Search

CN-121982868-A - Intelligent electric power fire early warning system with self-learning capability

CN121982868ACN 121982868 ACN121982868 ACN 121982868ACN-121982868-A

Abstract

The invention relates to the technical field of safety monitoring of electric power systems, and provides an intelligent electric power fire early warning system with self-learning capability, which comprises a site sensing layer, a gateway layer, a server layer and an interaction layer; the on-site sensing layer comprises an electric parameter acquisition module, a thermal parameter acquisition module and a fault characteristic acquisition module, the gateway layer comprises a data processing module, a characteristic extraction module, a history database, a self-learning analysis module, a treatment suggestion generation module and a human-computer interaction interface, the gateway layer comprises a data acquisition module and a interaction layer comprises a linkage control interface and a user terminal, and the data acquisition module is responsible for collecting, converting, synchronizing and uploading all sensor data of the on-site sensing layer. The invention fundamentally solves the two core defects of single monitoring dimension and static rigidification of the early warning model of the traditional electric power fire early warning system through two key technologies of multi-parameter fusion and closed loop self-learning.

Inventors

  • MA ZHEXUAN
  • NIU LIJUAN

Assignees

  • 中环低碳节能技术(北京)有限公司
  • 中数信科信息科技(北京)有限公司

Dates

Publication Date
20260505
Application Date
20260213

Claims (10)

  1. 1. An intelligent electric fire early warning system with self-learning capability is characterized by comprising a site perception layer, a gateway layer, a server layer and an interaction layer; The field perception layer comprises The electric parameter acquisition module is used for acquiring current and voltage parameters of the electric equipment; the thermal parameter acquisition module is used for acquiring real-time temperature parameters of the power equipment; The fault characteristic acquisition module is used for acquiring real-time fault parameters of the power equipment; The gateway layer comprises The data acquisition module is responsible for collecting, converting, synchronizing and uploading the data of each sensor of the field sensing layer; The server layer comprises The data processing module is used for carrying out filtering, denoising and normalization preprocessing on the acquired original data; The feature extraction module is used for extracting time domain, frequency domain and time domain features to form feature vectors; The historical database is used for structurally storing all real-time monitoring data, characteristic data, model versions, early warning records and treatment feedback results; The self-learning analysis module is used for constructing a prediction model, connecting the prediction model with a historical database, optimizing model parameters, and dynamically calculating and adjusting early warning threshold values of all monitoring parameters through algorithm learning; the treatment suggestion generation module is used for matching and generating targeted treatment suggestions according to the identified fault modes; the man-machine interaction interface is used for displaying real-time state, early warning information and treatment suggestions and allowing a user to confirm and feed back treatment results, wherein the feedback results are further used for optimizing the model; The interaction layer comprises The linkage control interface is used for realizing automatic emergency treatment in an emergency situation; And the user terminal is used for receiving early warning by operation and maintenance personnel, checking suggestions and managing the entrance of the system.
  2. 2. The intelligent power fire early warning system with self-learning capability according to claim 1, wherein the electrical parameter acquisition module comprises The current transformer is used for measuring the amplitude, waveform and phase of the loop current and judging equipment load, fault state and energy flow; And the voltage sensor is used for measuring the amplitude, waveform, frequency and phase of the power supply voltage and evaluating the power quality and the system stability.
  3. 3. The intelligent power fire early warning system with self-learning capability of claim 1, wherein the thermal parameter acquisition module comprises The wireless temperature measurement sensor is used for directly, continuously and accurately monitoring the contact type temperature of key easy hot spots of the power equipment; and the thermal imaging camera is used for carrying out two-dimensional temperature field scanning imaging on a large-area equipment area, the whole equipment or a part where the point sensor is difficult to install in a non-contact mode.
  4. 4. The intelligent power fire early warning system with self-learning capability according to claim 1, wherein the fault feature acquisition module comprises The arc detector is used for monitoring unique light radiation and strong electromagnetic radiation generated by an electric arc of the power equipment, and distinguishing fault electric arc from electric welding, illumination or other light sources by analyzing the frequency and intensity modes of the light signals so as to reduce false alarm; the partial discharge sensor is used for detecting weak and continuous pulse discharge occurring at the insulation defect position in the power equipment, and aims at early warning and early warning of insulation fault in a latent period; an ultrasonic sensor for detecting high frequency sound waves that are inaudible to the human ear, generated by discharge, arcing, or overheat faults.
  5. 5. The intelligent power fire early warning system with self-learning capability according to claim 1, wherein the self-learning analysis module comprises A basic fire prediction model, namely an initial model constructed by adopting a machine learning algorithm; The continuous learning unit is connected with the historical database, and periodically or triggerably retrains the basic fire prediction model by using newly-added monitoring data, optimizes model parameters, and enables the model to adapt to long-term evolution of equipment aging, environmental change and load fluctuation; The self-adaptive threshold adjusting unit dynamically calculates and adjusts the early warning threshold of each monitoring parameter according to the statistical analysis results of the current and historical normal operation data of the equipment; And the fault mode knowledge base learns and generalizes different types of typical fault modes and characteristics thereof from the historical abnormal data through the combination of unsupervised learning and supervised learning.
  6. 6. The intelligent power fire early warning system with self-learning capability according to claim 5, wherein the basic fire prediction model adopts a deep neural network algorithm to construct an initial model, and the initial model comprises a model input formula, a deep neural network forward propagation formula, a loss function formula, a back propagation and parameter optimization formula.
  7. 7. The intelligent power fire early warning system with self-learning capability according to claim 6, wherein the model input formula is: ; Wherein, the And the fusion information of different sensors at the time t.
  8. 8. The intelligent power fire early warning system with self-learning capability according to claim 6, wherein the deep neural network forward propagation formula is: wherein Θ= { W [1], b [1], W [ L ], b [ L ] } represent all parameter sets in the neural network that need to be learned; as a probability of risk of fire, , wherein, Is a Sigmoid function.
  9. 9. The intelligent power fire early warning system with self-learning capability of claim 6, wherein the loss function formula is: where m is the number of samples in a training batch (batch), i is the sample index, and the function gives a tremendous penalty to the model in case of misprediction.
  10. 10. The intelligent power fire early warning system with self-learning capability according to claim 6, wherein the back propagation and parameter optimization formula is: ; ; ; where α is the learning rate, and the step size of each update is controlled.

Description

Intelligent electric power fire early warning system with self-learning capability Technical Field The invention relates to the technical field of safety monitoring of power systems, in particular to an intelligent power fire early warning system with self-learning capability. Background The electric power system is an indispensable infrastructure of the modern society, and the safe and stable operation of the electric power system is directly related to national life and national economy. During long-term operation of power equipment (such as transformers, switch cabinets, cable joints, etc.), abnormal high temperatures and even fires may occur due to overload, insulation aging, poor contact, arc faults, etc. The electric fire disaster not only can cause huge direct economic loss, resulting in large-range power failure accidents, but also can cause serious secondary accidents, and constitutes a great threat to personnel safety and public safety. With the development of internet of things (IoT) and big data technologies, more advanced systems began to integrate a variety of sensors (e.g., electrical, thermal, fault signature sensors) and attempt to improve early warning capabilities using simple data analysis rules (e.g., trend analysis). However, these systems still have significant drawbacks in dealing with complex and variable operating environments of the power system, as follows: First, existing systems mostly make decisions based on a single type of parameter. For example, only the temperature or only the current is monitored. Early signs of power equipment failure tend to be multifaceted, and single parameter monitoring does not fully capture complex failure modes. For example, a poorly contacted joint may have a partial discharge or a minor arc generated before its temperature increases significantly. Depending on temperature monitoring alone, early warning delay can be caused, optimal treatment time is missed, and missing report (multi-dimensional fault characteristics cannot be identified) and false report (normal fluctuation without faults is misjudged as faults) are easily caused; Second, existing systems lack the ability to learn from historical data. They cannot utilize the accumulated mass monitoring data and the treatment feedback results to optimize their own decision logic. The analysis model and rules of the system are basically static after deployment, and cannot be self-perfected along with the evolution of the state of the equipment and the occurrence of a novel fault mode, so that the system cannot adapt to the long-term aging process of the equipment, and the training cannot be absorbed from the past false alarm and missing report. Over time, the accuracy of system early warning can gradually decrease, and long-term and sustainable safety guarantee cannot be realized; In view of the above, the invention provides an intelligent electric fire early warning system with self-learning capability. Disclosure of Invention The invention provides an intelligent electric fire early warning system with self-learning capability, which solves the problem that the prior art lacks the capability of learning from historical data. The intelligent electric fire early warning system with the self-learning capability comprises a site sensing layer, a gateway layer, a server layer and an interaction layer; The field perception layer comprises The electric parameter acquisition module is used for acquiring current and voltage parameters of the electric equipment; the thermal parameter acquisition module is used for acquiring real-time temperature parameters of the power equipment; The fault characteristic acquisition module is used for acquiring real-time fault parameters of the power equipment; The gateway layer comprises The data acquisition module is responsible for collecting, converting, synchronizing and uploading the data of each sensor of the field sensing layer; The server layer comprises The data processing module is used for carrying out filtering, denoising and normalization preprocessing on the acquired original data; The feature extraction module is used for extracting time domain, frequency domain and time domain features to form feature vectors; the history database is used for structurally storing all real-time monitoring data, characteristic data, model version, early warning record and treatment feedback result The self-learning analysis module is used for constructing a prediction model, connecting the prediction model with a historical database, optimizing model parameters, and dynamically calculating and adjusting early warning threshold values of all monitoring parameters through algorithm learning; the treatment suggestion generation module is used for matching and generating targeted treatment suggestions according to the identified fault modes; the man-machine interaction interface is used for displaying real-time state, early warning information and treatment suggestions and all