to be executed by the one or more processors, and when the one or more programs are executed by the processor, the method provided by the present application is implemented.

[0250] An embodiment of the present application provides a computer-readable storage medium, on which a computer program is stored. When the computer program runs on a processor, the processor can execute the corresponding content in the aforementioned method embodiment.

[0251] The above description is only a partial implementation method of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. An AI-assisted fully automatic weather radar spare parts performance detection and evaluation method, characterized in that: The method comprises: Acquire spare part attribute information of a weather radar spare part to be evaluated, test response data of the weather radar spare part to be evaluated, and a performance priori tag set corresponding to multiple radar spare part performance types; Performing implicit representation mining on the spare part attribute information according to a first machine learning algorithm to obtain an implicit representation of the spare part attribute information; Performing implicit representation mining on the test response data according to a second machine learning algorithm to obtain an implicit representation of the test response; Performing implicit representation mining on the performance priori mark set according to a third machine learning algorithm to obtain implicit representations of performance priori marks, wherein the first machine learning algorithm, the second machine learning algorithm, and the third machine learning algorithm are obtained by debugging together according to the same batch of debugging example templates; Determining the tendency support coefficient corresponding to each radar spare part performance type of the weather radar spare part to be evaluated according to the spare part attribute information implicit representation, the test response implicit representation and the performance priori mark implicit representation; The spare parts performance evaluation result of the weather radar spare parts to be evaluated is determined according to the tendency support coefficient.

2. The method according to claim 1, characterized in that The second machine learning algorithm includes a first branch algorithm and a second branch algorithm, and performing implicit representation mining on the test response data according to the second machine learning algorithm to obtain the test response implicit representation includes: Performing a feature expansion operation on the implicit representation to be executed corresponding to the test response data according to the first branch algorithm to obtain an expanded implicit representation; The extended implicit representation is implicitly represented and mapped according to the second branch algorithm to obtain a test response implicit representation.

3. The method according to claim 2, characterized in that The step of performing a feature expansion operation on the implicit representation to be executed corresponding to the test response data according to the first branch algorithm to obtain an expanded implicit representation includes: Acquire continuous data items in the test response data, and perform feature expansion operations on the continuous data items according to the first linear weighted array and the first deviation array to obtain a continuous array; Obtaining discrete data items in the test response data, and performing a feature expansion operation on the implicit representation to be executed corresponding to the discrete data items according to the second linear weighted array and the second deviation array to obtain a discrete array; The continuous array and the discrete array are concatenated to obtain an extended implicit representation.

4. The method according to claim 3, characterized in that The step of obtaining discrete data items in the test response data and performing a feature expansion operation on the implicit representation to be executed corresponding to the discrete data items according to the second linear weighted array and the second deviation array to obtain a discrete array includes: Obtaining discrete data items in the test response data, and performing one-hot encoding on the discrete data items to obtain encoded continuous data items; Performing a feature expansion operation on the encoded continuous data item according to the second linear weighted array and the second deviation array to obtain a discrete array; The determining, according to the implicit representation of the spare part attribute information, the implicit representation of the test response and the implicit representation of the performance priori mark, the tendency support coefficient corresponding to the weather radar spare part to be evaluated and each radar spare part performance type comprises: implicitly interacting the implicit representation of the spare part attribute information with the implicit representation of the test response to obtain an interactive implicit representation; Determining a calculated value of a dot product of the interaction implicit representation and the performance priori mark implicit representation; The tendency support coefficient corresponding to the weather radar spare part to be evaluated and each radar spare part performance type is determined according to the quantity product calculation value.

5. The method according to claim 4, characterized in that The implicitly interacting the implicit representation of the spare part attribute information with the implicit representation of the test response to obtain the interactive implicit representation includes: Respectively obtaining importance parameters corresponding to the implicit representation of the spare part attribute information and the implicit representation of the test response; The implicit representation of the spare part attribute information and the implicit representation of the test response are implicitly interacted according to the importance parameter to obtain an interactive implicit representation.

6. The method according to claim 5, characterized in that The respectively obtaining the importance parameters corresponding to the implicit representation of the spare part attribute information and the implicit representation of the test response includes: Performing a spare part basic performance condition evaluation on the spare part attribute information to obtain a spare part basic performance condition evaluation result; Determining, according to the spare part basic performance condition evaluation result, the importance parameter corresponding to the implicit representation of the spare part attribute information and the implicit representation of the test response; The determining of the calculated value of the scalar product of the implicit representation of the interaction and the implicit representation of the priori mark of the performance includes: Performing a uniform dimensional transformation on the interactive implicit representation according to the implicit representation dimension of the performance priori labeled implicit representation to obtain a target interactive implicit representation; Determine the scalar product of the target interaction implicit representation and the performance priori mark implicit representation to obtain a scalar product calculation value; The step of performing implicit representation mining on the spare part attribute information according to the first machine learning algorithm to obtain implicit representation of the spare part attribute information includes: Extracting the spare part attribute information according to a preset data capacity to obtain target spare part attribute data; The target spare part attribute data is subjected to implicit representation mining according to a first machine learning algorithm to obtain an implicit representation of the spare part attribute information.

7. The method according to claim 6, characterized in that The step of extracting the spare part attribute information according to the preset data capacity to obtain target spare part attribute data includes: If the data capacity of the spare part attribute information is less than the preset data capacity, fill the data items of the spare part attribute information according to the preset data capacity to obtain target spare part attribute data; If the data capacity of the spare part attribute information is greater than or equal to the preset data capacity, data items of the spare part attribute information are deleted according to the preset data capacity to obtain target spare part attribute data.

8. The method according to claim 1, characterized in that Before acquiring spare part attribute information of the weather radar spare part to be evaluated, test response data of the weather radar spare part to be evaluated, and a performance priori mark set corresponding to multiple radar spare part performance types, the method further includes: Obtain a debugging example template, wherein the debugging example template includes template spare part attribute information of a template weather radar spare part, template test response data of the template weather radar spare part, a template radar spare part performance type priori mark of the template weather radar spare part, and a performance priori mark set corresponding to multiple radar spare part performance types; Performing implicit representation mining on the template spare part attribute information according to the first machine learning algorithm to obtain implicit representation of the template spare part attribute information; Performing implicit representation mining on the template test response data according to the second machine learning algorithm to obtain an implicit representation of the template test response; Performing implicit expression interaction on the template spare part attribute information implicit expression and the template test response implicit expression to obtain a template interaction implicit expression; Performing implicit representation mining on the performance priori mark set according to the third machine learning algorithm to obtain an implicit representation of the template performance priori mark; determining a debugging error value according to the template interaction implicit representation, the template performance a priori mark implicit representation, and the template radar spare part performance type a priori mark, and optimizing algorithm configuration variables of the first machine learning algorithm, the second machine learning algorithm, and the third machine learning algorithm according to the debugging error value; Iterate the steps of obtaining a debugging example template and optimizing the algorithm configuration variables of the first machine learning algorithm, the second machine learning algorithm, and the third machine learning algorithm according to the obtained debugging example template, and stop when the first machine learning algorithm, the second machine learning algorithm, and the third machine learning algorithm reach a preset debugging stop condition.

9. The method according to claim 8, characterized in that The step of determining a debugging error value according to the template interaction implicit representation, the template performance a priori mark implicit representation, and the template radar spare part performance type a priori mark, and optimizing algorithm configuration variables of the first machine learning algorithm, the second machine learning algorithm, and the third machine learning algorithm according to the debugging error value includes: Dimensionally adjusting the template interaction implicit representation according to a fourth machine learning algorithm to obtain a target template interaction implicit representation; Determine the quantitative product of the target template interactive implicit representation and the template performance priori marked implicit representation to obtain a template spare part performance evaluation result; determining a debugging error value according to the template spare part performance evaluation result and the template radar spare part performance type priori mark, and optimizing the algorithm configuration variables of the first machine learning algorithm, the second machine learning algorithm, the third machine learning algorithm, and the fourth machine learning algorithm according to the debugging error value; The iterative step of acquiring a debugging example template and optimizing the algorithm configuration variables of the first machine learning algorithm, the second machine learning algorithm, and the third machine learning algorithm according to the acquired debugging example template, and stopping when the first machine learning algorithm, the second machine learning algorithm, and the third machine learning algorithm reach a preset debugging stop condition, includes: Iterate the steps of obtaining a debugging example template and optimizing the algorithm configuration variables of the first machine learning algorithm, the second machine learning algorithm, the third machine learning algorithm, and the fourth machine learning algorithm according to the obtained debugging example template, and stop when the first machine learning algorithm, the second machine learning algorithm, the third machine learning algorithm, and the fourth machine learning algorithm reach a preset debugging stop condition.

10. An AI-assisted fully automatic weather radar spare parts performance detection and evaluation device, characterized in that: The device comprises: A data acquisition module, used to acquire spare part attribute information of a weather radar spare part to be evaluated, test response data of the weather radar spare part to be evaluated, and a performance priori mark set corresponding to multiple radar spare part performance types; A first implicit representation module, used for performing implicit representation mining on the spare part attribute information according to a first machine learning algorithm to obtain an implicit representation of the spare part attribute information; A second implicit representation module, configured to perform implicit representation mining on the test response data according to a second machine learning algorithm to obtain an implicit representation of the test response; a third implicit representation module, configured to perform implicit representation mining on the performance priori mark set according to a third machine learning algorithm to obtain an implicit representation of the performance priori mark, wherein the first machine learning algorithm, the second machine learning algorithm and the third machine learning algorithm are obtained by debugging together according to the same batch of debugging example templates; A support coefficient determination module, used for determining the tendency support coefficient corresponding to the weather radar spare part to be evaluated and each radar spare part performance type according to the implicit representation of the spare part attribute information, the implicit representation of the test response and the implicit representation of the performance priori mark; The spare parts performance evaluation module is used to determine the spare parts performance evaluation result of the weather radar spare parts to be evaluated according to the tendency support coefficient.

Images