Energy internet device fault prediction method, device, and storage medium
By deploying edge sensing devices at energy internet terminals, collecting multimodal data, and using high-throughput Kalman filtering and ALBERT models for fault prediction, the problem of real-time monitoring and early warning of energy internet equipment is solved, and efficient fault identification and online diagnosis are achieved.
Patent Information
- Application Number
- CN202311160151.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-09-08
- Publication Date
- 2026-05-15
- Estimated Expiration
- 2043-09-08
AI Technical Summary
Existing technologies have not yet been able to effectively predict the failures of energy internet equipment, especially the real-time monitoring and early warning of abnormal equipment.
By deploying edge sensing devices at the energy internet terminal, multimodal real-time operating data is collected. High-throughput Kalman filtering combined with the ALBERT autoencoder model is used for data fusion. The ALBERT model and feedforward neural network are used for fault identification, and fault identification vectors are output to achieve fault prediction.
It enables real-time status monitoring and online diagnostic early warning of energy internet equipment, improves the accuracy and efficiency of fault prediction, and supports cloud-edge collaboration and edge platform control.
Smart Images
Figure CN117235622B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of prediction technology, and more specifically, to a method, device, and storage medium for predicting equipment failures in the energy internet. Background Technology
[0002] The Energy Internet is a network that integrates advanced power electronics, information technology, and intelligent management technology to interconnect a large number of energy nodes, such as power networks, oil networks, and natural gas networks, which consist of distributed energy harvesting devices, distributed energy storage devices, and various types of loads, in order to achieve bidirectional energy flow, peer-to-peer energy exchange, and sharing.
[0003] The Energy Internet is essentially a new type of information-energy integrated "wide area network" built on the Internet concept. It uses the large power grid as the "backbone network" and microgrids as "local area networks," employing an open and peer-to-peer integrated information-energy architecture to truly achieve bidirectional, on-demand energy transmission and dynamic balanced use, maximizing the adaptability to the integration of new energy sources. Microgrids are a fundamental component of the Energy Internet, forming "local area networks" through new energy power generation, the collection, aggregation, and sharing of micro-energy, and energy storage or consumption within the microgrid. The large power grid still possesses unparalleled advantages in transmission efficiency and will remain the "backbone network" of the Energy Internet in the future.
[0004] The energy internet contains a wide variety of devices, making real-time monitoring of device status, timely detection of abnormal devices, and fault prediction particularly important. However, currently, there is no method capable of detecting abnormal devices in the energy internet or predicting their faults. Summary of the Invention
[0005] This invention provides a method, device, and storage medium for predicting equipment failures in the energy internet.
[0006] According to an embodiment of the present invention, a method for predicting faults in energy internet equipment is provided, comprising the following steps:
[0007] S1, deploy edge sensing devices on various types of energy internet terminal devices to collect multimodal real-time energy internet operation data; wherein, the multimodal real-time energy internet operation data is energy internet node data or user terminal sensing device data;
[0008] S2 uses high-throughput Kalman filtering combined with an ALBERT-based autoencoder model to fuse the collected data in order to handle the differences in sensor reading parameters of different types of energy Internet devices.
[0009] S3. Obtain the corresponding parameter vector based on the fused data, input the parameter vector into different ALBERT models and a feedforward neural network respectively, and finally output the fault identification vector through the Softmax function to complete the fault prediction of energy Internet equipment.
[0010] Furthermore, in S1, the multimodal real-time energy internet operation data includes: device data, transmission data, and environmental data; in S3, the parameter vector includes: device parameter vector, transmission parameter vector, and environmental parameter vector.
[0011] Furthermore, the various types of equipment in the energy internet terminal include: wind turbines, photovoltaic facilities, and geothermal facilities; the sensing device includes: sensors;
[0012] The equipment data is related to the types of new energy equipment deployed at the current energy grid nodes, including: wind turbine speed, wake velocity, and photovoltaic panel surface temperature;
[0013] The transmitted data is used to determine whether the current Internet node is operating normally, and includes: power consumption, transmission rate, latency, and data packet loss rate;
[0014] The environmental data includes temperature and humidity sensing terminals and micro-dust sensing terminals, enabling data acquisition in different modalities.
[0015] Furthermore, obtaining the corresponding parameter vector based on the fused data specifically involves taking the data changes over a certain period of time as a vector for each fused data set to obtain the corresponding parameter vector.
[0016] Furthermore, each element in the fault identification vector represents the probability of a type of fault.
[0017] Furthermore, step S3 also includes:
[0018] The ALBERT model outputs fault types manually labeled based on fault identification vectors. Different types of faults are identified and labeled for different types of energy terminal equipment. The ALBERT model is trained through backpropagation and gradient descent, and the trained model can be used for fault prediction in the multimodal energy internet.
[0019] Furthermore, step S3 includes: assigning a fault diagnosis task to three different ALBERT models, with the three different ALBERT models respectively performing fault diagnosis on the operating status of the equipment, fault diagnosis on the data transmitted by the current node of the energy network, and identification and diagnosis of abnormal environmental parameters.
[0020] Further, step S3 includes:
[0021] An alert is issued when any one of the three different ALBERT models identifies an anomaly.
[0022] According to another embodiment of the present invention, an apparatus is provided, the apparatus including a processor and a memory coupled to the processor, wherein,
[0023] The memory stores program instructions for implementing the energy internet equipment fault prediction method.
[0024] The processor is used to execute the program instructions stored in the memory to predict energy internet equipment failures.
[0025] A storage medium storing processor-executable program instructions for executing the energy internet device fault prediction method.
[0026] This invention can provide real-time status, statistical information and historical details of equipment. For abnormal equipment, this application can perform online diagnosis and early warning, cloud-edge collaboration and edge platform control to achieve "controllable" equipment. Attached Figure Description
[0027] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:
[0028] Figure 1 This is a flowchart illustrating the energy internet equipment fault prediction method of the present invention.
[0029] Figure 2 This is a schematic diagram of the device structure according to an embodiment of this application;
[0030] Figure 3 This is a schematic diagram of the structure of the storage medium according to an embodiment of this application. Detailed Implementation
[0031] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.
[0032] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.
[0033] Example 1
[0034] Please see Figure 1 A flowchart illustrating the energy internet equipment fault prediction method of this invention is provided for detailed explanation:
[0035] Step S1 involves deploying edge sensing devices on various types of energy internet terminal equipment to collect multimodal real-time energy internet operation data. Specifically:
[0036] The various types of equipment in the energy internet terminal include: wind turbines, photovoltaic facilities, geothermal facilities, etc.
[0037] The sensing device includes: different types of sensors.
[0038] The multimodal real-time energy internet operation data can be either energy internet node data or user terminal sensing device data, including: device data, transmission data, and environmental data.
[0039] The equipment data is related to the type of new energy equipment deployed at the current energy grid node (such as wind power, photovoltaic, geothermal), and includes: wind turbine speed, wake velocity, photovoltaic panel surface temperature, etc.
[0040] The transmitted data is mainly used to determine whether the current Internet node is operating normally, and includes: power consumption, transmission rate, latency, data packet loss rate, etc.
[0041] The environmental data includes temperature and humidity sensing terminals and dust sensing terminals, enabling data acquisition in different modes; among which, different modes include: equipment operation mode, data transmission mode, and environmental parameter mode.
[0042] Step S2 involves fusing the collected data using a high-throughput Kalman filter combined with an ALBERT-based autoencoder model to address the differences in sensor reading parameters across various types of energy internet devices. Specifically:
[0043] The ALBERT model is a variant of the BERT model, primarily achieving a lightweight BERT model through parameter sharing. In the traditional BERT-Transformer model, parameters are not shared between layers; however, in ALBERT, the parameters of fully connected layers and attention layers are shared. That is, ALBERT still has multiple deep connections, but the parameters between each layer are the same.
[0044] The advantages of the ALBERT model include:
[0045] 1. Word Embedding Parameter Factorization
[0046] ALBERT's word embeddings only remember a relatively small amount of word information; more semantic and syntactic information is remembered by the hidden layers. Therefore, the dimension of the word embeddings does not need to be the same as the dimension of the hidden layers; the number of parameters can be reduced by decreasing the dimension of the word embeddings. Assume the vocabulary size is V, the dimension of the word embeddings is E, and the dimension of the hidden layers is H.
[0047] In BERT, E = H; ALBERT's approach is to reduce E and add a project layer between the word embedding and the hidden layer to connect the two layers. Let's analyze the number of parameters in the embedding layer in both cases.
[0048] 1) BERT: ParameterNumBERT = E * V = H * V. Typically, V is very large; in the Chinese BERT model, V is approximately 30,000. In BERT_base, H = 1024: ParameterNumBERT = 30,000 * 1024
[0049] 2) ALBERT: ParameterNumAL = (V + H) * E In ALBERT, E = 128; H = 1024: ParameterNumAL = 30000 * 128 + 128 * 1024 ParameterNumAL / ParameterNumAL L = 7.7 From the above analysis, it can be seen that by factoring the parameters of the embedding layer, the parameters of the embedding layer are successfully reduced to 1 / 8 of the original.
[0050] 2. Parameter sharing in hidden layers
[0051] BERT_base contains 12 hidden layers in between; BERT_large contains 24 hidden layers in between; parameters are not shared between the layers.
[0052] Parameter sharing can significantly reduce the number of parameters. Parameter sharing can be divided into parameter sharing in fully connected layers and attention layers. In ALBERT, the parameters of fully connected layers and attention layers are shared. That is, ALBERT still has multiple deep connections, but the parameters of each layer are the same. Obviously, in this way, the number of parameters in the hidden layers of ALBERT becomes 1 / 12 or 1 / 24 of the original.
[0053] 3. Sentence order prediction
[0054] In BERT, the task of determining the relationship between sentences is next sentence prediction (NSP), which involves feeding two sentences into the model and predicting whether the second sentence is the next sentence after the first sentence.
[0055] In ALBERT, the task of predicting sentence order is sentence-order prediction (SOP), which involves giving the model two sentences and asking it to predict their order. The paper explains that SOP is a more complex task than NSP, and that the model can learn more semantic relationships between sentences through the SOP task compared to NSP.
[0056] ALBERT utilizes two methods—word embedding parameter factorization and hidden layer parameter sharing—to significantly reduce the number of model parameters without sacrificing model performance. Hidden layer parameter sharing can greatly reduce model parameters and also helps improve model training speed.
[0057] ALBERT combines two parameter reduction techniques to eliminate major obstacles in pre-training models. The first is decompositional embedding parameterization: by decomposing a large vocabulary embedding matrix into two smaller matrices, the size of the hidden layer is separated from the size of the word embedding layer. This decomposition makes it easier to increase the size of the hidden layers without significantly increasing the size of the vocabulary embedding parameters. The second technique is cross-layer parameter sharing: this technique prevents the parameters from growing with the depth of the network. ALBERT configurations similar to BERT-large have 18x fewer parameters and training speed is improved by approximately 1.7x. Parameter reduction techniques also act as a form of regularization, stabilizing training and aiding generalization.
[0058] By using high-throughput Kalman filtering combined with an ALBERT-based autoencoder model to fuse the collected data, the differences in sensor reading parameters of different types of energy internet devices can be addressed, thereby reducing model training time and improving fault prediction accuracy.
[0059] Step S3: Based on the fused data, three types of parameter vectors are obtained. These three types of parameter vectors are then input into three different ALBERT models and a three-layer feedforward neural network. Finally, a fault identification vector is output through the Softmax function to complete the fault prediction of energy internet equipment. Specifically:
[0060] Based on the fused data, three types of parameter vectors are obtained. These three types of parameter vectors are then input into three different ALBERT models. The outputs of the three models are then input into a three-layer feedforward DNN neural network. Finally, a fault identification vector is output through the Softmax function to complete the fault prediction of energy internet equipment.
[0061] The method of obtaining three types of parameter vectors based on the fused data is as follows: For the fused data, the parameter changes over a period of time are taken as a vector to obtain three types of parameter vectors.
[0062] Each element in the fault identification vector represents the probability of a type of fault.
[0063] The model output consists of manually labeled fault types. Different types of faults are identified and labeled for different types of energy terminal equipment. The model is trained through backpropagation and gradient descent, and the trained model can be used for multimodal energy internet fault prediction.
[0064] In this embodiment, the three types of parameter vectors are encoded and trained separately, so that the final classification network (DNN) can better capture the semantic relationship between the three types of vectors (for example, the environmental parameter vector represents that the wind is strong, but the wind turbine unit senses that the wind turbine speed is too low). The two conditions are combined to determine that the wind turbine unit has a fault. However, using only the wind turbine unit data may not be able to identify the fault. Mixing all parameters together may cause model confusion and reduce accuracy.
[0065] Specifically, this embodiment distributes a large fault diagnosis task across three different ALBERT models. These three models respectively diagnose faults related to equipment operating status, data transmission at the current node of the energy network, and environmental parameter anomalies. A comprehensive model then provides an alert when any of these three models identifies an anomaly during the actual diagnosis process. By combining these three models into a single large model (a three-layer feedforward DNN neural network), the overall fault severity of the energy internet can be comprehensively assessed by learning the weights of these three types of faults. Based on the output assessment results, heuristic energy network control rules are designed.
[0066] For example, if only the data transmission is abnormal, it will not affect the operation of the current node's renewable energy equipment, and the system can still operate normally and generate electricity. However, if the equipment's operating parameters malfunction, it is necessary to take measures such as shutdown or power reduction based on the severity of the fault analyzed by the model.
[0067] Example 2
[0068] Please see Figure 2 This is a schematic diagram of the device structure according to an embodiment of this application. The device 50 includes a processor 51 and a memory 52 coupled to the processor 51.
[0069] The memory 52 stores program instructions for implementing the above-mentioned energy internet equipment fault prediction method.
[0070] The processor 51 is used to execute program instructions stored in the memory 52 to predict energy internet equipment failures.
[0071] The processor 51 can also be referred to as a CPU (Central Processing Unit). The processor 51 may be an integrated circuit chip with signal processing capabilities. The processor 51 can also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), an off-the-shelf programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. A general-purpose processor can be a microprocessor or any conventional processor.
[0072] Example 3
[0073] Please see Figure 3 This is a schematic diagram of the structure of the storage medium according to an embodiment of this application. The storage medium of this embodiment stores a program file 61 capable of implementing all the above methods. This program file 61 can be stored in the storage medium in the form of a software product, including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods of various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks, or devices such as computers, servers, mobile phones, and tablets.
[0074] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.
[0075] In the above embodiments of the present invention, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.
[0076] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The system embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection of units or modules may be electrical or other forms.
[0077] The units described as separate components may or may not be physically separate. Similarly, the components shown as units may or may not be physical units; they may be located in one place or distributed across multiple units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.
[0078] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.
[0079] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.
[0080] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.
Claims
1. A method for predicting equipment failures in an energy internet, characterized in that, Includes the following steps: S1, deploy edge sensing devices on various types of energy internet terminal devices to collect multimodal real-time energy internet operation data; wherein, the multimodal real-time energy internet operation data is energy internet node data or user terminal sensing device data; S2 uses high-throughput Kalman filtering combined with an ALBERT-based autoencoder model to fuse the collected data in order to handle the differences in sensor reading parameters of different types of energy Internet devices. S3, based on the fused data, obtain the corresponding parameter vector, and input the parameter vector into different ALBERT models and a feedforward neural network respectively. Finally, output the fault identification vector through the Softmax function to complete the fault prediction of energy internet equipment; each element in the fault identification vector represents the probability of a type of fault; wherein: Step S3 includes: assigning a fault diagnosis task to three different ALBERT models, with the three different ALBERT models respectively implementing fault diagnosis of equipment operating status, fault diagnosis of data transmitted by the current node of the energy network, and identification and diagnosis of abnormal environmental parameters. The ALBERT model outputs fault types manually labeled based on fault identification vectors. Different types of faults are identified and labeled for different types of energy terminal equipment. The ALBERT model is trained through backpropagation and gradient descent, and the trained model can be used for fault prediction in the multimodal energy internet.
2. The energy internet equipment fault prediction method according to claim 1, characterized in that, In S1, the multimodal real-time energy internet operation data includes: device data, transmission data, and environmental data; in S3, the parameter vector includes: device parameter vector, transmission parameter vector, and environmental parameter vector.
3. The energy internet equipment fault prediction method according to claim 2, characterized in that: The various types of equipment in the energy internet terminal include: wind turbines, photovoltaic facilities, and geothermal facilities; the sensing devices include: sensors; The equipment data is related to the types of new energy equipment deployed at the current energy grid nodes, including: wind turbine speed, wake velocity, and photovoltaic panel surface temperature; The transmitted data is used to determine whether the current Internet node is operating normally, and includes: power consumption, transmission rate, latency, and data packet loss rate; The environmental data includes temperature and humidity sensing terminals and micro-dust sensing terminals, enabling data acquisition in different modalities.
4. The energy internet equipment fault prediction method according to claim 1, characterized in that: The step of obtaining the corresponding parameter vector based on the fused data specifically involves taking the data changes over a certain period of time as a vector for each fused data set to obtain the corresponding parameter vector.
5. The energy internet equipment fault prediction method according to claim 4, characterized in that: Step S3 includes: An alert is issued when any one of the three different ALBERT models identifies an anomaly.