Self-adaptive soft measurement method, device, equipment, storage medium and product

Abstract

The invention discloses a self-adaptive soft measurement method and device, equipment, a storage medium and a product, and relates to the technical field of computers, and the self-adaptive soft measurement method comprises the steps that a to-be-measured time sequence is acquired; performing time sequence distribution identification on the to-be-detected time sequence to obtain a time sequence distribution representative sequence; and inputting the time sequence distribution representative sequence into a preset soft measurement model, and performing soft measurement on the time sequence distribution representative sequence based on the soft measurement model to obtain a soft measurement result, the soft measurement model is obtained by performing iterative training on a preset to-be-trained model based on a time sequence distribution representative sequence training sample and a soft measurement result label of the time sequence distribution representative sequence training sample. According to the method, the time sequence distribution representative sequence is obtained by performing time sequence distribution identification on the to-be-measured time sequence, so that the influence of data offset caused by sensor degradation on the measurement precision is reduced, and the precision of a soft measurement technology is remarkably improved.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to an adaptive soft measurement method, device, equipment, storage medium and product. Background Art

[0002] With the rapid development of modern process industries and the continuous expansion of production scale, the importance of monitoring and controlling key parameters in the production process is increasing. Due to technical or budgetary limitations, it is difficult to directly detect variables in modern process industries through online physical sensors. Therefore, soft measurement technology is usually used to collect easy-to-measure process variables and estimate and predict physical and chemical quantities that are difficult to measure directly. The basic idea of ​​soft measurement technology is to select a set of easy-to-measure auxiliary variables that are closely related to the difficult-to-measure target variables as input variables according to a certain optimization criterion, and to use computer software to estimate the target variables by constructing a certain mathematical relationship, showing multiple advantages such as low cost, configuration flexibility, good real-time performance and simple maintenance.

[0003] Traditional soft measurement technology ignores the data offset caused by sensor degradation and assumes that the distribution of data input used for training and actual applications is consistent. It builds a soft measurement prediction model through a large amount of historical data and corresponding regression algorithms for soft measurement. Since the physical sensors that collect time series data will always have certain performance degradation, the statistical distribution of time series data will continue to shift, which will lead to a continuous decrease in the accuracy of soft measurement technology.

[0004] Therefore, how to improve the accuracy of soft measurement technology is a problem that needs to be solved urgently.

[0005] The above contents are only used to assist in understanding the technical solution of the present application and do not constitute an admission that the above contents are prior art. Summary of the invention

[0006] The main purpose of the present application is to provide an adaptive soft measurement method, device, equipment, storage medium and product, aiming to solve the technical problem of improving the accuracy of soft measurement technology.

[0007] To achieve the above object, the present application proposes an adaptive soft measurement method, which comprises:

[0008] Get the time series to be tested;

[0009] Performing time series distribution identification on the time series to be tested to obtain a time series distribution representative sequence;

[0010] The time series distribution representative sequence is input into a preset soft measurement model, and based on the soft measurement model, soft measurement is performed on the time series distribution representative sequence to obtain a soft measurement result, wherein the soft measurement model is obtained by iteratively training a preset model to be trained based on the time series distribution representative sequence training samples and the soft measurement result labels of the time series distribution representative sequence training samples.

[0011] In one embodiment, the step of performing time series distribution identification on the time series to be tested to obtain a time series distribution representative sequence includes:

[0012] According to the preset value, the time series to be measured is evenly segmented to obtain a segmented time series;

[0013] Sampling the time series to be tested to obtain a sample time series;

[0014] Calculate the dissimilarity between each of the segmented time series and each of the sample time series to obtain a dissimilarity set;

[0015] Analyze the dissimilarity set to obtain the time series distribution of the time series to be tested;

[0016] Based on the time series distribution of the time series to be tested, the segmented time series that meet the preset conditions are selected and combined to obtain a time series distribution representative sequence.

[0017] In one embodiment, the soft measurement model includes a time series prediction model, the soft measurement result includes a target prediction result, and the step of inputting the time series distribution representative sequence into a preset soft measurement model, and performing soft measurement on the time series distribution representative sequence based on the soft measurement model to obtain the soft measurement result includes:

[0018] The representative sequence of the time series distribution is input into a preset time series prediction model, and based on the time series prediction model, the representative sequence of the time series distribution is predicted to obtain a target prediction result.

[0019] In one embodiment, before the step of inputting the time series distribution representative sequence into a preset time series prediction model, and predicting the time series distribution representative sequence based on the time series prediction model to obtain a target prediction result, the step further includes:

[0020] Obtaining training samples representing a time series distribution;

[0021] Determine the target label of the time series distribution representative sequence training sample;

[0022] Based on the time series distribution representative sequence training samples and the soft measurement result labels of the time series distribution representative sequence training samples, the preset model to be trained is iteratively trained to obtain a time series prediction model.

[0023] In one embodiment, the step of iteratively training a preset model to be trained based on a time series distribution representative sequence training sample and a soft measurement result label of the time series distribution representative sequence training sample to obtain a time series prediction model includes:

[0024] Inputting the time series distribution representative sequence training samples into a preset model to be trained to obtain sample prediction results;

[0025] Based on the sample prediction result and the target label, a prediction error value is calculated using a loss function;

[0026] Determining whether the predicted error value meets the error standard indicated by a preset threshold range;

[0027] If the error result does not meet the error standard, return to the step of inputting the time series distribution representative sequence training sample into the preset model to be trained to obtain the sample prediction result, and stop training until the prediction error value meets the error standard to obtain the time series prediction model.

[0028] In one embodiment, the step of calculating the prediction error value using a loss function based on the sample prediction result and the target label includes:

[0029] Determine the importance weight vector;

[0030] Based on the sample prediction result and the target label, a prediction error value is calculated using a loss function, wherein the loss function weights all hidden states by the importance weight vector.

[0031] In addition, to achieve the above-mentioned purpose, the present application also proposes an adaptive soft measurement device, the adaptive soft measurement device comprising:

[0032] An acquisition module is used to obtain the time series to be tested;

[0033] An identification module, used for performing time series distribution identification on the time series to be tested to obtain a representative sequence of the time series distribution;

[0034] A measurement module is used to input the time series distribution representative sequence into a preset soft measurement model, and based on the soft measurement model, perform soft measurement on the time series distribution representative sequence to obtain a soft measurement result, wherein the soft measurement model is obtained by iteratively training a preset model to be trained based on the time series distribution representative sequence training samples and the soft measurement result labels of the time series distribution representative sequence training samples.

[0035] In addition, to achieve the above objectives, the present application also proposes an adaptive soft measurement device, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the adaptive soft measurement method described above.

[0036] In addition, to achieve the above purpose, the present application also proposes a storage medium, which is a computer-readable storage medium, and stores a computer program on the storage medium. When the computer program is executed by a processor, the steps of the adaptive soft measurement method described above are implemented.

[0037] In addition, to achieve the above-mentioned purpose, the present application also provides a computer program product, wherein the computer program product includes a computer program, and when the computer program is executed by a processor, the steps of the adaptive soft-sensing method described above are implemented.

[0038] One or more technical solutions proposed in this application have at least the following technical effects:

[0039] Compared with the related art, the traditional soft measurement technology ignores the data offset caused by sensor degradation, assumes that the distribution of data input for training and practical application is consistent, and constructs a soft measurement prediction model for soft measurement through a large amount of historical data and corresponding regression algorithms. In this scheme, the time series to be measured is obtained; the time series distribution of the time series to be measured is identified to obtain a representative sequence of the time series distribution; the representative sequence of the time series distribution is input into a preset soft measurement model, and the representative sequence of the time series distribution is soft-measured based on the soft measurement model to obtain a soft measurement result, wherein the soft measurement model is obtained by iteratively training the preset model to be trained based on the training samples of the representative sequence of the time series distribution and the soft measurement result labels of the training samples of the representative sequence of the time series distribution. In this scheme, the representative sequence of the time series distribution is obtained by iteratively training the preset model to be trained based on the training samples of the representative sequence of the time series distribution and the soft measurement result labels of the training samples of the representative sequence of the time series distribution, and the adaptive measurement is realized when the time series distribution of the training set and the test set are inconsistent, and the influence of the data offset caused by sensor degradation on the measurement accuracy is reduced. Based on the trained soft measurement model, the representative sequence of the time series distribution is soft-measured to obtain the soft measurement result, which significantly improves the accuracy of the soft measurement technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0041] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0042] Figure 1 A schematic diagram of a process flow provided for the first embodiment of the adaptive soft sensing method of the present application;

[0043] Figure 2 A schematic diagram of a flow chart provided for the second embodiment of the adaptive soft-sensing method of the present application;

[0044] Figure 3 A schematic diagram of a training set for the adaptive soft-sensing method provided in Example 2 of the present application;

[0045] Figure 4 This is a schematic diagram of the module structure of the adaptive soft measurement device according to an embodiment of the present application;

[0046] Figure 5 Schematic diagram of the device structure of the hardware operating environment involved in the adaptive soft measurement method in the embodiment of the present application.

[0047] The purpose, features and advantages of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0048] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application and are not used to limit the present application.

[0049] In order to better understand the technical solution of the present application, a detailed description will be given below in conjunction with the accompanying drawings and specific implementation methods.

[0050] The main solution of the embodiment of the present application is: obtaining a time series to be measured;

[0051] Performing time series distribution identification on the time series to be tested to obtain a time series distribution representative sequence;

[0052] The time series distribution representative sequence is input into a preset soft measurement model, and based on the soft measurement model, soft measurement is performed on the time series distribution representative sequence to obtain a soft measurement result, wherein the soft measurement model is obtained by iteratively training a preset model to be trained based on the time series distribution representative sequence training samples and the soft measurement result labels of the time series distribution representative sequence training samples.

[0053] In this embodiment, the present application uses an adaptive soft measurement device as an execution subject. For ease of description, it is specifically described below as “device”.

[0054] Since existing technologies often ignore the fact that sensor degradation can cause data offset and assume that the distribution of data input used for training and actual applications remains consistent, a soft measurement prediction model is constructed through a large amount of historical data and corresponding regression algorithms for soft measurement.

[0055] The present application provides a solution, which obtains a representative sequence of the time series distribution by performing time series distribution identification on the time series to be measured, realizes adaptive measurement when the time series distribution of the training set and the test set are inconsistent, reduces the impact of data offset caused by sensor degradation on measurement accuracy, and performs soft measurement on the representative sequence of the time series distribution based on the trained soft measurement model to obtain soft measurement results, thereby significantly improving the accuracy of the soft measurement technology.

[0056] It should be noted that the execution subject of this embodiment can be a computing service device with data processing, network communication and program running functions, such as a tablet computer, a personal computer, a mobile phone, etc., or an electronic device capable of realizing the above functions, an adaptive soft measurement device, etc. The following takes the adaptive soft measurement device as an example to illustrate this embodiment and the following embodiments.

[0057] Based on this, the embodiment of the present application provides an adaptive soft measurement method, referring to Figure 1 , Figure 1 This is a flow chart of the first embodiment of the adaptive soft-sensing method of the present application.

[0058] In this embodiment, the adaptive soft-sensing method includes steps S10 to S30:

[0059] Step S10, obtaining a time series to be measured;

[0060] It should be noted that the time series to be measured is a data sequence of a group of measurable variables used as input in the soft measurement model.

[0061] It can be understood that the time series to be measured is a sequence of measurable variable data related to the production process, which is arranged in chronological order during the process production, is usually collected by sensors or other measuring equipment, and is used to monitor and optimize the production process.

[0062] In the specific implementation, the massive data in the process production are collected by sensors, and the time series to be measured is obtained through data processing.

[0063] The above is only a feasible implementation of step S10 provided in this embodiment, and this embodiment does not specifically limit the specific implementation of step S10.

[0064] Step S20, performing time series distribution identification on the time series to be tested to obtain a time series distribution representative sequence;

[0065] It should be noted that the time series to be measured is a data sequence of a group of measurable variables used as input in the soft measurement model, the time series distribution identification is to identify a representative subset of the time series, and the time series distribution representative sequence is a sequence representing the distribution of the time series to be measured.

[0066] It is understandable that in some cases, time series data may contain a large number of points, and processing and analyzing this data may be very time-consuming and resource-intensive. To simplify the problem, one approach is to extract a representative subset from the entire time series, which is the representative sequence of the time series distribution. After obtaining the representative sequence of the time series distribution, further analysis, such as modeling, prediction, etc., can be performed on this smaller and more representative data set, thereby improving computational efficiency while ensuring a certain degree of accuracy.

[0067] In a specific implementation, step S20 includes:

[0068] According to the preset value, the time series to be measured is evenly segmented to obtain a segmented time series;

[0069] Sampling the time series to be tested to obtain a sample time series;

[0070] Calculate the dissimilarity between each of the segmented time series and each of the sample time series to obtain a dissimilarity set;

[0071] Analyze the dissimilarity set to obtain the time series distribution of the time series to be tested;

[0072] Based on the time series distribution of the time series to be tested, the segmented time series that meet the preset conditions are selected and combined to obtain a time series distribution representative sequence.

[0073] It should be noted that the time series to be tested is a series of data points arranged in chronological order, usually the values ​​obtained by continuous measurement. The segmented time series is multiple sub-time series obtained after uniform segmentation. Each segmented time series represents a specific time period in the original time series and can be analyzed separately to study the characteristics and trends of the data in that time period.

[0074] In addition, it should be noted that the sample time series is a time series obtained through sampling. The sample time series usually contains fewer data points and is a simplified representation of the original time series for rapid analysis and modeling. The dissimilarity set is a set composed of multiple dissimilarity calculation results, which contains the calculated values ​​of the dissimilarity between all segmented time series and the sample time series.

[0075] In addition, it should be noted that the representative sequence of the time series distribution is the final selected representative time series, which is selected from the segmented time series by analyzing the dissimilarity set and based on certain preset conditions, aiming to reflect the main characteristics and trends of the original time series to the greatest extent.

[0076] It can be understood that the whole process is a conversion process from the original time series to the representative time series, which aims to reduce the complexity of the data through steps such as segmentation, sampling and dissimilarity calculation, while retaining the most critical and representative information for more efficient subsequent analysis.

[0077] Understandably, traditional soft sensing techniques often ignore the problem of sensor degradation and assume that the distribution of data input used for training and actual applications is consistent. Figure 3 ,This scheme adopts time series distribution identification and matching, identifies and matches the ,continuous data statistical distribution, and selects easy to measure ,auxiliary variables that are closely related to the difficult to measure ,target variables as input variables to achieve the estimation of the target ,variable.

[0078] In the specific implementation, in order to divide the time series into K segments with basically stable distribution, this case proposes a novel time series segmentation method, which uses a subset selection method based on dissimilarity to segment the time series by selecting model representatives to describe the time series:

[0079] First, the training set Input variable of length n Evenly divided into m segments, a segmented time series is obtained, each segment is represented by a model M j , where j = 1, .., m. The length of each segment is equal, and m>K must be guaranteed.

[0080] Use the maximum mean difference to measure and Every two samples M in the two sets j With x i The difference between them is expressed as:

[0081] d ji =MMD(M j ,x i )

[0082] Among them, x i is a single sample, i.e., a sample time series, and M j It is a single sample set, that is, a segmented time series. Therefore, the conventional Euclidean distance alone cannot measure the dissimilarity between them, while MMD can well quantify the dissimilarity between samples and sets.

[0083] Then, the relationship between each two samples is expressed in matrix form:

[0084]

[0085] For the matrix D, the goal is to get Find a target subset from the set to represent the entire time series Each model M in the target subset can be used to represent a sub-time series.

[0086] Therefore, based on the difference d ji and the associated assignment variable z ji You can create an optimization program. Assign the variable z ji Also expressed in matrix form:

[0087]

[0088] Among them, z ji Used to represent M j Can it represent x? i If possible, ji That is, it is equal to 1; otherwise, it is equal to 0. Therefore, A subset of The cost function can be expressed by the dissimilarity d ji express:

[0089]

[0090] In the process of subset selection, the ideal model representative M needs to have as small a difference as possible between the samples in a subsequence it represents.

[0091] Based on the above objectives, we can construct an objective function for subset selection:

[0092]

[0093]

[0094]

[0095]

[0096] Limit 1 means assigning a value to variable z ji and b c It can only take the values ​​of 0 or 1; Restriction 2 means that there can only be one model representative for each sample; Restriction 3 is used to limit the total number of selected representatives K.

[0097] Use a common linear programming solver to optimize the above objective function. We can get a Subset of as well as and The mapping relationship of samples in each model M k The subsequence represented can be considered a stable subsequence

[0098] In this embodiment, the time series to be measured is uniformly segmented according to preset values ​​to obtain a segmented time series; the time series to be measured is sampled to obtain a sample time series; the dissimilarity between each segmented time series and each sample time series is calculated to obtain a dissimilarity set; the dissimilarity set is analyzed to obtain the time series distribution of the time series to be measured; based on the time series distribution of the time series to be measured, the segmented time series that meets the preset conditions are selected and combined to obtain a representative sequence of the time series distribution, which solves the problem of being unable to measure when the time series distributions of the training set and the test set are inconsistent, and eliminates the impact of data drift caused by sensor degradation on the predicted input data.

[0099] Step S30, input the time series distribution representative sequence into a preset soft measurement model, and based on the soft measurement model, perform soft measurement on the time series distribution representative sequence to obtain a soft measurement result, wherein the soft measurement model is obtained by iteratively training a preset model to be trained based on the time series distribution representative sequence training samples and the soft measurement result labels of the time series distribution representative sequence training samples.

[0100] It should be noted that the time series distribution representative sequence is a representative time series subset extracted from the original time series data, which is obtained through steps such as segmentation, sampling and dissimilarity calculation, aiming to simplify the original data while retaining its key features and trends.

[0101] In addition, it should be noted that the soft measurement model is a model constructed using mathematical and statistical methods, which is used to estimate variables that are difficult to measure directly or are very costly to measure. It usually relies on easily accessible related auxiliary variables to infer the value of the variable to be measured. The soft measurement result is the result obtained after analyzing the time series distribution representative sequence using the soft measurement model, which is an estimate or prediction of the variable in the original time series. The training sample is a data set used to build and train the soft measurement model, and the soft measurement result label is the actual measurement value or known result corresponding to each time series distribution representative sequence in the training sample, which helps the model learn how to predict the output soft measurement result based on the input time series distribution representative sequence.

[0102] In addition, it should be noted that the preset model to be trained is the initial model architecture selected before starting model training, which contains a series of parameters that need to be adjusted and optimized through the training process to achieve the best prediction effect. The iterative training is the process of model training, which usually involves multiple iterations. In each iteration, the model predicts the training sample based on the current parameters, and then compares the prediction results with the soft measurement result labels, and adjusts the model parameters according to the differences. This process is repeated until the model performance reaches a satisfactory level.

[0103] It can be understood that, through the process of preprocessing time series data to training and applying the soft measurement model, the time series data is segmented, sampled and the dissimilarity is calculated to obtain a representative sequence of the time series distribution, which is then input into the soft measurement model obtained by iterative training, and finally the estimation or prediction result of the measured variable is obtained.

[0104] In a specific implementation, step S30 includes:

[0105] The representative sequence of the time series distribution is input into a preset time series prediction model, and based on the time series prediction model, the representative sequence of the time series distribution is predicted to obtain a target prediction result.

[0106] It should be noted that the time series prediction model is a model constructed using mathematical and statistical methods, which is used to estimate the input variables of the time series, and the target prediction result is the estimated or predicted result of the variable to be measured.

[0107] It can be understood that through the process of preprocessing time series data to training and application of soft measurement models, segmentation, sampling and dissimilarity calculation of time series data are used to obtain a representative sequence of time series distribution, which is then input into the time series prediction model obtained through iterative training, and finally the estimation or prediction result of the measured variable is obtained.

[0108] In a specific implementation, the step of inputting the time series distribution representative sequence into a preset time series prediction model, predicting the time series distribution representative sequence based on the time series prediction model, and obtaining a target prediction result includes:

[0109] Obtaining training samples representing a time series distribution;

[0110] Determine the target label of the time series distribution representative sequence training sample;

[0111] Based on the time series distribution representative sequence training samples and the soft measurement result labels of the time series distribution representative sequence training samples, the preset model to be trained is iteratively trained to obtain a time series prediction model.

[0112] It should be noted that the time series distribution representative sequence training samples are representative subsets extracted from the original time series data and used to train the time series prediction model. The target label is that when training the time series prediction model, each training sample needs to have a corresponding target value, that is, the answer or result of the model prediction task. The preset model to be trained is the initial model architecture selected before starting the model training. This model usually contains a series of parameters that need to be adjusted and optimized through the training process to achieve the best prediction effect.

[0113] In addition, it should be noted that the iterative training is a process of model training, which usually involves multiple iterations. In each iteration, the model predicts the training samples based on the current parameters, and then compares the prediction results with the target labels, and adjusts the model parameters according to the differences until the model performance reaches a satisfactory level.

[0114] It can be understood that the entire training process is a machine learning process, including obtaining training samples representing the time series distribution, determining the target label, performing iterative training based on the training samples and labels, and using the trained time series prediction model for future predictions.

[0115] In a specific implementation, the step of iteratively training a preset model to be trained based on a time series distribution representative sequence training sample and a soft measurement result label of the time series distribution representative sequence training sample to obtain a time series prediction model includes:

[0116] Inputting the time series distribution representative sequence training samples into a preset model to be trained to obtain sample prediction results;

[0117] Based on the sample prediction result and the target label, a prediction error value is calculated using a loss function;

[0118] Determining whether the predicted error value meets the error standard indicated by a preset threshold range;

[0119] If the error result does not meet the error standard, return to the step of inputting the time series distribution representative sequence training sample into the preset model to be trained to obtain the sample prediction result, and stop training until the prediction error value meets the error standard to obtain the time series prediction model.

[0120] It should be noted that the time series distribution representative sequence training samples are representative subsets extracted from the time series and are used to train the time series prediction model. The preset model to be trained is the initial model architecture selected before starting the model training, which usually includes a series of parameters that need to be adjusted and optimized through the training process to achieve the best prediction effect.

[0121] In addition, it should be noted that the sample prediction result is the prediction value output by the model after the training sample representing the time series distribution is input into the model to be trained. It is the estimate made by the model based on the current parameter settings and input data. The loss function is a function used to measure the difference between the model prediction result and the target label. The smaller the value of the loss function, the closer the model's prediction result is to the actual target value. The prediction error value is a specific value calculated by the loss function, which indicates the degree of difference between the model prediction result and the target label. It is one of the key indicators for evaluating model performance. The error standard indicated by the preset threshold range is the maximum acceptable prediction error value set in advance, which is used to determine whether the performance of the model meets the requirements.

[0122] It can be understood that the whole process is a machine learning training cycle, including model prediction, error calculation, performance evaluation, and parameter adjustment until the model reaches a satisfactory performance standard.

[0123] In a specific implementation, the step of calculating the prediction error value using a loss function based on the sample prediction result and the target label includes:

[0124] Determine the importance weight vector;

[0125] Based on the sample prediction result and the target label, a prediction error value is calculated using a loss function, wherein the loss function weights all hidden states by the importance weight vector.

[0126] It should be noted that the importance weight vector is a set of numerical values, which is usually associated with each hidden state or feature in the sequence data, and is used to measure the influence of each hidden state on the final prediction result. By adjusting the weight vector, the model can better focus on those states or features that have the greatest impact on the prediction results, thereby improving the accuracy of the prediction.

[0127] In the specific implementation, all hidden states are weighted by the determined importance weight vector. When calculating the loss function, the contribution of each hidden state is weighted according to the importance weight vector, which means that the loss function not only considers the difference between the predicted value and the actual value, but also considers the contribution of different hidden states to this difference. Therefore, the model can adjust its internal parameters more finely to more accurately capture the dynamic changes in time series data.

[0128] In the specific implementation, after obtaining the representative sequence of the time series distribution, the optimal model parameters are learned through a method similar to domain generalization, and the basic framework of the recurrent neural network (RNN) is used to build a time series prediction model:

[0129] To develop the soft sensor model, the loss function of the prediction It can be expressed as:

[0130]

[0131] in, Indicates that it comes from the cycle of labeled samples, is the loss function, such as mean squared error, and θ represents the model parameters. However, minimizing There is no guarantee that the differences in distributions between time periods will decrease.

[0132] Distribution matching is performed by introducing the Maximum Mean Discrepancy (MMD) as a regularization term. Therefore, it is defined that In the case of , the distribution matching loss on the hidden state is:

[0133]

[0134] Where d(·,·) represents the MMD distance, which can be estimated by the following formula:

[0135]

[0136] Among them, all hidden states of RNN can be obtained through standard RNN calculation.

[0137] To further improve the model performance, a weight vector α is introduced to learn the relative importance of hidden states, where all hidden states are weighted by a normalized α. The following formula defines the time distribution matching loss:

[0138]

[0139] Combination and The final loss function of the time distribution matching of an RNN layer can be expressed as:

[0140]

[0141] Here, λ is a trade-off hyperparameter.

[0142] The weight vector α is determined by the boosting algorithm that guides the importance weight vector. In the initial stage, the model is pre-trained on the historical training data to obtain the initial parameters θ. Then, for each RNN layer, the weight vector α is updated to reduce the distribution difference. The α learned in the pre-trained model is labeled as α 0 . After that, for each layer of RNN, α i The initial value of is set to αi,j ={1 / T} T , T is the total number of time steps. The boosting algorithm adjusts the importance coefficient by comparing the distribution differences in different periods. If the distribution distance in the n+1th round is greater than that in the nth round, the importance coefficient is increased. The formula is as follows:

[0143]

[0144]

[0145] Where G(·) is the update function, which is calculated at different learning stages through distribution matching loss; and They represent the distribution distances in the n+1th round and the nth round respectively; σ is the sigmoid function in the activation function, which is used to ensure that the value of G(·) is greater than 1 and to ensure the increasing importance; α is the weight vector.

[0146] This embodiment provides an adaptive soft measurement method. This scheme obtains a representative sequence of the time series distribution by performing time series distribution identification on the time series to be measured, thereby realizing adaptive measurement when the time series distributions of the training set and the test set are inconsistent, reducing the impact of data offset caused by sensor degradation on measurement accuracy, and performing soft measurement on the representative sequence of the time series distribution based on the trained soft measurement model to obtain a soft measurement result, thereby significantly improving the accuracy of the soft measurement technology.

[0147] For example, to help understand the implementation process of the adaptive soft-sensing method obtained by combining this embodiment with the above-mentioned embodiment 1, please refer to Figure 2 , Figure 2 A brief flow chart of an adaptive soft-sensing method is provided, specifically:

[0148] By reading the process data, the time series to be measured is obtained;

[0149] Performing time series distribution identification on the time series to be tested, segmenting the time series by selecting a model representative to describe the time series based on a subset selection method of dissimilarity, and obtaining a time series distribution representative sequence;

[0150] The representative sequence of the time series distribution is input into a preset time series prediction model, and based on the time series prediction model, a time series prediction is performed on the representative sequence of the time series distribution to obtain a time series prediction result.

[0151] This solution obtains the time series to be measured by reading process data, identifies the time series distribution of the time series to be measured, and obtains a representative sequence of the time series distribution. This realizes adaptive measurement when the time series distributions of the training set and the test set are inconsistent, reduces the impact of data offset caused by sensor degradation on measurement accuracy, and performs time series prediction on the representative sequence of the time series distribution based on the trained time series prediction model to obtain a time series prediction result, which significantly improves the prediction performance of the model and thereby improves the accuracy of soft measurement technology.

[0152] It should be noted that the above examples are only used to understand the present application and do not constitute a limitation on the adaptive soft measurement method of the present application. More simple transformations based on this technical concept are all within the protection scope of the present application.

[0153] This application also provides an adaptive soft measurement device, please refer to Figure 4 , the adaptive soft measurement device comprises:

[0154] An acquisition module 10 is used to acquire a time series to be measured;

[0155] An identification module 20 is used to identify the time series distribution of the time series to be tested and obtain a representative sequence of the time series distribution;

[0156] The measurement module 30 is used to input the time series distribution representative sequence into a preset soft measurement model, and based on the soft measurement model, perform soft measurement on the time series distribution representative sequence to obtain a soft measurement result, wherein the soft measurement model is obtained by iteratively training a preset model to be trained based on the time series distribution representative sequence training samples and the soft measurement result labels of the time series distribution representative sequence training samples.

[0157] And / or, the identification module 20 includes:

[0158] A segmentation module, used for evenly segmenting the time series to be measured according to a preset value to obtain a segmented time series;

[0159] A sampling module, used for sampling the time series to be tested to obtain a sample time series;

[0160] A calculation module, used for calculating the dissimilarity between each of the segmented time series and each of the sample time series to obtain a dissimilarity set;

[0161] An analysis module, used for analyzing the dissimilarity set to obtain the time series distribution of the time series to be tested;

[0162] The combination module is used to select the segmented time series that meet the preset conditions based on the time series distribution of the time series to be tested, and combine them to obtain a representative sequence of the time series distribution.

[0163] And / or, the measuring module 30 includes:

[0164] The prediction module is used to input the time series distribution representative sequence into a preset time series prediction model, and based on the time series prediction model, predict the time series distribution representative sequence to obtain a target prediction result.

[0165] And / or, the prediction module includes:

[0166] A first acquisition module is used to acquire a training sample of a representative sequence of time series distribution;

[0167] A first determination module is used to determine the target label of the training sample of the time series distribution representative sequence;

[0168] The first training module is used to iteratively train a preset model to be trained based on a time series distribution representative sequence training sample and a soft measurement result label of the time series distribution representative sequence training sample to obtain a time series prediction model.

[0169] And / or, the first training module includes:

[0170] A first input module, used to input the time series distribution representative sequence training sample into a preset to-be-trained model to obtain a sample prediction result;

[0171] A first calculation module, configured to calculate a prediction error value using a loss function based on the sample prediction result and the target label;

[0172] A first judgment module, used to judge whether the prediction error value meets the error standard indicated by a preset threshold range;

[0173] The first loop module is used to return to the step of inputting the time series distribution representative sequence training sample into the preset model to be trained to obtain the sample prediction result if the error result does not meet the error standard, and stop training after the prediction error value meets the error standard to obtain the time series prediction model.

[0174] And / or, the first calculation module includes:

[0175] A second determination module is used to determine an importance weight vector;

[0176] The second calculation module is used to calculate the prediction error value based on the sample prediction result and the target label by using a loss function, wherein the loss function weights all hidden states by the importance weight vector.

[0177] The adaptive soft measurement device provided by the present application adopts the adaptive soft measurement method in the above embodiment, which can solve the technical problem of low precision of soft measurement technology. Compared with the prior art, the beneficial effects of the adaptive soft measurement device provided by the present application are the same as the beneficial effects of the adaptive soft measurement method provided by the above embodiment, and other technical features in the adaptive soft measurement device are the same as the features disclosed in the above embodiment method, which will not be repeated here.

[0178] The present application provides an adaptive soft measurement device, which includes: at least one processor; and a memory connected to the at least one processor in communication; wherein the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the adaptive soft measurement method in the above-mentioned embodiment 1.

[0179] Reference below Figure 5 , which shows a schematic diagram of the structure of an adaptive soft measurement device suitable for implementing the embodiment of the present application. The adaptive soft measurement device in the embodiment of the present application may include but is not limited to mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 5 The adaptive soft measurement device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0180] like Figure 5As shown, the adaptive soft measurement device may include a processing device 1001 (such as a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM: Read Only Memory) 1002 or a program loaded from a storage device 1003 to a random access memory (RAM: Random Access Memory) 1004. In RAM1004, various programs and data required for the operation of the xxx device are also stored. The processing device 1001, ROM1002 and RAM1004 are connected to each other through a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Generally, the following systems can be connected to the I / O interface 1006: an input device 1007 including, for example, a touch screen, a touch pad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; an output device 1008 including, for example, a liquid crystal display (LCD: Liquid Crystal Display), a speaker, a vibrator, etc.; a storage device 1003 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 1009. The communication device 1009 can allow the adaptive soft measurement device to communicate with other devices wirelessly or wired to exchange data. Although the figure shows an adaptive soft measurement device with various systems, it should be understood that it is not required to implement or have all the systems shown. More or fewer systems can be implemented or provided alternatively.

[0181] In particular, according to the embodiments disclosed in the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device, or installed from a storage device 1003, or installed from a ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present application are executed.

[0182] The adaptive soft measurement device provided by the present application adopts the adaptive soft measurement method in the above embodiment, which can solve the technical problem of low precision of soft measurement technology. Compared with the prior art, the beneficial effects of the adaptive soft measurement device provided by the present application are the same as the beneficial effects of the adaptive soft measurement method provided by the above embodiment, and other technical features in the adaptive soft measurement device are the same as the features disclosed in the method of the previous embodiment, which will not be repeated here.

[0183] It should be understood that the various parts disclosed in this application can be implemented by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0184] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0185] The present application provides a computer-readable storage medium having computer-readable program instructions (ie, computer programs) stored thereon, and the computer-readable program instructions are used to execute the adaptive soft-sensing method in the above-mentioned embodiment.

[0186] The computer-readable storage medium provided in the present application may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, systems or devices, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, system or device. The program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination of the above.

[0187] The computer-readable storage medium may be included in the adaptive soft measurement device; or may exist independently without being assembled into the adaptive soft measurement device.

[0188] The computer-readable storage medium carries one or more programs. When the one or more programs are executed by the adaptive soft measurement device, the adaptive soft measurement device: obtains a time series to be measured; performs time series distribution identification on the time series to be measured to obtain a time series distribution representative sequence; inputs the time series distribution representative sequence into a preset soft measurement model, and performs soft measurement on the time series distribution representative sequence based on the soft measurement model to obtain a soft measurement result, wherein the soft measurement model is obtained by iteratively training a preset model to be trained based on a time series distribution representative sequence training sample and a soft measurement result label of the time series distribution representative sequence training sample.

[0189] Computer program code for performing the operations of the present application may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0190] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present application. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0191] The modules involved in the embodiments described in this application may be implemented by software or hardware, wherein the name of the module does not constitute a limitation on the unit itself in some cases.

[0192] The readable storage medium provided by the present application is a computer-readable storage medium, which stores computer-readable program instructions (i.e., computer programs) for executing the above-mentioned adaptive soft measurement method, and can solve the technical problem of low precision of soft measurement technology. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided by the present application are the same as the beneficial effects of the adaptive soft measurement method provided by the above-mentioned embodiment, and will not be repeated here.

[0193] The present application also provides a computer program product, including a computer program, and when the computer program is executed by a processor, the steps of the above-mentioned adaptive soft-sensing method are implemented.

[0194] The computer program product provided by the present application can solve the technical problem of low precision of soft measurement technology. Compared with the prior art, the beneficial effects of the computer program product provided by the present application are the same as the beneficial effects of the adaptive soft measurement method provided by the above embodiment, which will not be repeated here.

[0195] The above descriptions are only some embodiments of the present application, and are not intended to limit the patent scope of the present application. All equivalent structural changes made using the contents of the present application specification and drawings under the technical concept of the present application, or direct / indirect applications in other related technical fields are included in the patent protection scope of the present application.

Claims

1. An adaptive soft-sensing method, characterized in that: The method includes: Get the time series to be tested; Performing time series distribution identification on the time series to be tested to obtain a time series distribution representative sequence; The time series distribution representative sequence is input into a preset soft measurement model, and based on the soft measurement model, soft measurement is performed on the time series distribution representative sequence to obtain a soft measurement result, wherein the soft measurement model is obtained by iteratively training a preset model to be trained based on the time series distribution representative sequence training samples and the soft measurement result labels of the time series distribution representative sequence training samples.

2. The method according to claim 1, characterized in that The step of performing time series distribution identification on the time series to be tested to obtain a time series distribution representative sequence comprises: According to the preset value, the time series to be measured is evenly segmented to obtain a segmented time series; Sampling the time series to be tested to obtain a sample time series; Calculate the dissimilarity between each of the segmented time series and each of the sample time series to obtain a dissimilarity set; Analyze the dissimilarity set to obtain the time series distribution of the time series to be tested; Based on the time series distribution of the time series to be tested, the segmented time series that meet the preset conditions are selected and combined to obtain a time series distribution representative sequence.

3. The method according to claim 1, characterized in that The soft measurement model includes a time series prediction model, the soft measurement result includes a target prediction result, and the step of inputting the time series distribution representative sequence into a preset soft measurement model, and performing soft measurement on the time series distribution representative sequence based on the soft measurement model to obtain the soft measurement result includes: The representative sequence of the time series distribution is input into a preset time series prediction model, and based on the time series prediction model, the representative sequence of the time series distribution is predicted to obtain a target prediction result.

4. The method according to claim 3, characterized in that Before the step of inputting the time series distribution representative sequence into a preset time series prediction model and predicting the time series distribution representative sequence based on the time series prediction model to obtain a target prediction result, the step further includes: Obtaining training samples representing a time series distribution; Determine the target label of the time series distribution representative sequence training sample; Based on the time series distribution representative sequence training samples and the soft measurement result labels of the time series distribution representative sequence training samples, the preset model to be trained is iteratively trained to obtain a time series prediction model.

5. The method according to claim 4, characterized in that The step of iteratively training a preset model to be trained based on the time series distribution representative sequence training samples and the soft measurement result labels of the time series distribution representative sequence training samples to obtain a time series prediction model includes: Inputting the time series distribution representative sequence training samples into a preset model to be trained to obtain sample prediction results; Based on the sample prediction result and the target label, a prediction error value is calculated using a loss function; Determining whether the predicted error value meets the error standard indicated by a preset threshold range; If the error result does not meet the error standard, return to the step of inputting the time series distribution representative sequence training sample into the preset model to be trained to obtain the sample prediction result, and stop training until the prediction error value meets the error standard to obtain the time series prediction model.

6. The method according to claim 5, characterized in that The step of calculating the prediction error value using a loss function based on the sample prediction result and the target label includes: Determine the importance weight vector; Based on the sample prediction result and the target label, a prediction error value is calculated using a loss function, wherein the loss function weights all hidden states by the importance weight vector.

7. An adaptive soft measurement device, characterized in that: The device comprises: An acquisition module is used to obtain the time series to be tested; An identification module, used for performing time series distribution identification on the time series to be tested to obtain a representative sequence of the time series distribution; A measurement module is used to input the time series distribution representative sequence into a preset soft measurement model, and based on the soft measurement model, perform soft measurement on the time series distribution representative sequence to obtain a soft measurement result, wherein the soft measurement model is obtained by iteratively training a preset model to be trained based on the time series distribution representative sequence training samples and the soft measurement result labels of the time series distribution representative sequence training samples.

8. An adaptive soft measurement device, characterized in that: The device comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the adaptive soft-sensing method according to any one of claims 1 to 6.

9. A storage medium, characterized in that: The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the adaptive soft-sensing method according to any one of claims 1 to 6 are implemented.

10. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the steps of the adaptive soft-sensing method according to any one of claims 1 to 6 are implemented.

Images