Mine area power distribution network adaptive protection method and system based on dynamic risk assessment

Abstract

The present application belongs to the technical field of power distribution network protection, and particularly relates to a mine area power distribution network adaptive protection method and system based on dynamic risk assessment. The original current and voltage waveforms of the power distribution network are obtained, and the original impedance characteristics are extracted; based on deep learning technology, the predicted impedance of the power distribution network at a future time is obtained in combination with the original impedance characteristics; a misoperation risk model and a refusal operation risk model are established, and the misoperation risk value and the refusal operation risk value are calculated in combination with the predicted impedance; based on the misoperation risk value and the refusal operation risk value, the risk benefit ratio is calculated, and the dynamic weight of the impedance criterion and the dynamic weight of the current criterion are calculated using the risk benefit ratio; the dynamic weight of the impedance criterion and the dynamic weight of the current criterion obtained are used to weight and fuse the impedance criterion and the current criterion, and the final composite criterion is obtained; based on the composite criterion, adaptive protection and intelligent decision-making are performed on the power distribution network. The present application can realize dynamic adaptation of the protection strategy, and improve the reliability and speed of the protection system.

Description

Technical Field

[0001] This invention belongs to the field of power distribution network protection technology, and particularly relates to an adaptive protection method and system for mining area power distribution networks based on dynamic risk assessment. Background Technology

[0002] The proper configuration of relay protection in the power distribution system of a mining area is crucial for maintaining safe production. Currently, the protection configuration of mining power distribution networks is mainly divided into two categories: current protection and impedance protection, with the traditional three-stage current protection technology being widely adopted. The main technical solutions include: Technical Solution 1: Traditional Three-Stage Current Protection Stage I: Instantaneous overcurrent protection, the action criterion is I>I set.I ; Stage II: Time-limited instantaneous overcurrent protection, the operation criterion is I>I. set.II Fixed delay; Stage III: Time-limit overcurrent protection, operation criterion is I>I set.III , graded delay.

[0003] Where I represents the measured current, I set.I Indicates the current action threshold of segment I; I set.II Indicates the current action threshold of segment II; I set.III This indicates the threshold for the current action of segment III.

[0004] Technical Solution 2: Protection based on impedance principle: Distance protection, action criterion is Z <Z set Where Z represents impedance, Z set This indicates the impedance operating threshold. Since line impedance is proportional to its length, the measured impedance directly reflects the electrical distance from the fault point to the protection installation location.

[0005] Distance protection systems have a preset operating direction discrimination zone (called "operating characteristic"). The system determines whether the measured impedance angle falls within the preset positive operating zone by calculating the measured impedance angle in real time. During a positive fault, the phase difference between the measured voltage and current (i.e., the measured impedance angle) primarily reflects the impedance angle of the line between the fault point and the protection installation point. During a reverse fault, the calculated measured impedance angle falls into a region completely different from the positive direction.

[0006] Based on the above two technical solutions, the existing technical methods have the following main problems and shortcomings: 1. Fixed protection settings and poor adaptability: The protection settings are fixed after being set according to the most stringent conditions, and cannot adapt to changes in the power grid operation mode; 2. Prominent contradictions and difficulties in coordination: The contradiction between sensitivity and reliability is difficult to reconcile; 3. Poor action delay and speed: The II and III protection stages have a fixed delay and cannot achieve full-line speed action; 4. Limited functionality and insufficient intelligence: It relies solely on the current amplitude as a criterion and lacks intelligent decision-making capabilities. Summary of the Invention

[0007] To overcome the shortcomings of the prior art, this invention provides an adaptive protection method and system for mining area power distribution networks based on dynamic risk assessment. This method can adapt to the special and complex operating conditions of mining area power distribution networks, provide intelligent protection for the mining area power distribution networks, and realize dynamic adaptation of protection strategies, thereby significantly improving the reliability and speed of the protection system.

[0008] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions: The first aspect of this invention provides an adaptive protection method for mining area power distribution networks based on dynamic risk assessment.

[0009] An adaptive protection method for mining area power distribution networks based on dynamic risk assessment includes the following steps: Obtain the original current and voltage waveforms of the distribution network and extract the original impedance characteristics; Based on deep learning technology and combined with the original impedance characteristics, the predicted impedance of the distribution network in the future time is obtained; Establish false trigger risk models and false trigger risk models, and calculate the false trigger risk value and false trigger risk value by combining the predicted impedance; Based on the risk values ​​of false tripping and failure to tripping, the risk-benefit ratio is calculated, and the dynamic weights of the impedance criterion and the current criterion are calculated using the risk-benefit ratio. By utilizing the dynamic weights of the impedance criterion and the current criterion, the impedance criterion and the current criterion are weighted and fused to obtain the final composite criterion. Based on composite criteria, adaptive protection and intelligent decision-making are performed on the distribution network.

[0010] The second aspect of the present invention provides an adaptive protection system for mining area power distribution networks based on dynamic risk assessment.

[0011] An adaptive protection system for mining area power distribution networks based on dynamic risk assessment includes: The impedance feature extraction module is configured to: acquire the original current and voltage waveforms of the distribution network and extract the original impedance features; The impedance prediction module is configured to: obtain the predicted impedance of the distribution network in the future time based on deep learning technology and combined with the original impedance characteristics; The risk value calculation module is configured to: establish a false activation risk model and a false activation risk model, and calculate the false activation risk value and the false activation risk value by combining the predicted impedance; The dynamic weight calculation module is configured to: calculate the risk-benefit ratio based on the false trip risk value and the failure to trip risk value, and use the risk-benefit ratio to calculate the dynamic weight of the impedance criterion and the dynamic weight of the current criterion; The dynamic fusion module is configured to: use the dynamic weights of the obtained impedance criterion and the dynamic weights of the current criterion to perform weighted fusion of the impedance criterion and the current criterion to obtain the final composite criterion; The decision-making module is configured to perform adaptive protection and intelligent decision-making for the distribution network based on composite criteria. The above one or more technical solutions have the following beneficial effects: This invention provides an adaptive protection method and system for mining power distribution networks based on dynamic risk assessment. Compared with the technical defects of traditional three-stage current protection, such as rigid setting values, high false trip and failure to trip rates, and slow response speed, this invention effectively solves the fundamental problem that traditional protection cannot adapt to the complex and variable operating conditions of mining power distribution networks due to fixed setting values ​​by introducing deep learning prediction and dynamic risk optimization mechanisms.

[0012] This invention utilizes an LSTM-Attention prediction engine to proactively predict system impedance trends and combines this with a dynamic risk assessment model to quantify the risks of false tripping and failure to trip in real time, thus achieving a shift from passive response to proactive prediction. Measured and simulation data show that this method can reduce the false tripping rate from 2.1% in traditional protection to below 0.5%, the failure to trip rate from 1.8% to below 0.3%, and the average action time from 20.8ms to 12.5ms, achieving a synergistic improvement in the three core indicators of reliability, sensitivity, and speed of action.

[0013] This invention, through risk-benefit ratio decision theory, transforms the irreconcilable contradiction between sensitivity and reliability in traditional protection into a dynamically optimizable mathematical problem. The system can automatically adjust the weighting of protection criteria based on real-time operating conditions: in situations prone to maloperation, such as motor starting, it automatically enhances the risk assessment of maloperation, thereby increasing reliability; in situations prone to failure to operate, such as high-resistance grounding faults, it strengthens the role of impedance criteria, improving sensitivity. This risk-aware adaptive mechanism enables the protection system to make intelligent decisions in response to different operating conditions, completely freeing it from the constraints of fixed values.

[0014] This invention calculates the risk-benefit ratio based on the risk values ​​of false activation and failure to activate, and uses this ratio to calculate the dynamic weights of the impedance criterion and the current criterion. Using these dynamic weights, the impedance and current criteria are weighted and fused to obtain the final composite criterion. Compared to using a single current amplitude criterion, the solution provided by this invention enables more scientific and accurate intelligent decision-making.

[0015] Furthermore, this invention possesses a long-term evolutionary capability not found in traditional protection systems. Through performance monitoring and online optimization, the system can continuously improve predictive models and risk parameters using operational data, enabling protection performance to continuously optimize over time and adapt to gradual changes in the power grid structure. This self-optimizing characteristic not only extends the technical lifespan of the equipment but also significantly reduces the repeated setting costs associated with system upgrades. It is estimated to reduce power outage losses by 500,000 to 1,000,000 RMB per year and prevent equipment damage by 300,000 to 800,000 RMB per year, with an investment recovery period controllable within 2-3 years.

[0016] This invention not only surpasses traditional protection in terms of technical performance, but also achieves a fundamental transformation in system architecture from static protection to dynamic optimization and from fixed strategies to adaptive decision-making, providing a brand-new technical solution for the safe and stable operation of mining power distribution networks.

[0017] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0018] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0019] Figure 1 This is a flowchart of the method in Example 1.

[0020] Figure 2 This is a risk assessment matrix diagram for Example 1.

[0021] Figure 3 This is a flowchart of the intelligent decision-making and criterion fusion process in Example 1.

[0022] Figure 4 This is a timing diagram for task scheduling in Example 1.

[0023] Figure 5 This is a complete signal processing flowchart of the protection device in Embodiment 1. Detailed Implementation

[0024] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0025] It should be noted that the terminology used herein is for the purpose of describing particular implementations only and is not intended to limit the exemplary implementations of the present invention.

[0026] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0027] Example 1 This embodiment discloses an adaptive protection method for mining area power distribution networks based on dynamic risk assessment. Based on LSTM-Attention neural network and dynamic risk optimization, it fundamentally solves the technical bottlenecks of traditional protection methods in complex mining conditions, such as rigid setpoints, prominent contradictions between sensitivity and reliability, and slow response speed.

[0028] like Figure 1 As shown, the adaptive protection method for mining area power distribution networks based on dynamic risk assessment includes the following steps: Obtain the original current and voltage waveforms of the distribution network and extract the original impedance characteristics; Based on deep learning technology and combined with the original impedance characteristics, the predicted impedance of the distribution network in the future time is obtained; Establish false trigger risk models and false trigger risk models, and calculate the false trigger risk value and false trigger risk value by combining the predicted impedance; Based on the risk values ​​of false tripping and failure to tripping, the risk-benefit ratio is calculated, and the dynamic weights of the impedance criterion and the current criterion are calculated using the risk-benefit ratio. By utilizing the dynamic weights of the obtained impedance criterion and current criterion, the impedance criterion and current criterion are weighted and fused to obtain the final composite criterion. Based on composite criteria, adaptive protection and intelligent decision-making are performed on the distribution network.

[0029] To overcome the technical deficiencies of traditional protection methods, this invention provides an intelligent protection method capable of adapting to the complex operating conditions of mining area power distribution networks. The objectives include: (1) To achieve dynamic self-adaptation of protection strategies and eliminate the contradictions caused by fixed protection settings; (2) Significantly reduces the rate of false tripping and failure to trip of protection systems, and improves reliability; (3) Shorten the fault clearing time and improve the protection speed; (4) Enhance the intelligence level of the protection system and enable it to make predictions and decisions.

[0030] The core of this invention is to construct a closed-loop intelligent protection system of "sensing-prediction-evaluation-decision": This system first senses the voltage and current signals of the power grid in real time through high-speed sampling and signal processing, and extracts key features such as impedance; Subsequently, a specially designed LSTM-Attention neural network prediction engine was used to accurately predict the future development trend of system impedance, achieving a leap from "passive response" to "active prediction". Based on this, a dynamic risk assessment system is innovatively introduced. By constructing a quantitative mathematical model of the risk of malfunction and the risk of refusal to operate, and combining it with the real-time status of the system (such as motor start-up and failure), the comprehensive risk of the current decision is calculated. Finally, based on the risk-benefit ratio, the fusion weights of the impedance criterion and the current criterion are dynamically and adaptively adjusted to form the optimal composite protection criterion. This significantly improves the sensitivity and speed of protection while ensuring reliability.

[0031] This embodiment guides current decisions by predicting future states and dynamically optimizes decision-making strategies through risk assessment, forming a new generation of adaptive protection system that can self-perceive, intelligently predict, and dynamically adjust. It is particularly suitable for mining area power distribution networks with frequent load shocks and complex operating conditions.

[0032] Specifically, this embodiment includes the following eight core processing steps: Step 1: High-frequency signal acquisition and synchronous measurement.

[0033] Precise decision-making in protection systems begins with high-quality data sources. In mining power distribution networks, the initiation of impulsive loads and the occurrence of faults trigger complex electromagnetic transient processes containing abundant high-frequency components, which traditional sampling rates struggle to capture.

[0034] Therefore, a 10kHz (kilohertz) high-frequency synchronous sampling technique is adopted, with the core consideration being to achieve the optimal balance between data fidelity and system real-time performance. Although theoretically a 100Hz (hertz) sampling rate is sufficient for power frequency requirements, increasing it to 10kHz (200 points per cycle) can accurately capture transient high-frequency components up to 2kHz, laying a high-resolution data foundation for subsequent in-depth analysis. At the same time, this rate is within the range easily supported by existing hardware technology.

[0035] In this embodiment, the raw data used includes current data and voltage data to facilitate the extraction of impedance characteristics.

[0036] In this embodiment, high-precision sensors deployed on the protected object synchronously collect the instantaneous values ​​of three-phase voltage and three-phase current at key nodes of the mining area's power distribution network. The collected raw voltage and current instantaneous data are immediately transmitted to the subsequent processing module, primarily for the following two key purposes: (1) Real-time feature extraction: The data will be fed into "Step Two: Real-time Impedance Characteristic Calculation Based on Sliding Window". From the original instantaneous value sequence, key characteristic quantities such as voltage, current, and impedance characteristics that can characterize the system's operating state are extracted in real time.

[0037] (2) Constructing a historical sequence: The extracted feature data will be cached in a first-in-first-out (FIFO) historical data buffer, forming a continuous temporal feature sequence. This sequence is the input to "Step Four: LSTM-Attention Prediction Engine" to train and run the prediction model, enabling intelligent prediction of the future state development trend of the system.

[0038] Step 2: Calculation of real-time impedance characteristics based on a sliding window.

[0039] After obtaining the original waveform, it is necessary to extract the characteristic quantity that can characterize the essential state of the system - the fundamental impedance.

[0040] Directly using instantaneous values ​​or simple calculations is susceptible to harmonic and aperiodic component interference, leading to unreliable protection criteria. A sliding window FFT algorithm is employed to calculate the system's fundamental impedance in real time. A 20ms window width (one power frequency cycle) ensures accurate fundamental impedance extraction and effectively filters out interference; while a 10ms step size (half-cycle overlap) guarantees real-time feature updates, preventing the missed detection of rapidly developing faults. Impedance is calculated using Ohm's law. :

[0041] in, and These are the extracted fundamental voltage and current phasors, where R and X are the resistive and reactive components of the impedance. This is the imaginary part.

[0042] The resulting impedance amplitude | Z | and angle θ Z It forms the cornerstone of subsequent intelligent analysis.

[0043] Step 3: Historical data buffering and sequence construction.

[0044] A single moment-in-time feature cannot reflect the dynamic behavior trend of the system. To enable the protection system to have "predictive" capabilities, it must be provided with historical context. This embodiment constructs a continuously updated historical data buffer, storing the feature sequence of the most recent 100ms (corresponding to 10 calculation steps, with 10ms corresponding to one calculation step):

[0045] in, This represents the characteristic sequence of the most recent 100ms; This represents the feature sequence whose step size was most recently calculated. This represents the feature sequence with the second most recent calculation step size; This represents the feature sequence of the tenth most recent calculation step.

[0046] The choice of 100ms is based on observations of typical fault development processes in mining power grids. Common fault development processes or motor starting processes in mining power distribution networks typically exhibit clear evolutionary patterns within this timescale (100ms). This timescale is sufficient to cover the complete process from fault initiation to significant change, providing the necessary temporal context for predictive models to understand the system's "behavioral patterns."

[0047] In this embodiment, the "historical sequence data" specifically refers to a structured, time-ordered sequence of multidimensional feature vectors. Each data point in the sequence... x t Represents a specific moment t The captured multidimensional feature vector, which comprehensively characterizes the operating state of the power system, consists of the core components: the real and imaginary parts of the system impedance and the effective value of the current, calculated in real time from step two, as well as the rate of change of the impedance over time, obtained by differentiating the impedance amplitude. These data points are continuously stored at 10ms intervals.

[0048] Step 4: Prospective computation of the LSTM-Attention prediction engine.

[0049] Traditional protection systems always lag behind fault development, which is the root cause of the contradiction between speed and reliability. This embodiment introduces an LSTM-Attention hybrid prediction model to achieve a leap from passive response to active prediction.

[0050] LSTM (Long Short-Term Memory) networks can effectively learn long-term dependencies in power system time-series data due to their gating mechanism (input gate, forget gate, output gate). They can learn and memorize long-term time-series patterns such as "motor startup is a slow process" and "fault is a rapid drop".

[0051] Building on this, the Attention mechanism is used to address the problem of information focus. It automatically identifies the most critical moments for prediction in the historical sequence (such as the fault initiation point).

[0052] Its core calculations are as follows:

[0053] in, It is a scoring function that calculates a specific moment in a historical sequence. Hidden state With the current (most recent) moment t Hidden state h t The strength of the correlation or association between them h k Represents the loop variable for the summation term. kThis is the index of the loop variable that iterates through all historical moments during the summation process. Through this mechanism, the model can assign high weights to critical historical moments such as the initiation point of a coal mine power grid fault. This generates a context vector that focuses on the most critical information. c .

[0054] The idea behind this step is to allow the model to automatically learn and assign higher weights to the most critical historical moments for predicting the future (such as the precise starting point of a failure), thereby converging into a more informative context vector. This mechanism not only improves the accuracy of impedance prediction but also makes the model's decision-making process transparent and interpretable.

[0055] Specifically, the historical feature sequence from step three is input into the prediction model to generate a predicted value for the system impedance state one calculation step later (i.e., 10ms later), specifically the predicted real and imaginary parts of the impedance. The specific data flow follows the forward propagation process of a neural network: (1) Sequence embedding: The input historical feature sequence is first fed into the LSTM network layer. The LSTM uses its internal gating mechanism (input gate, forget gate, output gate) to perform deep processing on the sequence and output the hidden state corresponding to each time step, thereby capturing the long-term dependencies in the sequence. (2) Attention Focus: Subsequently, the hidden states of all time steps are fed into the multi-head attention layer. This mechanism calculates the correlation between the current final state and each historical state, assigns a dynamic attention weight to each historical time step, and enables the model to automatically focus on the most critical historical moments for predicting the future, and generates a weighted summary context vector.

[0056] (3) Information integration and prediction: The context vector weighted by the attention mechanism contains the most critical information for prediction. It is then further integrated through a global average pooling layer, and finally transformed nonlinearly through a fully connected neural network layer to generate the final output—the real and imaginary parts of the predicted impedance at future time.

[0057] Step 5: Dual dynamic risk assessment and system state adaptation.

[0058] Traditional protection methods lack risk awareness, and their "set values" are rigid, failing to intelligently weigh the pros and cons of actions under complex and ever-changing operating conditions. To quantify the potential risks of "false activation" and "failure to activate" behind protection decisions, and to enable risk assessment to adapt to different system operating conditions, this embodiment establishes a dynamic risk assessment system that precisely quantifies the abstract risks of "false activation" and "failure to activate."

[0059] The system comprises two core models: The erroneous activation risk model is used to assess the risk of protecting against unnecessary risks.

[0060] in, R mis This is the risk value for accidental activation. To predict impedance, Z set The impedance action threshold, k 1 represents the risk factor. S factor This is the system state factor. Indicates the predicted impedance magnitude With protection setting impedance The magnitude of the deviation between the two values ​​is measured to ensure that the measured "deviation" is always a non-negative value, thus guaranteeing the rationality of the model output.

[0061] When predicting impedance The closer to the action threshold Z set The more the protection device is in the "critical ambiguity zone," the greater the possibility of malfunction and the higher the risk of malfunction. R mis It grows exponentially.

[0062] The refusal-to-move risk model is used to quantify the risk of not moving:

[0063] in, R fail To determine the risk value for refusal to move, k 2 represents the risk factor. S factor This is the system state factor.

[0064] When the measured current exceeds the set value, but the measured impedance remains high, exceeding the threshold (e.g., high-resistance grounding), there is a risk of failure to operate. R fail It increases saturatedly with increasing impedance deviation. Among them... k 1. k 2 represents the risk coefficient, the system state factor. S factor This involves injecting the impact of operating conditions into the model, transforming the abstract "operating conditions" into specific risk adjustment coefficients, and achieving a deep coupling between risk assessment and reality.

[0065] In this embodiment, the system state factor S factorThis is a scalar adjustment coefficient used to dynamically quantify the impact of different operating conditions on protection risks. Its core function is to adaptively adjust the assessed values ​​of maloperation risk and failure to operate risk based on the real-time operating status identified by the system (such as normal, motor start, fault, or recovery). This factor is obtained without complex formula calculations; instead, it is based on a logical comparison between a preset event time series (such as motor start time, fault start time, and fault clearing time) and the current time point to determine the real-time status and directly map it to a predefined value. For example: The value is assigned to 1.0 when the system is in a normal state; The value is increased to 1.3 during motor startup to enhance the prevention of malfunction risk; The value drops to 0.7 during a failure to encourage correct action; The recovery period is set to 1.1 to maintain appropriate vigilance, thereby enabling intelligent adaptation of risk assessment and changes in operating conditions.

[0066] Step 6: Optimize and adjust the weights based on the risk-benefit ratio.

[0067] Traditional multi-criteria fusion methods typically use fixed weights, making adaptive optimization impossible when faced with the critical question of "which criterion should be trusted more." To address this issue, the weights of impedance and current criteria in the final decision are dynamically adjusted based on real-time risk assessment results. The core idea behind this step is that protection decisions should be based on a trade-off between risk and benefit.

[0068] This embodiment introduces the concept of Risk-Benefit Ratio (RBR) to dynamically guide weight allocation:

[0069] in B The operational efficiency is defined as the weighted sum of two normalized indices: the degree of current over-limit and the degree of impedance reduction. It is calculated internally based on the real-time electrical quantities of the system.

[0070] in: I and | Z | These represent the currently measured current value and impedance amplitude, respectively; I set and Z set These are preset current and impedance operating thresholds. The first value measures the relative extent to which the current exceeds the threshold; a larger value indicates a more severe overcurrent and a higher urgency for operation. The second value measures the relative extent to which the impedance falls below the threshold; a larger value indicates a more severe impedance drop and a greater likelihood of failure.

[0071] If the current decision carries high risk and low benefit, this embodiment should favor a more conservative and reliable criterion (such as overcurrent); conversely, a more sensitive criterion (such as impedance) can be used. Based on RBR, weights are intelligently allocated: ,

[0072] Its underlying logic is to restrict the weights to the range (0,1) and sum them to 1, and to continuously and smoothly transfer the confidence level according to RBR: When risk is low and benefit is high (RBR→0), trust in accurate impedance criteria ( →1); When the risk is high and the benefit is low (RBR→∞), a reliable current criterion is relied upon. →1).

[0073] in, For the dynamic weights of the impedance criterion, The dynamic weights for the current criterion.

[0074] Furthermore, to suppress severe weight oscillations caused by instantaneous fluctuations in electrical quantities or measurement noise, ensure the smoothness and stability of the protection strategy, and avoid frequent maloperation of the protection device or strategy swaying, a low-pass smoothing filter is introduced: , in, This represents the unfiltered raw impedance weights calculated directly from the latest risk-benefit ratio (RBR) at the current time t. It reflects the system's most direct response to the current instantaneous state. This represents the smoothed impedance weights used at the current time t after low-pass filtering. It is a weighted average of the smoothed weights from the previous time step and the original weights at the current time step. express t The dynamic weights of the impedance criterion optimized at time -1.

[0075] Smoothing coefficient α Taking 0.8 ensures the stability of the strategy and reflects its practicality in engineering.

[0076] Meanwhile, the dynamic weights of the current criterion Synchronous smoothing filtering is performed to obtain the optimized dynamic weights of the current criterion. .

[0077] like Figure 2 The diagram shown is the risk assessment matrix for this embodiment. After predicting the impedance trend: When the predicted impedance shows a downward trend, there is a high risk of failure to operate; in this case, the impedance weight should be increased. When the predicted impedance shows an upward trend, there is a high risk of false tripping; in this case, the current weight should be increased. When the predicted impedance shows a stable trend, it indicates that the system is in a low-risk state, and the weights should be balanced.

[0078] Step 7: Multi-criteria intelligent fusion and decision-making.

[0079] A single criterion has significant limitations, and multiple criteria with fixed logic cannot adapt to all operating conditions, highlighting the trade-off between sensitivity and reliability. The final protection decision requires comprehensive information from multiple sources, such as... Figure 3 The diagram shown illustrates the intelligent decision-making and criterion fusion process in this embodiment. This embodiment integrates multiple protection criteria, optimized by weights, into a unified, optimal composite criterion, and makes the final tripping decision. To avoid simplistic logical judgments, the optimized weights and carefully designed criteria are intelligently fused.

[0080] First, there is the more directional impedance criterion:

[0081] in, As an impedance criterion, This represents the magnitude of the system's fundamental impedance calculated at the current moment; Indicates the impedance action threshold; Indicates impedance angle deviation. This represents the impedance angle deviation, which is the absolute value of the difference between the current system impedance angle and the preset reference direction angle. D factor Represents the directional factor, which is a binary logic judgment result, taking the value of 0 or 1.

[0082] The design of this criterion incorporates three layers of information: The impedance amplitude ratio reflects the fault distance, and the exponential term introduces the impedance angle deviation. Δ θ To identify the nature of the fault (resistive / inductive) and the directionality factor. D factor Ensure selectivity. Directional factor. D factor It is a binary logic judgment result used to ensure that the protection device only operates on forward faults (i.e., the fault point is located on the downstream line side pointed to by the protection device), while reliably blocking reverse faults (the fault point is located on the upstream side of the power supply). It is a switching quantity with a value of 0 or 1. It is the currently measured system impedance angle. With preset reference direction angle The absolute value of the difference, the reference direction angle is a fixed setting value, usually based on the inherent impedance angle of the protected line.

[0083] Combining the normalized current criterion:

[0084] in, As a current criterion, Represents current. This represents the current action threshold. The current criterion adopts a normalized design for easy generalization, ultimately achieving intelligent integration.

[0085] The final composite criterion is formed as follows:

[0086] The final criterion is no longer fixed; it is a composite criterion. C compound This represents the optimal combination of judgments under the current risk environment.

[0087] like Figure 3 As shown, after initiating the decision-making process, the current measurement data is first acquired, and the directional impedance criterion and current criterion are calculated separately. Simultaneously, dynamic weights are acquired, namely the dynamic weights of the optimized impedance criterion and the optimized current criterion. These optimized dynamic weights are then used to weight and fuse the aforementioned impedance and current criteria to obtain a composite criterion. C compound .

[0088] When making decisions, first set a threshold for compound actions. C set (e.g., 0.8), when the calculated composite criterion C compound Greater than the composite action threshold C set When a fault is detected in the distribution network, protective actions are executed to achieve adaptive protection. For example, trip signals are generated, and dynamic events are recorded.

[0089] When the calculated composite criterion C compound Less than the composite action threshold C set During this period, maintain monitoring, update historical events, and finally conclude the decision-making process.

[0090] In this embodiment, a composite action threshold is set. C set It is 0.8. (Using...) C compoundA confidence threshold of >0.8 represents the optimal balance between sensitivity and reliability. It sets a high confidence threshold, ensuring that action is only taken when the probability of a true failure is extremely high. This provides ample safety margin for various measurement noises, prediction errors, and transient fluctuations in the system, thereby fundamentally eliminating false triggering in critical states.

[0091] Step 8: Performance monitoring and online optimization.

[0092] Traditional protection systems, once operational, have fixed performance, unable to learn from operational experience or cope with long-term changes in the power grid structure. This step enables the protection system to continuously learn and self-optimize, adapting to grid changes and constantly improving performance. The core idea is to introduce closed-loop feedback, where the system continuously collects performance data during operation, especially from "edge cases." This data can be used to incrementally train the LSTM-Attention prediction model, fine-tuning its parameters; it can also be used to adjust risk coefficients. k 1. k 2. Online calibration makes risk assessments more closely aligned with actual operating environments. This transforms the system from a device with fixed performance after commissioning into a living entity capable of learning from experience and continuously adapting to changes in the power grid, realizing the long-term value and ultimate goal of adaptive protection.

[0093] like Figure 4 The diagram shown is a task scheduling sequence diagram for this embodiment. It includes several stages: data acquisition, feature calculation, LSTM prediction, risk assessment, weight adjustment, criterion fusion, and decision execution. Figure 4 The leftmost timeline unit is ms (milliseconds): Signal acquisition and ADC conversion are performed within 0-1ms to complete the data acquisition phase. Feature extraction calculation is performed within 1-3ms, completing the feature calculation stage; LSTM-Attention prediction is performed within 3-6ms to complete the LSTM prediction stage; in addition, risk assessment calculation and prediction results input are performed to complete the risk assessment stage. The weights are dynamically adjusted within 6-8ms to complete the weight adjustment phase; at the same time, the criteria are calculated. Fusion decision is made in 8-9ms; The execution status is updated within 9-10ms; Once one monitoring cycle is completed, the monitoring for the next cycle will continue.

[0094] like Figure 5 The diagram shown is a complete signal processing flowchart of the protection device in this embodiment.

[0095] In this embodiment, a current sampling circuit is set on the secondary side of the CT and a voltage sampling circuit is set on the secondary side of the PT. The current sampling circuit and the voltage sampling circuit are connected to an analog multiplexer to achieve high-speed sampling of current and voltage, and analog-to-digital conversion is achieved based on an ADC converter.

[0096] The analog-to-digital conversion result from the ADC converter is then input to the DSP processor. One path sends the result to the host computer via the communication interface for monitoring, while the other path executes the protection algorithm of this embodiment. The process involves signal preprocessing, feature calculation, LSTM prediction, risk assessment, weight adjustment, and criterion fusion. Based on the final composite criterion, it is determined whether to execute a protection action. If a protection action is to be executed, the target relay is output based on the trip logic to perform the circuit breaker disconnection operation.

[0097] Example 2 This embodiment discloses an adaptive protection system for mining area power distribution networks based on dynamic risk assessment.

[0098] An adaptive protection system for mining area power distribution networks based on dynamic risk assessment includes: The impedance feature extraction module is configured to: acquire the original current and voltage waveforms of the distribution network and extract the original impedance features; The impedance prediction module is configured to: obtain the predicted impedance of the distribution network in the future time based on deep learning technology and combined with the original impedance characteristics; The risk value calculation module is configured to: establish a false activation risk model and a false activation risk model, and calculate the false activation risk value and the false activation risk value by combining the predicted impedance; The dynamic weight calculation module is configured to: calculate the risk-benefit ratio based on the false trip risk value and the failure to trip risk value, and use the risk-benefit ratio to calculate the dynamic weight of the impedance criterion and the dynamic weight of the current criterion; The dynamic fusion module is configured to: use the dynamic weights of the obtained impedance criterion and the dynamic weights of the current criterion to perform weighted fusion of the impedance criterion and the current criterion to obtain the final composite criterion; The decision-making module is configured to perform adaptive protection and intelligent decision-making for the distribution network based on composite criteria. Those skilled in the art will understand that the modules or steps of the present invention described above can be implemented using general-purpose computer devices. Optionally, they can be implemented using computer-executable program code, thereby allowing them to be stored in a storage device for execution by a computer device, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. The present invention is not limited to any particular combination of hardware and software.

[0099] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. An adaptive protection method for mining area power distribution networks based on dynamic risk assessment, characterized in that, Includes the following steps: Obtain the original current and voltage waveforms of the distribution network and extract the original impedance characteristics; Based on deep learning technology and combined with the original impedance characteristics, the predicted impedance of the distribution network in the future time is obtained; Establish false trigger risk models and false trigger risk models, and calculate the false trigger risk value and false trigger risk value by combining the predicted impedance; Based on the risk values ​​of false tripping and failure to tripping, the risk-benefit ratio is calculated, and the dynamic weights of the impedance criterion and the current criterion are calculated using the risk-benefit ratio. By utilizing the dynamic weights of the impedance criterion and the current criterion, the impedance criterion and the current criterion are weighted and fused to obtain the final composite criterion. Based on composite criteria, adaptive protection and intelligent decision-making are performed on the distribution network; The erroneous operation risk model is as follows: ; The refusal-to-move risk model is as follows: ; in, R mis This is the risk value for accidental activation. R fail To determine the risk value for refusal to move, To predict impedance, Z set The impedance action threshold, k 1. k 2 represents the risk factor. S factor For system state factors; Indicates the predicted impedance magnitude With protection setting impedance The magnitude of the deviation between the two values ​​is measured to ensure that the measured "deviation" is always a non-negative value, thus guaranteeing the rationality of the model output.

2. The adaptive protection method for mining area power distribution networks based on dynamic risk assessment as described in claim 1, characterized in that, The specific formula for calculating the risk-benefit ratio is as follows: ； Where B represents the action benefit; RBR represents the risk-benefit ratio. The dynamic weights of the impedance criterion and the current criterion are as follows: ， ； in, For the dynamic weights of the impedance criterion, The dynamic weights for the current criterion.

3. The adaptive protection method for mining area power distribution networks based on dynamic risk assessment as described in claim 1, characterized in that, The impedance criterion and current criterion are calculated as follows: ； ； in, As an impedance criterion, Indicates the original impedance magnitude; Indicates the impedance action threshold; Indicates impedance angle deviation. It is the absolute value of the difference between the current system impedance angle and the preset reference direction angle; D factor Represents the directional factor, which is a binary logic judgment result, taking the value of 0 or 1; As a current criterion, Represents current. This indicates the current action threshold.

4. The adaptive protection method for mining area power distribution networks based on dynamic risk assessment as described in claim 2, characterized in that, This also includes optimizing the dynamic weights of the obtained impedance criterion and current criterion using a first-order low-pass smoothing filter, resulting in optimized dynamic weights for the impedance criterion and current criterion. ； in, The dynamic weight of the optimized impedance criterion at time t is represented by α, where α is the smoothing coefficient. This represents the dynamic weight of the optimized impedance criterion at time t-1; Represents the dynamic weights of the impedance criterion; The dynamic weights of the current criterion are simultaneously smoothed and filtered to obtain the optimized dynamic weights of the current criterion. .

5. The adaptive protection method for mining area power distribution networks based on dynamic risk assessment as described in claim 4, characterized in that, The final composite criterion is as follows: ； in, C compound It is a composite criterion; As an impedance criterion, As a current criterion, This represents the dynamic weights of the optimized impedance criterion. This represents the dynamic weight of the optimized current criterion.

6. The adaptive protection method for mining area power distribution networks based on dynamic risk assessment as described in claim 1, characterized in that, Based on composite criteria, adaptive protection and intelligent decision-making for distribution networks are performed, specifically including: Set a threshold for composite actions; When the calculated composite criterion is greater than the composite action threshold, a fault is determined to have occurred in the distribution network, and protective actions are performed on the distribution network to achieve adaptive protection.

7. The adaptive protection method for mining area power distribution networks based on dynamic risk assessment as described in claim 1, characterized in that, Based on deep learning technology and combined with the original impedance characteristics, the predicted impedance of the distribution network in the future is obtained, specifically including: Introduce an LSTM-Attention hybrid prediction model; A continuously updated historical data buffer is constructed to store the historical sequence data of the most recent 100ms, providing the context information required for inference; the historical sequence data is input into the LSTM-Attention hybrid prediction model to obtain the predicted impedance of the distribution network in the future. The historical sequence data is a multi-dimensional feature vector that comprehensively characterizes the operating state of the power system, including: the real and imaginary parts of the system impedance, the effective value of the current, and the rate of change of time obtained by differentiating the impedance amplitude in real time.

8. The adaptive protection method for mining area power distribution networks based on dynamic risk assessment as described in claim 7, characterized in that, The information transfer process of the LSTM-Attention hybrid prediction model is as follows: in, Indicates time The hidden state; h t Indicates the current time t The hidden state; h k Represents the loop variable for the summation term. k The index of the loop variable used to iterate through all historical moments during the summation process; It is a scoring function that calculates the time intervals in the historical sequence. τ Hidden state h τ With the current moment t Hidden state h t The strength of the correlation or association between them; Indicates attention weights. c This represents the context vector.

9. A mine area distribution network adaptive protection system based on dynamic risk assessment, employing the mine area distribution network adaptive protection method based on dynamic risk assessment as described in any one of claims 1-8, characterized in that, include: The impedance feature extraction module is configured to: acquire the original current and voltage waveforms of the distribution network and extract the original impedance features; The impedance prediction module is configured to: obtain the predicted impedance of the distribution network in the future time based on deep learning technology and combined with the original impedance characteristics; The risk value calculation module is configured to: establish a false activation risk model and a false activation risk model, and calculate the false activation risk value and the false activation risk value by combining the predicted impedance; The dynamic weight calculation module is configured to: calculate the risk-benefit ratio based on the false trip risk value and the failure to trip risk value, and use the risk-benefit ratio to calculate the dynamic weight of the impedance criterion and the dynamic weight of the current criterion; The dynamic fusion module is configured to: use the dynamic weights of the obtained impedance criterion and the dynamic weights of the current criterion to perform weighted fusion of the impedance criterion and the current criterion to obtain the final composite criterion; The decision-making module is configured to perform adaptive protection and intelligent decision-making for the distribution network based on composite criteria.

Images