Energy station high-frequency data acquisition and abnormal state identification method based on cloud edge collaboration

Abstract

The application discloses an energy station high-frequency data acquisition and abnormal state identification method based on cloud edge cooperation, comprising the following steps: acquiring vibration, sound and pressure wave data of energy station equipment by installing vibration sensors, sound sensors and pressure wave sensors at the energy station equipment; collecting relevant energy station equipment data information by a data acquisition module in an edge computing platform to obtain high-frequency data of the energy station equipment; after clustering learning of historical high-frequency data of the energy station equipment by a cloud platform, labeling the high-frequency data as normal and abnormal data, and establishing a high-frequency data anomaly detection model, the high-frequency data anomaly detection model is issued to the edge computing platform to detect real-time high-frequency data of the collected energy station equipment, and if the detected data is abnormal high-frequency data, the abnormal high-frequency data is uploaded to the cloud platform; after receiving the abnormal high-frequency data of the energy station equipment, the cloud platform performs energy station equipment anomaly identification through a pre-established energy station equipment anomaly identification model to obtain an energy station equipment anomaly type.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of comprehensive energy stations, and particularly relates to an energy station high-frequency data acquisition and abnormal state identification method based on cloud-edge collaboration. BACKGROUND

[0002] In the construction of a comprehensive energy station in a comprehensive park or development zone, various energies (gas, water, steam, heat energy, and electric energy) and various energy supply and energy saving equipment (generator sets, waste heat boilers, refrigeration units, solar panels, etc.) are combined, and production and energy transmission are centrally managed and controlled through the Internet to create an independent energy island and realize resource integration and comprehensive utilization of regional energy.

[0003] The Internet of Things is a fusion application of intelligent sensing and identification technology, ubiquitous computing, and ubiquitous network. In the 5G era of Internet of Everything, the Internet of Things uses different types of sensors to perceive surrounding objects and physical environments to provide a basis for data analysis in the application layer of the Internet of Things. The sensor acquisition system monitors, senses, and acquires various environmental or monitoring object information in real time, realizes the exchange between various elements of the physical world, information space, and human society, and has different scales and different acquisition frequencies. Different sensors require different acquisition frequencies, and high-frequency acquisition of different sensors requires the sensor acquisition system to have large-scale, high-bandwidth real-time transmission, computing, storage, and processing capabilities.

[0004] At present, the types of equipment involved in the comprehensive energy station are diverse and have strong coupling, and it is particularly important to ensure the normal operation of each device in the energy station. At present, most energy station device operation monitoring usually uses Internet of Things technology to collect and identify temperature and flow, and there is little research on high-frequency data such as vibration data, sound data, and pressure wave data. In addition, each device in the energy station has a long service life, and the time between two failures is long. Long-term unfiltered collection, transmission, and storage of high-frequency data have high requirements for hardware devices, but such high-frequency data can more timely and sensitively identify devices and help operators analyze and maintain energy station devices in advance, which is an important guarantee for the unattended scenario of the energy station. Therefore, how to optimize the collection method of high-frequency data according to the characteristics of the energy station equipment, effectively collect and detect data, and identify the normal and safe operation of the energy station equipment is a problem that needs to be solved at present.

[0005] Based on the above technical problems, a new energy station high-frequency data acquisition and abnormal state identification method based on cloud-edge collaboration is needed. SUMMARY

[0006] The technical problems to be solved by the present application are to overcome the shortcomings of the prior art and provide an energy station high-frequency data acquisition and abnormal state identification method based on cloud edge collaboration.

[0007] To solve the above technical problems, the technical scheme of the present application is:

[0008] The present application provides an energy station high-frequency data acquisition and abnormal state identification method based on cloud edge collaboration, comprising: a high-frequency data acquisition stage and an abnormal state identification stage.

[0009] The high-frequency data acquisition stage comprises:

[0010] By installing vibration sensors, sound sensors and pressure wave sensors at energy station equipment, vibration data, sound data and pressure wave data information of the energy station equipment are obtained.

[0011] The data acquisition module in the edge computing platform acquires relevant energy station equipment data information according to the pre-defined energy station equipment acquisition object, high-frequency data acquisition content, high-frequency data acquisition frequency, high-frequency data acquisition start and end conditions and high-frequency data acquisition priority, and obtains energy station equipment high-frequency data; the energy station equipment high-frequency data at least includes sound data, vibration data and pressure wave data of water pumps, fans, pipelines and compressors in the energy station.

[0012] The abnormal state identification stage comprises:

[0013] After clustering learning of the historical high-frequency data of the energy station equipment by the cloud platform, the high-frequency data is labeled as normal and abnormal data, and the historical high-frequency data with labels is trained to establish a high-frequency data abnormal detection model, which is then issued to the edge computing platform.

[0014] The high-frequency data abnormal detection model in the edge computing platform detects the collected real-time high-frequency data of the energy station equipment, and if the detection is abnormal high-frequency data, it is uploaded to the cloud platform; otherwise, the high-frequency data is directly stored in the edge computing platform.

[0015] After receiving the abnormal high-frequency data of the energy station equipment by the cloud platform, the energy station equipment abnormal identification model is used to identify the abnormality of the energy station equipment, and the abnormal type of the energy station equipment is obtained.

[0016] Further, the data collection module in the edge computing platform collects relevant energy station device data information according to the pre-defined energy station device collection object, high-frequency data collection content, high-frequency data collection frequency, high-frequency data collection start-stop condition and high-frequency data collection priority, and obtains energy station device high-frequency data, including:

[0017] The data collection module in the edge computing platform receives a preset data collection definition file, including the energy station device collection object, high-frequency data collection content, high-frequency data collection frequency, high-frequency data collection start-stop condition and high-frequency data collection priority;

[0018] Each energy station device collection object is virtually modeled as a collection element model, and a collection element model ID is identified and registered in the edge computing platform. After successful registration, it is detected whether the collection element model ID record is owned, and after confirmation, data collection can be performed. The collection element model includes the high-frequency data collection content, high-frequency data collection frequency, high-frequency data collection start-stop condition and high-frequency data collection priority related to the energy station device collection object;

[0019] According to the high-frequency data collection priority, high-priority, medium-priority and low-priority energy station devices are selected in stages, and according to the high-frequency data collection start-stop condition, when the energy station device state meets the start collection condition, the high-frequency data collection frequency is tracked to collect the high-frequency data state change of the corresponding energy station device object at a preset high-frequency sampling frequency, and the high-frequency data collection content related energy station device high-frequency data is obtained. When the energy station device state meets the stop collection condition, the high-frequency data collection of the corresponding energy station device object is stopped, and the energy station device high-frequency data is generated;

[0020] The setting of the high-frequency data collection priority includes obtaining an energy station device operation expert experience value variable interval, the experience value variable interval is [0, 1], 0 is the lowest priority energy station device object, and 1 is the highest priority energy station device object; according to the energy station device operation expert experience value variable interval, the collection data priority is set, including high priority, medium priority and low priority;

[0021] The high-frequency data collection start-stop condition includes a collection start identifier, a collection end identifier, a pause collection identifier and a resume collection identifier.

[0022] Further, the edge computing platform further comprises a high-precision timing module, a mobile communication module, a WIFI communication module, a sensor network communication module, a data processing and control module, an external interface module, a device location module and a storage module; the high-precision timing module is used to provide high-precision time synchronization function for the energy station device; the mobile communication module is used to support communication with the 5G mobile network; the WIFI communication module is used to provide access capability of the wireless local area network; the sensor network communication module is used to provide communication with the high-speed sensor network and the device; the data processing and control module is used to provide high-performance processing capability and perform data fusion and mining processing; the external interface module is used to provide an interface for connecting various sensors; the device location module is used to provide device location information through the GPS or Beidou device; and the storage module is used to provide a specific storage structure and store high-frequency data.

[0023] Further, after the cloud platform performs clustering learning on the historical high-frequency data of the energy station device, the high-frequency data is labeled as normal and abnormal data, and the historical high-frequency data with the label is trained to establish a high-frequency data anomaly detection model, comprising:

[0024] The cloud platform acquires the device high-frequency data of the energy station device in a preset historical time period and performs preprocessing;

[0025] A clustering algorithm is used to calculate the distance of each energy station device high-frequency data to the initial center, the data is divided into the cluster with the smallest distance, and iteration is performed;

[0026] According to the clustering result, the cluster with the smallest data volume is defined as an abnormal data cluster, and the normal high-frequency data cluster and the abnormal high-frequency data cluster are labeled respectively;

[0027] The neural network parameters are set, and the neural network is established;

[0028] After the energy station device high-frequency data with the label after clustering is input into the neural network for training, the high-frequency data anomaly detection model is established.

[0029] Further, the clustering algorithm uses a K-means clustering algorithm;

[0030] The distance of each energy station device high-frequency data to the initial center is calculated and expressed as:

[0031]

[0032] The energy station device high-frequency data set is defined as X m×n =[X 1,n ,X 2,n ,…,X m,n] is composed of high-frequency data collected by m energy station device sensors at n time points; X i,n is the data collected by the i th energy station device sensor at n time points; C s,n is the initial cluster center of the energy station device high-frequency data;

[0033] The normal high-frequency data cluster and the abnormal high-frequency data cluster are respectively labeled, represented as:

[0034]

[0035] X m×n = [X 1,n ,X 2,n ,…,X m,n , y i ] = [(X 1,n ,y i ), (X 1,n ,y i ), …, (X m,n ,y i )].

[0036] Further, the neural network adopts an improved BP neural network, by introducing BAS moth antennae search algorithm in the global search process of FPA flower pollination algorithm, adopting mutation differential strategy in the local search process to form BAS-FPA optimization algorithm, and then using BAS-FPA optimization algorithm to optimize the initial weight and threshold parameters of BP neural network, forming BAS-FPA-BP network model;

[0037] In the BAS-FPA optimization algorithm, each pollen particle itself is a solution in the solution space, representing a possible combination of BP neural network weight and threshold value;

[0038] The initial BP neural network weight and threshold value are encoded and processed, the mapping relationship between BP neural network weight, threshold value and BAS-FPA pollen particle is established through vector encoding method, the best weight and threshold value are obtained through optimization solution, and the optimal BP neural network is obtained.

[0039] Further, after receiving the abnormal high-frequency data of the energy station device through the cloud platform, the energy station device abnormal identification is carried out through the pre-established energy station device abnormal identification model, and the energy station device abnormal type is obtained, including:

[0040] Obtain an abnormal data sample based on a historical energy station device abnormal type database in the cloud platform; the abnormal data sample includes an energy station device abnormal type corresponding to abnormal high-frequency data, and the abnormal type at least includes pump internal foreign matter impact damage, surge, part loosening, insufficient flow, insufficient fan air volume, excessive vibration, overheat, part wear, compressor stall, low exhaust pressure, overcurrent, part failure, high ambient temperature, pipeline leakage, and damage;

[0041] Establish an improved generative adversarial network model: taking a generative adversarial network GAN as a model subject, including a generator network G and a discriminator network D; the generator network G includes a decoder and an encoder connected in sequence, both of which adopt independent recurrent neural network modules; the discriminator network D includes linear layers, activation function layers, linear layers, and activation function layers connected in sequence; introduce a gradient penalty into the generative adversarial network GAN, and the loss function of the generator network G is defined as: f w (x) is an independent recurrent neural network, p g is generated data, and x is a high-frequency data sample; the loss function of the discriminator network D is defined as:

[0042]

[0043] λ is a gradient penalty function; is a function for gradient calculation; ||·||2 is the two-norm of a matrix;

[0044] After the abnormal data sample is input into the improved generative adversarial network model for training, an energy station device abnormality identification model is established;

[0045] When the cloud platform receives energy station device abnormal high-frequency data uploaded by the edge computing platform, energy station device abnormality identification is performed through the energy station device abnormality identification model, and an energy station device abnormal type is output.

[0046] Further, the improved generative adversarial network model further includes: introducing a double-time scale update rule to balance the update speed of the generator network G and the discriminator network D, which is represented as: l d 、l g are learning rates of the discriminator network D and the generator network G, respectively; n is the number of iterations; h n (d), h n (g) are gradients of the discriminator network D and the generator network G, respectively; β、 are update coefficients of the discriminator network D and the generator network G, respectively.

[0047] Further, after the cloud platform learns the historical high-frequency data of the energy station equipment, the high-frequency data is labeled as normal and abnormal data, and the method that can also be used includes:

[0048] Obtain the high-frequency data of the equipment in a preset historical time period of the energy station equipment through the cloud platform, and perform preprocessing;

[0049] Feature extraction is performed on the high-frequency data of each energy station equipment to obtain high-frequency characteristic value data;

[0050] The clustering algorithm is used to cluster the high-frequency characteristic value data of the normal state and the abnormal state of the energy station equipment preconfigured to obtain the normal state clustering center and the abnormal state clustering center;

[0051] The distance between the high-frequency characteristic value data and the normal state clustering center and the abnormal state clustering center is calculated, and is represented as:

[0052]

[0053]

[0054] (x3,y3) is the high-frequency characteristic value data point; D A D is the distance between the high-frequency characteristic value data point (x3, y3) represented by the current high-frequency data and the normal state clustering center A (x1, y1); D B D is the distance between the high-frequency characteristic value data point (x3, y3) represented by the current high-frequency data and the abnormal state clustering center B (x2, y2);

[0055] If D A <D B , it indicates that no abnormality occurs; if D A >D B , it indicates that an abnormality occurs;

[0056] According to the clustering result, the high-frequency data corresponding to the abnormality is defined as an abnormal data cluster, and the normal high-frequency data cluster and the abnormal high-frequency data cluster are labeled respectively.

[0057] Further, the edge computing platform and the cloud platform communicate securely through identity authentication and encryption and decryption algorithms: the cloud platform generates a unique identity for each edge computing platform and encrypts and transmits it to the corresponding edge computing platform; the edge computing platform obtains its own identity by decryption, attaches the identity to the communication data, and encrypts and transmits it to the cloud platform; after receiving the encrypted data packet, the cloud platform verifies whether the identity is legal by decryption, and processes the data subsequently after the identity is legal.

[0058] The beneficial effects of the present application are:

[0059] The application installs vibration sensors, sound sensors and pressure wave sensors at the energy station equipment, obtains vibration data, sound data and pressure wave data information of the energy station equipment; the related energy station equipment data information is collected by a data collection module in the edge computing platform according to the pre-defined energy station equipment collection object, high-frequency data collection content, high-frequency data collection frequency, high-frequency data collection start and stop conditions and high-frequency data collection priority, and energy station equipment high-frequency data is obtained; the energy station equipment high-frequency data at least includes sound data, vibration data and pressure wave data of the water pump, fan, pipeline and compressor in the energy station; after the cloud platform clusters and learns the historical high-frequency data of the energy station equipment, the high-frequency data is labeled as normal and abnormal data, and the historical high-frequency data with labels is trained to establish a high-frequency data anomaly detection model, which is issued to the edge computing platform; the collected real-time high-frequency data of the energy station equipment is detected by the high-frequency data anomaly detection model in the edge computing platform, if the detection is abnormal high-frequency data, it is uploaded to the cloud platform; otherwise, the high-frequency data is directly stored in the edge computing platform; after the cloud platform receives the abnormal high-frequency data of the energy station equipment, the energy station equipment anomaly recognition model is used for energy station equipment anomaly recognition, and the energy station equipment anomaly type is obtained; on the one hand, a high-frequency collection data service platform is constructed based on edge computing to collect, store, upload, calculate and analyze sensing data, which meets the needs of long-time high-frequency collection of high-density sensors in the energy station; on the other hand, the main task of the edge computing platform is to collect high-frequency data of the energy station equipment, and to detect which high-frequency data of the energy station equipment is abnormal, and to upload the abnormal high-frequency data to the cloud platform, and the training of the anomaly detection model and the anomaly type recognition are in the cloud platform. The purpose of this is to reduce the traffic cost of the Internet of Things, and if the high-frequency data of the energy station equipment is uploaded for a month, it will be several hundred G, and if we select only the abnormal high-frequency data for uploading through abnormal detection, it has actual economic significance for computing resources, storage resources and traffic fees; therefore, the edge computing platform and the cloud platform operate cooperatively, which can fully utilize the data collection and local data detection and processing capabilities of the edge computing platform, and can utilize the big data processing, model training and data sharing capabilities of the cloud platform, and finally realize the collection, abnormal detection and abnormal state recognition of high-frequency data.

[0060] Other features and advantages will be set forth in the descriptions that follow, and in part will be apparent from the description, or can be learned by practice of the application. The purposes and other advantages of the present application will be realized and attained by the structure particularly pointed out in the written description and claims hereof.

[0061] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the following preferred embodiments are specifically described below, and the accompanying drawings are as follows. BRIEF DESCRIPTION OF DRAWINGS

[0062] In order to more clearly illustrate the technical solutions in the specific embodiments or prior art of the present application, the drawings required to be used in the description of the specific embodiments or prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0063] Figure 1 A flow chart of a high-frequency data acquisition and abnormal state identification method for an energy station based on cloud-edge collaboration is provided.

[0064] Figure 2 A schematic block diagram of the principle of a high-frequency data acquisition and abnormal state identification method for an energy station based on cloud-edge collaboration is provided. DETAILED DESCRIPTION

[0065] In order to make the purpose, technical solutions and advantages of the embodiments of the present application more clear, the technical solutions of the present application will be described clearly and completely below in conjunction with the drawings. Obviously, the described embodiments are some embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0066] Embodiment 1

[0067] Figure 1 A flow chart of a high-frequency data acquisition and abnormal state identification method for an energy station based on cloud-edge collaboration is provided.

[0068] Figure 2 A schematic block diagram of the principle of a high-frequency data acquisition and abnormal state identification method for an energy station based on cloud-edge collaboration is provided.

[0069] As shown in Figure 1 , 2 , the present embodiment 1 provides a high-frequency data acquisition and abnormal state identification method for an energy station based on cloud-edge collaboration, which includes a high-frequency data acquisition stage and an abnormal state identification stage.

[0070] The high-frequency data acquisition stage includes:

[0071] By installing vibration sensors, sound sensors and pressure wave sensors at the energy station equipment, vibration data, sound data and pressure wave data information of the energy station equipment are obtained.

[0072] The data acquisition module in the edge computing platform collects relevant energy station equipment data information according to the pre-defined energy station equipment collection object, high-frequency data collection content, high-frequency data collection frequency, high-frequency data collection start and stop conditions and high-frequency data collection priority, and obtains energy station equipment high-frequency data; the energy station equipment high-frequency data at least includes sound data, vibration data and pressure wave data of a water pump, a fan, a pipeline and a compressor in the energy station;

[0073] The abnormal state identification stage includes:

[0074] After the cloud platform clusters and learns the historical high-frequency data of the energy station equipment, the high-frequency data is labeled as normal and abnormal data, and the historical high-frequency data with labels is trained to establish a high-frequency data anomaly detection model, which is then distributed to the edge computing platform;

[0075] The high-frequency data anomaly detection model in the edge computing platform detects the collected real-time high-frequency data of the energy station equipment, and if the detection is abnormal high-frequency data, the data is uploaded to the cloud platform; otherwise, the high-frequency data is directly stored in the edge computing platform;

[0076] After the cloud platform receives the abnormal high-frequency data of the energy station equipment, the pre-established energy station equipment anomaly identification model is used to identify the energy station equipment anomaly, and the energy station equipment anomaly type is obtained.

[0077] It should be noted that the number of energy stations is multiple, and the number of edge computing platforms is at least one. The relationship between the edge computing platform and the energy station is one-to-many or one-to-one. According to the distance between each energy station, one edge computing platform can correspond to multiple energy stations, or one edge computing platform can correspond to one energy station. Through the corresponding edge computing platform, the high-frequency data can be collected and detected nearby. If the detection is abnormal high-frequency data, it indicates that the running state of the equipment in the energy station has a certain degree of abnormality, which belongs to the local identification of each edge computing platform to the corresponding energy station. There is a certain correlation between the abnormal high-frequency data and the running state of the energy station equipment. The cloud platform further identifies the abnormal high-frequency data to obtain the energy station equipment anomaly type. The energy station equipment anomaly identification model established by the cloud platform belongs to global identification, and the abnormal high-frequency data in the cloud platform is a big data sample of each energy station equipment.

[0078] In the embodiment, the data acquisition module in the edge computing platform collects relevant energy station equipment data information according to the pre-defined energy station equipment collection object, high-frequency data collection content, high-frequency data collection frequency, high-frequency data collection start and stop conditions and high-frequency data collection priority, and obtains energy station equipment high-frequency data, including:

[0079] The preset data collection definition file is received by a data collection module in the edge computing platform, including energy station equipment collection objects, high-frequency data collection content, high-frequency data collection frequency, high-frequency data collection start and stop conditions, and high-frequency data collection priority;

[0080] Each energy station equipment collection object is virtually modeled as a collection element model, and a collection element model ID is identified and registered in the edge computing platform. After successful registration, it is detected whether the collection element model ID record is owned. After confirmation, data collection can be performed. The collection element model includes high-frequency data collection content, high-frequency data collection frequency, high-frequency data collection start and stop conditions, and high-frequency data collection priority related to the energy station equipment collection object.

[0081] According to the high-frequency data collection priority, energy station equipment with high, medium, and low priority is selected step by step. According to the high-frequency data collection start and stop conditions, when the energy station equipment state meets the start collection condition, the high-frequency data state change of the corresponding energy station equipment object is tracked and collected according to the high-frequency data collection frequency at a preset high-frequency sampling frequency, the high-frequency data of the energy station equipment related to the high-frequency data collection content is obtained, and when the energy station equipment state meets the stop collection condition, the high-frequency data collection of the corresponding energy station equipment object is stopped, and the energy station equipment high-frequency data is generated.

[0082] The setting of the high-frequency data collection priority includes obtaining an energy station equipment operation expert experience value variable interval, the experience value variable interval is [0, 1], 0 is the lowest priority energy station equipment object, and 1 is the highest priority energy station equipment object; according to the energy station equipment operation expert experience value variable interval, the collection data priority is set, including high, medium, and low priority.

[0083] The high-frequency data collection start and stop conditions include collection start identification, collection end identification, pause collection identification, and resume collection identification.

[0084] In the embodiment, the edge computing platform further comprises a high-precision timing module, a mobile communication module, a WIFI communication module, a sensor network communication module, a data processing and control module, an external interface module, a device location module and a storage module; the high-precision timing module is configured to provide high-precision time synchronization function for the energy station device; the mobile communication module is configured to support communication with a 5G mobile network; the WIFI communication module is configured to provide access capability of a wireless local area network; the sensor network communication module is configured to provide communication with a high-speed sensor network and a device; the data processing and control module is configured to provide high-performance processing capability and perform data fusion and mining processing; the external interface module is configured to provide an interface for connecting various sensors; the device location module is configured to provide device location information through a GPS or Beidou device; and the storage module is configured to provide a specific storage structure and store high-frequency data.

[0085] In the embodiment, after the cloud platform performs clustering learning on the historical high-frequency data of the energy station device, the high-frequency data is labeled as normal and abnormal data, and the historical high-frequency data with labels is trained to establish a high-frequency data anomaly detection model, including:

[0086] The cloud platform acquires device high-frequency data of the energy station device in a preset historical time period and performs preprocessing;

[0087] A clustering algorithm is used to calculate the distance of each energy station device high-frequency data to an initial center, data is divided into a cluster with the smallest distance according to the distance, and iteration is performed;

[0088] According to the clustering result, a cluster with the smallest data volume is defined as an abnormal data cluster, and normal high-frequency data clusters and abnormal high-frequency data clusters are labeled respectively;

[0089] Neural network parameters are set, and a neural network is established;

[0090] After the energy station device high-frequency data with labels after clustering is input into the neural network for training, a high-frequency data anomaly detection model is established.

[0091] It should be noted that in the energy station equipment high frequency data, K initial cluster centers are randomly selected, the distance of the initial high frequency data to the centroid is compared and classified, and the cluster center point is updated by calculating the average distance of the high frequency data in the cluster, and iteration is performed until the final classification result meets the minimum distance of the data in the cluster and the maximum distance between clusters, and the iteration process is terminated. As an unsupervised learning method in machine learning, K-means clustering does not require prior knowledge of data, and the calculation is simple, which meets the needs of energy station equipment abnormal high frequency data detection. According to the small proportion of sensor abnormal high frequency data compared with normal data, when the K-means clustering algorithm is applied to energy station equipment abnormal high frequency data detection, we can identify the cluster with the least data in the cluster after classification as the abnormal cluster, and all high frequency data in the cluster are abnormal data.

[0092] The combination of K-means clustering and neural network is used to realize efficient detection of energy station equipment high frequency data. Among them, K-means clustering is mainly used to realize the normal and abnormal marking of energy station equipment high frequency data, and the clustering results are used as sample data set to effectively realize the modeling analysis of neural network. The algorithm first classifies the energy station equipment high frequency data by using the K-means clustering algorithm in unsupervised learning, and labels different categories, and then trains the neural network with the labeled energy station equipment high frequency data to obtain the energy station equipment high frequency data anomaly detection model and issue it to the edge computing platform. The model does not need to be clustered again in subsequent detection, and can directly detect the collected energy station equipment high frequency data.

[0093] The main task of the edge computing platform is to detect which energy station equipment high frequency data is abnormal, and the abnormal high frequency data needs to be uploaded to the cloud platform. The training of the anomaly detection model and the identification of the abnormal type are in the cloud platform. The purpose of this is to reduce the traffic cost of the Internet of Things. If the energy station equipment high frequency data is uploaded for a month, it will be several hundred G. If we select abnormal high frequency data for uploading by some detection, it has practical economic significance for computing resources, storage resources and traffic cost.

[0094] In the embodiment, the clustering algorithm adopts K-means clustering algorithm;

[0095] The distance of each energy station equipment high frequency data to the initial center is represented as:

[0096]

[0097] The energy station equipment high frequency data set is defined as X m×n =[X 1,n ,X 2,n ,…,X m,n], which is composed of high-frequency data collected by m energy station device sensors at n time points; X i,n is the data collected by the i th energy station device sensor at n time points; C s,n is the initial cluster center of the energy station device high-frequency data;

[0098] The normal high-frequency data cluster and the abnormal high-frequency data cluster are respectively labeled, denoted as:

[0099]

[0100] X m×n = [X 1,n ,X 2,n ,…,X m,n , y i ] = [(X 1,n ,y i ),(X 1,n ,y i ),…,(X m,n ,y i )].

[0101] In this embodiment, the neural network adopts an improved BP neural network, which is formed by introducing a BAS moth search algorithm in the global search process of the FPA flower pollination algorithm, adopting a mutation differential strategy in the local search process to form a BAS-FPA optimization algorithm, and then optimizing the initial weight and threshold parameters of the BP neural network by using the BAS-FPA optimization algorithm to form a BAS-FPA-BP network model.

[0102] In the BAS-FPA optimization algorithm, each pollen particle itself is a solution in the solution space, representing a possible combination of the weights and thresholds of a group of BP neural networks.

[0103] The initial weights and thresholds of the BP neural network are encoded and processed, the mapping relationship between the weights and thresholds of the BP neural network and the BAS-FPA pollen particles is established through vector encoding, the best weights and thresholds are obtained through optimization and solution, and the optimal BP neural network is obtained.

[0104] It should be noted that the FPA flower pollination algorithm is a swarm intelligence optimization algorithm inspired by the pollination process of natural flowering plants. To improve efficiency, the algorithm simplifies the flower pollination process, assuming that each plant only has one flower, and each flower only produces one pollen particle. A flower or a pollen particle can be mapped to a candidate solution in the solution space, and each flower reproduces offspring by performing cross-pollination operation through conversion probability p, or self-pollination operation with probability 1-p. Cross-pollination can be regarded as the global detection behavior of the algorithm, and self-pollination can be regarded as the local search process of the algorithm. The former refers to the transfer of pollen from one flower to another flower on another plant, and the latter refers to the transfer of pollen from one flower to a different flower on the same plant.

[0105] The principle of the FPA algorithm is based on the above process, that is, through pollen transmission, the more suitable flowers in the population survive, and the main principles include:

[0106] Cross-pollination belongs to the global search process, and global search is based on Levy distribution;

[0107] Self-pollination belongs to the local search process;

[0108] The probability of successful pollen pollination is positively correlated with the similarity of the two flowers;

[0109] The mutual switching of global search and local search is controlled by conversion probability p.

[0110] The BAS algorithm does not need to know the specific form of the function, but only needs an individual, which greatly reduces the computational load compared to other swarm intelligence algorithms. The beetle is a kind of beetle, most of which have long antennae for detecting food odor. The bionics principle is inspired by the process of the beetle moving according to the strength of the food odor to determine the position of the food during foraging. Each beetle has two antennae extending to the left and right directions, and the next moving path of the beetle will always be towards the direction with stronger odor intensity. The food odor can be regarded as a function, and the goal of the beetle is to find the point with the maximum global odor intensity, i.e. the point with the maximum function value.

[0111] The flower pollination algorithm based on beetle antenna search improves the convergence speed of the FPA algorithm by introducing the BAS algorithm into the global search process of the FPA algorithm. By using the mutation difference strategy in the local search process, the ability of the FPA algorithm to jump out of the local optimum is improved, and thus the BAS-FPA algorithm is finally formed. In the improved global search process, after the pollen particle is updated according to the FPA algorithm, each pollen particle can be regarded as a beetle, and the new solution after updating is calculated and compared with the solution before updating. If the updated solution is better, the solution is updated and retained, otherwise the update is abandoned.

[0112] In the embodiment, after the cloud platform receives the high-frequency data of the abnormal energy station equipment, the energy station equipment abnormality recognition model is used to recognize the abnormal energy station equipment, and the energy station equipment abnormal type is obtained, including:

[0113] Based on the historical energy station equipment abnormal type database in the cloud platform, an abnormal data sample is obtained; the abnormal data sample includes an energy station equipment abnormal type corresponding to the abnormal high-frequency data, and the abnormal type at least includes internal foreign object impact damage of a water pump, surge, loose parts, insufficient flow, insufficient fan air volume, excessive vibration, overheat, component wear, compressor stall, low exhaust pressure, overcurrent, component failure, high ambient temperature, pipeline leakage and damage;

[0114] An improved generative adversarial network model is established: taking a generative adversarial network GAN as a model main body, including a generator network G and a discriminator network D; the generator network G includes a decoder and an encoder connected in sequence, both of which adopt independent recurrent neural network modules; the discriminator network D includes linear layers, activation function layers, linear layers and activation function layers connected in sequence; the gradient penalty is introduced into the generative adversarial network GAN, and the loss function of the generator network G is defined as: f w (x) is an independent recurrent neural network, p g is generated data, and x is a high-frequency data sample; the loss function of the discriminator network D is defined as:

[0115]

[0116] λ is a gradient penalty function; is a function for gradient calculation; ||·||2 is the two-norm of a matrix;

[0117] After the abnormal data sample is input into the improved generative adversarial network model for training, an energy station equipment abnormality recognition model is established;

[0118] When the cloud platform receives the high-frequency data of the abnormal energy station equipment uploaded by the edge computing platform, the energy station equipment abnormality recognition model is used to recognize the abnormal energy station equipment, and the energy station equipment abnormal type is output.

[0119] It should be noted that the gradient penalty is applied to the training of the generative adversarial network (GAN), which not only improves the network convergence performance and the quality of the generated data samples, but also makes the network training more stable. The unsupervised anomaly recognition is performed by using the generative adversarial network structure. The independent recurrent neural network module is used in the generation network part, which can effectively improve the depth of the network structure, and the non-saturation function is used as the activation function, so that the problems of gradient disappearance and gradient explosion can be avoided. Therefore, the improved generative adversarial network model of the present application solves the problems of poor convergence, gradient explosion and poor quality of generated data samples in the existing training based on the generative adversarial network.

[0120] In the present embodiment, the improved generative adversarial network model further comprises: introducing a double time scale update rule to balance the update speed of the generator network G and the discriminator network D, which is represented as: l d 、l g are the learning rates of the discriminator network D and the generator network G respectively; n is the iteration number; h n (d), h n (g) are the gradients of the discriminator network D and the generator network G respectively; β, are the update coefficients of the discriminator network D and the generator network G respectively.

[0121] It should be noted that the improved generative adversarial network model uses the same learning rate for training G and D, but the feature of the network itself is that the update speed of D cannot keep up with the update speed of G. During model training, the update speed between the two should be balanced as much as possible to prevent the network from collapsing. Therefore, the double time scale update rule is introduced to balance the network update problem, and different learning rates are set for D and G in the WGAN-GP network. The purpose is to speed up the convergence speed of D, that is, the learning rate of D is greater than that of G. The double time scale update rule allows G and D to update at a speed of 1:1, which produces better results under the same time conditions, making the generative adversarial network model more stable and reducing the running time.

[0122] In the present embodiment, after the cloud platform learns the historical high-frequency data of the energy station equipment, the high-frequency data is labeled as normal and abnormal data, and the method that can also be used includes:

[0123] Obtain the high-frequency data of the equipment in the preset historical time period of the energy station equipment through the cloud platform, and perform preprocessing;

[0124] Extract features from each energy station equipment high-frequency data to obtain high-frequency feature value data;

[0125] The clustering algorithm is used to cluster the pre-configured high-frequency characteristic value data of the normal state and the abnormal state of the energy station equipment to obtain a normal state clustering center and an abnormal state clustering center;

[0126] The distance between the high-frequency characteristic value data and the normal state clustering center and the abnormal state clustering center is calculated and represented as:

[0127]

[0128]

[0129] (x3,y3) is a high-frequency characteristic value data point; D A D is the distance between the high-frequency characteristic value data point (x3,y3) represented by the current high-frequency data and the normal state clustering center A(x1,y1); B D is the distance between the high-frequency characteristic value data point (x3,y3) represented by the current high-frequency data and the abnormal state clustering center B(x2,y2);

[0130] If D A <D B , it indicates that no abnormality occurs; if D A >D B , it indicates that an abnormality occurs.

[0131] According to the clustering result, the high-frequency data corresponding to the abnormality is defined as an abnormal data cluster, and the normal high-frequency data cluster and the abnormal high-frequency data cluster are labeled respectively.

[0132] In this embodiment, the edge computing platform and the cloud platform communicate securely through identity verification and encryption and decryption algorithms: the cloud platform generates a unique identity for each edge computing platform and encrypts the identity and transmits it to the corresponding edge computing platform; the edge computing platform obtains its own identity by decryption, attaches the identity to the communication data, and encrypts and transmits the data to the cloud platform; after receiving the encrypted data packet, the cloud platform verifies whether the identity is legal by decryption, and processes the data subsequently after the identity is legal.

[0133] In several embodiments provided in the present application, it should be understood that the disclosed system and method can also be implemented by other manners. The system embodiments described above are only illustrative, for example, the flowcharts and block diagrams in the drawings show the possible implementation architecture, function and operation of the system, method and computer program product according to the embodiments of the present application. In this regard, each block in the flowchart or block diagram can represent a module, a program segment or a part of code, which includes one or more executable instructions for implementing the specified logic function. It should also be noted that in some alternative implementations, the functions noted in the blocks can occur in different order from that shown in the drawings. For example, two consecutive blocks can actually be executed substantially in parallel, and sometimes they can be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, and the combination of blocks in the block diagram and / or flowchart, can be implemented by a dedicated hardware-based system for performing the specified functions or actions, or can be implemented by a combination of dedicated hardware and computer instructions.

[0134] In addition, each functional module in the embodiments of the present application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part. When the functions are realized in the form of software functional modules and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or part of the technical solutions can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for making a computer device (which can be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the embodiments of the present application methods. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0135] Based on the above ideal embodiments according to the present application, through the above description, relevant personnel can make various changes and modifications without deviating from the scope of the technical idea of the present application. The technical scope of the present application is not limited to the contents in the specification, and must be determined according to the scope of the claims.

Claims

1. A cloud-edge collaboration-based energy station high-frequency data acquisition and abnormal state identification method, characterized in that, include: High-frequency data acquisition phase and abnormal state identification phase; The high-frequency data acquisition stage includes: Vibration sensors, sound sensors, and pressure wave sensors are installed at the energy station equipment to acquire vibration data, sound data, and pressure wave data information of the energy station equipment. The data acquisition module in the edge computing platform collects relevant energy station equipment data information according to predefined energy station equipment acquisition objects, high-frequency data acquisition content, high-frequency data acquisition frequency, high-frequency data acquisition start and end conditions, and high-frequency data acquisition priority to obtain high-frequency data of energy station equipment; the high-frequency data of energy station equipment includes at least the sound data, vibration data, and pressure wave data of water pumps, fans, pipelines, and compressors in the energy station; The abnormal state identification stage includes: After clustering and learning the historical high-frequency data of energy station equipment through the cloud platform, the high-frequency data is labeled as normal and abnormal data. After training the labeled historical high-frequency data, a high-frequency data anomaly detection model is established and sent to the edge computing platform. The high-frequency data collected from the energy station equipment is detected by the high-frequency data anomaly detection model in the edge computing platform. If the high-frequency data is detected as abnormal, it is uploaded to the cloud platform; otherwise, the high-frequency data is directly stored in the edge computing platform. After receiving high-frequency data on energy station equipment anomalies through the cloud platform, the system uses a pre-established energy station equipment anomaly identification model to identify the types of anomalies, including: Based on the historical energy station equipment anomaly type database in the cloud platform, anomaly data samples are obtained; the anomaly data samples include energy station equipment anomaly types corresponding to high-frequency abnormal data, and the anomaly types include at least the following: water pump internal foreign object impact damage, surge, loose parts, insufficient flow, insufficient fan air volume, excessive vibration, overheating, component wear, compressor stall, low exhaust pressure, overcurrent, component failure, high ambient temperature, pipeline leakage, and damage. Establish an improved generative adversarial network model: take the generative adversarial network (GAN) as the model subject, including the generator network G and the discriminator network D; the generator network G includes a decoder and an encoder connected in sequence, both of which adopt independent recurrent neural network modules; the discriminator network D includes linear layers, activation function layers, linear layers and activation function layers connected in sequence; the gradient penalty is introduced into the generative adversarial network GAN, and the loss function of the generator network G is defined as: f w (x) is an independent recurrent neural network, p g is generated data, x is a high-frequency data sample; the loss function of the discriminator network D is defined as: λ is a gradient penalty function; is a function to compute the gradient; || · ||2is the matrix 2-norm; An abnormal data sample is input into an improved generative adversarial network model for training, and an abnormal identification model for energy station equipment is established. When the cloud platform receives high-frequency abnormal data of energy station equipment uploaded by the edge computing platform, it identifies the abnormality of the energy station equipment through the energy station equipment abnormality identification model and outputs the abnormality type of the energy station equipment.

2. The energy plant high frequency data collection and abnormal condition identification method of claim 1, wherein, The data acquisition module in the edge computing platform collects relevant energy station equipment data information according to predefined energy station equipment acquisition objects, high-frequency data acquisition content, high-frequency data acquisition frequency, high-frequency data acquisition start and end conditions, and high-frequency data acquisition priority, to obtain high-frequency data of the energy station equipment, including: The data acquisition module in the edge computing platform receives a preset data acquisition definition file, which includes the energy station equipment acquisition objects, high-frequency data acquisition content, high-frequency data acquisition frequency, high-frequency data acquisition start and end conditions, and high-frequency data acquisition priority. Each energy station equipment collection object is virtually modeled as a collection meta-model, and a collection meta-model ID is identified, registered in an edge computing platform, and after successful registration, it is detected whether the collection meta-model ID record is owned, and after confirmation, data collection can be performed thereon; the collection meta-model includes high-frequency data collection content, high-frequency data collection frequency, high-frequency data collection start and stop conditions, and high-frequency data collection priority related to the energy station equipment collection object; According to the high-frequency data collection priority, energy station equipment with high, medium and low priority is selected step by step, and according to the high-frequency data collection start and stop conditions, when the energy station equipment state meets the start collection condition, the high-frequency data state change of the corresponding energy station equipment object is tracked and collected according to the high-frequency data collection frequency at a preset high-frequency sampling frequency, and the high-frequency data of the energy station equipment related to the high-frequency data collection content is obtained, and when the energy station equipment state meets the stop collection condition, the high-frequency data collection of the corresponding energy station equipment object is stopped, and the high-frequency data of the energy station equipment is generated; The setting of the high-frequency data collection priority includes: obtaining an energy station equipment operation expert experience value variable interval, the experience value variable interval is [0, 1], 0 is the lowest priority energy station equipment object, and 1 is the highest priority energy station equipment object; according to the energy station equipment operation expert experience value variable interval, the collection data priority is set, including high, medium and low priority; The high-frequency data collection start and stop conditions include collection start identification, collection end identification, pause collection identification and resume collection identification.

3. The energy plant high frequency data collection and abnormal condition identification method of claim 1, wherein, The edge computing platform further includes a high-precision time service module, a mobile communication module, a WIFI communication module, a sensor network communication module, a data processing and control module, an external interface module, a device location module and a storage module; the high-precision time service module is used to provide high-precision time synchronization function for the energy station equipment; the mobile communication module is used to support communication with the 5G mobile network; the WIFI communication module is used to provide the access ability of the wireless local area network; the sensor network communication module is used to provide communication with the high-speed sensor network and the device; the data processing and control module is used to provide high-performance processing capability for data fusion and mining processing; the external interface module is used to provide an interface for connecting various sensors; the device location module is used to provide device location information through GPS or Beidou device; the storage module is used to provide a specific storage structure for storing high-frequency data.

4. The energy plant high frequency data collection and abnormal condition identification method of claim 1, wherein, After the cloud platform clusters and learns the historical high-frequency data of the energy station equipment, the high-frequency data is labeled as normal and abnormal data, and the historical high-frequency data with labels is trained to establish a high-frequency data anomaly detection model, including: The cloud platform obtains the device high-frequency data of the energy station equipment in a preset historical time period, and performs preprocessing; A clustering algorithm is used to calculate the distance of each energy station equipment high-frequency data to the initial center, and the data is divided into the cluster with the smallest distance according to the distance, and iteration is performed; According to the clustering result, the cluster with the smallest data amount is defined as an abnormal data cluster, and the normal high-frequency data cluster and the abnormal high-frequency data cluster are labeled respectively; The neural network parameters are set, and the neural network is established; After the labeled high-frequency data of the energy station equipment after clustering is input into the neural network for training, the high-frequency data anomaly detection model is established.

5. The energy plant high frequency data collection and abnormal condition identification method of claim 4, wherein, The clustering algorithm adopts a K-means clustering algorithm; The distance of each high-frequency data of the energy station equipment to the initial center is calculated and represented as: Energy station equipment high-frequency data set definition is X m×n = [X 1,n , X 2,n , …, X m,n ] composed of high-frequency data collected by m energy station equipment sensors in n time; X i,n is the data collected by the i th energy station equipment sensor in n time; C s,n is the initial cluster center of energy station equipment high-frequency data; The normal high-frequency data cluster and the abnormal high-frequency data cluster are labeled respectively, and represented as: X m×n = [X 1,n , X 2,n , …, X m,n , y i ] = [(X 1,n , y i ), (X 2,n , y i ), …, (X m,n , y i )].

6. The energy plant high frequency data collection and abnormal condition identification method of claim 4, wherein, The neural network adopts an improved BP neural network, a BAS moth antennae search algorithm is introduced in the global search process of the FPA flower pollination algorithm, a mutation differential strategy is adopted in the local search process to form a BAS-FPA optimization algorithm, and then the initial weight and threshold parameters of the BP neural network are optimized by using the BAS-FPA optimization algorithm to form a BAS-FPA-BP network model; In the BAS-FPA optimization algorithm, each pollen particle itself is a solution in the solution space, representing a possible combination of the weight and threshold of the BP neural network. The weight and threshold of the initial BP neural network are encoded, the mapping relationship between the BP neural network weight, threshold and BAS-FPA pollen particle is established through vector encoding, the best weight and threshold are obtained by optimization and solution, and the optimal BP neural network is obtained.

7. The energy plant high frequency data collection and abnormal condition identification method of claim 1, wherein, The improved generative adversarial network model further comprises: introducing a double time scale update rule to balance the update speed of the generator network G and the discriminator network D, denoted as: l d 、l g are learning rates of the discriminator network D and the generator network G respectively; n is the number of iterations; h n (d)、h n (g) are gradients of the discriminator network D and the generator network G respectively; β、 are update coefficients of the discriminator network D and the generator network G respectively.

8. The energy plant high frequency data collection and abnormal condition identification method of claim 4, wherein, After the cloud platform learns the historical high-frequency data of the energy station equipment, the high-frequency data are labeled as normal and abnormal data, and the following methods can also be used: The cloud platform obtains the high-frequency data of the energy station equipment in a preset historical time period, and performs preprocessing; The high-frequency data of each energy station equipment are feature extracted to obtain high-frequency feature value data; The clustering algorithm is used to cluster the high-frequency feature value data of the normal state and the abnormal state of the energy station equipment preconfigured to obtain normal state clustering centers and abnormal state clustering centers; The distance between the high-frequency feature value data and the normal state clustering centers and the abnormal state clustering centers is calculated and represented as: (x3, y3) is a high-frequency characteristic value data point; D A D is the distance between the high-frequency characteristic value data point (x3, y3) represented by the current high-frequency data and the normal state clustering center A (x1, y1); B D is the distance between the high-frequency characteristic value data point (x3, y3) represented by the current high-frequency data and the abnormal state clustering center B (x2, y2). If D A <D B , then it indicates that no abnormality occurs; if D A >D B , then it indicates that an abnormality occurs; According to the clustering result, the high-frequency data corresponding to the abnormality are defined as an abnormal data cluster, and the normal high-frequency data cluster and the abnormal high-frequency data cluster are labeled respectively.

9. The energy plant high frequency data collection and abnormal condition identification method of claim 1, wherein, The edge computing platform and the cloud platform communicate securely through identity verification and encryption and decryption algorithms: the cloud platform generates a unique identity for each edge computing platform and encrypts it and transmits it to the corresponding edge computing platform; the edge computing platform obtains its own identity by decryption, attaches the identity to the communication data, and encrypts and transmits it to the cloud platform; after receiving the encrypted data packet, the cloud platform verifies whether the identity is legal by decryption, and processes the data subsequently after the identity is legal.

Images