Thermal power plant safety monitoring and early warning management method and system based on artificial intelligence

Abstract

The invention relates to a thermal power plant safety monitoring and early warning management method and system based on artificial intelligence. The method comprises the following steps: acquiring vibration data of equipment of a thermal power plant at each moment within a preset time interval; dividing the vibration data into a plurality of clusters based on a clustering algorithm; generating vibration data features based on the plurality of clusters; inputting the vibration data features into a trained artificial intelligence model to obtain a fault prediction result; and generating a corresponding maintenance strategy based on the fault prediction result, the maintenance strategy being used for indicating to maintain the equipment. According to the method, the equipment vibration data is analyzed and subjected to feature extraction through a clustering algorithm, complex modes and internal structures in the data can be better captured, more accurate information is provided for subsequent model training and fault prediction, and therefore the accuracy and reliability of safety monitoring and early warning are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field related to artificial intelligence, and in particular relates to a safety monitoring and early warning management method and system for a thermal power plant based on artificial intelligence. Background Art

[0002] In the operation of thermal power plants, the safe and stable operation of equipment is of vital importance. Thermal power plant equipment such as steam turbines, generators, fans, coal mills, etc., operate for a long time under complex and harsh working conditions such as high temperature, high pressure, and high speed, and are prone to various faults. If these faults cannot be monitored and warned in time, they may cause equipment damage, shutdown, and even cause serious safety accidents, bringing huge economic losses and safety hazards to power production.

[0003] Traditional thermal power plant equipment safety monitoring methods are mainly based on single parameter threshold judgments or simple rule systems. For example, by setting a fixed threshold for vibration amplitude, an alarm is issued when the vibration value exceeds the threshold; or the equipment is inspected based on the equipment operation time and the maintenance cycle set by experience. However, this method has many limitations. On the one hand, single parameter threshold judgments are often too sensitive or not sensitive enough, because the state of the equipment in actual operation is complex and changeable, and the parameters may also fluctuate within a certain range due to changes in operating conditions (such as load fluctuations, ambient temperature changes, etc.) during normal operation, which can easily lead to false alarms or missed alarms. On the other hand, the inspection method based on a fixed maintenance cycle cannot capture early signs of sudden equipment failures in real time, and cannot meet the requirements of modern thermal power plants for refined equipment management and high-reliability operation.

[0004] With the development of technology, some monitoring methods based on data analysis and machine learning have been gradually applied to the field of equipment fault diagnosis. These methods use equipment operation data to build models and try to mine patterns in the data to predict faults.

[0005] When processing these complex data, current machine learning models often fail to effectively extract and utilize the key information contained therein. Especially in the feature extraction stage, most methods simply use some common statistical features (such as mean, variance, etc.) or directly use the original data as model input without deeply exploring the inherent structure and complex patterns in the data, resulting in insufficient accuracy and robustness of the model when facing complex thermal power plant equipment fault diagnosis problems. Therefore, there is an urgent need for a technical solution that can more effectively process the complex operating data of thermal power plant equipment and accurately extract features to improve fault monitoring and early warning capabilities. Summary of the invention

[0006] Based on this, it is necessary to provide an artificial intelligence-based thermal power plant safety monitoring and early warning management method, system, computer equipment and storage medium that can accurately extract features to improve fault monitoring and early warning capabilities in response to the above technical problems.

[0007] In the first aspect, the present application provides a safety monitoring and early warning management method for a thermal power plant based on artificial intelligence, comprising:

[0008] Obtain vibration data of equipment in thermal power plants at each time within a preset time interval;

[0009] Based on the clustering algorithm, the vibration data is divided into multiple clusters;

[0010] Generate vibration data features based on multiple clusters;

[0011] Input the vibration data features into the trained artificial intelligence model to obtain the fault prediction results;

[0012] Based on the fault prediction results, a corresponding maintenance strategy is generated, and the maintenance strategy is used to instruct the maintenance of the equipment.

[0013] In the second aspect, the present application also provides a thermal power plant safety monitoring and early warning management system based on artificial intelligence, including:

[0014] A vibration data acquisition module is used to acquire vibration data of the equipment in the thermal power plant at each time within a preset time interval;

[0015] A data clustering module, used to divide the vibration data into multiple clusters based on a clustering algorithm;

[0016] A vibration data feature generation module, used for generating vibration data features based on multiple clusters;

[0017] The fault prediction module is used to input the vibration data features into the trained artificial intelligence model to obtain the fault prediction results;

[0018] The management module is used to generate a corresponding maintenance strategy based on the fault prediction result, and the maintenance strategy is used to instruct the maintenance of the equipment.

[0019] In a third aspect, the present application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, it implements an artificial intelligence-based thermal power plant safety monitoring and early warning management method as described in the first aspect.

[0020] In a fourth aspect, the present application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a thermal power plant safety monitoring and early warning management method based on artificial intelligence as in the first aspect.

[0021] The above-mentioned artificial intelligence-based thermal power plant safety monitoring and early warning management method, system, computer equipment and storage medium, which analyzes and extracts features from the equipment's vibration data through a clustering algorithm, can better capture the complex patterns and internal structures in the data, and provide more accurate information for subsequent model training and fault prediction, thereby improving the accuracy and reliability of safety monitoring and early warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related technologies, the drawings required for use in the embodiments or the related technical descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0023] Figure 1 A schematic diagram of a process flow of a thermal power plant safety monitoring and early warning management method based on artificial intelligence provided by the present invention;

[0024] Figure 2 A schematic structural diagram of an artificial intelligence-based thermal power plant safety monitoring and early warning management system provided by the present invention. DETAILED DESCRIPTION

[0025] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0026] refer to Figure 1 , which presents a flow chart of a thermal power plant safety monitoring and early warning management method based on artificial intelligence provided by the present application, the method comprising the following steps:

[0027] S101. Obtain vibration data of equipment in a thermal power plant at each moment within a preset time interval.

[0028] Specifically, the thermal power plant equipment that needs to be monitored may include but is not limited to key mechanical devices such as steam turbines, generators, various pumps, fans, etc. The setting of the preset time interval can comprehensively consider the equipment operation characteristics, data collection costs, and the requirements for timely fault detection. For example, it can be set to collect the equipment's operation data within 10 minutes every 5 minutes to ensure that the vibration of the equipment under different operating conditions can be captured. By installing high-precision vibration sensors at key parts of the equipment, such as bearing seats and casings, the sensors can convert the vibration signals monitored in real time into digital signals, which are vibration data. The data acquisition system is responsible for collecting the vibration data from each sensor at each moment within the preset time interval, and performing preliminary sorting and storage.

[0029] S102: Divide the vibration data into multiple clusters based on a clustering algorithm.

[0030] Specifically, the clustering algorithms that can be used in this step include K-Means algorithm, hierarchical clustering algorithm, etc. Through the clustering process, the vibration data is divided into multiple clusters with similar characteristics, and each cluster can represent a specific operating state or state range of the equipment.

[0031] S103: Generate vibration data features based on multiple clusters.

[0032] Specifically, for each cluster, a variety of features that can reflect the operating status of the equipment are extracted. For example, the statistical features such as the mean, variance, maximum value, and minimum value of the vibration data within the cluster can be calculated. These features can describe the average level, degree of fluctuation, and extreme conditions of the vibration under the operating status of the equipment represented by the cluster. At the same time, the frequency characteristics of the vibration data can also be analyzed, such as obtaining the amplitude distribution of the vibration signal at different frequencies through fast Fourier transform (FFT), and extracting the main frequency components and their corresponding amplitudes and other features. In addition, considering the timing of the equipment's operating status, some timing features, such as the autocorrelation function value, can also be calculated to reflect the correlation of the vibration data in the time series. These features extracted from each cluster are combined and sorted to form a vector or matrix that can comprehensively characterize the characteristics of the equipment's vibration data.

[0033] S104: Input the vibration data features into the trained artificial intelligence model to obtain a fault prediction result.

[0034] Specifically, the trained artificial intelligence model can be a neural network model, a support vector machine model, etc. Taking the neural network model as an example, it is usually composed of an input layer, multiple hidden layers and an output layer. The vibration data features generated in S103 are input to the input layer of the neural network model, and the number of neurons in the input layer corresponds to the dimension of the vibration data features. Then, the data undergoes complex nonlinear transformation and feature learning in the hidden layer. The structure and number of neurons in the hidden layer can be designed and adjusted according to the actual situation. The weights and biases of the neural network are continuously optimized through a large amount of training data and back propagation algorithms, so that it can accurately learn the mapping relationship between the vibration data features and the equipment failure. Finally, the fault prediction result is obtained in the output layer. The number of neurons and the representation method of the output layer depend on the type and number of faults to be predicted. For example, if it is to predict whether the equipment fails (binary classification problem), the output layer can be a neuron, and the output 0 indicates normal and 1 indicates fault; if it is to predict multiple specific fault types (multi-classification problem), the output layer can have multiple neurons, each neuron corresponds to the probability or category identification of a fault type, and the possible fault conditions of the equipment are determined by analyzing and judging the output results.

[0035] S105. Generate a corresponding maintenance strategy based on the fault prediction result, where the maintenance strategy is used to instruct maintenance of the equipment.

[0036] Specifically, the corresponding maintenance strategy is formulated according to the different situations of the fault prediction results. If the fault prediction results show that the equipment is in normal operation, then the maintenance strategy can be just to continue routine monitoring and maintenance, such as regularly checking the working status of the sensor, cleaning the dust on the surface of the equipment, etc. If a minor fault is predicted in the equipment, the maintenance strategy can include local inspection and adjustment of the equipment, such as lubricating a bearing, tightening loose connectors, etc., and shortening the time interval for the next monitoring to track the development of the fault in time. When a serious fault is predicted in the equipment, the maintenance strategy needs to formulate a detailed downtime maintenance plan, which can include organizing professional maintenance personnel, preparing maintenance tools and replacing parts, etc., determining the time node and maintenance process for maintenance, and ensuring that the equipment can be repaired as soon as possible and restored to normal operation.

[0037] The above-mentioned artificial intelligence-based thermal power plant safety monitoring and early warning management method analyzes and extracts features from equipment vibration data through a clustering algorithm, which can better capture the complex patterns and internal structures in the data, and provide more accurate information for subsequent model training and fault prediction, thereby improving the accuracy and reliability of safety monitoring and early warning.

[0038] In an optional embodiment, obtaining vibration data of equipment in a thermal power plant at each time within a preset time interval includes the following steps:

[0039] Obtain the vibration amplitude value A of the equipment in the thermal power plant at each moment within the preset time interval i and the vibration frequency f i ; Wherein, i represents the sequence number of the moment, i=1, 2, ..., n, and n represents the total number of moments within the preset time interval;

[0040] The vibration amplitude value A i and the vibration frequency f i Composition data set P i =(A i , f i ), and the data set P i As vibration data; wherein each moment within the preset time interval corresponds to a data group P i =(A i , f i ).

[0041] Specifically, it is assumed that there are n moments in the preset time interval, i = 1, 2, ..., n, representing the sequence number of the moment. At each moment i, the data acquisition system obtains the vibration amplitude value A from the sensor installed on the equipment i and the vibration frequency f i . Vibration amplitude value A i It reflects the strength of the equipment vibration, and the vibration frequency f i It reflects the fast and slow rhythm of vibration. For example, for a normally operating steam turbine, its vibration amplitude may fluctuate within a small range, and the vibration frequency is relatively stable; when the equipment fails, the vibration amplitude may suddenly increase, and the vibration frequency may also change.

[0042] For the vibration amplitude value A collected at each moment within the preset time interval i and the vibration frequency f i , combining them into a data set P i =(A i , f i ). The purpose of this is to associate the two key vibration parameters at the same time so as to more comprehensively describe the vibration state of the equipment at that time. For example, at a certain moment, the vibration amplitude value is 0.5mm and the vibration frequency is 50Hz, then the data set P i =(0.5, 50).

[0043] Each moment corresponds to such a data set P i, these data groups constitute the vibration data sequence of the device within the preset time interval. These data can be stored in the form of arrays, matrices or database records. For example, the vibration data over a period of time can be stored in a two-dimensional array, where each row represents a data group at a moment, the first column stores the vibration amplitude value, and the second column stores the vibration frequency value. Such a data structure facilitates data access, query and analysis. In this way, the vibration data of the device at each moment within the preset time interval can be fully obtained.

[0044] In an optional embodiment, the clustering algorithm is a K-Means algorithm; based on the clustering algorithm, the vibration data is divided into a plurality of clusters, including the following steps:

[0045] Step 1: Determine the number of clusters K by the elbow method; wherein the number of clusters K is used to divide the vibration data at n moments within a preset time interval into K clusters, K = 1, 2, ..., n; among the K clusters, the set of vibration data contained in the kth cluster is C k , k=1,2,…,K.

[0046] Step 2: From n data sets P i Randomly select K data sets P from i , as cluster centers; where cluster centers are E1, E2, …, E K , the cluster center of the kth cluster is

[0047] Step 3: Group each data set P i =(A i , f i ) is standardized to According to Z i Calculate each data set P i To each cluster center E k The distance d(Z i , E k ), the calculation formula is:

[0048] Step 4: Group each data set P i Assign to the distance d(Z i , E k ) The nearest cluster center E k The cluster C k ;in,

[0049] Step 5: For each cluster C k , recalculate and update its cluster center in, n kis cluster C k The data set P contained in i The number of

[0050] Step 6: Repeat steps 3 to 5 until all cluster centers are Get K different clusters C k ;in, is the updated cluster center, is the cluster center before updating, ∈ is the preset threshold.

[0051] Specifically, for step one:

[0052] In the K-Means algorithm of this step, the number of clusters K determines how many different categories or groups the vibration data of n moments within a preset time interval are divided into. Here, the value range of K is from 1 to n, covering all possible situations from classifying all data into one category to each data being a category in itself. For example, if K = 1, it means that the vibration data of all moments are regarded as a whole without subdivision; and if K = n, it is equivalent to the vibration data of each moment becoming a cluster separately, which usually has no practical significance. In practical applications, the value of K is usually a suitable value between 2 and n.

[0053] The elbow method is a common method for determining the K value. The basic idea is to determine the appropriate K by calculating the clustering error under different K values ​​and then observing the trend of the error changing with the K value. Specifically, for each K value (starting from 1 and gradually increasing), run the K-Means algorithm to calculate the sum of the distances from the data points to the cluster centers to which they belong. This sum can be used as a measure of the clustering error. When K is small, the clustering error is large because each cluster contains more data points and the data points are more dispersed relative to the cluster center; as K increases, the clustering error will gradually decrease because each cluster contains fewer data points and the data points are closer to the cluster center. However, when K increases to a certain extent, the reduction in clustering error will gradually decrease. The K value at this time corresponds to the elbow point, which is considered to be a more appropriate number of clusters. For example, we calculated the clustering error from K = 2 to K = 10 and found that when K = 3, the error curve had an obvious turning point, similar to the shape of an elbow, so K = 3 may be a more appropriate clustering number, which means that dividing the vibration data into 3 clusters can better reflect the different operating conditions of the equipment.

[0054] For step 2:

[0055] At the beginning of the K-Means algorithm, K data groups need to be randomly selected as the initial clustering centers. i =(A i , fi ) is randomly selected from the vibration data of n moments previously obtained. For example, if n = 100 (i.e., there are 100 moments of vibration data), and K = 3 is determined, then 3 are randomly selected from these 100 data groups as the initial cluster centers. For example, the selected data group is P 20 =(A 20 , f 20 ), P 50 =(A 50 , f 50 ) and P 20 =(A 80 , f 80 ).

[0056] The selected K cluster centers are denoted as E1, E2, ..., E K , where the kth cluster center here represents the value of the kth cluster center in the vibration amplitude dimension, Represents the value in the vibration frequency dimension. Initially, they are equal to the vibration amplitude value A in the corresponding data group randomly selected i and the vibration frequency value f i For example, for the randomly selected P 20 =(A 20 , f 20 ) as the first cluster center E1, then These initial cluster centers will serve as reference points for subsequent data assignment, and data groups will be assigned to corresponding clusters based on their distances from these cluster centers.

[0057] For step three:

[0058] Data standardization is to make data of different dimensions (vibration amplitude and vibration frequency) comparable and avoid the influence of different dimensions on clustering results. There are many common data standardization methods. Here we use a simple standardization method, such as subtracting the mean of each dimension from the data of that dimension and then dividing it by the standard deviation of that dimension. i , calculate its mean and standard deviation The standardized vibration amplitude value is Similarly, for the vibration frequency value f i , calculate its mean and standard deviation Normalized vibration frequency value The standardized data set is obtained

[0059] Using the Euclidean distance formula To calculate the data set Z i To cluster center E k This distance measures the difference between the data set and the cluster center in terms of vibration amplitude and vibration frequency. For example, if a data set is standardized to Z i =(0.5, 0.3), a cluster center E k =(0.4, 0.2), then according to the distance formula, we can get The smaller the distance, the closer the data group is to the cluster center, and the more likely it is to be assigned to the cluster to which the cluster center belongs.

[0060] For step four:

[0061] For each data set P i (Normalized to Z i ), calculate it to K cluster centers E1, E2, ..., E K The distance d(Z i , E1), d(Z i , E2),…,d(Z i , E k Then, the data set P i Assign to the cluster C to which the cluster center with the smallest distance belongs k That is, find The k value will be i Assigned to C k For example, if the data set Z is calculated i The distance to E1 is 0.5, the distance to E2 is 0.3, and the distance to E3 is 0.4, so k = 2. Since 0.3 is the minimum distance, P i Assigned to cluster C2.

[0062] After this step, all n data sets P i are assigned to one of the K clusters. Each cluster C k Contains several clusters that are considered to be related to the cluster center E k The data group with the closest distance. In this way, K clusters are initially formed, but the cluster centers at this time may not be optimal, because the initial cluster centers are randomly selected, so the cluster centers need to be further adjusted.

[0063] For step five:

[0064] In order to make the clustering more reasonable, it is necessary to recalculate the cluster center based on the existing data groups in each cluster. k , whose cluster centers The update formula is and where n k is cluster C k The data set P contained in i The meaning of this formula is that the value of the new cluster center in the vibration amplitude dimension is the average value of the vibration amplitude values ​​of all data groups in the cluster, and the value in the vibration frequency dimension is the average value of the vibration frequency values ​​of all data groups in the cluster. For example, if there are 10 data groups in cluster C2, their vibration amplitude values ​​are A 30 , A 35 , …, A 44 ,but Similarly, a similar calculation is performed for the vibration frequency value. Thus updating the cluster center

[0065] By updating the cluster center, it can better represent the average characteristics of the data group within the cluster. In this way, in the subsequent data allocation process, the data group can be more accurately allocated to the appropriate cluster. With the continuous update of the cluster center, the clustering results will gradually stabilize, making the data group within each cluster more compact and the differences between different clusters more obvious, thereby better reflecting the vibration data characteristics of the equipment under different operating conditions.

[0066] For step six:

[0067] for The Euclidean norm (L2 norm) can be used to calculate The calculation formula is:

[0068]

[0069] in, is the updated cluster center, is the cluster center before the update, and ∈ is the threshold. This threshold ∈ determines when the algorithm stops iterating. When the change of the cluster center is less than this threshold, it is considered that the cluster has reached a stable state and the algorithm stops iterating. By continuously repeating the iterative process, the cluster center and the division of clusters will be continuously optimized, and finally a more ideal clustering result will be obtained.

[0070] In an optional embodiment, generating vibration data features based on multiple clusters includes the following steps:

[0071] Calculate each cluster C k The data set P i The ratio of the number of data sets to the total number of data sets k , we get the cluster number ratio feature vector R = (r1, r2, ..., r k ,…,rK );in,

[0072] Calculate each cluster C k The average vibration amplitude and the average vibration frequency Get the cluster average vibration amplitude feature vector and cluster average vibration frequency eigenvector in, and

[0073] Calculate each cluster C k The data set P i To the cluster center E to which it belongs k The average distance Get the average distance feature vector in,

[0074] The cluster number ratio feature vector R, the cluster average vibration amplitude feature vector A, the cluster average vibration frequency feature vector F and the average distance feature vector D are combined to form a combined feature vector And the combined feature vector X is used as the vibration data feature.

[0075] Specifically, for each cluster C k , calculate the number of data groups n it contains k Ratio to the total number of data sets n This ratio r k It can reflect the relative size of each cluster in the entire data set. For example, if there are 100 data groups in total (n=100), and there are 30 data groups in cluster C1 (n1=30), then By calculating the proportion of the number of each cluster, the cluster quantity proportion feature vector R can provide a macro information about the distribution of data in each cluster. Different proportions can imply the time proportion or frequency of occurrence of the equipment in different operating states, which is helpful for the subsequent analysis of the overall operating status of the equipment and fault prediction.

[0076] The calculated proportion r of each cluster k Arrange them in sequence to form the cluster quantity ratio feature vector R = (r1, r2, ..., r K ). This vector, as part of the overall vibration data feature, provides the distribution feature information of the data at the cluster level for subsequent analysis. For example, if there are 3 clusters (K = 3), and r1 = 0.3, r2 = 0.4, r3 = 0.3 are calculated, then R = (0.3, 0.4, 0.3).

[0077] For each cluster C k , calculate the average value of the vibration amplitude of all data groups within it The specific calculation formula is: This means that cluster C k The vibration amplitude value A of each data group i Add them together and divide by the number of data groups in the cluster n k For example, there are 5 data groups in cluster C2, and their vibration amplitude values ​​are 0.2, 0.3, 0.25, 0.35, and 0.28 respectively. Each element in the average vibration amplitude eigenvector A It can reflect the average level of vibration amplitude in the corresponding cluster. The difference in average vibration amplitude of different clusters may be related to the vibration intensity characteristics of the equipment under different operating conditions, which can be used for subsequent analysis of whether the operating status of the equipment is normal and the possible fault types.

[0078] Similarly, calculate each cluster C k The average vibration frequency The formula is It is cluster C k The average value of the vibration frequency values ​​of all data groups in the cluster. For example, for the above cluster C2, if its vibration frequency values ​​are 45, 48, 46, 47, and 46.5 respectively, then Elements in the cluster average vibration frequency eigenvector F It provides average characteristic information about the vibration frequency of each cluster. The change of vibration frequency is often related to factors such as the operating speed of the equipment and the resonance of components. By analyzing these average vibration frequency characteristics, we can further understand the operating status and potential failures of the equipment.

[0079] For each cluster C k , calculate all the internal data groups P i To the cluster center E k The distance d(Z i , E k ). First, according to the distance formula defined previously Calculate the distance from each data group to the cluster center (here it is assumed that the data group has been standardized to The cluster center is These distances are then added together and divided by the number of data groups in the cluster, n. k , and get the average distance For example, there are 8 data groups in cluster C3, and their distances to the cluster center are 0.1, 0.12, 0.08, 0.15, 0.09, 0.11, 0.13, and 0.1 respectively.

[0080] The average distance between the elements in the feature vector D It reflects the compactness or discreteness of the data groups within each cluster. A smaller average distance means that the data groups within the cluster are relatively tightly around the cluster center, indicating that the operating state of the equipment represented by the cluster is relatively stable and consistent; while a larger average distance may indicate that the data groups within the cluster are more dispersed, and the vibration characteristics of the equipment under this operating state vary greatly, and there may be some potential unstable factors or abnormal conditions. By analyzing the average distance feature vector, we can further understand the changes in the operating state of the equipment from the perspective of the compactness of the data distribution, providing more basis for fault prediction.

[0081] By combining the four eigenvectors obtained by the above calculation in a specific order, a combined eigenvector X containing rich information is formed. This combined eigenvector integrates information on the number distribution of clusters, the average level of vibration amplitude, the average level of vibration frequency, and the compactness of internal data groups. For example, if there are 3 clusters (K=3), the combined eigenvector X can be X=(0.3, 0.4, 0.3, 0.25, 0.32, 0.28, 45.5, 47.2, 46.8, 0.08, 0.12, 0.1). This combination method can comprehensively describe the characteristics of vibration data at the cluster level, providing sufficient information basis for subsequent input into the artificial intelligence model for fault prediction, so that the model can comprehensively consider multiple factors to accurately judge the operating status of the equipment and predict possible faults.

[0082] The combined feature vector X, as the final vibration data feature, will be input into the trained artificial intelligence model. In the model, these features will be associated and learned with the failure mode of the equipment. The model predicts equipment failure by analyzing the complex relationship between these features and their potential connection with the failure. For example, the model may learn that when there is a certain change in the proportion of cluster numbers, the average vibration amplitude exceeds a certain threshold, the average vibration frequency is within a certain range, and the average distance fluctuates abnormally, the equipment may have a certain type of failure. In this way, based on these features, the model can make accurate predictions about the future operation of the equipment.

[0083] In an optional embodiment, the training method of the artificial intelligence model includes the following steps:

[0084] Based on the historical data of the equipment in the thermal power plant, a training set is obtained;

[0085] Select the MLP model as the neural network model to build the artificial intelligence model;

[0086] Based on the back propagation algorithm, the MLP model is trained using the training set to obtain a trained artificial intelligence model.

[0087] Specifically, a large amount of data is first collected from the equipment operation history records of the thermal power plant. These historical data packets may include various operating parameters of the equipment in different time periods in the past and the corresponding operating status information, such as normal operation of the equipment, occurrence of different types of faults, etc. In addition to the vibration data mentioned above (vibration amplitude value and vibration frequency), data of other related parameters such as temperature, pressure, flow, etc. may also be involved. The collected data needs to be sorted and preprocessed to ensure the accuracy and completeness of the data. For example, check whether there are missing values ​​or outliers in the data, correct or delete the outliers, and use mean filling, interpolation and other methods to process the missing values.

[0088] The historical data can be labeled according to the actual operating status of the equipment and divided into different categories. For example, the data when the equipment is operating normally is labeled as the "normal" category, and the data when the equipment has a specific fault (such as bearing wear, blade damage, etc.) is labeled as the corresponding fault category. These labeled historical data will serve as the basis for training artificial intelligence models. Through such data classification, the model can learn the characteristic patterns of data under different operating conditions so that it can accurately identify the status of the equipment in subsequent predictions. Then, a part of these labeled data is extracted as a training set. The size and composition of the training set need to be reasonably determined according to the actual situation. Generally speaking, it is necessary to ensure that the training set can cover all possible operating conditions and fault conditions of the equipment to improve the generalization ability of the model.

[0089] The MLP (Multi-layer Perceptron) model is a commonly used neural network structure that is very suitable for processing complex nonlinear relationship data such as thermal power plant equipment monitoring data. It consists of multiple neurons arranged in layers, usually including an input layer, one or more hidden layers, and an output layer. The number of neurons in the input layer depends on the number of features of the input data. For example, if the input vibration data feature vector X contains m elements (such as the proportion of cluster numbers, average vibration amplitude, and other features), then the input layer has m neurons to receive these data features. The number of hidden layers and the number of neurons in each hidden layer can be adjusted according to the complexity of the data and the needs of the problem. The neurons in the hidden layer are connected to the neurons in the adjacent layers through weights and biases, and different activation functions (such as Sigmoid, ReLU, etc.) can be used to introduce nonlinear characteristics, so that the model can learn the complex nonlinear relationships in the data. The number of neurons in the output layer depends on the type of result to be predicted. For example, if the prediction is whether a device fails (a binary classification problem), the output layer can have 1 neuron, which outputs 0 or 1 to indicate normal or faulty. If the prediction is for multiple specific fault types (a multi-classification problem), the output layer can have the same number of neurons as the fault types, and each neuron outputs the probability or category identifier of the corresponding fault type.

[0090] After determining to select the MLP model, the structure of the MLP model is constructed according to the input data features and prediction targets analyzed above. For example, assuming that the input vibration data feature vector X has 12 features (such as when K=3, the cluster quantity ratio feature vector has 3 elements, the cluster average vibration amplitude feature vector has 3 elements, the cluster average vibration frequency feature vector has 3 elements, and the average distance feature vector has 3 elements), the input layer is set to 12 neurons. Select 2 hidden layers, the first hidden layer is set to 20 neurons, using the ReLU activation function, and the second hidden layer is set to 10 neurons, also using the ReLU activation function. If it is to predict whether the equipment is faulty (binary classification problem), the output layer is set to 1 neuron, and the Sigmoid activation function is used to convert the output into a probability value between 0 and 1, indicating the probability of equipment failure. In this way, an MLP model structure suitable for equipment failure prediction in thermal power plants is constructed, ready for subsequent training.

[0091] The back propagation algorithm is a key algorithm for training neural networks. It is based on the idea of ​​gradient descent. It calculates the gradient of the loss function (such as the cross entropy loss function for classification problems) for each parameter (weight and bias) in the model, and then updates the parameters according to the direction and size of the gradient to gradually reduce the value of the loss function, so that the prediction result of the model is closer to the true value. During the training process, the input data in the training set (i.e., the historical vibration data characteristics of the equipment, etc.) is first input into the MLP model, and the predicted output of the model is obtained after the forward propagation calculation of the model. Then, the predicted output is compared with the actual annotation result (the actual operating status of the equipment) to calculate the value of the loss function. Then, through the back propagation algorithm, starting from the output layer, the gradient of the loss function for the weight and bias of each hidden layer and input layer neuron is calculated in turn. These gradients reflect the degree of influence of each parameter on the loss function. The parameters are updated according to the opposite direction of the gradient so that the model can reduce the error in the next prediction.

[0092] When training the MLP model using the training set, multiple iterations are usually performed. In each iteration, all data samples in the training set are traversed, and the above-mentioned forward propagation, loss calculation, and back propagation parameter update process are performed on each sample. In order to improve the training effect and prevent overfitting, some optimization techniques can also be used, such as stochastic gradient descent (SGD) and its variants (such as Adagrad, Adadelta, Adam, etc.) to adjust the learning rate and control the step size of parameter updates; regularization methods (such as L1 regularization, L2 regularization) are used to constrain the complexity of the model to avoid overfitting caused by excessive complexity of the model, so that the model can be better generalized to unseen data. Through continuous iterative training until certain stopping conditions are met, such as reaching a preset number of iterations, the value of the loss function converges to a stable minimum value, or the accuracy on the validation set no longer improves, the MLP model obtained at this time is a trained artificial intelligence model, which can accurately predict the failure of the equipment based on the input equipment vibration data characteristics and other information.

[0093] In an optional embodiment, obtaining a training set based on historical data of equipment in a thermal power plant includes the following steps:

[0094] Based on the historical data of the equipment in the thermal power plant, the historical vibration data of the equipment at m preset time intervals is obtained. and historical fault records; where m = 1, 2, ..., M, M is the number of samplings, m is the sample number, is the vibration amplitude value of the mth sample at the i-th moment, f i m is the historical vibration frequency of the mth sample at the i-th moment;

[0095] Based on historical vibration data According to the above method, the combined feature vector of each sample is obtained And the combined feature vector X m As the input of model training; where X m Represents the combined feature vector of the mth sample;

[0096] Determine each combined feature vector X based on historical fault records m The corresponding fault type is encoded using one-hot encoding as Y m As the output label; where t1 to t B Characterize different fault types, Or 1, b = 1, 2, ..., B, if This means that the fault type of the mth sample is t b The type of fault characterized;

[0097] Based on the input X m and output label Y m , and obtain the training set of the artificial intelligence model {(X1, Y1), (X2, Y2), …, (X M , Y M )}.

[0098] Specifically, thermal power plants usually record the operation data of equipment for a long time, and can obtain the operation information of equipment within m preset time intervals from its historical database. For each preset time interval (for example, every hour, day, or week, etc., set according to actual conditions), collect the vibration amplitude value of the equipment at each moment and the vibration frequency f i m , forming historical vibration data These data can be stored in different tables or files in the database. At the same time, the corresponding historical fault records are obtained, which record in detail whether the equipment has failed in each time period and the specific type of fault. For example, the historical fault records may indicate that a bearing of the steam turbine has worn out or the generator has overheated in a certain time period.

[0099] For m preset time intervals (m=1, 2, ..., M, where M is the total number of sampling times), each interval is treated as an independent sample. In each sample, there are vibration data at n moments (assuming that data is collected at n moments in each preset time interval), thus constructing a series of historical vibration data samples. For example, for the first sample (m=1), at the first moment i=1 there is a vibration amplitude value and vibration frequency At the second moment i=2, and And so on, the historical vibration data of sample 1 is formed The historical fault records correspond to these samples one by one and are used to subsequently determine the fault type label of each sample.

[0100] For each sample's historical vibration data The vibration data features are generated according to the method mentioned above. First, the number of clusters K is determined by the elbow method (the same K value can be used for all samples, or it can be adjusted appropriately according to the characteristics of the samples). Then, from the n data groups P of the samples, i m Randomly select K as cluster centers, then standardize the data group, calculate the distance from each data group to each cluster center, assign the data group to the cluster to which the nearest cluster center belongs, recalculate and update the cluster center, repeat these steps until the convergence condition is met, and get K clusters. Afterwards, calculate each cluster The proportion of the number of data groups included in the sample to the total number of data groups Get the cluster quantity ratio feature vector Calculate each cluster The average vibration amplitude and the average vibration frequency , and obtain the cluster average vibration amplitude feature vector and cluster average vibration frequency eigenvector Calculate each cluster The average distance from the data group in to the cluster center Get the average distance feature vector

[0101] The number of clusters accounts for the feature vector R m , cluster average vibration amplitude feature vector A m , cluster average vibration frequency eigenvector F m and the average distance feature vector D m Combine them in a specific order to form the combined feature vector of sample m:

[0102]

[0103] This combined feature vector comprehensively summarizes the vibration data features of sample m, and uses it as the input for model training, so that the model can learn the relationship between different operating states and faults of the equipment based on these features. For example, if the average vibration amplitude of a cluster in the combined feature vector of sample m is high and the average distance is large, it may indicate that the equipment has a potential fault risk during the operating time period corresponding to the sample. The model has the ability to predict faults by learning the association between a large number of such feature vectors and fault types.

[0104] According to the historical fault records, find the corresponding combined feature vector X m ) to determine the specific fault type. Assume that the device may have B different fault types, represented by t1, t2, …, t B Then, one-hot encoding is used to convert the fault type into a vector form. The principle of one-hot encoding is to create a vector Y with a length of B for B fault types. m , if the fault type of sample m is t b , then in the vector Y m In The remaining elements For example, if the equipment has three fault types (B=3), namely "bearing fault", "blade fault" and "overheating fault", for a sample with the fault type "blade fault", its one-hot encoded output label Y m =(0, 1, 0).

[0105] Through one-hot encoding, the categorical variable of fault type is converted into a numerical form suitable for processing by the neural network model. During the model training process, the model can be trained based on the input combined feature vector X m and the corresponding output label Y m The mapping relationship between features and fault types is learned. This encoding method makes it clear that different fault types can be distinguished, making it easier for the model to accurately adjust parameters during training to predict the correct fault type. For example, during training, the model will gradually learn that when the input combined feature vector has certain specific feature values, the corresponding one-hot encoding representation of the fault type should be output, thereby improving the accuracy of fault prediction.

[0106] The combined feature vector X of each sample obtained previously m (m=1, 2, ..., M) as input, and the corresponding one-hot encoded output label Y mCombined in pairs. This forms a series of input-output pairs, which constitute the training set of the artificial intelligence model. For example, the combined feature vector X1 of the first sample and its corresponding output label Y1 constitute a pair of data (X1, Y1) in the training set, and the corresponding data of the second sample constitutes (X2, Y2), and so on. This training set covers the information of the equipment in different operating states (represented by the combined feature vector) and the corresponding fault type (represented by one-hot encoding), providing a data basis for the subsequent use of the back-propagation algorithm to train artificial intelligence models such as the MLP model.

[0107] In an optional embodiment, the vibration data features are input into a trained artificial intelligence model to obtain a fault prediction result, including the following steps:

[0108] Input the combined feature vector X into the artificial intelligence model and output the vector

[0109] The fault type corresponding to the element in the vector Y whose value is greater than the preset value is confirmed as a potential fault, and the potential fault is used as the fault prediction result.

[0110] Specifically, the combined feature vector X is input into the artificial intelligence model, and the output vector First, the combined feature vector generated in the previous step As the input of a trained artificial intelligence model (such as an MLP model). In the MLP model, the neurons in the input layer receive the individual elements of the combined feature vector. The data then undergoes a complex nonlinear transformation through the hidden layer. The neurons in the hidden layer perform weighted summation on the input data based on the pre-learned weights and biases, and introduce nonlinear characteristics through activation functions (such as ReLU, etc.). For example, the neurons in the first hidden layer will calculate the sum of the product of the input data and its own weight, add the bias, and then get the output through the activation function. This output is used as the input of the next hidden layer or output layer, and so on. After multiple layers of calculation and transmission, a vector is finally obtained in the output layer. Here B represents the number of fault types that the model can predict. (b=1,2,…,B) represents the corresponding fault type t b The predicted value of , the size and meaning of which depends on the output layer settings and training method of the model. For example, if the output layer uses the Sigmoid activation function, then The value will be between 0 and 1, indicating the type of failure that occurred on the device probability.

[0111] Based on the knowledge and patterns learned in the training phase, the model predicts the possible fault types and their corresponding probabilities of the equipment according to the input combined feature vector X. Each element in the output vector Y Reflects the model's response to equipment failure type t b For example, if (Assuming B = 3, corresponding to three different fault types), this means that the model believes that the probability of the first fault type occurring in the device is high (0.8), the probability of the second fault type occurring is low (0.1), and the probability of the third fault type occurring is very low (0.05). Through such an output vector, the model provides a basis for subsequent determination of potential faults, which can help operation and maintenance personnel understand the risks that the equipment may face in advance.

[0112] The preset value can usually be selected through the analysis of a large amount of historical data and the evaluation of model performance. For example, if the model outputs the probability of failure (between 0 and 1), the preset value can be set to 0.5, which means that when the model predicts that the probability of a certain fault type is greater than 0.5, the fault type is considered to be a potential fault. Of course, different preset values ​​can also be set for different fault types according to the importance and risk level of different fault types. For example, for some fault types that seriously affect the safe operation of equipment, the preset value can be set relatively low so that potential risks can be discovered more promptly.

[0113] After the preset value is determined, the output vector Y is checked. For each element in vector Y if If it is greater than the preset value, the fault type t b Confirmed as a potential fault. For example, continuing with the above output vector Y = (0.8, 0.1, 0.05) as an example, if the preset value is 0.5, then the first fault type will be confirmed as a potential fault. These potential faults can be used as fault prediction results to promptly notify the operation and maintenance personnel of the thermal power plant. The operation and maintenance personnel can take corresponding measures based on the prediction results, such as further checking the equipment, arranging preventive maintenance, preparing spare parts for repair, etc., so as to improve the reliability and safety of equipment operation, avoid downtime and production losses caused by equipment failure, and realize effective safety monitoring and early warning management of thermal power plant equipment. At the same time, the fault prediction results can also serve as an important basis for subsequent analysis of equipment operation status change trends and optimization of equipment maintenance strategies, and continuously improve and perfect the equipment management system.

[0114] In an embodiment of the present application, a thermal power plant safety monitoring and early warning management method based on artificial intelligence is provided. The method installs sensors at key parts of thermal power plant equipment to collect vibration amplitude values ​​and frequency data at each moment within a preset time interval, processes and extracts multi-dimensional features through a clustering algorithm to form a combined feature vector, and uses a training set containing historical vibration data and fault records to train the MLP model. Finally, the combined feature vector obtained in real time is input into the trained model to set a preset value to determine that the fault type corresponding to the value in the output vector is greater than the value as a potential fault. This method realizes the accurate prediction of thermal power plant faults, reduces the probability of sudden faults, improves equipment reliability, enables operation and maintenance personnel to perform maintenance in advance, optimizes maintenance strategies, reduces costs, and at the same time improves the safety of thermal power plants and ensures production operations and personnel safety.

[0115] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0116] Based on the same inventive concept, the embodiment of the present application also provides a system for implementing the above-mentioned artificial intelligence-based thermal power plant safety monitoring and early warning management method. The implementation scheme for solving the problem provided by the system is similar to the implementation scheme recorded in the above-mentioned method, so the specific limitations in the embodiments of one or more artificial intelligence-based thermal power plant safety monitoring and early warning management systems provided below can be referred to the above limitations on an artificial intelligence-based thermal power plant safety monitoring and early warning management method, which will not be repeated here.

[0117] In an exemplary embodiment, Figure 2 As shown, a thermal power plant safety monitoring and early warning management system 200 based on artificial intelligence is provided, comprising:

[0118] The vibration data acquisition module 201 is used to acquire vibration data of the equipment in the thermal power plant at each moment within a preset time interval.

[0119] The data clustering module 202 is used to divide the vibration data into multiple clusters based on a clustering algorithm.

[0120] The vibration data feature generation module 203 is used to generate vibration data features based on multiple clusters.

[0121] The fault prediction module 204 is used to input the vibration data features into the trained artificial intelligence model to obtain a fault prediction result.

[0122] The management module 205 is used to generate a corresponding maintenance strategy based on the fault prediction result, and the maintenance strategy is used to instruct to repair the equipment.

[0123] Optionally, the vibration data acquisition module 201 includes:

[0124] The data acquisition unit is used to obtain the vibration amplitude value A of the equipment in the thermal power plant at each moment within a preset time interval. i and the vibration frequency f i ; Wherein, i represents the sequence number of the moment, i=1, 2, ..., n, and n represents the total number of moments within the preset time interval.

[0125] Data combination unit, used to convert the vibration amplitude value A i and the vibration frequency f i Composition data set P i =(A i , f i ), and the data set P i As vibration data; wherein each moment within the preset time interval corresponds to a data group P i =(A i , f i ).

[0126] Optionally, the clustering algorithm used by the data clustering module 202 is a K-Means algorithm, and the data clustering module 202 is specifically used to perform the following operations:

[0127] Step 1: Determine the number of clusters K by the elbow method; wherein the number of clusters K is used to divide the vibration data at n moments within a preset time interval into K clusters, K = 1, 2, ..., n; among the K clusters, the set of vibration data contained in the kth cluster is C k , k=1,2,…,K.

[0128] Step 2: From n data sets P i Randomly select K data sets P from i , as cluster centers; where cluster centers are E1, E2, …, E K , the cluster center of the kth cluster is

[0129] Step 3: Group each data set P i =(A i , f i) is standardized to According to Z i Calculate each data set P i To each cluster center E k The distance d(Z i , E k ), the calculation formula is:

[0130] Step 4: Group each data set P i Assign to the distance d(Z i , E k ) The nearest cluster center E k The cluster C k ;in,

[0131] Step 5: For each cluster C k , recalculate and update its cluster center in, n k is cluster C k The data set P contained in i The number of

[0132] Step 6: Repeat steps 3 to 5 until all cluster centers are Get K different clusters C k ;in, is the updated cluster center, is the cluster center before updating, ∈ is the preset threshold.

[0133] Optionally, the vibration data feature generation module 203 includes:

[0134] Cluster quantity ratio feature generation unit, used to calculate each cluster C k The data set P i The ratio of the number of data sets to the total number of data sets k , we get the cluster number ratio feature vector R = (r1, r2, ..., r k ,…,r K );in,

[0135] Cluster average parameter feature generation unit, used to calculate each cluster C k The average vibration amplitude and the average vibration frequency Get the cluster average vibration amplitude feature vector and cluster average vibration frequency eigenvector in, and

[0136] Cluster average distance feature acquisition unit, used to calculate each cluster C k The data set P i To the cluster center E to which it belongs k The average distance Get the average distance feature vector

[0137] in,

[0138] The feature combination unit is used to combine the cluster number proportion feature vector R, the cluster average vibration amplitude feature vector A, the cluster average vibration frequency feature vector F and the average distance feature vector D to form a combined feature vector And the combined feature vector X is used as the vibration data feature.

[0139] Optionally, the fault prediction module 204 includes:

[0140] The training set generating unit is used to obtain a training set based on historical data of equipment in the thermal power plant.

[0141] The model building unit is used to select the MLP model as the neural network model to build the artificial intelligence model.

[0142] The model training unit is used to train the MLP model using the training set based on the back propagation algorithm to obtain a trained artificial intelligence model.

[0143] Optionally, the training set generating unit is specifically used to perform the following operations:

[0144] Step 1: Based on the historical data of the equipment in the thermal power plant, obtain the historical vibration data of the equipment at m preset time intervals and historical fault records; where m = 1, 2, ..., M, M is the number of samplings, m is the sample number, is the vibration amplitude value of the mth sample at the i-th moment, f i m is the historical vibration frequency of the mth sample at the ith moment.

[0145] Step 2: Based on historical vibration data According to the above method, the combined feature vector of each sample is obtained And the combined feature vector X m As the input of model training; where X m Represents the combined feature vector of the mth sample.

[0146] Step 3: Determine each combined feature vector X based on historical fault records m The corresponding fault type is encoded using one-hot encoding as Y m As the output label; where t1 to t B Characterize different fault types, Or 1, b = 1, 2, ..., B, if This means that the fault type of the mth sample is t b Characterizes the type of fault.

[0147] Step 4: Based on the input X m and output label Y m , and obtain the training set of the artificial intelligence model {(X1, Y1), (X2, Y2), …, (X M , Y M )}.

[0148] Optionally, the fault prediction module 204 includes:

[0149] The data feature processing unit is used to input the combined feature vector X into the artificial intelligence model and output the vector

[0150] The fault prediction result generating unit is used to identify the fault type corresponding to the element in the vector Y whose value is greater than the preset value as a potential fault, and use the potential fault as the fault prediction result.

[0151] An embodiment of the present application further provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps in the above-mentioned method embodiments are implemented.

[0152] The embodiments of the present application further provide a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0153] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can refer to the partial description of the method embodiments. The device embodiments described above are only schematic, wherein the components described as separate parts may or may not be physically separated, and the parts displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the disclosed solution. A person of ordinary skill in the art can understand and implement it without paying any creative work.

[0154] The above-mentioned embodiments only express several implementation methods of the embodiments of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the patent of the embodiments of the present application. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the embodiments of the present application, and these all belong to the protection scope of the embodiments of the present application.

Claims

1. A safety monitoring and early warning management method for a thermal power plant based on artificial intelligence, characterized in that: The method comprises: Obtain vibration data of equipment in thermal power plants at each time within a preset time interval; Based on a clustering algorithm, dividing the vibration data into a plurality of clusters; generating a vibration data signature based on a plurality of the clusters; Inputting the vibration data features into a trained artificial intelligence model to obtain a fault prediction result; Based on the fault prediction result, a corresponding maintenance strategy is generated, where the maintenance strategy is used to instruct maintenance of the equipment.

2. The method according to claim 1, characterized in that The step of obtaining vibration data of the equipment in the thermal power plant at each time within a preset time interval includes: Obtain the vibration amplitude value A of the equipment of the thermal power plant at each moment within the preset time interval i and the vibration frequency f i ; Wherein, i represents the sequence number of the moment, i=1,2,…,n, and n represents the total number of moments within the preset time interval; The vibration amplitude value A i and the vibration frequency f i Composition data set P i =(A i ,f i ), and the data set P i As the vibration data; wherein each moment within the preset time interval corresponds to a data group P i =(A i ,f i ).

3. The method according to claim 2, characterized in that: The clustering algorithm is K-Means algorithm; The vibration data is divided into a plurality of clusters based on a clustering algorithm, including: Step 1: Determine the number of clusters K by the elbow method; The cluster number K is used to divide the vibration data at n moments within the preset time interval into K clusters, K = 1, 2, ..., n; among the K clusters, the set of vibration data contained in the kth cluster is C k , k=1,2,…,K; Step 2: From n data sets P i Randomly select K of the data sets P i , as cluster centers; wherein the cluster centers are E1, E2, …, E K , the cluster center of the kth cluster is Step 3: Each of the data groups P i =(A i ,f i ) is standardized to According to Z i Calculate each of the data sets P i To each of the cluster centers E k The distance d(Z i ,E k ), the calculation formula is: Step 4: Group each of the data sets P i Assigned to the distance d(Z i ,E k ) The nearest cluster center E k The cluster C k ;in, Step 5: For each cluster C k , recalculate and update its cluster center in, n k For the cluster C k The data set P contained in i the number of Step 6: Repeat steps 3 to 5 until all the cluster centers are Get K different clusters C k ;in, is the updated cluster center, is the cluster center before updating, ∈ is the preset threshold.

4. The method according to claim 3, characterized in that The generating of vibration data features based on the plurality of clusters comprises: Calculate each cluster C k The data set P includes i The ratio of the number of data sets to the total number of data sets is r k , we get the cluster number ratio feature vector R=(r1,r2,…,r k ,…,r K );in, Calculate each cluster C k The average vibration amplitude and the average vibration frequency Get the cluster average vibration amplitude feature vector and cluster average vibration frequency eigenvector in, and Calculate each cluster C k The data set P i To the cluster center E to which it belongs k The average distance Get the average distance feature vector in, The cluster number proportion feature vector R, the cluster average vibration amplitude feature vector A, the cluster average vibration frequency feature vector F and the average distance feature vector D are combined to form a combined feature vector And the combined feature vector X is used as the vibration data feature.

5. The method according to claim 4, characterized in that The training method of the artificial intelligence model includes: Based on the historical data of the equipment in the thermal power plant, a training set is obtained; Selecting the MLP model as the neural network model to construct the artificial intelligence model; Based on the back propagation algorithm, the MLP model is trained using the training set to obtain the trained artificial intelligence model.

6. The method according to claim 5, characterized in that The training set is obtained based on the historical data of the equipment in the thermal power plant, including: Based on the historical data of the equipment in the thermal power plant, the historical vibration data of the equipment at m preset time intervals is obtained. and historical fault records; where m = 1, 2, ..., M, M is the number of samplings, m is the sample number, is the vibration amplitude value of the mth sample at the i-th moment, is the historical vibration frequency of the mth sample at the i-th moment; Based on the historical vibration data According to the above method, the combined feature vector of each sample is obtained And the combined feature vector X m As the input of model training; where X m Represents the combined feature vector of the mth sample; Determine each combined feature vector X according to the historical fault records m The corresponding fault type is encoded using one-hot encoding as Y m As the output label; where t1 to t B Characterize different fault types, Or 1, b=1,2,…,B, if It means that the fault type of the mth sample is t b The type of fault characterized; Based on the input quantity X m and the output label Y m , and obtain the training set of the artificial intelligence model {(X1,Y1),(X2,Y2),…,(X M ,Y M )}.

7. The method according to claim 6, characterized in that The step of inputting the vibration data features into a trained artificial intelligence model to obtain a fault prediction result includes: The combined feature vector X is input into the artificial intelligence model, and the output vector The fault type corresponding to the element in the vector Y whose value is greater than a preset value is confirmed as a potential fault, and the potential fault is used as the fault prediction result.

8. A thermal power plant safety monitoring and early warning management system based on artificial intelligence, characterized in that: The system comprises: A vibration data acquisition module is used to acquire vibration data of the equipment in the thermal power plant at each time within a preset time interval; A data clustering module, used for dividing the vibration data into a plurality of clusters based on a clustering algorithm; A vibration data feature generation module, used for generating vibration data features based on the plurality of clusters; A fault prediction module, used to input the vibration data features into a trained artificial intelligence model to obtain a fault prediction result; A management module is used to generate a corresponding maintenance strategy based on the fault prediction result, and the maintenance strategy is used to instruct maintenance of the equipment.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images