A method and system for predicting reactive power of a DC arc furnace

Abstract

The present invention discloses a method and system for predicting reactive power of a DC arc furnace, which relates to the technical field of load forecasting. The method includes: collecting high-frequency harmonic indicators in the arc-breaking reactive power transition stage, constructing a harmonic indicator library, quantifying the sensitivity of each harmonic reactive power to arc breaking, screening out harmonic sensitivity factors and corresponding harmonic sensitivity vectors; constructing an arc-breaking classification model, mapping the harmonic sensitivity vectors collected in real time to a three-dimensional early warning space, judging whether it is in the arc-breaking intervention space, constructing a harmonic prediction and analysis model, outputting the predicted harmonic sensitivity vector, performing similarity analysis with the arc-breaking intervention space characteristics to obtain harmonic sensitivity similarity, and formulating a compensation strategy. The system includes a harmonic acquisition module, a sensitivity analysis module, a mutation classification module, a similarity recognition module, and an intervention prediction module. The present invention improves the accuracy of reactive power mutation prediction, which is beneficial to reducing power grid disturbances and equipment losses.

Description

Technical Field

[0001] The present invention relates to the technical field of load forecasting, and in particular to a method and system for predicting reactive power of a DC arc furnace. Background Art

[0002] In the modern metallurgical industry, DC arc furnaces, core equipment for smelting high-alloy steel and specialty metals, offer advantages such as high energy conversion efficiency and flexible process control. However, frequent arc interruptions during operation can trigger transient changes in reactive power, leading to grid voltage flicker, accelerated equipment insulation aging, and even unplanned downtime.

[0003] Arc interruption is essentially the transient collapse of the arc plasma channel. The resulting nonlinear harmonic components (primarily the 3rd, 5th, and 7th harmonics) provide early indicators of arc stability. For example, when the plasma ionization level drops suddenly, the 3rd harmonic impedance exhibits a characteristic decrease due to stray capacitance coupling, a change that precedes changes in the fundamental reactive power. Existing fundamental wave analysis methods struggle to capture the characteristic changes in high-frequency harmonics, making it impossible to predict reactive power changes in advance, limiting effective risk management of DC arc furnace operations.

[0004] To this end, the present invention provides a method and system for predicting reactive power of a DC arc furnace. Summary of the Invention

[0005] The object of the present invention is to provide a method and system for predicting reactive power of a DC arc furnace to solve at least one of the above-mentioned problems in the prior art.

[0006] A method for predicting reactive power of a DC arc furnace comprises the following steps:

[0007] Collect high-frequency harmonic indicators during the arc-breaking reactive power transition phase of a DC arc furnace and build a harmonic indicator library;

[0008] Based on the harmonic index library, the sensitivity of each harmonic reactive power to arc failure is quantified. Based on the sensitivity and each harmonic, screening is performed to obtain the harmonic sensitivity factor and the corresponding harmonic sensitivity vector;

[0009] An arc-breaking classification model is constructed to perform three-dimensional classification mapping of arc-breaking reactive power mutations. Harmonic sensitive vectors are collected in real time and input into the arc-breaking classification model to determine whether the reactive power mutation level of the DC arc furnace is within the arc-breaking intervention space. If not, a prediction and analysis signal is generated.

[0010] Based on the received prediction analysis signal, a harmonic prediction analysis model is constructed and input into the real-time harmonic sensitivity vector, and the predicted harmonic sensitivity vector is output. The predicted harmonic sensitivity vector is analyzed for similarity with the harmonic sensitivity vector in the arc interruption intervention space to obtain the harmonic sensitivity similarity;

[0011] Based on the similarity of harmonic sensitivity, a prediction analysis is performed to obtain the approach time of the predicted harmonic sensitivity vector to the arc-breaking intervention space, and a compensation strategy is formulated.

[0012] Furthermore, the high-frequency harmonic indicators include:

[0013] The amplitude mutation rate, phase deviation rate, and harmonic phase jump ratio of the reactive power amplitude of different subharmonic components.

[0014] Furthermore, the harmonic sensitivity factor and the corresponding harmonic sensitivity vector are obtained as follows:

[0015] Extract the amplitude mutation rate, phase deviation rate, and harmonic phase jump ratio of each subharmonic from the harmonic index library to establish a sub-index vector;

[0016] Based on the secondary index vector, the sensitivity of each harmonic reactive power to arc failure is quantified, and the harmonic entropy density of the secondary harmonics corresponding to different indicators is obtained as the sensitivity of each harmonic reactive power to arc failure;

[0017] The sensitivity of each harmonic reactive power to arc failure corresponding to different high-frequency harmonic index categories is sorted to obtain the harmonic sensitivity factor;

[0018] Based on the harmonic sensitivity factor, the index corresponding to the harmonic sensitivity factor is obtained, and the harmonic sensitivity vector is constructed.

[0019] Furthermore, the sensitivity of each harmonic reactive power to arc failure is quantified as follows:

[0020] The Euclidean distance between each harmonic and the high-frequency harmonic index of other harmonics is calculated. The probability distribution of each harmonic index is calculated through the exponential function and the preset kernel bandwidth parameter to obtain the harmonic entropy density.

[0021] Furthermore, the method for judging whether the reactive power mutation level of the DC arc furnace is within the arc breaking intervention space is as follows:

[0022] An arc-breaking classification model is constructed to perform three-dimensional classification mapping of arc-breaking reactive power mutations, establish a three-dimensional early warning space, collect the harmonic sensitive vectors of the DC arc furnace in real time, and perform normalized mapping to the three-dimensional early warning space;

[0023] Compare the coordinates of the mapped three-dimensional warning space to determine whether the coordinates are in the broken arc intervention space.

[0024] Furthermore, the arc-breaking classification model is constructed to perform three-dimensional classification mapping on the arc-breaking reactive power mutation as follows:

[0025] Obtain the harmonic sensitive vectors of historical arc-breaking events and normal operating conditions, perform coordinate conversion, and obtain three-dimensional coordinate points;

[0026] Set hard threshold boundaries based on 3D coordinate points to divide the 3D area;

[0027] Based on the divided three-dimensional areas, three-dimensional warning space modeling is carried out to realize the construction of the arc fault classification model.

[0028] Furthermore, the prediction analysis signal is generated in the following manner:

[0029] If it is not in the arc-breaking intervention space, calculate the proximity ratio of the harmonic sensitive vector in the three-dimensional warning space to the hard threshold boundary, and obtain the mutation proximity ratio of the vector amplitude mutation rate, the deviation proximity ratio of the phase deviation rate, and the jump proximity ratio of the phase jump ratio;

[0030] The mutation proximity ratio, deviation proximity ratio, and jump proximity ratio are summed to obtain the intervention threshold value;

[0031] A comparative analysis is performed based on the intervention threshold to determine whether a predictive analysis signal is generated.

[0032] Furthermore, the harmonic sensitivity similarity is obtained as follows:

[0033] Obtain harmonic sensitive vectors within N monitoring periods and construct a harmonic measurement group;

[0034] A harmonic prediction and analysis model is constructed using a long short-term memory neural network algorithm. The harmonic measurement group is input into the prediction and analysis model to output the harmonic sensitivity vector within the future monitoring period.

[0035] Map the harmonic sensitive vector to the three-dimensional warning space to determine whether it is in the arc-breaking intervention space;

[0036] If not, the intervention threshold is calculated, and whether the intervention threshold is higher than or equal to the preset intervention threshold is determined again;

[0037] If the intervention limit value is again determined to be higher than or equal to the preset intervention limit threshold, harmonic sensitivity similarity analysis is performed to obtain the harmonic sensitivity similarity.

[0038] Furthermore, the harmonic sensitivity similarity analysis is performed as follows:

[0039] Obtain the harmonic sensitivity vector corresponding to the hard threshold boundary that is not in the arc-breaking intervention space and construct a similar benchmark group;

[0040] Obtain harmonic sensitivity vectors within N future monitoring periods and construct similarity comparison groups;

[0041] Calculate the Mahalanobis distance between each harmonic sensitive vector in the similar comparison group and the mean vector of the similar benchmark group;

[0042] The cost matrix is constructed by combining the Mahalanobis distance from each harmonic sensitive vector to the mean vector, and the DTW distance of the cost matrix is calculated, and the DTW distance is used as the harmonic sensitive similarity.

[0043] A DC arc furnace reactive power prediction system includes the following modules:

[0044] Harmonic acquisition module: used to collect high-frequency harmonic indicators during the arc-breaking reactive power transition stage of the DC arc furnace and build a harmonic indicator library;

[0045] Sensitivity analysis module: Based on the harmonic index library, it is used to quantify the sensitivity of each harmonic reactive power to arc failure. Based on the sensitivity and each harmonic, it performs screening and processing to obtain the harmonic sensitivity factor and the corresponding harmonic sensitivity vector;

[0046] Sudden change classification module: This module is used to build an arc-breaking classification model to perform three-dimensional classification mapping of arc-breaking reactive sudden changes. It collects harmonic sensitive vectors in real time and inputs them into the arc-breaking classification model to determine whether the reactive sudden change level of the DC arc furnace is within the arc-breaking intervention space. If not, a prediction and analysis signal is generated.

[0047] Similarity recognition module: Based on the received prediction analysis signal, it is used to build a harmonic prediction analysis model and input the real-time harmonic sensitivity vector, output the predicted harmonic sensitivity vector, and perform similarity analysis between the predicted harmonic sensitivity vector and the harmonic sensitivity vector in the arc interruption intervention space to obtain the harmonic sensitivity similarity;

[0048] Intervention prediction module: Based on the similarity of harmonic sensitivity, prediction analysis is performed to obtain the approach time of the predicted harmonic sensitivity vector to the arc-breaking intervention space, and a compensation strategy is formulated.

[0049] Beneficial effects of the present invention:

[0050] 1. Starting from the high-frequency harmonic characteristics composed of amplitude mutation rate, phase deviation rate, and harmonic phase jump ratio, the reactive power mutation mechanism is analyzed. The early indicators of arc stability contained in nonlinear harmonic components such as multiple harmonics (such as the characteristic decrease of the third harmonic impedance) are used to capture abnormal conditions before arc breaking in advance and improve prediction accuracy.

[0051] 2. By constructing a three-dimensional hierarchical model, the harmonic sensitive vector is mapped to a space with amplitude mutation rate, phase deviation rate, and harmonic phase jump ratio as dimensions, and the space is divided into low-risk area, medium-risk area, and arc-breaking intervention area to achieve real-time judgment and visual warning of reactive power mutation level, and support step-by-step intervention strategy.

[0052] 3. Build a harmonic prediction and analysis model based on the LSTM neural network. Through historical data training, capture the temporal evolution of harmonic characteristics, predict the harmonic sensitivity vectors of future cycles, and calculate the harmonic sensitivity similarity by combining the Mahalanobis distance and DTW distance of similar benchmark groups. The predicted harmonic sensitivity vector approaches the arc-failure intervention space, and compensation strategies (such as dynamic power adjustment and triggering pre-intervention measures) are formulated in advance to reduce grid disturbances (such as voltage flicker) caused by arc failure and equipment losses caused by insulation aging and unplanned downtime. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0054] Figure 1 This is a flow chart of a method for predicting reactive power of a DC arc furnace provided by the present invention;

[0055] Figure 2 This is a flow chart of constructing an arc breaking classification model provided by the present invention;

[0056] Figure 3 is a flow chart of prediction analysis signal generation provided by the present invention;

[0057] Figure 4 The present invention provides a schematic structural diagram of a DC arc furnace reactive power prediction system. DETAILED DESCRIPTION

[0058] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0059] In the modern metallurgical industry, DC arc furnaces, with their efficient energy conversion and flexible process control, have become core equipment for the smelting of high-alloy steel and specialty metals. However, frequent arc interruptions during operation can trigger transient surges in reactive power, leading to grid voltage flicker, accelerated equipment insulation aging, and even unplanned downtime.

[0060] Arc breaking is essentially a transient collapse process of the arc plasma channel. The nonlinear harmonic components it produces (mainly the 3rd, 5th, and 7th harmonics) contain early indicators of arc stability. For example, when the plasma ionization degree drops sharply, the 3rd harmonic impedance will show a characteristic decrease due to stray capacitance coupling, which is earlier than the fundamental reactive power change. Therefore, breaking through the limitations of traditional fundamental wave analysis and analyzing the reactive power mutation mechanism starting from the high-frequency harmonic characteristics will help improve prediction accuracy, reduce grid disturbances during DC arc furnace operation, and improve equipment safety.

[0061] Example 1:

[0062] like Figure 1 As shown, a method for predicting reactive power of a DC arc furnace provided by an embodiment of the present invention specifically includes the following steps:

[0063] Step 1: Collect high-frequency harmonic indicators during the arc-breaking reactive power transition phase of the DC arc furnace and build a harmonic indicator library;

[0064] Among them, the method of collecting the high-frequency harmonic indicators of the DC arc furnace during the arc-breaking reactive power transition stage is:

[0065] Preferably, the voltage and current signals of the DC arc furnace are obtained and an arc furnace work log is generated through a 16-bit high-speed analog-to-digital conversion module in the DC arc furnace;

[0066] Obtaining a monitoring period of the arc-breaking reactive power transition phase, and dividing the monitoring period into a pre-monitoring period, a sudden change transition period, and a post-monitoring period;

[0067] The voltage and current signals of the DC arc furnace's historical pre-arc monitoring period are obtained from the arc furnace work log. The voltage and current signals of the pre-arc monitoring period are subjected to synchronous compression wavelet transform (SWT) to perform time-frequency decomposition of the voltage and current signals.

[0068] For example, the time period of 100ms before the DC arc furnace is broken is used as the pre-monitoring period, and the time period of 200ms after the DC arc furnace is broken is used as the post-monitoring period;

[0069] Extract the reactive power amplitude, phase and time-frequency energy distribution characteristics of different subharmonic components after time-frequency decomposition, and calculate the amplitude mutation rate, phase deviation rate and harmonic phase jump ratio of the reactive power amplitude of different subharmonic components;

[0070] The amplitude mutation rate, phase deviation rate and harmonic phase jump ratio are used as high-frequency harmonic indicators;

[0071] A harmonic index library is constructed, which includes the amplitude mutation rate, phase deviation rate, and harmonic phase jump ratio of the reactive power amplitude of harmonic components of different sub-orders.

[0072] It can be understood that the purpose of building a harmonic indicator library is:

[0073] Function 1: Provide core data support for quantifying the sensitivity of each harmonic reactive power to arc failure. By collecting high-frequency harmonic indicators during the arc failure reactive power transition phase and storing them in a structured manner, a historical data set containing multi-dimensional features is formed. The various indicators of each subharmonic are extracted from the harmonic indicator library to establish a sub-indicator vector, and the sensitivity of each harmonic to arc failure is further evaluated. In this way, the harmonic sensitivity factors and corresponding feature vectors that respond significantly to arc failure events are screened out, laying a data foundation for subsequent sensitive feature analysis.

[0074] Function 2: Provide historical samples and three-dimensional mapping basis for building arc fault classification models. The harmonic index library stores historical arc fault events and harmonic sensitivity vectors of normal operating conditions, which can be converted into three-dimensional coordinate points through coordinate transformation. These points are used to divide the three-dimensional warning space, providing data support for visual warning of arc fault risks and step-by-step intervention strategies.

[0075] Function 3: Provide time-series training samples for harmonic prediction and analysis models, supporting future trend forecasting. The high-frequency harmonic indicator sequences recorded by monitoring period in the harmonic indicator library serve as training data for the long-short-term memory (LSTM) neural network. By learning the temporal evolution of harmonic characteristics in historical data, a model capable of predicting harmonic sensitivity vectors within future monitoring periods is constructed. The predicted results can be further analyzed for similarity with the arc-failure intervention space characteristics to quantify the proximity between the current operating conditions and the pre-arc-failure state. This allows the predicted time at which the harmonic sensitivity vector approaches the arc-failure intervention space to be predicted and a dynamic compensation strategy to be formulated, improving the foresight and accuracy of reactive power mutation predictions.

[0076] Step 2: Based on the harmonic index library, quantify the sensitivity of each harmonic reactive power to arc failure, and perform screening based on the sensitivity and each harmonic to obtain the harmonic sensitivity factor and the corresponding harmonic sensitivity vector;

[0077] Among them, the method of quantifying the sensitivity of each harmonic reactive power to arc failure is:

[0078] Extract the amplitude mutation rate, phase deviation rate, and harmonic phase jump ratio of each subharmonic from the harmonic index library and establish the sub-index vector XL h ;

[0079] For example, for the hth harmonic, the secondary index vector is ,in 、 、 is the amplitude mutation rate, phase deviation rate, and harmonic phase jump ratio of the hth harmonic, h is the number of the harmonic suborder, and k is the category of the high-frequency harmonic index of the sub-index vector;

[0080] For example, k=1 represents the amplitude mutation rate of the secondary index vector, k=2 represents the phase deviation rate, and k=3 represents the harmonic phase jump ratio;

[0081] By formula: , get the Euclidean distance of k types of high-frequency harmonic indicators from the hth harmonic to other i-th harmonics ;

[0082] It should be noted that the other i-th harmonics do not include the h-th harmonic itself;

[0083] Among them, the value range of i is [1,H], H is the highest harmonic;

[0084] By formula: Get the harmonic entropy density S of k-type indicators H ;

[0085] in, , is the preset kernel bandwidth parameter;

[0086] The harmonic entropy density of the subharmonics corresponding to different indicators is used as the sensitivity of each harmonic reactive power to arc failure;

[0087] Among them, the categories of each harmonic and high-frequency harmonic index are screened based on sensitivity, and the method of obtaining the harmonic sensitivity index is as follows:

[0088] Obtain the sensitivity of each harmonic reactive power to arc failure corresponding to different high-frequency harmonic index categories, and sort the harmonics of each high-frequency harmonic index category in descending order according to the sensitivity value;

[0089] Select the harmonic sub-order with the highest index sensitivity value as the harmonic sensitivity factor;

[0090] It can be understood that since there are three types of high-frequency harmonic indicators, there are three types of harmonic sensitivity factors;

[0091] Based on the harmonic sensitivity factor, the high-frequency harmonic index corresponding to the harmonic sensitivity factor is obtained, and the harmonic sensitivity vector is constructed.

[0092] It needs to be explained that the purpose of constructing the harmonic sensitivity vector is:

[0093] Function 1: Provide real-time input features for the arc-breaking classification model to achieve three-dimensional mapping and risk warning of reactive power mutation levels. By extracting the high-frequency harmonic index corresponding to the harmonic sensitivity factor to construct a vector, it is normalized and mapped to a three-dimensional warning space with three-dimensional indicators as the dimension. Real-time comparison is performed with the coordinate division area of historical arc-breaking events and normal operating conditions to determine whether the reactive power mutation level of the current DC arc furnace has reached the arc-breaking intervention threshold, providing an intuitive spatial coordinate basis for immediately triggering intervention measures or generating predictive analysis signals;

[0094] Function 2: Provide time series feature input for the harmonic prediction and analysis model, supporting advanced prediction of the arc-failure intervention space state. Real-time harmonic sensitivity vectors are used as components of historical time series data and input into a prediction model constructed using a long short-term memory (LSTM) neural network. By learning the dynamic evolution of harmonic characteristics, a prediction vector for the future monitoring period is output. This prediction vector is further analyzed for similarity with the arc-failure intervention space characteristics to quantify the evolution trend of the current operating condition toward the arc-failure state. This allows the predicted time when the harmonic sensitivity vector approaches the arc-failure intervention space to be predicted and a targeted compensation strategy to be formulated, enabling advanced prediction and proactive control of reactive power mutations.

[0095] The technical solution of this embodiment is as follows: high-frequency harmonic indicators are collected during the reactive power transition phase of a DC arc furnace during arc breaking to construct a harmonic indicator library; based on the harmonic indicator library, the sensitivity of each harmonic reactive power to arc breaking is quantified, and screening and processing are performed based on the sensitivity and each harmonic to obtain a harmonic sensitivity factor and a corresponding harmonic sensitivity vector; real-time judgment and visual early warning of the reactive power mutation level are achieved, supporting a step-by-step intervention strategy.

[0096] Example 2:

[0097] like Figure 1 As shown, a method for predicting reactive power of a DC arc furnace further includes the following steps:

[0098] Step 3: Construct an arc-breaking classification model to perform three-dimensional classification mapping of arc-breaking reactive power mutations. Real-time acquisition of harmonic sensitive vectors is input into the arc-breaking classification model to determine whether the reactive power mutation level of the DC arc furnace is within the arc-breaking intervention space. If not, a prediction and analysis signal is generated.

[0099] Among them, the method of constructing the arc-breaking classification model to perform three-dimensional classification mapping of arc-breaking reactive power mutation is as follows:

[0100] Based on the amplitude mutation rate, phase deviation rate, and harmonic phase jump ratio, they correspond to the three dimensions of the three-dimensional warning space respectively;

[0101] Among them, the X-axis is the amplitude mutation rate of the sensitive harmonic sub-order, which reflects the degree of mutation of the reactive power amplitude;

[0102] The Y-axis is the phase deviation rate of the sensitive harmonic secondary, which reflects the degree of deviation of the harmonic phase from the normal state;

[0103] The Z axis is the harmonic phase jump ratio of the sensitive harmonic suborder, reflecting the frequency of phase mutation;

[0104] Convert the harmonic sensitivity vectors (including the values of three types of indicators) collected in real time into coordinate points in three-dimensional space;

[0105] like Figure 2 As shown in Figure 2, the way to construct the arc breaking classification model is:

[0106] A1. Obtain the harmonic sensitive vectors of historical arc-breaking events and normal operating conditions, perform coordinate conversion, and obtain three-dimensional coordinate points;

[0107] Obtain the historical harmonic sensitive vectors of the pre-monitoring period of the historical arc-breaking event, and convert the amplitude mutation rate, phase deviation rate, and harmonic phase jump ratio of the historical harmonic sensitive vectors into three-dimensional coordinate points;

[0108] Obtain the amplitude mutation rate, phase deviation rate, and harmonic phase jump ratio under normal operating conditions of a DC arc furnace and convert them into three-dimensional coordinate points;

[0109] A2. Setting hard threshold boundaries based on 3D coordinate points to divide the 3D area;

[0110] By statistically analyzing the three-dimensional coordinate points of historical arc-breaking events and normal operating conditions, three-dimensional areas with different risk levels are divided;

[0111] Preferably, the three-dimensional mean vector (μ x ,μᵧ,μz) and covariance matrix, a basic safety sphere with a radius of 1 times the standard deviation is constructed with the mean as the center as the low-risk area;

[0112] Analyze the distribution characteristics of historical coordinate points before arc failure, and define a circular medium-risk space of 1-2 times the standard deviation and a peripheral high-risk space of 2 times the standard deviation on the periphery of the safety sphere, as well as an arc failure intervention space;

[0113] Set a hard threshold boundary and mark the area that meets the hard threshold boundary as the arc-breaking intervention space;

[0114] For example, the hard threshold boundaries may be an amplitude mutation rate > 30%, a phase deviation rate > 15°, and a phase jump ratio > 12%;

[0115] A3. Based on the divided three-dimensional areas, three-dimensional warning space modeling is carried out to build a fault arc classification model;

[0116] Preferably, the divided three-dimensional area is converted into an executable real-time warning model. First, a three-dimensional coordinate mapping module is created in the industrial control system. The amplitude mutation rate, phase deviation rate, and phase jump ratio of the harmonic sensitive vector collected in real time are normalized and then projected into the three-dimensional warning space.

[0117] It can be understood that the physical significance of constructing a three-dimensional warning space is to achieve visual classification and real-time warning of reactive power mutation risks;

[0118] Using amplitude mutation rate, phase deviation rate, and harmonic phase jump ratio as three-dimensional coordinate dimensions, the harmonic sensitivity vector of the DC arc furnace is mapped to a concrete three-dimensional space. By dividing the space into low-risk areas, medium-risk areas, and arc-failure intervention areas, the abstract reactive power mutation risk is converted into an intuitively observable coordinate position relationship, realizing a "spatial" early warning of arc-failure risk.

[0119] Establish quantitative judgment benchmarks and logical chains for arc interruption intervention;

[0120] Three-dimensional coordinate points are constructed using harmonic sensitivity vectors from historical arcing events and normal operating conditions. Hard threshold boundaries are set based on statistical analysis, and the criteria for arcing intervention are converted into geometric boundaries in the coordinate space. When a real-time mapped coordinate point enters the arcing intervention space, intervention measures are directly triggered. If not, the intervention threshold value is generated by calculating the percentage of the coordinate point's proximity to the hard threshold boundary. This provides a quantitative basis for subsequent predictive analysis signal generation and critical time prediction, forming a standardized risk assessment logic that integrates spatial mapping, boundary comparison, and threshold judgment.

[0121] like Figure 3 As shown, the harmonic sensitive vector of the DC arc furnace is collected in real time, and normalized and mapped to the three-dimensional warning space to obtain the coordinates of the harmonic sensitive vector mapped to the three-dimensional warning space;

[0122] Based on the coordinates of the harmonic sensitive vector mapped to the three-dimensional warning space, it is judged whether the coordinates are in the arc-breaking intervention space. If so, arc-breaking mutation measures are taken to intervene;

[0123] It should be noted that when a DC arc furnace's reactive power fluctuation level is determined to be within the arc-breaking intervention range, the system immediately triggers arc-breaking intervention measures. Specifically, these measures include: real-time adjustment of reactive power output through dynamic reactive compensation devices, rapidly suppressing transient reactive power fluctuations caused by arc breaking, stabilizing the grid voltage and reducing voltage flicker; At the same time, the system automatically adjusts the DC arc furnace's power control parameters (such as current and voltage setpoints) or triggers arc stabilization control strategies (such as optimizing electrode lifting speeds and adjusting melting process parameters) to enhance the stability of the arc plasma channel and prevent further deterioration of the arc-breaking event. These intervention measures aim to limit reactive power fluctuations to a safe range through real-time closed-loop control, reduce equipment insulation aging losses, prevent unplanned downtime, and achieve proactive management and control of DC arc furnace operational risks.

[0124] If it is not in the arc-breaking intervention space, calculate the proximity ratio of the harmonic sensitive vector in the three-dimensional warning space to the hard threshold boundary, and obtain the mutation proximity ratio of the vector amplitude mutation rate, the deviation proximity ratio of the phase deviation rate, and the jump proximity ratio of the phase jump ratio;

[0125] It should be explained that the proximity ratio is obtained by calculating the ratio of the distance from each indicator in the harmonic sensitivity vector to the corresponding hard threshold boundary;

[0126] The mutation proximity ratio, deviation proximity ratio, and jump proximity ratio are summed to obtain the intervention threshold value;

[0127] The physical meaning of the intervention limit value is to quantify the potential degree of the current harmonic characteristics tending towards the arc-breaking risk, which serves as a key intermediate indicator for determining whether to start the predictive analysis process. The intervention limit value calculates the mutation proximity ratio, deviation proximity ratio, and jump proximity ratio of the harmonic sensitive vector in the three-dimensional warning space to the hard threshold boundary, and linearly sums the trend of each dimensional indicator in the three-dimensional space deviating from the normal state to form a quantitative value that comprehensively reflects the "degree of approach" of the current working condition to the arc-breaking intervention space. The larger the intervention limit value, the more significant the pre-arcing characteristics such as amplitude mutation, phase deviation, and phase jump, and the closer the system is to the critical state requiring intervention.

[0128] Perform comparative analysis based on intervention thresholds to determine whether a predictive analysis signal is generated;

[0129] If the intervention definition value is higher than or equal to the preset intervention definition threshold, a predictive analysis signal is generated;

[0130] If the intervention limit value is lower than the preset intervention limit threshold, the amplitude mutation rate, phase deviation rate and harmonic phase jump ratio of the DC arc furnace are continuously monitored;

[0131] Step 4: Based on the received prediction analysis signal, a harmonic prediction analysis model is constructed and the real-time harmonic sensitivity vector is input. The predicted harmonic sensitivity vector is output, and similarity analysis is performed between the predicted harmonic sensitivity vector and the harmonic sensitivity vector of the arc interruption intervention space to obtain the harmonic sensitivity similarity;

[0132] Among them, the method of constructing the harmonic prediction analysis model is:

[0133] Based on the duration of the pre-monitoring period, a monitoring period is established, harmonic sensitivity vectors within N monitoring periods are obtained, and a harmonic measurement group is constructed;

[0134] Preferably, N≥10;

[0135] A harmonic prediction and analysis model is constructed using a long short-term memory neural network algorithm. The harmonic measurement group is input into the prediction and analysis model to output the harmonic sensitivity vector within the future monitoring period.

[0136] It will be understood by those skilled in the art that when a long short-term memory neural network (LSTM) is used to construct a harmonic prediction and analysis model, a three-dimensional harmonic sensitivity vector containing N monitoring periods, such as a harmonic measurement group of [N, 3] dimensions, is converted into the three-dimensional tensor [number of samples, time step, number of features] format required by the LSTM, where the time step is N and the number of features is 3 (corresponding to the amplitude mutation rate, phase deviation rate, and phase jump ratio);

[0137] The harmonic prediction and analysis model structure usually includes: an input layer to receive sequence data, an LSTM layer (1-2 layers, 64-128 hidden units) to capture the long-term dependence of the harmonic characteristic evolution trend in the time series before the sudden change of the arc furnace reactive power;

[0138] The fully connected layer (Dense layer) outputs a prediction vector (the future value of the three indicators) of the same dimension as the input through a linear activation function;

[0139] When training the harmonic prediction analysis model, the vectors of N consecutive monitoring cycles in the historical data are used as input, and the vector of the next cycle is used as the label. The model parameters are optimized through gradient descent to minimize the mean square error between the predicted value and the actual value.

[0140] Among them, the harmonic measurement group obtains the harmonic sensitivity vector in the future monitoring period through the harmonic prediction analysis model as follows:

[0141] When the real-time harmonic measurement set is input into the trained model, the LSTM memory unit processes the harmonic features of each cycle step by step, using the "forget gate," "input gate," and "output gate" mechanisms to filter the amplitude mutation rate fluctuation pattern of the previous 10 cycles. Combined with the latest features of the current input, a linear transformation is performed at the fully connected layer to generate a three-dimensional harmonic sensitivity vector prediction value for the next monitoring cycle, enabling time-series prediction of the dynamic evolution of harmonic features.

[0142] For example, when a three-dimensional sensitive vector containing 10 monitoring cycles is collected in real time to form a [10,3]-dimensional harmonic measurement group and input into the trained LSTM model, the memory unit analyzes the amplitude mutation rate, phase deviation rate, and phase jump ratio of each cycle time step by time:

[0143] First, the "forget gate" is used to filter out historical fluctuations in the first nine cycles that are not related to the current working conditions (such as accidental amplitude fluctuations that are not arc-breaking), and only the key patterns that continuously affect the current state are retained (such as a continuous increase of 15% in the amplitude mutation rate in the past three cycles);

[0144] The "input gate" integrates the characteristics of the latest cycle (such as a sudden increase of 8° in the phase deviation rate) with the historical information in the memory cell, updating the internal state to strengthen the feature association related to the arc break;

[0145] The "output gate" generates hidden layer outputs based on the updated memory state. After linear transformation by the fully connected layer, it outputs a three-dimensional prediction vector for the next monitoring cycle. For example, based on the trend of the amplitude mutation rate fluctuating between 5% and 20% in the first 10 cycles and continuously increasing in the past three cycles, it predicts that the amplitude mutation rate in the next cycle will be 25%, the phase deviation rate will be 18°, and the phase jump ratio will be 14%. This quantifies the dynamic evolution trend of the harmonic characteristics and provides a time series prediction basis for arc fault risk assessment.

[0146] Obtain the harmonic sensitivity vectors within the future monitoring period output by the harmonic prediction and analysis model, map them to the three-dimensional warning space, and determine whether they are in the arc-breaking intervention space;

[0147] If it is in the arc-breaking intervention space, then formulate arc-breaking reactive power mutation intervention measures in advance; if it is not, calculate the intervention limit value and judge again whether the intervention limit value is higher than or equal to the preset intervention limit threshold;

[0148] If the intervention limit value is again judged to be higher than or equal to the preset intervention limit threshold, harmonic sensitivity similarity analysis is performed;

[0149] The method for performing similarity analysis between the predicted harmonic sensitivity vector and the harmonic sensitivity vector of the arc-breaking intervention space is as follows:

[0150] Extract the harmonic sensitivity vectors corresponding to the hard threshold boundaries of the arc-breaking intervention space that were not within the N historical monitoring periods before the historical arc-breaking event from the arc furnace work log, and construct a similar benchmark group B;

[0151] From the harmonic prediction analysis model, obtain the harmonic sensitivity vectors in N future monitoring periods and construct a similarity comparison group P;

[0152] Calculate the mean vector and covariance matrix of the similar benchmark group, and calculate the Mahalanobis distance from each harmonic sensitive vector to the mean vector in the similar comparison group based on the mean vector and covariance matrix of the similar benchmark group;

[0153] The cost matrix is constructed by combining the Mahalanobis distance from each harmonic sensitive vector to the mean vector, and the dynamic time warping (DTW) distance of the cost matrix is calculated. The DTW distance is used as the harmonic sensitive similarity.

[0154] Those skilled in the art will understand that, the harmonic sensitive vectors corresponding to the hard threshold boundaries that are not in the arc-breaking intervention space are obtained to construct a similarity benchmark group, the mean vector and covariance matrix of the similarity benchmark are calculated, the Mahalanobis distance of each vector in the similarity comparison group to the similarity benchmark group is calculated, the Mahalanobis distance of each vector to the mean vector is obtained, and a cost matrix is constructed with the Mahalanobis distance as an element; then, the cost matrix is processed using the dynamic time warping (DTW) algorithm, and the DTW distance is obtained by finding the optimal time alignment path of the two groups of vector sequences and accumulating the Mahalanobis distances on the path. The smaller the DTW distance, the higher the similarity between the predicted harmonic sensitive vector and the arc-breaking intervention space feature, and this is used as the harmonic sensitivity similarity measure to quantify the degree of proximity of the current operating condition to the arc-breaking state.

[0155] Step 5: Perform prediction analysis based on harmonic sensitivity similarity to obtain the predicted approach time of the harmonic sensitivity vector to the arc-breaking intervention space, and formulate a compensation strategy;

[0156] Among them, the fundamental harmonic sensitivity similarity is predicted and analyzed, and the predicted harmonic sensitivity vector tends to approach the arc-breaking intervention space in the following way:

[0157] Preferably, the harmonic sensitivity similarity of each cycle in the similar comparison group and the similar benchmark group is calculated to construct a time series prediction sequence S(t);

[0158] The time series prediction sequence is used to predict the approach time of the predicted harmonic sensitive vector to the arc-breaking intervention space through the long short-term memory neural network algorithm;

[0159] It can be understood by those skilled in the art that when the time series prediction sequence is input into the LSTM time series prediction model, the model captures the evolution of similarity over time through memory units (such as the accelerated growth trend of the time series prediction sequence before arc breaking);

[0160] When training the LSTM time series prediction model, a sliding window is used to intercept sequence fragments (such as the similarity of the previous w periods) as input to predict the time series prediction sequence S(t+1) of the next period. The loss function is the mean square error (MSE);

[0161] In the prediction phase of the LSTM time series prediction model, the trained intervention threshold LSTM intervention threshold is used to perform rolling predictions on the current sequence (after each prediction, the input window is updated and new prediction values are added), generating similarity prediction values for future cycles in real time.

[0162] Obtain the harmonic sensitive vector of the period corresponding to the similarity prediction value and calculate the intervention threshold value. When the intervention threshold value is higher than or equal to the preset intervention threshold value, the periodic time when the predicted harmonic sensitive vector approaches the arc-breaking intervention space is obtained;

[0163] A compensation strategy is formulated based on the predicted approach time of the harmonic sensitive vector to the arc-breaking intervention space.

[0164] It needs to be explained that, by analyzing the time series prediction sequence of harmonic sensitivity similarity through the long short-term memory neural network algorithm, the cycle time when the intervention limit value is higher than or equal to the preset threshold is determined as the approach time of the predicted harmonic sensitivity vector tending to the arc breaking intervention space. Based on this, the power control parameters of the DC arc furnace (such as current and voltage setting values) are dynamically adjusted in advance, and the dynamic reactive power compensation device is switched on and off in real time to balance the reactive power fluctuations. At the same time, the arc stability pre-intervention measures (such as optimizing the electrode lifting and lowering rates, adjusting the smelting process parameters) are triggered to actively suppress the reactive power mutation before the arc breaking risk reaches a critical state, thereby reducing the risk of grid voltage flicker, equipment insulation aging and unplanned shutdown, and realizing forward-looking regulation of the DC arc furnace operation status.

[0165] The technical solution of this embodiment is as follows: an arc-breaking classification model is constructed to perform three-dimensional classification mapping of arc-breaking reactive power mutations, harmonic sensitivity vectors are collected in real time and input into the arc-breaking classification model to determine whether the reactive power mutation level of the DC arc furnace is within the arc-breaking intervention space. If not, a prediction and analysis signal is generated; based on the received prediction and analysis signal, a harmonic prediction and analysis model is constructed and input into the real-time harmonic sensitivity vector, a predicted harmonic sensitivity vector is output, and similarity analysis is performed between the predicted harmonic sensitivity vector and the harmonic sensitivity vector in the arc-breaking intervention space to obtain harmonic sensitivity similarity; prediction and analysis are performed based on the harmonic sensitivity similarity to obtain the predicted approach time of the harmonic sensitivity vector to the arc-breaking intervention space, and a compensation strategy is formulated; the predicted approach time of the harmonic sensitivity vector to the arc-breaking intervention space is predicted, and a compensation strategy (such as dynamic power adjustment and triggering of pre-intervention measures) is formulated in advance to reduce grid disturbances (such as voltage flicker) caused by arc breaking and equipment losses caused by insulation aging and unplanned downtime.

[0166] Example 3:

[0167] like Figure 4 As shown, a DC arc furnace reactive power prediction system includes the following modules:

[0168] Harmonic acquisition module: used to collect high-frequency harmonic indicators during the arc-breaking reactive power transition stage of the DC arc furnace and build a harmonic indicator library;

[0169] Sensitivity analysis module: Based on the harmonic index library, it is used to quantify the sensitivity of each harmonic reactive power to arc failure. Based on the sensitivity and each harmonic, it performs screening and processing to obtain the harmonic sensitivity factor and the corresponding harmonic sensitivity vector;

[0170] Sudden change classification module: This module is used to build an arc-breaking classification model to perform three-dimensional classification mapping of arc-breaking reactive sudden changes. It collects harmonic sensitive vectors in real time and inputs them into the arc-breaking classification model to determine whether the reactive sudden change level of the DC arc furnace is within the arc-breaking intervention space. If not, a prediction and analysis signal is generated.

[0171] Similarity recognition module: Based on the received prediction analysis signal, it is used to build a harmonic prediction analysis model and input the real-time harmonic sensitivity vector, output the predicted harmonic sensitivity vector, and perform similarity analysis between the predicted harmonic sensitivity vector and the harmonic sensitivity vector in the arc interruption intervention space to obtain the harmonic sensitivity similarity;

[0172] Intervention prediction module: performs prediction analysis based on harmonic sensitivity similarity to obtain the predicted time when the harmonic sensitivity vector approaches the arc-breaking intervention space, and formulates compensation strategies;

[0173] The above is a detailed description of an embodiment of the present invention. However, the content described is only a preferred embodiment of the present invention and should not be considered to limit the scope of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A method for predicting reactive power of a DC arc furnace, characterized in that: The following steps are involved: Collect high-frequency harmonic indicators during the arc-breaking reactive power transition phase of a DC arc furnace and build a harmonic indicator library; Based on the harmonic index library, the sensitivity of each harmonic reactive power to arc failure is quantified. Based on the sensitivity and each harmonic, screening is performed to obtain the harmonic sensitivity factor and the corresponding harmonic sensitivity vector; The method of obtaining the harmonic sensitivity factor and the corresponding harmonic sensitivity vector is: Extract the amplitude mutation rate, phase deviation rate, and harmonic phase jump ratio of each subharmonic from the harmonic index library to establish a sub-index vector; Based on the secondary index vector, the sensitivity of each harmonic reactive power to arc failure is quantified, and the harmonic entropy density of the secondary harmonics corresponding to different indicators is obtained as the sensitivity of each harmonic reactive power to arc failure; The sensitivity of each harmonic reactive power to arc failure corresponding to different high-frequency harmonic index categories is sorted to obtain the harmonic sensitivity factor; Based on the harmonic sensitivity factor, the index corresponding to the harmonic sensitivity factor is obtained, and the harmonic sensitivity vector is constructed; An arc-breaking classification model is constructed to perform three-dimensional classification mapping of arc-breaking reactive power mutations. Harmonic sensitive vectors are collected in real time and input into the arc-breaking classification model to determine whether the reactive power mutation level of the DC arc furnace is within the arc-breaking intervention space. If not, a prediction and analysis signal is generated. Based on the received prediction analysis signal, a harmonic prediction analysis model is constructed and input into the real-time harmonic sensitivity vector, and the predicted harmonic sensitivity vector is output. The predicted harmonic sensitivity vector is analyzed for similarity with the harmonic sensitivity vector in the arc interruption intervention space to obtain the harmonic sensitivity similarity; The method for obtaining the harmonic sensitivity similarity is: Obtain harmonic sensitive vectors within N monitoring periods and construct a harmonic measurement group; Where N is the number of monitoring cycles; A harmonic prediction and analysis model is constructed using a long short-term memory neural network algorithm. The harmonic measurement group is input into the prediction and analysis model to output the harmonic sensitivity vector within the future monitoring period. Map the harmonic sensitive vector to the three-dimensional warning space to determine whether it is in the arc-breaking intervention space; If not, the intervention threshold is calculated, and whether the intervention threshold is higher than or equal to the preset intervention threshold is determined again; If the intervention threshold value is again determined to be higher than or equal to the preset intervention threshold value, harmonic sensitivity similarity analysis is performed to obtain harmonic sensitivity similarity; The method for performing the harmonic sensitivity similarity analysis is: Obtain the harmonic sensitivity vector corresponding to the hard threshold boundary that is not in the arc-breaking intervention space and construct a similar benchmark group; Obtain harmonic sensitivity vectors within N future monitoring periods and construct similarity comparison groups; Calculate the Mahalanobis distance between each harmonic sensitive vector in the similar comparison group and the mean vector of the similar benchmark group; The cost matrix is constructed by combining the Mahalanobis distance from each harmonic sensitive vector to the mean vector, and the DTW distance of the cost matrix is calculated, and the DTW distance is used as the harmonic sensitive similarity; Based on the similarity of harmonic sensitivity, a prediction analysis is performed to obtain the approach time of the predicted harmonic sensitivity vector to the arc-breaking intervention space, and a compensation strategy is formulated.

2. A DC arc furnace reactive power prediction method according to claim 1, characterized in that: The high-frequency harmonic indicators include: The amplitude mutation rate, phase deviation rate, and harmonic phase jump ratio of the reactive power amplitude of different subharmonic components.

3. A DC arc furnace reactive power prediction method according to claim 1, characterized in that: The method of quantifying the sensitivity of each harmonic reactive power to arc failure is: The Euclidean distance between each harmonic and the high-frequency harmonic index of other harmonics is calculated. The probability distribution of each harmonic index is calculated through the exponential function and the preset kernel bandwidth parameter to obtain the harmonic entropy density.

4. A DC arc furnace reactive power prediction method according to claim 1, characterized in that: The method for judging whether the reactive power mutation level of the DC arc furnace is within the arc breaking intervention space is as follows: An arc-breaking classification model is constructed to perform three-dimensional classification mapping of arc-breaking reactive power mutations, establish a three-dimensional early warning space, collect the harmonic sensitive vectors of the DC arc furnace in real time, and perform normalized mapping to the three-dimensional early warning space; Compare the coordinates of the mapped three-dimensional warning space to determine whether the coordinates are in the broken arc intervention space.

5. A DC arc furnace reactive power prediction method according to claim 4, characterized in that: The method of constructing the arc-breaking classification model to perform three-dimensional classification mapping on arc-breaking reactive power mutation is as follows: Obtain the harmonic sensitive vectors of historical arc-breaking events and normal operating conditions, perform coordinate conversion, and obtain three-dimensional coordinate points; Set hard threshold boundaries based on 3D coordinate points to divide the 3D area; Based on the divided three-dimensional areas, three-dimensional warning space modeling is carried out to realize the construction of the arc fault classification model.

6. A DC arc furnace reactive power prediction method according to claim 1, characterized in that: The method of generating the prediction analysis signal is as follows: If it is not in the arc-breaking intervention space, calculate the proximity ratio of the harmonic sensitive vector in the three-dimensional warning space to the hard threshold boundary, and obtain the mutation proximity ratio of the vector amplitude mutation rate, the deviation proximity ratio of the phase deviation rate, and the jump proximity ratio of the phase jump ratio; The mutation proximity ratio, deviation proximity ratio, and jump proximity ratio are summed to obtain the intervention threshold value; A comparative analysis is performed based on the intervention threshold to determine whether a predictive analysis signal is generated.

7. A DC arc furnace reactive power prediction system, used to implement the DC arc furnace reactive power prediction method according to any one of claims 1 to 6, characterized in that: Includes the following modules: Harmonic acquisition module: used to collect high-frequency harmonic indicators during the arc-breaking reactive power transition stage of the DC arc furnace and build a harmonic indicator library; Sensitivity analysis module: Based on the harmonic index library, it is used to quantify the sensitivity of each harmonic reactive power to arc failure. Based on the sensitivity and each harmonic, it performs screening and processing to obtain the harmonic sensitivity factor and the corresponding harmonic sensitivity vector; Sudden change classification module: This module is used to build an arc-breaking classification model to perform three-dimensional classification mapping of arc-breaking reactive sudden changes. It collects harmonic sensitive vectors in real time and inputs them into the arc-breaking classification model to determine whether the reactive sudden change level of the DC arc furnace is within the arc-breaking intervention space. If not, a prediction and analysis signal is generated. Similarity recognition module: Based on the received prediction analysis signal, it is used to build a harmonic prediction analysis model and input the real-time harmonic sensitivity vector, output the predicted harmonic sensitivity vector, and perform similarity analysis between the predicted harmonic sensitivity vector and the harmonic sensitivity vector in the arc interruption intervention space to obtain the harmonic sensitivity similarity; Intervention prediction module: Based on the similarity of harmonic sensitivity, prediction analysis is performed to obtain the approach time of the predicted harmonic sensitivity vector to the arc-breaking intervention space, and a compensation strategy is formulated.

Images