Cement rotary kiln state recognition method and device, computer equipment and storage medium

Abstract

The embodiment of the application belongs to the field of artificial intelligence, and relates to a cement rotary kiln state identification method and device, computer equipment and a storage medium. The method comprises the following steps: obtaining running data of a cement rotary kiln; inputting the running data into an initial self-encoder to obtain encoded data and reconstruction data; calculating reconstruction error according to the running data and the reconstruction data, and generating fusion data according to the reconstruction error and the encoded data; inputting the fusion data into an initial clustering model to obtain clustering prediction results and associating the running data with kiln state based on the clustering prediction results; calculating a joint loss based on the running data, the reconstruction data, the encoded data and the clustering prediction results, adjusting the initial self-encoder and the initial clustering model according to the joint loss to obtain a self-encoder and a clustering model, and constructing a state identification model; and inputting to-be-identified data into the state identification model to obtain a state identification result. The application improves the accuracy of cement rotary kiln state identification.

Description

Technical Field

[0001] This application relates to the fields of artificial intelligence and cement production technology, and in particular to a method, apparatus, computer equipment and storage medium for identifying the status of a cement rotary kiln. Background Technology

[0002] The cement firing process exhibits characteristics such as nonlinearity, multivariability, strong coupling, large time delay, and time-varying properties, making the kiln conditions of cement rotary kilns extremely complex. The cement rotary kiln is a key piece of equipment in cement production. During the cement firing process, it is necessary to identify the kiln's condition to control the rotary kiln and improve the cement's yield. Simultaneously, with increasingly tight energy supplies and soaring prices of electricity, coal, and other energy sources required for cement production, optimizing the control of the cement rotary kiln based on its condition is of paramount importance for improving energy conservation and reducing consumption in this core aspect of cement production.

[0003] However, traditional cement rotary kiln condition identification technology mainly relies on the manual experience of on-site operators. Due to the complexity of kiln conditions, the accuracy of manual identification of cement rotary kiln conditions is low, and it is becoming increasingly unsuitable for today's cement rotary kiln systems. Summary of the Invention

[0004] The purpose of this application is to provide a method, apparatus, computer equipment, and storage medium for identifying the status of a cement rotary kiln, in order to solve the problem of low accuracy in identifying the status of a cement rotary kiln.

[0005] To address the aforementioned technical problems, this application provides a method for identifying the state of a cement rotary kiln, employing the following technical solution:

[0006] Obtain operating data of cement rotary kilns;

[0007] The running data is input into the initial autoencoder to obtain encoded data and reconstructed data;

[0008] The reconstruction error is calculated based on the running data and the reconstruction data, and fused data is generated based on the reconstruction error and the encoded data.

[0009] The fused data is input into the initial clustering model to obtain the clustering prediction result, and the operating data is correlated with the kiln condition based on the reconstruction error of the clustering prediction result.

[0010] Calculate the joint loss based on the running data, the reconstructed data, the encoded data, and the clustering prediction results;

[0011] The initial autoencoder and the initial clustering model are adjusted according to the joint loss until the joint loss satisfies the training stopping condition, thereby obtaining the autoencoder and the clustering model, and a state recognition model is constructed based on the autoencoder and the clustering model.

[0012] The data to be identified is input into the state recognition model to obtain the state recognition result.

[0013] To address the aforementioned technical problems, this application also provides a cement rotary kiln status identification device, which employs the following technical solution:

[0014] Run the acquisition module to obtain the operating data of the cement rotary kiln;

[0015] The reconstruction module is used to input the running data into the initial autoencoder to obtain encoded data and reconstructed data;

[0016] The calculation module is used to calculate the reconstruction error based on the running data and the reconstruction data, and to generate fused data based on the reconstruction error and the encoded data;

[0017] The fusion input module is used to input the fused data into the initial clustering model to obtain the clustering prediction result, and to associate the operating data with the kiln condition based on the reconstruction error of the clustering prediction result.

[0018] The loss calculation module is used to calculate the joint loss based on the running data, the reconstructed data, the encoded data, and the clustering prediction results;

[0019] The model training module is used to adjust the initial autoencoder and the initial clustering model according to the joint loss until the joint loss meets the training stopping condition, thereby obtaining the autoencoder and the clustering model, and constructing a state recognition model based on the autoencoder and the clustering model.

[0020] The state recognition module is used to input the data to be recognized into the state recognition model to obtain the state recognition result.

[0021] To address the aforementioned technical problems, this application also provides a computer device that employs the following technical solution:

[0022] Obtain operating data of cement rotary kilns;

[0023] The running data is input into the initial autoencoder to obtain encoded data and reconstructed data;

[0024] The reconstruction error is calculated based on the running data and the reconstruction data, and fused data is generated based on the reconstruction error and the encoded data.

[0025] The fused data is input into the initial clustering model to obtain the clustering prediction result, and the operating data is correlated with the kiln condition based on the reconstruction error of the clustering prediction result.

[0026] Calculate the joint loss based on the running data, the reconstructed data, the encoded data, and the clustering prediction results;

[0027] The initial autoencoder and the initial clustering model are adjusted according to the joint loss until the joint loss satisfies the training stopping condition, thereby obtaining the autoencoder and the clustering model, and a state recognition model is constructed based on the autoencoder and the clustering model.

[0028] The data to be identified is input into the state recognition model to obtain the state recognition result.

[0029] To address the aforementioned technical problems, this application also provides a computer-readable storage medium, employing the technical solution described below:

[0030] Obtain operating data of cement rotary kilns;

[0031] The running data is input into the initial autoencoder to obtain encoded data and reconstructed data;

[0032] The reconstruction error is calculated based on the running data and the reconstruction data, and fused data is generated based on the reconstruction error and the encoded data.

[0033] The fused data is input into the initial clustering model to obtain the clustering prediction result, and the operating data is correlated with the kiln condition based on the reconstruction error of the clustering prediction result.

[0034] Calculate the joint loss based on the running data, the reconstructed data, the encoded data, and the clustering prediction results;

[0035] The initial autoencoder and the initial clustering model are adjusted according to the joint loss until the joint loss satisfies the training stopping condition, thereby obtaining the autoencoder and the clustering model, and a state recognition model is constructed based on the autoencoder and the clustering model.

[0036] The data to be identified is input into the state recognition model to obtain the state recognition result.

[0037] Compared with the prior art, the embodiments of this application have the following main advantages: acquiring the operating data of a cement rotary kiln; inputting the operating data into an initial autoencoder for feature extraction to obtain encoded data, and then reconstructing and fitting to obtain reconstructed data, thereby extracting the most informative features from the operating data; calculating the reconstruction error based on the operating data and the reconstructed data, generating fused data based on the reconstruction error and the encoded data, inputting the fused data into an initial clustering model, transforming state prediction into a clustering problem, obtaining clustering prediction results, and the clustering prediction results being a set of operating data under different kiln conditions. Operating data under different kiln conditions have different reconstruction errors, therefore... This method can determine the kiln condition corresponding to the operating data, thereby learning the data distribution of the operating data under each kiln condition. Based on the operating data, reconstructed data, encoded data, and clustering prediction results, the joint loss is calculated, thereby adjusting the initial autoencoder and initial clustering model until the autoencoder and clustering model are obtained, and the autoencoder and clustering model are constructed into a state recognition model. The data to be identified is input into the state recognition model to obtain the state recognition result. This application does not require manual intervention. Through clustering, data feature extraction and reasoning judgment are performed, which can mine data patterns that are helpful for clustering analysis from top to bottom, thus improving the accuracy of cement rotary kiln state recognition. Attached Figure Description

[0038] To more clearly illustrate the solutions in this application, the accompanying drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0039] Figure 1 This is an exemplary system architecture diagram to which this application can be applied;

[0040] Figure 2 This is a flowchart of an embodiment of the cement rotary kiln state identification method according to this application;

[0041] Figure 3 This is a schematic diagram of a self-encoder according to an embodiment of this application;

[0042] Figure 4 This is a schematic diagram of a structure of an embodiment of the cement rotary kiln status identification device according to this application;

[0043] Figure 5 This is a schematic diagram of the structure of one embodiment of the computer device according to this application. Detailed Implementation

[0044] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein in the specification of the application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings of this application, are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings of this application are used to distinguish different objects, not to describe a particular order.

[0045] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0046] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0047] like Figure 1 As shown, system architecture 100 may include terminal devices 101, 102, and 103, a network 104, and a server 105. Network 104 serves as the medium for providing communication links between terminal devices 101, 102, and 103 and server 105. Network 104 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0048] Users can use terminal devices 101, 102, and 103 to interact with server 105 via network 104 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 101, 102, and 103, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social media platform software, etc.

[0049] Terminal devices 101, 102, and 103 can be various electronic devices with displays and support web browsing, including but not limited to smartphones, tablets, e-book readers, MP3 players (Moving Picture Experts Group Audio Layer III), MP4 players (Moving Picture Experts Group Audio Layer IV), laptops, and desktop computers, etc.

[0050] Server 105 can be a server that provides various services, such as a backend server that supports the pages displayed on terminal devices 101, 102, and 103.

[0051] It should be noted that the cement rotary kiln status identification method provided in this application embodiment is generally executed by a server, and correspondingly, the cement rotary kiln status identification device is generally installed in the server.

[0052] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0053] Continue to refer to Figure 2 The diagram illustrates a flowchart of an embodiment of the cement rotary kiln condition identification method according to this application. The cement rotary kiln condition identification method includes the following steps:

[0054] Step S201: Obtain the operating data of the cement rotary kiln.

[0055] In this embodiment, the cement rotary kiln status identification method operates on electronic equipment (e.g., Figure 1 The server shown can communicate with the terminal via wired or wireless connection. It should be noted that the aforementioned wireless connection methods may include, but are not limited to, 3G / 4G / 5G connections, WiFi connections, Bluetooth connections, WiMAX connections, Zigbee connections, UWB (ultra wideband) connections, and other currently known or future wireless connection methods.

[0056] Specifically, during the operation of a cement rotary kiln, relevant data from kiln condition monitoring points are recorded to obtain operational data. For example, operational data may include kiln current, secondary and tertiary air temperature, and decomposition furnace outlet temperature from the monitoring points.

[0057] In one embodiment, the data acquisition frequency f1 is determined based on the cement production conditions and the control system requirements, and the operating data of the cement rotary kiln is collected according to the acquisition frequency f1; simultaneously, the kiln condition identification frequency f2 is determined. The acquisition frequency can be on the order of seconds, and the identification frequency can be on the order of minutes.

[0058] The operational data includes stable kiln operation data, excellent kiln operation data, and suboptimal kiln operation data, which correspond to three states of kiln conditions during cement firing: stable, excellent, and suboptimal. These represent stable, good, and bad kiln conditions, respectively. Generally, excellent and suboptimal kiln conditions can be categorized as unstable kiln conditions.

[0059] Step S202: Input the running data into the initial autoencoder to obtain the encoded data and the reconstructed data.

[0060] The initial autoencoder can be an autoencoder that has not yet been trained. An autoencoder is a deep autoencoder that can perform unsupervised learning and consists of three parts: an encoder, a decoder, and a loss function.

[0061] Specifically, the running data is input into the initial autoencoder, which reconstructs the data. The encoder in the initial autoencoder downsamples the running data to obtain compressed features, which are denoted as encoded data. Then, the decoder in the initial autoencoder upsamples these compressed features (i.e., decodes them) to obtain the reconstructed data. The dimensionality of the reconstructed data is the same as the running data input to the encoder.

[0062] In one embodiment, when inputting operational data into the initial autoencoder, the operational data needs to be sliced. When the acquisition frequency f1 and the kiln condition identification frequency f2 are different, the operational data can be sliced ​​according to the kiln condition identification frequency f2, and the operational data within the time window of the kiln condition identification frequency f2 is input into the encoder of the initial autoencoder one by one. The encoder uses a one-dimensional convolution method for feature extraction to obtain a compressed feature vector. To prevent gradient vanishing, gradient exploding, and degradation problems, a residual method is used for cross-layer connections. Similarly, the decoder also uses a residual method for cross-layer connections, thereby combining features from different modalities and preventing feature loss.

[0063] Step S203: Calculate the reconstruction error based on the running data and the reconstruction data, and generate fused data based on the reconstruction error and the encoded data.

[0064] Specifically, the reconstructed data x' is obtained by reconstructing the running data x. The two have the same dimensions, but their numerical values ​​differ. From this, the reconstruction error z can be calculated. r =||x-x'|| 2 The reconstruction error z r The encoded data z output by the encoder c By splicing the data together, we obtain the fused data z.

[0065] Step S204: Input the fused data into the initial clustering model to obtain the clustering prediction results, and associate the operating data with the kiln condition based on the reconstruction error of the clustering prediction results.

[0066] The initial clustering model can be a clustering model that has not yet been trained.

[0067] Specifically, the various states of kiln conditions are known, including stable kiln conditions, excellent kiln conditions, and suboptimal kiln conditions, and the distribution of operational data differs in each state. Therefore, the identification of kiln conditions can be transformed into a clustering problem.

[0068] The fused data is input into the initial clustering model, which then clusters the operational data to obtain cluster prediction results. These predictions include different clusters, each a set of operational data under a specific kiln condition. For operational data under different kiln conditions, the reconstruction error calculated after passing through the initial autoencoder is different. Therefore, based on the reconstruction error of each cluster's operational data after clustering, the kiln condition corresponding to each cluster can be determined. The initial autoencoder and the initial clustering model correlate the operational data with its corresponding kiln condition, thereby learning and mining the data distribution and patterns of the operational data under each kiln condition.

[0069] Typically, among the sampled operational data, the most frequent data is from stable kiln operation, followed by data from excellent kiln operation, and the least frequent is data from suboptimal kiln operation. When the initial autoencoder reconstructs the various types of operational data, the larger the sample size for the type of data, the smaller the reconstruction error for that type. Therefore, the reconstruction error is smallest for stable kiln operation data, followed by excellent kiln operation data, and largest for suboptimal kiln operation data.

[0070] In one embodiment, since the clusters obtained by clustering contain multiple running data samples, the reconstruction error of each running data sample can be calculated separately, and then the average value can be calculated to compare the clusters.

[0071] In one embodiment, the initial clustering model can be a model consisting of multiple fully connected layers, with the activation function of the last layer being softmax, outputting the probability p of the running data belonging to each state, and thus obtaining the clustering prediction result.

[0072] Step S205: Calculate the joint loss based on the running data, reconstructed data, encoded data, and clustering prediction results.

[0073] Specifically, the initial autoencoder has a model loss calculated based on the running data, reconstructed data, and encoded data; the initial clustering model also has a model loss calculated based on the fused data and clustering prediction results, while the fused data is calculated based on the error between the running data and the reconstructed data, and the encoded data. The joint loss can be calculated from the model losses of the initial autoencoder and the initial clustering model, representing the total loss incurred in reconstructing and clustering the running data.

[0074] Step S206: Adjust the initial autoencoder and initial clustering model according to the joint loss until the joint loss meets the training stopping condition to obtain the autoencoder and clustering model, and build a state recognition model based on the autoencoder and clustering model.

[0075] Specifically, with the goal of reducing the joint loss, the model parameters of the initial autoencoder and the initial clustering model are adjusted based on the joint loss. The initial autoencoder and the initial clustering model are then iteratively trained using running data until the obtained joint loss meets the training stopping condition. At this point, training stops, and the initial autoencoder at the time of training stopping is taken as the final autoencoder, and the initial clustering model at the time of training stopping is taken as the final clustering model. The training stopping condition can be that the joint loss is less than a preset loss threshold.

[0076] State recognition models can be composite models, built on autoencoders and clustering models, with the autoencoders and clustering models serving as sub-models within the state recognition model.

[0077] Step S207: Input the data to be identified into the state recognition model to obtain the state recognition result.

[0078] Specifically, when applying the model, data to be identified is acquired. This data can be operational data generated by a cement rotary kiln. The data is input into the state identification model, which processes it to identify the kiln condition and obtain the state identification result.

[0079] In this embodiment, the operating data of a cement rotary kiln is acquired; the operating data is input into an initial autoencoder for feature extraction to obtain encoded data, and then reconstructed and fitted to obtain reconstructed data, thereby extracting the most informative features from the operating data; the reconstruction error is calculated based on the operating data and the reconstructed data, and fused data is generated based on the reconstruction error and the encoded data. The fused data is input into an initial clustering model, transforming state prediction into a clustering problem, and obtaining clustering prediction results. The clustering prediction results are a set of operating data under different kiln conditions. The operating data under different kiln conditions have different reconstruction errors, thus determining the kiln condition corresponding to the operating data, thereby learning the data distribution of operating data under each kiln condition; the joint loss is calculated based on the operating data, reconstructed data, encoded data, and clustering prediction results, thereby adjusting the initial autoencoder and the initial clustering model until the autoencoder and the clustering model are obtained, and the autoencoder and the clustering model are constructed into a state recognition model; the data to be identified is input into the state recognition model to obtain the state recognition result. This application does not require manual intervention. By performing data feature extraction and reasoning judgment through clustering, it can mine data patterns that are helpful for cluster analysis from top to bottom, improving the accuracy of cement rotary kiln state recognition.

[0080] Furthermore, before step S202 above, the method may include: detecting abnormal samples in the operating data using a standard deviation algorithm to remove abnormal samples from the operating data; calculating the correlation coefficient between various process features in the operating data after removing abnormal samples to perform dimensionality reduction processing on the operating data after removing abnormal samples based on the correlation coefficient; and standardizing the operating data after dimensionality reduction processing.

[0081] Specifically, before inputting the running data into the initial autoencoder, the running data can be preprocessed. First, the running data is processed using a standard deviation algorithm (the 3 sigma rule in statistics, used to detect outliers in a sample) to detect and remove abnormal samples. These abnormal samples may be generated due to equipment malfunctions or other reasons.

[0082] The operational data contains relevant data on various process characteristics, such as kiln current and blast temperature. Dimensionality reduction can be performed on the operational data. During dimensionality reduction, the Spearman and Pearson algorithms can be used to calculate the correlation coefficient between pairwise process characteristics. The correlation coefficient measures the strength of the correlation between process characteristics. When the relationship between process characteristics is non-linear, the Spearman algorithm can be used; when the relationship is linear, the Pearson algorithm can be used.

[0083] The process features that are paired together are identified based on the correlation coefficient, and then dimensionality reduction is performed on them. In one embodiment, the process features can also be dimensionality reduced using the PCA (principal components analysis) algorithm.

[0084] Finally, the dimensionality-reduced operating data is standardized by using the z-score algorithm to convert it into a standard normal distribution, eliminating errors caused by different dimensions of process features. The z-score algorithm is a linear transformation that compresses the feature vector x of the operating data proportionally before shifting it.

[0085] In this embodiment, the running data is subjected to anomaly detection, dimensionality reduction, and standardization, making the running data more standardized and ensuring the accuracy of subsequent running data processing.

[0086] Furthermore, step S202 may include: inputting the running data into the encoder in the initial autoencoder to encode the running data through each convolutional layer and each bottleneck layer in the encoder to obtain encoded data; inputting the encoded data into the decoder in the initial autoencoder to decode the encoded data through each deconvolutional layer in the decoder to obtain reconstructed data.

[0087] Specifically, Figure 3This is a schematic diagram of an autoencoder / initial autoencoder in one embodiment, see reference. Figure 3 The initial autoencoder includes an encoding process and a decoding process. The encoder compresses the running data x into the latent space z to obtain the encoded data (i.e., the compressed features). The decoder reconstructs the data x' based on the latent state. The error between x' and x is fed back to the network for adjustment and optimization.

[0088] When the acquisition frequency f1 and the kiln condition identification frequency f2 are different, the operating data can be sliced ​​according to the kiln condition identification frequency f2, and the operating data within the time window of the kiln condition identification frequency f2 can be input into the encoder in the initial autoencoder. The encoder can contain multiple convolutional layers and bottleneck layers, each with its own network structure parameters. The convolutional layers are one-dimensional convolutional layers (Conv1D). The encoder uses one-dimensional convolution to extract features and obtain compressed feature vectors. To prevent gradient vanishing, gradient explosion, and degradation problems, a residual method is used for cross-layer connections to combine features from different modalities.

[0089] The decoder contains multiple deconvolutional layers, each with its own network structure parameters; these deconvolutional layers can be one-dimensional deconvolutional layers (DeConv1D). Similarly, the decoder uses residual methods for cross-layer connections.

[0090] The encoder encodes high-dimensional input into low-dimensional latent variables, thereby forcing the neural network to learn the most informative features; the decoder restores the latent variables of the hidden layer to the initial dimension. Ideally, the output of the decoder can perfectly or approximately recover the original input.

[0091] An autoencoder uses the input data x itself as supervision to guide the neural network in attempting to learn a mapping relationship, thereby obtaining a reconstructed output x'. In the scenario of time series anomaly detection, anomalies are rare compared to normal data. Therefore, it can be assumed that if the difference between the output x' reconstructed by the autoencoder and the original input exceeds a certain threshold, the original time series contains an anomaly. This difference can be the reconstruction error or the reconstruction loss calculated based on the reconstruction error and the encoded data.

[0092] Therefore, after obtaining the reconstructed data, the difference between the operating data and the reconstructed data is compared to determine whether there are any anomalies in the operating data. When there are no anomalies in the operating data, the cement rotary kiln can be considered to be in a stable kiln condition; when there are anomalies in the operating data, the cement rotary kiln can be considered to be in a non-stable kiln condition.

[0093] In this embodiment, an initial autoencoder is used for encoding and decoding to reconstruct the running data, obtaining encoded data and reconstructed data, thereby learning the most informative features in the running data and preparing data for kiln condition identification.

[0094] Furthermore, the initial autoencoder includes a first autoencoder and a second autoencoder, and the initial clustering model includes a first clustering model and a second clustering model. The first autoencoder is connected to the first clustering model, and the second autoencoder is connected to the second clustering model. Step S204 may include: during the first stage of training, inputting the fused data output by the first autoencoder into the first clustering model to obtain the first clustering prediction result; the first autoencoder is used to process the full amount of running data; based on the reconstruction error of the first clustering prediction result, the running data is divided into mass samples and niche samples, and the first autoencoder and the first clustering model are instructed to label the mass samples as kiln condition stable samples; during the second stage of training, inputting the fused data output by the second autoencoder into the second clustering model to obtain the second clustering prediction result; the second autoencoder is used to process the niche samples; based on the reconstruction error of the second clustering prediction result, the niche samples are divided into kiln condition excellent samples and kiln condition substandard samples, and the second autoencoder and the second clustering model are instructed to associate the niche samples with kiln condition excellent samples or kiln condition substandard samples.

[0095] Specifically, this application transforms the prediction problem into a clustering problem. The kiln condition includes a stable kiln condition and a non-stationary kiln condition. The non-stationary kiln condition can be further divided into an excellent kiln condition and a poor kiln condition. The distribution of operating data is different when the cement rotary kiln is in a stable kiln condition and a non-stationary kiln condition; similarly, the distribution of operating data also differs when the cement rotary kiln is in an excellent kiln condition and a poor kiln condition.

[0096] To more accurately separate operational data in different states, this application employs a clustering model. The training and application process can be divided into two stages, setting up two clustering sub-models: a first clustering model and a second clustering model. Each sub-model contains multiple fully connected layers, with the last layer using softmax as the activation function. Each sub-model performs binary classification, outputting the probability p of a sample data belonging to one of the two categories. Furthermore, the initial autoencoder in this application includes two parallel first and second autoencoders. The first autoencoder is connected to the first clustering model, and its result is input into the first clustering model. The second autoencoder is connected to the second clustering model, and its result is input into the second clustering model. The network structures of the first and second autoencoders can be identical, and their network structure parameters can also be identical initially. The first and second autoencoders perform the same processing on the operational data, but their network structure parameters differentiate during training.

[0097] In the first training phase, the acquired full set of operational data is input into the first autoencoder to obtain fused data. This fused data is then input into the first clustering model, which calculates the probability that the operational data belongs to one of two categories, yielding the first clustering prediction result. The first training phase separates stationary kiln condition samples from non-stationary kiln condition samples, with stationary samples having a numerical advantage. As described earlier in this application, the first autoencoder reconstructs the stationary samples, which have a numerical advantage, resulting in a smaller reconstruction error. However, since non-stationary kiln condition samples are fewer in number than stationary samples, their reconstruction error is relatively larger. Calculating the reconstruction error between the two clusters of operational data allows for the differentiation between stationary and non-stationary kiln condition samples. Because the number of stationary kiln condition samples exceeds that of non-stationary samples, the stationary samples are classified as common samples, and the non-stationary samples are classified as niche samples.

[0098] The second round of training targets a minority of samples. These minority samples are input into a second autoencoder to obtain fused data, which is then input into a second clustering model to obtain the second clustering prediction results. The second training phase separates samples with excellent kiln conditions from those with suboptimal kiln conditions. Since the number of excellent kiln condition samples is greater than that of suboptimal samples, the reconstruction error of excellent kiln condition samples after reconstruction by the second autoencoder is relatively small, while the reconstruction error of suboptimal kiln condition samples is relatively large. Calculating the reconstruction error between the two clusters in the second clustering result allows for the differentiation between excellent and suboptimal kiln condition samples.

[0099] In the first stage, the first autoencoder and the first clustering model associate the status label of the "stable kiln condition sample" with the operating data representing the stable kiln condition sample; in the second stage, the second autoencoder and the second clustering model associate the status label of the "excellent kiln condition sample" with the operating data representing the excellent kiln condition sample, and associate the status label of the "second-worst kiln condition sample" with the operating data representing the second-worst kiln condition sample, thereby uncovering hidden data patterns in the operating data to improve the accuracy of clustering.

[0100] In this embodiment, different types of operational data are separated by different clustering models, and the kiln condition corresponding to the separated operational data is determined based on the reconstruction error, so that the model can identify operational data under different kiln condition conditions.

[0101] Furthermore, step S205 may include: calculating the reconstruction loss based on the running data, reconstructed data, and encoded data; calculating the sample distribution information of the running data based on the clustering prediction results; calculating the sample energy of the running data based on the sample distribution information; and calculating the joint loss based on the reconstruction loss and the sample energy.

[0102] Specifically, the joint loss is calculated from the reconstruction loss and the sample energy, where the reconstruction loss L1 is expressed as follows:

[0103] L1=θ1||x-x'|| 2 +θ2KL(p(z c |x)p(z0)) (1)

[0104] Where x represents the running data, x' represents the reconstructed data, θ1 represents the reconstruction error loss weight, and ||x-x'|| 2 θ2 is the reconstruction error, i.e., the error between the running data x and the reconstructed data x'; θ2 is the regularization term loss weight, KL represents the divergence, which measures the degree of approximation between the reconstructed data x' and the running data x; p(z) c |x) represents the encoded data output by the encoder, and p(z0) is a standard normal distribution, i.e., p(z0) = normal(0,1). The existence of the regularization term aims to make p(z0) = normal(0,1). c |x) is also close to a normal distribution; if there is no regularization term, the model will continuously reduce randomness, i.e. the variance of the encoder output, in order to reduce reconstruction loss. After the randomness is reduced, the autoencoder will be unable to generate various data.

[0105] Clustering prediction results typically consist of two probability values ​​p representing the probability that a cement rotary kiln belongs to a certain kiln condition. Based on the clustering prediction results, the sample distribution information of the operating data can be calculated.

[0106]

[0107]

[0108]

[0109] Where, p ik This represents the probability value of belonging to each kiln condition, where N is the sample size and k takes a value of 2. μ represents the estimated variance of the distribution for each category. k S represents the estimated mean of the distribution for each category. k Let T be the covariance matrix, and T denote the transpose.

[0110] Then calculate the sample energy (i.e., the maximum likelihood value of the sample) E(z) of the running data:

[0111]

[0112] The joint loss is expressed as:

[0113]

[0114] Where ω is the loss weight of the clustering model.

[0115] Since the training can be divided into two stages, the first stage separates the stationary kiln condition samples and the non-stationary kiln condition samples through the first clustering model. After obtaining the clustering prediction results of the first clustering model, the joint loss can be calculated to train the first autoencoder and the first clustering model. Then, the second stage is carried out, in which the excellent kiln condition samples and the inferior kiln condition samples in the non-stationary kiln condition samples are separated through the second clustering model. After obtaining the clustering prediction results of the second clustering model, the joint loss is calculated to train the second autoencoder and the second clustering model.

[0116] In this embodiment, the reconstruction loss and sample energy are calculated separately, and the joint loss is calculated based on the reconstruction loss and sample energy. This comprehensively considers the losses of the initial autoencoder and the initial clustering model, thus improving the accuracy of the loss calculation.

[0117] Furthermore, step S207 may include: inputting the data to be identified into the first autoencoder in the state recognition model to obtain fused data output by the first autoencoder; inputting the fused data into the first clustering model in the state recognition model to obtain a first clustering result; when it is determined from the first clustering result that the data to be identified belongs to the mass sample, generating a state recognition result based on the first clustering result; when it is determined from the first clustering result that the data to be identified belongs to the niche sample, inputting the data to be identified into the second autoencoder in the state recognition model to obtain fused data output by the second autoencoder; inputting the fused data into the second clustering model in the state recognition model to obtain a second clustering result, and generating a state recognition result based on the second clustering result.

[0118] Specifically, in application, data is collected from a cement rotary kiln to obtain data to be identified. This data is then input into an autoencoder in the state recognition model. The encoder in the autoencoder encodes the data to be identified to obtain encoded data, and the decoder in the autoencoder decodes the encoded data to obtain reconstructed data. The reconstruction error is calculated based on the data to be identified and the reconstructed data, and the encoded data and the reconstruction error are directly concatenated to obtain fused data.

[0119] Since the autoencoder includes a first autoencoder and a second autoencoder, in application, the data to be identified can be input into the first autoencoder and the second autoencoder respectively to obtain the fused data output by each. In one embodiment, the data to be identified is first input into the first autoencoder to obtain the fused data output by the first autoencoder; when the data to be identified is determined to belong to a minority sample according to the first clustering model, the data to be identified is then input into the second autoencoder to obtain the fused data output by the second autoencoder.

[0120] Then, the fused data output by the first autoencoder is input into the first clustering model in the state recognition model for processing to obtain the first clustering result.

[0121] The first clustering result can show the probability that the data to be identified belongs to a stable kiln condition sample or a non-stationary kiln condition sample. Stable kiln condition samples are also known as common samples. When the probability that the data to be identified belongs to a stable kiln condition sample is greater than the probability that the data to be identified belongs to a non-stationary kiln condition sample, it is determined that the data to be identified belongs to the common samples. This generates a state identification result for a cement rotary kiln in a stable kiln condition, and the state identification result can also include the corresponding probability value.

[0122] When the first clustering result indicates that the data to be identified belongs to a minority sample, i.e., a non-stationary kiln condition sample, the fused data output from the second autoencoder is input into the second clustering model for further identification, resulting in the second clustering result. The second clustering result can show the probability that the data to be identified belongs to a sample with excellent kiln condition or a sample with poor kiln condition. When the probability that the data to be identified belongs to a sample with excellent kiln condition is greater than the probability that it belongs to a sample with poor kiln condition, a state identification result of "cement rotary kiln in excellent kiln condition" is generated; otherwise, a state identification result of "cement rotary kiln in poor kiln condition" is generated.

[0123] In this embodiment, the data to be identified is input into the state recognition model. First, the data is reconstructed through an autoencoder to extract the most informative features. Then, the state is identified through the first clustering model and the second clustering model, which improves the accuracy of the state recognition of the cement rotary kiln.

[0124] Furthermore, after step S207, the process may further include: determining state indication parameters based on the state identification results; calculating state control parameters based on the state indication parameters and the reconstruction error; and controlling the cement rotary kiln based on the state control parameters.

[0125] Specifically, the status indication parameter is a number used to identify the kiln condition of the cement rotary kiln. Different kiln condition states correspond to different status indication parameters. In one embodiment, when the cement rotary kiln is in a stable kiln condition state, the status indication parameter can be 1, or no status indication parameter is required; when the cement rotary kiln is in a good kiln condition state, the status indication parameter can be 1; when the cement rotary kiln is in a suboptimal kiln condition state, the status indication parameter can be -1.

[0126] The state control parameter is a numerical value, which is a required parameter for process control of a cement rotary kiln. The state control parameter is calculated based on the state indication parameter and the reconstruction error. In one embodiment, the reconstruction error is standardized to the interval [0,1], and then the standardized reconstruction error is multiplied by the state indication parameter to obtain the state control parameter.

[0127] In this embodiment, the state indication parameters are determined based on the state identification results, and then the state control parameters are calculated with the reconstruction error, so as to perform process control on the cement rotary kiln based on the state control parameters.

[0128] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by instructing related hardware with computer-readable instructions. These computer-readable instructions can be stored in a computer-readable storage medium. When executed, the program can include the processes of the embodiments of the above methods. The aforementioned storage medium can be a non-volatile storage medium such as a magnetic disk, optical disk, or read-only memory (ROM), or random access memory (RAM).

[0129] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0130] Further reference Figure 4 As a response to the above Figure 2 The present application provides an embodiment of a cement rotary kiln status identification device, which is similar to the method described above. Figure 2 Corresponding to the method embodiments shown, this device can be specifically applied to various electronic devices.

[0131] like Figure 4 As shown, the cement rotary kiln state identification device 300 described in this embodiment includes: a running acquisition module 301, a running reconstruction module 302, a calculation module 303, a fusion input module 304, a loss calculation module 305, a model training module 306, and a state identification module 307, wherein:

[0132] The operation acquisition module 301 is used to acquire the operating data of the cement rotary kiln.

[0133] The reconstruction module 302 is used to input the running data into the initial autoencoder to obtain the encoded data and the reconstructed data.

[0134] The calculation module 303 is used to calculate the reconstruction error based on the running data and the reconstruction data, and to generate fused data based on the reconstruction error and the encoded data.

[0135] The fusion input module 304 is used to input the fused data into the initial clustering model to obtain the clustering prediction results, and to associate the operating data with the kiln condition based on the reconstruction error of the clustering prediction results.

[0136] The loss calculation module 305 is used to calculate the joint loss based on the running data, reconstructed data, encoded data and clustering prediction results.

[0137] The model training module 306 is used to adjust the initial autoencoder and the initial clustering model according to the joint loss until the joint loss meets the training stopping condition, thereby obtaining the autoencoder and the clustering model, and constructing a state recognition model based on the autoencoder and the clustering model.

[0138] The state recognition module 307 is used to input the data to be recognized into the state recognition model to obtain the state recognition result.

[0139] In this embodiment, the operating data of a cement rotary kiln is acquired; the operating data is input into an initial autoencoder for feature extraction to obtain encoded data, and then reconstructed and fitted to obtain reconstructed data, thereby extracting the most informative features from the operating data; the reconstruction error is calculated based on the operating data and the reconstructed data, and fused data is generated based on the reconstruction error and the encoded data. The fused data is input into an initial clustering model, transforming state prediction into a clustering problem, and obtaining clustering prediction results. The clustering prediction results are a set of operating data under different kiln conditions. The operating data under different kiln conditions have different reconstruction errors, thus determining the kiln condition corresponding to the operating data, thereby learning the data distribution of operating data under each kiln condition; the joint loss is calculated based on the operating data, reconstructed data, encoded data, and clustering prediction results, thereby adjusting the initial autoencoder and the initial clustering model until the autoencoder and the clustering model are obtained, and the autoencoder and the clustering model are constructed into a state recognition model; the data to be identified is input into the state recognition model to obtain the state recognition result. This application does not require manual intervention. By performing data feature extraction and reasoning judgment through clustering, it can mine data patterns that are helpful for cluster analysis from top to bottom, improving the accuracy of cement rotary kiln state recognition.

[0140] In some optional implementations of this embodiment, the cement rotary kiln status identification device 300 may further include: an anomaly rejection module, a dimensionality reduction processing module, and a standard processing module, wherein:

[0141] The anomaly removal module is used to detect and remove abnormal samples in the running data using the standard deviation algorithm.

[0142] The dimensionality reduction module is used to calculate the correlation coefficients between various process features in the operational data after removing abnormal samples, so as to perform dimensionality reduction on the operational data after removing abnormal samples based on the correlation coefficients.

[0143] The standard processing module is used to standardize the runtime data after dimensionality reduction.

[0144] In this embodiment, the running data is subjected to anomaly detection, dimensionality reduction, and standardization, making the running data more standardized and ensuring the accuracy of subsequent running data processing.

[0145] In some optional implementations of this embodiment, the reconstructing module 302 may include: an encoding input submodule and a decoding input submodule, wherein:

[0146] The encoding input submodule is used to input the running data into the encoder in the initial autoencoder, so that the running data can be encoded by each convolutional layer and each bottleneck layer in the encoder to obtain encoded data.

[0147] The decoding input submodule is used to input the encoded data into the decoder in the initial autoencoder, so that the encoded data can be decoded through each deconvolution layer in the decoder to obtain the reconstructed data.

[0148] In this embodiment, an initial autoencoder is used for encoding and decoding to reconstruct the running data, obtaining encoded data and reconstructed data, thereby learning the most informative features in the running data and preparing data for kiln condition identification.

[0149] In some optional implementations of this embodiment, the initial autoencoder includes a first autoencoder and a second autoencoder, the initial clustering model includes a first clustering model and a second clustering model, the first autoencoder is connected to the first clustering model, and the second autoencoder is connected to the second clustering model; the fusion input module 304 may include: a first clustering submodule, a stationary labeling submodule, a second clustering submodule, and a sample association submodule, wherein:

[0150] The first clustering submodule is used to input the fused data output by the first autoencoder into the first clustering model during the first stage of training to obtain the first clustering prediction result; the first autoencoder is used to process the full amount of running data.

[0151] The stable labeling submodule is used to divide the running data into mass samples and niche samples based on the reconstruction error of the first clustering prediction result, and instruct the first autoencoder and the first clustering model to label the mass samples as kiln condition stable samples.

[0152] The second clustering submodule is used to input the fused data output by the second autoencoder into the second clustering model during the second stage of training to obtain the second clustering prediction result; the second autoencoder is used to process minority samples.

[0153] The sample association submodule is used to classify the minority samples into kiln condition good samples and kiln condition bad samples based on the reconstruction error of the second clustering prediction result, and instruct the second autoencoder and the second clustering model to associate the minority samples with kiln condition good samples or kiln condition bad samples.

[0154] In this embodiment, different types of operational data are separated by different clustering models, and the kiln condition corresponding to the separated operational data is determined based on the reconstruction error, so that the model can identify operational data under different kiln condition conditions.

[0155] In some optional implementations of this embodiment, the loss calculation module 305 may include: a reconstruction loss calculation submodule, a distribution calculation submodule, an energy calculation submodule, and a joint loss calculation submodule, wherein:

[0156] The reconstruction loss calculation submodule is used to calculate the reconstruction loss based on the running data, reconstruction data, and encoded data.

[0157] The distribution calculation submodule is used to calculate the sample distribution information of the running data based on the clustering prediction results.

[0158] The energy calculation submodule is used to calculate the sample energy of the running data based on the sample distribution information.

[0159] The joint loss calculation submodule is used to calculate the joint loss based on the reconstruction loss and sample energy.

[0160] In this embodiment, the reconstruction loss and sample energy are calculated separately, and the joint loss is calculated based on the reconstruction loss and sample energy. This comprehensively considers the losses of the initial autoencoder and the initial clustering model, thus improving the accuracy of the loss calculation.

[0161] In some optional implementations of this embodiment, the state recognition module 307 may include: a first fusion submodule, a first clustering submodule, a first generation submodule, a second fusion submodule, and a second clustering submodule, wherein:

[0162] The first fusion submodule is used to input the data to be identified into the first autoencoder in the state recognition model to obtain the fused data output by the first autoencoder.

[0163] The first clustering submodule is used to input the fused data into the first clustering model in the state recognition model to obtain the first clustering result.

[0164] The first generation submodule is used to generate a state recognition result based on the first clustering result when it is determined that the data to be identified belongs to the mass sample based on the first clustering result.

[0165] The second fusion submodule is used to input the fused data into the second autoencoder in the state recognition model when it is determined from the first clustering result that the data to be identified belongs to a minority sample, so as to obtain the fused data output by the second autoencoder.

[0166] The second clustering submodule is used to input the fused data into the second clustering model in the state recognition model to obtain the second clustering result, and generate the state recognition result based on the second clustering result.

[0167] In this embodiment, the data to be identified is input into the state recognition model. First, the data is reconstructed through an autoencoder to extract the most informative features. Then, the state is identified through the first clustering model and the second clustering model, which improves the accuracy of the state recognition of the cement rotary kiln.

[0168] In some optional implementations of this embodiment, the cement rotary kiln status identification device 300 may further include: an indication determination module, a control calculation module, and a control module, wherein:

[0169] The indicator determination module is used to determine the status indicator parameters based on the status recognition results.

[0170] The control calculation module is used to calculate the state control parameters based on the state indication parameters and the reconstruction error.

[0171] The control module is used to control the cement rotary kiln based on state control parameters.

[0172] In this embodiment, the state indication parameters are determined based on the state identification results, and then the state control parameters are calculated with the reconstruction error, so as to perform process control on the cement rotary kiln based on the state control parameters.

[0173] To address the aforementioned technical problems, embodiments of this application also provide a computer device. Please refer to [link / reference needed]. Figure 5 , Figure 5 This is a basic structural block diagram of the computer device in this embodiment.

[0174] The computer device 4 includes a memory 41, a processor 42, and a network interface 43 that are interconnected via a system bus. It should be noted that only the computer device 4 with components 41-43 is shown in the figure; however, it should be understood that it is not required to implement all the shown components, and more or fewer components can be implemented alternatively. Those skilled in the art will understand that the computer device described here is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0175] The computer device can be a desktop computer, laptop, handheld computer, or cloud server, etc. The computer device can interact with the user via a keyboard, mouse, remote control, touchpad, or voice control.

[0176] The memory 41 includes at least one type of readable storage medium, including flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 41 may be an internal storage unit of the computer device 4, such as the hard disk or memory of the computer device 4. In other embodiments, the memory 41 may also be an external storage device of the computer device 4, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the computer device 4. Of course, the memory 41 may include both the internal storage unit and its external storage device of the computer device 4. In this embodiment, the memory 41 is typically used to store the operating system and various application software installed on the computer device 4, such as computer-readable instructions for a cement rotary kiln status identification method. In addition, the memory 41 can also be used to temporarily store various types of data that have been output or will be output.

[0177] In some embodiments, the processor 42 may be a central processing unit (CPU), controller, microcontroller, microprocessor, or other data processing chip. The processor 42 is typically used to control the overall operation of the computer device 4. In this embodiment, the processor 42 is used to execute computer-readable instructions stored in the memory 41 or to process data, for example, to execute computer-readable instructions for the cement rotary kiln status identification method.

[0178] The network interface 43 may include a wireless network interface or a wired network interface, which is typically used to establish communication connections between the computer device 4 and other electronic devices.

[0179] The computer device provided in this embodiment can execute the above-described cement rotary kiln status identification method. Here, the cement rotary kiln status identification method can be any of the cement rotary kiln status identification methods described in the various embodiments above.

[0180] In this embodiment, the operating data of a cement rotary kiln is acquired; the operating data is input into an initial autoencoder for feature extraction to obtain encoded data, and then reconstructed and fitted to obtain reconstructed data, thereby extracting the most informative features from the operating data; the reconstruction error is calculated based on the operating data and the reconstructed data, and fused data is generated based on the reconstruction error and the encoded data. The fused data is input into an initial clustering model, transforming state prediction into a clustering problem, and obtaining clustering prediction results. The clustering prediction results are a set of operating data under different kiln conditions. The operating data under different kiln conditions have different reconstruction errors, thus determining the kiln condition corresponding to the operating data, thereby learning the data distribution of operating data under each kiln condition; the joint loss is calculated based on the operating data, reconstructed data, encoded data, and clustering prediction results, thereby adjusting the initial autoencoder and the initial clustering model until the autoencoder and the clustering model are obtained, and the autoencoder and the clustering model are constructed into a state recognition model; the data to be identified is input into the state recognition model to obtain the state recognition result. This application does not require manual intervention. By performing data feature extraction and reasoning judgment through clustering, it can mine data patterns that are helpful for cluster analysis from top to bottom, improving the accuracy of cement rotary kiln state recognition.

[0181] This application also provides another embodiment, namely, a computer-readable storage medium storing computer-readable instructions that can be executed by at least one processor to cause the at least one processor to perform the steps of the cement rotary kiln state identification method as described above.

[0182] In this embodiment, the operating data of a cement rotary kiln is acquired; the operating data is input into an initial autoencoder for feature extraction to obtain encoded data, and then reconstructed and fitted to obtain reconstructed data, thereby extracting the most informative features from the operating data; the reconstruction error is calculated based on the operating data and the reconstructed data, and fused data is generated based on the reconstruction error and the encoded data. The fused data is input into an initial clustering model, transforming state prediction into a clustering problem, and obtaining clustering prediction results. The clustering prediction results are a set of operating data under different kiln conditions. The operating data under different kiln conditions have different reconstruction errors, thus determining the kiln condition corresponding to the operating data, thereby learning the data distribution of operating data under each kiln condition; the joint loss is calculated based on the operating data, reconstructed data, encoded data, and clustering prediction results, thereby adjusting the initial autoencoder and the initial clustering model until the autoencoder and the clustering model are obtained, and the autoencoder and the clustering model are constructed into a state recognition model; the data to be identified is input into the state recognition model to obtain the state recognition result. This application does not require manual intervention. By performing data feature extraction and reasoning judgment through clustering, it can mine data patterns that are helpful for cluster analysis from top to bottom, improving the accuracy of cement rotary kiln state recognition.

[0183] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0184] Obviously, the embodiments described above are only some embodiments of this application, not all embodiments. The accompanying drawings show preferred embodiments of this application, but do not limit the patent scope of this application. This application can be implemented in many different forms; rather, the purpose of providing these embodiments is to provide a more thorough and comprehensive understanding of the disclosure of this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments, or make equivalent substitutions for some of the technical features. Any equivalent structures made using the content of this application's specification and drawings, directly or indirectly applied to other related technical fields, are similarly within the scope of patent protection of this application.

Claims

1. A method for recognizing a state of a cement rotary kiln, characterized by, The method comprises the following steps: obtaining operation data of a cement rotary kiln, wherein the operation data comprises kiln current, secondary air temperature, tertiary air temperature and outlet temperature of a preheater of the kiln; inputting the operation data into an initial autoencoder to obtain encoded data and reconstructed data; calculating reconstruction error according to the operation data and the reconstructed data, and generating fusion data according to the reconstruction error and the encoded data; inputting the fusion data into an initial clustering model to obtain clustering prediction results, and associating the operation data with kiln status based on reconstruction error of the clustering prediction results; calculating joint loss based on the operation data, the reconstructed data, the encoded data and the clustering prediction results; adjusting the initial autoencoder and the initial clustering model according to the joint loss until the joint loss meets a training stop condition, to obtain an autoencoder and a clustering model, and constructing a status recognition model based on the autoencoder and the clustering model; inputting to-be-recognized data into the status recognition model to obtain status recognition results; the initial autoencoder comprises a first autoencoder and a second autoencoder, and the initial clustering model comprises a first clustering model and a second clustering model, the first autoencoder is connected with the first clustering model, and the second autoencoder is connected with the second clustering model; the step of inputting the fusion data into the initial clustering model to obtain clustering prediction results and associating the operation data with kiln status based on reconstruction error of the clustering prediction results comprises: in the first stage of training, inputting fusion data output by the first autoencoder into the first clustering model to obtain first clustering prediction results; the first autoencoder is used for processing all of the operation data; dividing the operation data into popular samples and unpopular samples according to reconstruction error of the first clustering prediction results, and instructing the first autoencoder and the first clustering model to mark the popular samples as kiln status stable samples; in the second stage of training, inputting fusion data output by the second autoencoder into the second clustering model to obtain second clustering prediction results; the second autoencoder is used for processing the unpopular samples; dividing the unpopular samples into kiln status excellent samples and kiln status substandard samples according to reconstruction error of the second clustering prediction results, and instructing the second autoencoder and the second clustering model to associate the unpopular samples with the kiln status excellent samples or the kiln status substandard samples.

2. The cement rotary kiln condition recognition method according to claim 1, characterized in that, Before the step of inputting the operation data into the initial autoencoder to obtain encoded data and reconstructed data, the method further comprises: detecting abnormal samples in the operation data by a standard deviation algorithm to eliminate the abnormal samples in the operation data; calculating correlation coefficients between various process characteristics in the operation data after the abnormal samples are eliminated, to perform dimension reduction processing on the operation data after the abnormal samples are eliminated according to the correlation coefficients; performing standardization processing on the operation data after the dimension reduction processing.

3. The cement rotary kiln condition identification method according to claim 1, characterized in that, The step of inputting the operation data into the initial autoencoder to obtain encoded data and reconstructed data comprises: inputting the running data into an encoder in an initial auto-encoder to perform encoding processing on the running data through each convolutional layer and each bottleneck layer in the encoder to obtain encoded data; inputting the encoded data into a decoder in the initial auto-encoder to perform decoding processing on the encoded data through each de-convolutional layer in the decoder to obtain reconstructed data.

4. The cement rotary kiln condition identification method according to claim 1, characterized in that, The step of calculating a joint loss based on the running data, the reconstructed data, the encoded data and the clustering prediction result comprises: calculating a reconstruction loss according to the running data, the reconstructed data and the encoded data; calculating sample distribution information of the running data based on the clustering prediction result; calculating sample energy of the running data according to the sample distribution information; calculating a joint loss according to the reconstruction loss and the sample energy.

5. The cement rotary kiln condition identification method according to any one of claims 1 to 4, characterized in that, The step of inputting the to-be-identified data into the state identification model to obtain a state identification result comprises: inputting the to-be-identified data into a first auto-encoder in the state identification model to obtain fusion data output by the first auto-encoder; inputting the fusion data into a first clustering model in the state identification model to obtain a first clustering result; when it is determined according to the first clustering result that the to-be-identified data belongs to the mass sample, generating a state identification result according to the first clustering result; when it is determined according to the first clustering result that the to-be-identified data belongs to the small sample, inputting the to-be-identified data into a second auto-encoder in the state identification model to obtain fusion data output by the second auto-encoder; inputting the fusion data into a second clustering model in the state identification model to obtain a second clustering result, and generating a state identification result according to the second clustering result.

6. The cement rotary kiln condition identification method according to claim 5, characterized in that, After the step of inputting the to-be-identified data into the state identification model to obtain a state identification result, the method further comprises: determining a state indication parameter according to the state identification result; calculating a state control parameter according to the state indication parameter and the reconstruction error; controlling the cement rotary kiln based on the state control parameter.

7. A cement rotary kiln condition recognition device, characterized by, The cement rotary kiln state identification device implements the cement rotary kiln state identification method according to any one of claims 1 to 6, and comprises: an operation acquisition module configured to acquire running data of a cement rotary kiln, wherein the running data comprises kiln current, secondary and tertiary air temperature and outlet temperature of a decomposition furnace of a kiln condition measuring point; an operation reconstruction module configured to input the running data into an initial auto-encoder to obtain encoded data and reconstructed data; a calculation module configured to calculate a reconstruction error according to the running data and the reconstructed data, and generate fusion data according to the reconstruction error and the encoded data; a fusion input module configured to input the fusion data into an initial clustering model to obtain a clustering prediction result, and associate the running data with a kiln condition state based on a reconstruction error of the clustering prediction result; a loss calculation module configured to calculate a joint loss based on the running data, the reconstructed data, the encoded data and the clustering prediction result; a loss calculation module configured to calculate a joint loss based on the running data, the reconstructed data, the encoded data and the clustering prediction result; a model training module, configured to adjust the initial autoencoder and the initial clustering model according to the joint loss until the joint loss meets a training stop condition, to obtain an autoencoder and a clustering model, and to construct a state recognition model based on the autoencoder and the clustering model; a state recognition module, configured to input to-be-recognized data into the state recognition model to obtain a state recognition result. 8.A computer device, comprising a memory and a processor, wherein the memory stores computer readable instructions, and the processor, when executing the computer readable instructions, implements steps of the cement rotary kiln state recognition method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer readable instructions, and the computer readable instructions, when executed by a processor, implement steps of the cement rotary kiln state recognition method according to any one of claims 1 to 6.

Images