New energy power distribution network intelligent power protection measurement and control system and method based on Internet of Things technology

Abstract

The invention relates to the technical field of power systems and automation thereof, in particular to a new energy power distribution network intelligent power protection measurement and control system and method based on the Internet of Things technology, and the system comprises a global sensing layer, an edge calculation layer, a cooperative control layer, a communication network layer and an application service layer. The system collects multi-dimensional data through a sensor array, performs real-time processing and feature extraction on an edge side, dynamically corrects a protection constant value by using a deep reinforcement learning model, generates a regional collaborative isolation strategy, and starts a two-stage network reconstruction algorithm to realize quick power recovery of a non-fault region. According to the method, the communication bandwidth requirement can be reduced by about 60%, the response speed is improved, the protection action accuracy reaches 98.5% or above, and the fault isolation time does not exceed 150 milliseconds.

Description

Technical Field

[0001] This invention belongs to the field of power system and its automation technology, specifically relating to an intelligent power protection and control system and method for new energy distribution networks based on Internet of Things (IoT) technology. Background Technology

[0002] As the global energy structure accelerates its transition to clean and low-carbon energy, the operation of renewable energy distribution networks, as key carriers for the integration of distributed photovoltaic, wind power, and other renewable energy sources, is undergoing profound changes. The high proportion of distributed power sources has transformed distribution networks from traditional unidirectional radial structures into complex networks with multiple sources, bidirectional connections, and weak inertia, resulting in highly dynamic operating conditions and significantly increased uncertainty. Against this backdrop, traditional power protection and control systems face severe challenges. Existing technologies generally employ protection strategies based on local electrical quantity measurement and fixed threshold logic, which are ill-suited to new characteristics such as fluctuations in renewable energy output, frequent reversals in power flow direction, and reduced fault current amplitudes. This leads to decreased protection selectivity, sensitivity, and reliability, and a significant increase in the risk of false trips and failures to trip. Simultaneously, traditional control systems rely on a single data acquisition dimension, primarily limited to basic electrical quantities such as current and voltage. They lack effective integration of multi-source heterogeneous information, including equipment status (such as temperature and vibration), environmental meteorological conditions (such as wind speed and sunlight), and dynamic load changes, limiting the accuracy of early fault identification, risk situation awareness, and self-healing decision-making. Furthermore, existing system architectures are mostly centralized, requiring a large amount of raw sensor data to be uploaded to the main station for centralized analysis. This not only strains communication bandwidth resources but also introduces significant transmission delays, making it difficult to meet millisecond-level response requirements in scenarios involving rapid fault isolation and recovery. More importantly, the current lack of effective coordination mechanisms among protection devices means that independent actions by each node can easily lead to expanded isolation ranges or suboptimal recovery paths, hindering the achievement of the intelligent self-healing goal of "minimum isolation, maximum power restoration." In addition, issues such as closed communication protocols and heterogeneous device interfaces further restrict the realization of cross-vendor and cross-regional collaborative protection functions. In conclusion, in complex distribution network environments with high penetration rates of renewable energy, there is an urgent need to construct a new power protection and control system that deeply integrates IoT technology, possesses full-domain perception capabilities, edge intelligent processing, adaptive decision-making, and regional collaborative execution capabilities. This system aims to overcome the bottlenecks of existing technologies in data processing efficiency, protection action accuracy, and fault isolation speed, effectively improving the safety, resilience, and intelligence level of the distribution network. Summary of the Invention

[0003] This invention addresses the shortcomings of existing protection and control technologies for new energy distribution networks, such as low data processing efficiency, low accuracy of protection actions, long fault isolation time, high communication bandwidth pressure, and lack of adaptability to the operating environment. It provides an intelligent power protection and control system and method for new energy distribution networks based on Internet of Things (IoT) technology. The system constructs a closed-loop architecture encompassing "perception—edge—coordination—execution—self-healing," deeply integrating multi-dimensional perception across the entire domain, real-time feature extraction at the edge, adaptive protection decision-making, regional collaborative isolation, and rapid network reconstruction capabilities. This solves the risks of false tripping, failure to tripping, and cascading faults caused by fixed settings, information silos, and response delays in traditional protection strategies under conditions of high-proportion distributed power source access.

[0004] This invention provides an intelligent power protection and control system for new energy distribution networks based on Internet of Things (IoT) technology, comprising five functional layers: a global perception layer, an edge computing layer, a collaborative control layer, a communication network layer, and an application service layer; wherein: The comprehensive sensing layer is deployed at key nodes of the distribution network, including substation outgoing lines, distributed power supply grid connection points, and important load connections, to synchronously collect three types of data: electrical quantities, equipment status, and environmental meteorological data. This layer includes current transformers, voltage transformers, temperature sensors, vibration sensors, and meteorological sensors. The current transformers measure three-phase line current signals with an accuracy class of 0.5 and a sampling frequency set to 4kHz. The voltage transformers monitor the amplitude and phase angle of the three-phase voltage at the nodes, with a measurement error not exceeding ±0.2%. The temperature sensors are surface-mounted on the surface of circuit breakers or transformers, covering a measurement range from -40℃ to 125℃. The vibration sensors are fixed to the main equipment housing and collect mechanical vibration spectra in the 0–2kHz frequency band. The meteorological sensors record light intensity (unit: W / m²) every 5 minutes. 2 The sensor outputs (wind speed, m / s), rainfall (mm / 5min), and ambient temperature (°C) are all uploaded as digital signals after local analog-to-digital conversion.

[0005] Furthermore, the edge computing layer is connected to the global perception layer and consists of multiple edge intelligent terminals. Each terminal integrates a data preprocessing module, a feature extraction module, and an adaptive protection module. After receiving the original sampling sequence, the data preprocessing module first extracts the 50Hz power frequency fundamental component using a second-order Butterworth low-pass filter, with a cutoff frequency set to 60Hz. Then, it uses the db4 wavelet basis function to perform a 5-level wavelet packet decomposition on the filtered signal, reconstructing the 5th level low-frequency coefficients to eliminate high-frequency electromagnetic interference and sensor noise. The feature extraction module performs a Discrete Fourier Transform (DFT) on the preprocessed three-phase current and voltage signals to calculate the fundamental effective value and the 2nd–13th harmonic content. Simultaneously, an improved mathematical morphology filter is used to construct multi-scale structural elements. This filter first processes the signal... Background estimation is obtained by performing open-closed and closed-open combination morphological operations respectively. Then calculate the morphological gradient And by setting a threshold Identify the fault transient traveling wave head, among which For 10 consecutive seconds under normal operating conditions Standard deviation of the sequence The value is set to 3.0; in addition, the feature extraction module also calculates the correlation coefficient of the traveling wave waveform between adjacent monitoring points through a sliding window. When the correlation coefficient is lower than 0.85, it is determined that there is a fault section between the two nodes; the adaptive protection module dynamically adjusts the overcurrent protection setting value based on the real-time extracted current surge, voltage drop rate, power direction and harmonic distortion rate. , directional element action boundary and reclosing delay The execution cycle of the protection logic is strictly controlled within 30 milliseconds.

[0006] Specifically, the adaptive protection module is implemented using a deep reinforcement learning algorithm, and its state space is a 15-dimensional vector, including: the current renewable energy penetration rate (defined as the ratio of total distributed power output to regional load), the average current surge over the previous 5 sampling periods, the average voltage drop rate over the previous 5 sampling periods, the load change gradient (unit: MW / s), the ambient temperature coefficient (normalized to [0,1]), the weather warning level G, and the line load rate. The following data are considered: number of false alarms and failures in 100 historical protection actions; current harmonic distortion (THD); equipment surface temperature; vibration energy entropy; and photovoltaic output correction factor. Wind power output correction factor Action space is defined as the protection parameter adjustment vector. ; ,in This represents the probability of refusal to move within the sliding window. This represents the probability of accidental activation. The time delay from fault occurrence to protection output (in milliseconds); weighting coefficient. , , Values ​​were taken within the ranges [0.5, 0.7], [0.2, 0.4], and [0.05, 0.15] respectively, and then... Cross-validation determines the optimal combination on the historical fault dataset; the trained policy network is quantized by 8 bits and deployed on the ARM Cortex-A72 processor of the edge terminal. The output Δ vector is smoothed by a first-order digital low-pass filter with a cutoff frequency of 10Hz and then superimposed on the basic set value, which is applied to the protection logic judgment unit.

[0007] Furthermore, the collaborative control layer connects to the edge computing layer and includes a regional coordination controller and a policy generation engine. The regional coordination controller synchronizes the clocks of each edge intelligent terminal using the IEEE 1588 Precise Time Protocol (PTP) to ensure that the timestamp error of all uploaded data is less than 1 millisecond. The policy generation engine receives fault characteristic information from multiple edge terminals and integrates the power grid topology and short-term load forecast values. Forecast values ​​of new energy power output and weather warning levels A weighted graph model is constructed; in this model, circuit breakers are abstracted as vertices, power distribution lines are modeled as edges, and edge weights are assigned. ,in Line impedance modulus (unit: ), The current load rate (normalized to [0,1]). The strategy generation engine uses a breadth-first search (BFS) algorithm to traverse the graph structure starting from the two vertices at the ends of the fault section, determine the minimum connected subgraph as the initial isolation range, and verify whether the power flow transfer after isolation causes the current of adjacent lines to exceed 80% of their long-term allowable current carrying capacity. If there is an over-limit, the isolation boundary is expanded until the thermal stability constraint is met.

[0008] Among them, the predicted output value of the new energy source The weather warning level G is dynamically adjusted; G is jointly determined by local meteorological sensors and the provincial meteorological warning platform API, and the maximum value of the two is taken; the local meteorological sensors sample wind speed and rainfall every minute, and output a warning when the wind speed at 10 consecutive sampling points is ≥15m / s or the cumulative rainfall in 24 hours is ≥50mm. Provincial API return ;when When the photovoltaic power output is predicted, it is multiplied by a correction factor of 0.9; when At that time, the predicted wind power output is multiplied by 0.85. Multiply by 0.75; if a provincial API experiences three consecutive HTTP timeouts (timeout threshold of 2 seconds), a 5-minute countdown will begin. If communication is not restored during the countdown, only [the following method will be used]. If local sensor data becomes invalid and more than 10 minutes have passed since the last valid value, then a forced setting will be implemented. And triggers a device self-test alarm.

[0009] Furthermore, the communication network layer spans the global perception layer, edge computing layer, and collaborative control layer, adopting a hierarchical heterogeneous networking architecture. The perception layer and edge layer utilize a LoRa industrial wireless sensor network, operating in the 470–510MHz frequency band, with a communication distance of up to 2 kilometers and a packet loss rate of less than 0.1%. The edge layer and collaborative control layer are connected via gigabit fiber optic Ethernet, with an end-to-end transmission latency of less than 10 milliseconds. The system supports dual protocol stacks of OPC UA and IEC 61850. IEC 61850 uses MMS and GOOSE service mapping to achieve millisecond-level transmission of protection trip commands, while OPC UA is used for non-real-time data interaction and docking with the dispatch master station.

[0010] The application service layer connects to the collaborative control layer, providing a web-based graphical human-machine interface and mobile monitoring applications. The interface displays the distribution network topology coloring status, electrical quantity curves of each node, protection action event list, fault waveform recordings, and meteorological early warning information in real time. It supports remote modification of protection settings, downloading COMTRADE format waveform recording files, configuring edge terminal parameters, and pushing alarm information via SMS or APP when protection anomalies or communication interruptions are detected.

[0011] The present invention also provides an intelligent power protection and control method based on the above system, comprising the following steps: S110: The sensor array deployed at each node of the power distribution network synchronously collects three-phase current, three-phase voltage, equipment surface temperature, 0–2kHz vibration spectrum, light intensity, wind speed, rainfall and ambient temperature. The sampled data is timestamped after local AD conversion. S120: Real-time processing of collected data at the edge intelligent terminal, specifically including: using DFT to calculate the fundamental effective value and the content of the 2nd to 13th harmonics; using an improved mathematical morphology filter to extract the arrival time of the transient traveling wave front of the fault; identifying the fault section by calculating the correlation coefficient (threshold 0.85) of the traveling wave waveforms of adjacent monitoring points; and simultaneously calculating the three-phase differential current, measuring the impedance component, and the instantaneous power direction characteristics. S130: An adaptive protection model is trained based on real-time operating status and historical fault data, dynamically correcting overcurrent protection settings, directional element operating zones, and reclosing delay parameters. The model employs a deep reinforcement learning architecture, with a 15-dimensional feature vector input to the state space and an output to the action space. Reward function middle, and Obtained through historical action statistics within a sliding window. The difference between the actual fault time and the protection exit time is provided by the digital simulation kernel; the model is deployed on the ARM Cortex-A72 processor after being quantized by 8 bits, and the output is applied to the protection logic after being filtered by a 10Hz low-pass filter. S140: When a fault event is detected, a differentiated regional collaborative isolation strategy is generated based on the fault type; for phase-to-phase short-circuit faults, the circuit breaker is directly controlled to trip based on the minimum isolation range determined by the BFS; for single-phase ground faults, the faulty feeder is confirmed by combining the small current line selection results (based on the comparison of zero-sequence current amplitude and phase) before isolation is performed; the tripping command is transmitted through the IEC 61850 GOOSE service, using CRC-32 checksum and a maximum of 3 retransmissions mechanism to ensure reliable delivery of the command; the total time from fault detection to completion of isolation is controlled within 150 milliseconds; S150: After performing fault isolation, initiate a two-stage network reconstruction algorithm; in the first stage, calculate the electrical centroid of each power-depleted island Ω. ,in For nodes Active load, This refers to the voltage amplitude; starting from each connected interconnecting switch. To find the shortest electrical distance path (defined as the sum of the magnitudes of the path impedances) using Dijkstra's algorithm, a candidate set of switching operations is generated. In the second stage, a mixed-integer programming (MIP) model is established within the candidate set, with the objective function being... ,in For nodes A binary variable determining whether to restore power. This represents the total number of switch operations. The constraints include: the network maintains a radial structure. (Branch switch status), voltage deviation at each node Line current After solving the MIP model, the optimal switching operation sequence is issued to ensure that the average power restoration time in non-faulty areas does not exceed 2 seconds. Compared with the prior art, the present invention has the following beneficial effects: This invention performs noise reduction, feature extraction, and local protection decision-making on the edge, only uploading feature quantities and alarm events, reducing communication bandwidth requirements by approximately 60%. The adaptive protection model dynamically adjusts setpoints based on a 15-dimensional state space, achieving a protection action accuracy rate of over 98.5%. The collaborative control layer integrates meteorological, topological, and forecast information to generate a minimum isolation strategy, achieving a fault location accuracy of 99.2% and an isolation time of ≤150 milliseconds. The network reconstruction algorithm restores power supply to non-faulty areas within 2 seconds. The system supports heterogeneous access via multiple protocols such as LoRa, fiber optic, OPC UA, and IEC 61850, and possesses the ability to coordinate and pre-control under extreme weather conditions such as typhoons, making it suitable for complex distribution network scenarios with a high proportion of renewable energy access. Attached Figure Description

[0012] Figure 1This is a schematic diagram of the overall technical solution architecture of a smart power protection and control system for new energy distribution networks based on Internet of Things technology proposed in this invention; Figure 2 This is a schematic diagram of the core principle framework of the adaptive protection model in this invention; Figure 3 This is a flowchart illustrating the data processing and feature extraction logic of the global perception layer and the edge computing layer in this invention. Figure 4 This is a flowchart illustrating the logical process of multi-source information integration and collaborative protection strategy generation in the collaborative control layer of this invention. Figure 5 This is a logical flow diagram of the fault detection and regional collaborative isolation strategy execution in this invention; Figure 6 This is a logical flowchart of the network reconstruction and power restoration process in this invention. Detailed Implementation

[0013] To further illustrate the technical means and effects adopted by the present invention to achieve the intended purpose, the specific embodiments according to the present invention will be described in detail below with reference to the accompanying drawings and preferred embodiments.

[0014] See Figure 1 The system architecture diagram shown illustrates that the intelligent power protection and control system for new energy distribution networks based on Internet of Things technology described in this invention consists of five functional layers: a global perception layer, an edge computing layer, a collaborative control layer, a communication network layer, and an application service layer. In a coastal distribution network demonstration area with a high proportion of distributed photovoltaic and wind power integration, the global perception layer is deployed at the substation outgoing lines, distributed power grid connection points, and important load connections, and includes current transformers, voltage transformers, temperature sensors, vibration sensors, and meteorological sensors. Specifically, the current transformers measure the three-phase line current signals in real time with an accuracy of 0.5, and the sampling frequency is set to 4kHz; the voltage transformers synchronously monitor the three-phase voltage amplitude and phase angle at each node, with the measurement error controlled within ±0.2%; the temperature sensors are patch-mounted on the surface of the circuit breaker operating mechanism housing, covering a measurement range from -40℃ to 125℃; the vibration sensors are fixed to the main transformer housing via magnetic bases, collecting the mechanical vibration spectrum in the 0–2kHz frequency band; and the meteorological sensors record the light intensity (W / m²) every 5 minutes. 2 The data includes wind speed (m / s), rainfall (mm / 5min), and ambient temperature (°C). All sensor data are converted from analog to digital locally, timestamped, and uploaded through the communication network layer.

[0015] like Figure 2As shown, the edge computing layer consists of multiple edge intelligent terminals, each integrating a data preprocessing module, a feature extraction module, and an adaptive protection module. When the raw sampled data enters the data preprocessing module, it first extracts the 50Hz power frequency fundamental component using a second-order Butterworth low-pass filter, with a cutoff frequency set to 60Hz. Then, it uses the db4 wavelet basis function to perform a 5-level wavelet packet decomposition on the filtered signal, reconstructing only the 5th level low-frequency coefficients to eliminate high-frequency electromagnetic interference and sensor noise. The feature extraction module performs a Discrete Fourier Transform (DFT) on the preprocessed three-phase current and voltage signals to calculate the fundamental effective value and the 2nd–13th harmonic content. Simultaneously, it uses an improved mathematical morphology filter to construct multi-scale structural elements: first, the signal… Background estimation is obtained by performing open-closed and closed-open combination morphological operations respectively. Then calculate the morphological gradient and through threshold Identify the fault transient traveling wave head, among which For 10 consecutive seconds under normal operating conditions Standard deviation of the sequence The value is set to 3.0. Furthermore, this module calculates the correlation coefficient of the traveling wave waveform between adjacent monitoring points using a sliding window. When the correlation coefficient is below 0.85, a fault section is determined to exist between the two nodes. The adaptive protection module 204 dynamically adjusts the overcurrent protection setting value based on real-time extracted current surges, voltage dip rates, power direction, and harmonic distortion rates. , directional element action boundary and reclosing delay The execution cycle of the protection logic is strictly controlled within 30 milliseconds.

[0016] Furthermore, such as Figure 2 As shown, the adaptive protection module is implemented using a deep reinforcement learning algorithm. Its state space is a 15-dimensional vector, including: the current renewable energy penetration rate (defined as the ratio of total output of distributed power sources to regional load), the average current surge rate of the previous 5 sampling periods, the average voltage drop rate of the previous 5 sampling periods, the load change gradient (unit: MW / s), the ambient temperature coefficient (normalized to [0,1]), and the weather warning level. Line load rate The following data are considered: number of false alarms and failures in 100 historical protection actions; current harmonic distortion (THD); equipment surface temperature; vibration energy entropy; and photovoltaic output correction factor. Wind power output correction factor ; Action space is defined as the protection parameter adjustment vector. The reward function is designed as follows: ,in This represents the probability of refusal to move within the sliding window. This represents the probability of accidental activation. The time delay from the occurrence of a fault to the protection exit (unit: ms); , , Values ​​were taken within the ranges [0.5, 0.7], [0.2, 0.4], and [0.05, 0.15] respectively, and then... Folded cross-validation determines the optimal combination on a historical failure dataset; Among them, the probability of refusal to move Defined as the ratio of the number of times a fault actually occurs within the sliding window but the protection fails to operate to the total number of faults, the probability of false operation. Defined as the ratio of the number of times the protection malfunctions under fault-free conditions to the total number of operating time windows.

[0017] The trained policy network is quantized by 8 bits and deployed on the ARM Cortex-A72 processor of the edge terminal 201. The output Δ vector is smoothed by a first-order digital low-pass filter with a cutoff frequency of 10Hz and then superimposed on the basic set value, which is used to act on the protection logic judgment unit.

[0018] The deep reinforcement learning model employs a Double Deep Q-Network (DQN) architecture, consisting of a main Q-network and a target Q-network, both using three fully connected neural networks with 128, 64, and 32 neurons per layer, respectively, and ReLU activation function. The training process utilizes an experience replay mechanism, randomly sampling transfer samples in batches of 64 from a historical fault sample database. Perform gradient descent updates.

[0019] like Figure 4 As shown, the collaborative control layer includes a regional coordination controller and a policy generation engine. The regional coordination controller 301 synchronizes the clocks of each edge intelligent terminal 201 via the IEEE 1588 Precise Time Protocol (PTP), ensuring that the timestamp error of all uploaded data is less than 1 millisecond. The policy generation engine receives fault characteristic information from multiple edge terminals, integrating the power grid topology and short-term load forecasts. Forecast values ​​of new energy power output Given a weather warning level G, a weighted graph model is constructed. In this model, circuit breakers are abstracted as vertices, power distribution lines are modeled as edges, and edge weights are assigned. ,,in The impedance modulus of the line (unit: Ω). The current load rate (normalized to [0,1]). The strategy generation engine 302 uses the breadth-first search (BFS) algorithm to traverse the graph structure starting from the two vertices at both ends of the fault section, determine the minimum connected subgraph as the initial isolation range, and verify whether the power flow transfer after isolation causes the current of the adjacent line to exceed 80% of its long-term allowable current carrying capacity. If there is an over-limit, the isolation boundary is expanded until the thermal stability constraint is met.

[0020] Among them, the predicted value of new energy power output The alert level is dynamically adjusted based on weather warning level G; G is jointly determined by local meteorological sensors and the provincial meteorological warning platform API, taking the maximum value of the two; local meteorological sensors sample wind speed and rainfall every minute, and output an alert when the wind speed at 10 consecutive sampling points is ≥15m / s or the cumulative rainfall in 24 hours is ≥50mm. Provincial API return ;when When the photovoltaic power output is predicted, it is multiplied by a correction factor of 0.9; when At that time, the predicted wind power output is multiplied by 0.85. Multiply by 0.75; if a provincial API experiences three consecutive HTTP timeouts (timeout threshold of 2 seconds), a 5-minute countdown will begin. If communication is not restored during the countdown, only [the following method will be used]. If local sensor data becomes invalid and more than 10 minutes have passed since the last valid value, then a forced setting will be implemented. And triggers a device self-test alarm.

[0021] The communication network layer spans the overall perception layer, edge computing layer, and collaborative control layer, adopting a hierarchical heterogeneous networking architecture. The perception layer and edge layer use a LoRa industrial wireless sensor network, operating in the 470–510MHz frequency band, with a communication distance of up to 2 kilometers and a packet loss rate of less than 0.1%. The edge layer and collaborative control layer are connected via gigabit fiber optic Ethernet, with an end-to-end transmission latency of less than 10 milliseconds. The system supports dual protocol stacks of OPC UA and IEC 61850. IEC 61850 uses MMS and GOOSE service mapping to achieve millisecond-level transmission of protection trip commands, while OPC UA is used for non-real-time data interaction and docking with the dispatch master station.

[0022] The application service layer provides a web-based graphical human-machine interface and mobile monitoring applications; the interface displays the distribution network topology coloring status, electrical quantity curves of each node, protection action event list, fault waveform recordings and meteorological warning information in real time; it supports remote modification of protection settings, downloading COMTRADE format waveform recording files, configuring edge terminal parameters, and pushing alarm information via SMS or APP when protection anomalies or communication interruptions are detected.

[0023] Combination Figure 3 , Figure 5 and Figure 6The specific operation flow of the method of the present invention is as follows: In the S110 phase, the sensor array synchronously collects three-phase current, three-phase voltage, equipment surface temperature, 0–2kHz vibration spectrum, light intensity, wind speed, rainfall and ambient temperature. The sampled data is timestamped after local AD conversion. In the S120 stage, the collected data is processed in real time at the edge intelligent terminal 201. Specifically, this includes: using DFT to calculate the fundamental effective value and the content of the 2nd to 13th harmonics; using an improved mathematical morphology filter to extract the arrival time of the fault transient traveling wave front; identifying the fault section by calculating the correlation coefficient (threshold 0.85) of the traveling wave waveforms of adjacent monitoring points; and simultaneously calculating the three-phase differential current, the measured impedance component, and the instantaneous power direction characteristics. In stage S130, an adaptive protection model is trained based on real-time operating status and historical fault data, and protection parameters are dynamically adjusted. The model adopts a deep reinforcement learning architecture, with a 15-dimensional feature vector as input to the state space and an output in the action space. Reward function middle, and Obtained through historical action statistics within a sliding window. The difference between the actual fault time and the protection exit time is provided by the digital simulation kernel; the model is deployed on the ARM Cortex-A72 processor after being quantized by 8 bits, and the output is applied to the protection logic after being filtered by a 10Hz low-pass filter. In the S140 phase, when a fault event is detected, a differentiated regional collaborative isolation strategy is generated based on the fault type. For phase-to-phase short-circuit faults, the circuit breaker is directly controlled to trip based on the minimum isolation range determined by the BFS. For single-phase ground faults, the faulty feeder is confirmed by combining the small current line selection results (based on the comparison of zero-sequence current amplitude and phase) before isolation is performed. The tripping command is transmitted through the IEC 61850 GOOSE service, using CRC-32 checksum and a maximum of 3 retransmissions mechanism to ensure reliable delivery of the command. The total time from fault detection to completion of isolation is controlled within 150 milliseconds. In phase S150, after fault isolation is performed, a two-stage network reconstruction algorithm is initiated; in the first stage, for each power-depleted island Ω, its electrical centroid is calculated. ,in For nodes Active load, This refers to the voltage amplitude; starting from each connected interconnecting switch. To find the shortest electrical distance path (defined as the sum of the magnitudes of the path impedances) using Dijkstra's algorithm, a candidate set of switching operations is generated. In the second stage, a mixed-integer programming (MIP) model is established within the candidate set, with the objective function being... ,in For nodes A binary variable determining whether to restore power. This represents the total number of switch operations. The constraints include: the network maintains a radial structure. (Branch switch status), voltage deviation at each node Line current After solving the MIP model, the optimal switching operation sequence is issued to ensure that the average power restoration time in non-faulty areas does not exceed 2 seconds.

Claims

1. An intelligent power protection measurement and control system for a new energy distribution network based on Internet of Things technology, characterized in that, The global perception layer, the edge computing layer, the collaborative control layer, the communication network layer and the application service layer are included. The global perception layer is deployed at key nodes of the power distribution network and includes current transformers, voltage transformers, temperature sensors, vibration sensors and weather sensors, which are used to synchronously collect three-phase currents, three-phase voltages, device surface temperatures, 0-2 kHz vibration frequency spectrums, light intensities, wind speeds, rainfall amounts and environmental temperatures. The edge computing layer includes a plurality of edge intelligent terminals, each of which integrates a data preprocessing module, a feature extraction module and an adaptive protection module. The collaborative control layer includes a regional coordination controller and a strategy generation engine. The communication network layer connects the global perception layer, the edge computing layer and the collaborative control layer. The application service layer connects the collaborative control layer and provides a graphical human-computer interaction interface and a mobile terminal monitoring application. 2.The IoT technology-based intelligent electric power protection monitoring and control system for new energy power distribution networks according to claim 1, characterized in that, The current transformer has a precision level of 0.5 and a sampling frequency of 4 kHz; the voltage transformer has a measurement error of not more than ±0.2%; the temperature sensor has a measurement range of -40℃ to 125℃; the vibration sensor collects a frequency band of 0-2 kHz; and the weather sensor records the light intensity, the wind speed, the rainfall amount and the environmental temperature every 5 minutes. 3.The IoT technology-based intelligent electric power protection monitoring and control system for new energy power distribution networks according to claim 1, characterized in that, The data preprocessing module is configured to sequentially perform second-order Butterworth low-pass filtering on the original sampling signals with a cutoff frequency of 60 Hz, and perform 5-layer wavelet packet decomposition using a db4 wavelet basis function and reconstruct only the 5th layer low-frequency coefficients. 4.The IoT technology-based intelligent electric power protection monitoring and control system of a new energy power distribution network according to claim 3, characterized in that, The feature extraction module is configured to perform a discrete Fourier transform on the preprocessed three-phase current and voltage signals to calculate fundamental effective values and 2-13 harmonic content rates, and to identify a fault transient traveling wave front using an improved mathematical morphological filter, wherein a morphological gradient , is a background estimation obtained by performing open-close and close-open combined morphological operations on the signal , respectively, and a threshold , is a standard deviation of a continuous 10-second sequence under normal operating conditions. 5.The IoT technology-based intelligent electric power protection monitoring and control system for new energy power distribution networks according to claim 4, characterized in that, The feature extraction module is further configured to calculate the correlation coefficient of the traveling wave waveform between adjacent monitoring points through a sliding window, and determine that there is a fault section between the two nodes when the correlation coefficient is lower than 0.

85. 6.The IoT technology-based intelligent electric power protection monitoring and control system for new energy power distribution networks according to claim 1, characterized in that, The adaptive protection module adopts a deep reinforcement learning algorithm, and a state space thereof is a 15-dimensional vector, including: a new energy penetration rate at a current moment, a mean value of current sudden change variables in the last 5 sampling periods, a mean value of voltage drop rates in the last 5 sampling periods, a load change gradient, an environmental temperature coefficient, a weather warning level , a line load rate , a misoperation frequency and a refusal operation frequency in the last 100 protection operations, a current harmonic distortion rate THD, a device surface temperature, a vibration energy entropy, a photovoltaic output correction coefficient , a wind power output correction coefficient ; and an action space thereof is a protection parameter adjustment vector . 7.The IoT technology-based intelligent electric power protection monitoring and control system for new energy power distribution networks according to claim 6, characterized in that, The reward function of the adaptive protection module is wherein is the probability of misoperation in the sliding window, is the probability of misoperation, is the delay from fault occurrence to protection export; the weight coefficient ; the trained strategy network is deployed on an ARM Cortex-A72 processor after 8-bit integer quantization, and the output delta vector is superimposed on the basic fixed value after being smoothed by a first-order digital low-pass filter with a cutoff frequency of 10 Hz. 8.The IoT technology-based intelligent electric power protection monitoring and control system for new energy power distribution networks according to claim 1, characterized in that, The regional coordinated controller synchronizes the clock of each edge intelligent terminal through IEEE 1588 precision time protocol, ensures that the timestamp error is less than 1 millisecond; the strategy generation engine constructs a weighted graph model, wherein the circuit breaker is a vertex, the power distribution line is an edge, and the edge weight , is the line impedance modulus, is the current load rate normalized to [0, 1], the strategy generation engine adopts a breadth-first search algorithm to determine the minimum connected subgraph as the initial isolation range from the vertices at both ends of the fault section, and checks whether the power flow transfer after isolation leads to the current of the adjacent line exceeding 80% of the long-term allowable load capacity, and if there is an overrun, the isolation boundary is expanded until the thermal stability constraint is met. 9.The intelligent electric power protection monitoring and control system based on new energy power distribution network and Internet of Things technology of claim 1, wherein, In the communication network layer, the LoRa industrial wireless sensor network is used between the global perception layer and the edge computing layer, and the working frequency band is 470-510 MHz; the gigabit optical fiber Ethernet is used between the edge computing layer and the collaborative control layer, and the end-to-end transmission delay is less than 10 milliseconds; the system supports the dual protocol stacks of OPC UA and IEC 61850, and the IEC 61850 uses MMS and GOOSE service mapping.

10. A new energy distribution network intelligent power protection measurement and control method based on Internet of Things technology, characterized in that, The method includes the following steps: S110, collecting multi-dimensional operation data including three-phase currents, three-phase voltages, device temperatures, vibration frequencies and environmental weather parameters through a sensor array deployed at each node of the power distribution network; S120, performing real-time processing on the collected data on the edge side, calculating current differential values, impedance components and power direction features, and detecting fault transient characteristics; S130, training an adaptive protection model based on real-time operation states and historical data, and dynamically correcting overcurrent protection setting values, directional element action zones and reclosing delay parameters; S140, when a fault event is detected, generating a regional collaborative isolation strategy according to fault type and location information, and sending a tripping instruction to a related circuit breaker through a communication network; S150, starting a network reconstruction algorithm after executing fault isolation, calculating an optimal power supply recovery path and controlling the action of sectional switches to realize fast power restoration of non-fault areas.

Images