Method for detecting abnormal state of aluminum alloy homogenizing furnace based on data driving

Abstract

The invention provides an aluminum alloy homogenizing furnace abnormal state detection method based on data driving, which comprises the following steps of: fitting heating section data and natural gas consumption data of an aluminum alloy homogenizing furnace in a normal state by using a time domain response curve of a second-order control system under a step signal in advance to obtain a corresponding standard control curve; acquiring natural gas consumption data and electric energy consumption data of the to-be-detected aluminum alloy homogenizing furnace in each production process; calculating an accumulative error between the heating section data and a preset heating electric energy standard control curve through a fixed sliding window, calculating an accumulative error between the natural gas consumption data and a preset natural gas standard control curve, and preliminarily judging that the data is abnormal when the accumulative error corresponds to a corresponding accumulative error threshold value; and re-checking the abnormal data by adopting an isolated forest anomaly detection model.

Description

Technical Field

[0001] This invention relates to the field of abnormal state detection in aluminum alloy smelting processes, and particularly to a data-driven method and system for detecting abnormal states in aluminum alloy homogenizing furnaces. Background Technology

[0002] Homogenization of aluminum alloys is a crucial step in hot working, aiming to eliminate compositional segregation and internal stresses generated during casting. As a typical energy conversion system, the natural gas and electricity consumed by the homogenizing furnace are primarily used to heat and hold the aluminum alloy ingots, causing the necessary physicochemical changes. When the homogenizing furnace is in ideal working condition, its natural gas and electricity consumption curves should be stable and repeatable when processing the same size and quantity of ingots. However, long-term operation of the homogenizing furnace under high temperature and multi-physics coupling environments is susceptible to the following factors: nonlinear changes in natural gas flow and temperature field distribution leading to abnormal energy consumption; accumulation of latent faults such as aging heating elements and decreased furnace sealing; the impact of grid voltage fluctuations and changes in ambient temperature and humidity on the electrothermal system; or improper operation leading to inappropriate heating rate control and homogenization time deviations from the process window.

[0003] Traditional aluminum alloy homogenizing furnace condition monitoring relies on manual inspections and threshold alarms, and energy dispatch is also largely manual, with outdated metering methods leading to high labor costs and energy waste. These methods are also lagging and fail to meet the high-precision, low-false-alarm industrial demands, necessitating the development of advanced energy monitoring systems to improve efficiency and reduce energy consumption. Therefore, this invention proposes a data-driven method for detecting abnormal conditions in aluminum alloy homogenizing furnaces by monitoring and analyzing the dynamic changes in energy consumption data. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by providing a data-driven method and system for detecting abnormal states in aluminum alloy homogenizing furnaces.

[0005] To achieve the above objectives, the first aspect of the present invention provides a data-driven method for detecting abnormal states in aluminum alloy homogenizing furnaces, the method comprising the following steps:

[0006] The heating section data of the aluminum alloy homogenizing furnace under normal conditions was pre-fitted with the time domain response curve of the second-order control system under a step signal to obtain the standard control curve of heating power.

[0007] Natural gas consumption data under normal conditions of aluminum alloy homogenizer were pre-fitted with the time-domain response curve of second-order control system under step signal to obtain natural gas standard control curve;

[0008] Obtain natural gas consumption and electricity consumption data for each production process of the aluminum alloy homogenizing furnace under test;

[0009] Extract heating section data from power consumption data, calculate the cumulative error between heating section data and preset heating power standard control curve through a fixed sliding window, and preliminarily determine that the current production process is abnormal when the cumulative error is greater than the set first cumulative error threshold.

[0010] And / or, calculate the cumulative error between natural gas consumption data and the preset natural gas standard control curve through a fixed sliding window, and preliminarily determine that the current production process is abnormal when the cumulative error is greater than the set second cumulative error threshold;

[0011] An isolated forest anomaly detection model was used to verify and issue alarms for natural gas consumption and electricity consumption data during this production process.

[0012] The above scheme treats the homogenizing furnace as an energy conversion system. The energy consumption data of the homogenizing furnace is a highly sensitive and reliable indirect observation variable for its internal thermal state and material phase transformation process. Detection is completed through a two-stage detection process. The first stage, based on an anomaly detection method using the control system response curve and cumulative error, detects the natural gas consumption data and electrical energy consumption data during each production process of the aluminum alloy homogenizing furnace, achieving high real-time detection of anomalies. The second stage uses an anomaly detection model based on isolated forests to detect the natural gas consumption data and electrical energy consumption data, achieving high-precision anomaly detection and providing auxiliary judgment and verification. The two stages are executed sequentially, jointly ensuring the speed, accuracy, and automation of the detection.

[0013] In one possible embodiment, the formula for calculating the first cumulative error threshold or the second cumulative error threshold is: T=μ+kσ; where μ is the cumulative mean error of the heating section data sample under normal conditions of the aluminum alloy homogenizing furnace, or the cumulative mean error of the natural gas consumption data sample; σ is the cumulative standard deviation of the cumulative error of the heating section data sample under normal conditions of the aluminum alloy homogenizing furnace, or the cumulative standard deviation of the cumulative error of the natural gas consumption data sample; and k is a preset safety factor.

[0014] The above scheme sets a reasonable error threshold to detect as many anomalies as possible while ensuring a low false alarm rate, avoiding misjudging normal fluctuations as abnormalities. Simultaneously, when setting the threshold, the peak value and overshoot of the standard response curve of the second-order control system must be considered. This is because the cumulative error between the tested sample and the standard sample curve is relatively large near the peak value; therefore, the threshold should be appropriately increased to reduce the possibility of false alarms.

[0015] The formula for calculating the length of a fixed sliding window is L∈[f s / 10, f s / 2];where f s This refers to the data sampling frequency.

[0016] In the above scheme, the sampling frequency f is used.s Set the window length appropriately to avoid it being too long or too short.

[0017] In one embodiment, an isolated forest anomaly detection model is used to review and issue alarms for natural gas consumption data and electricity consumption data during the current production process, including:

[0018] Energy consumption index features and data distribution index features are extracted from the natural gas consumption data and electricity consumption data of this production process as feature vectors, and then fed into the trained isolated forest anomaly detection model to identify normal data points and abnormal data points.

[0019] The energy consumption index features include total natural gas energy consumption, total electricity energy consumption, and the ratio of total natural gas energy consumption to total electricity energy consumption; the data distribution index features include the Gamma distribution, skewness, and kurtosis of natural gas consumption data and electricity consumption data.

[0020] The total energy consumption of natural gas, the total energy consumption of electricity, and the ratio of total energy consumption of natural gas to total energy consumption of electricity are selected as visualization feature dimensions, and normal data points and abnormal data points are projected onto the visualization space.

[0021] Calculate the centroids of all normal data points in the visualization space;

[0022] Calculate the Euclidean distance between the centroids of abnormal data points and normal data points. If the Euclidean distance is greater than the alarm threshold, the abnormal data point is determined to be a real anomaly, and an alarm is triggered. If the Euclidean distance is less than or equal to the alarm threshold, the abnormal data point is determined to be a false anomaly, and no alarm is triggered.

[0023] The training steps for the isolated forest anomaly detection model include:

[0024] Historical natural gas consumption and electricity consumption data of aluminum alloy homogenizing furnaces were obtained as a training set.

[0025] Extract the energy consumption index features and data distribution index features of each sample in the training set as feature vectors;

[0026] The feature vectors are used as input to train an isolated forest anomaly detection model.

[0027] The above scheme employs a feature system based on energy consumption and data distribution indicators to achieve high-precision identification of abnormal data points. To further differentiate samples and observe the characteristics of total energy consumption of natural gas and electricity during the operation of the aluminum alloy homogenizing furnace, three core energy consumption indicators are selected as coordinate axes to form a three-dimensional feature space. The identified abnormal data points are then projected onto this space for visualization. Furthermore, to suppress false alarms and missed alarms, an Euclidean distance metric is introduced: by calculating the distance from abnormal data points in the feature visualization space to the center of normal sample clusters and comparing it with a preset alarm threshold, the degree of abnormality can be intuitively quantified, thereby effectively improving detection reliability.

[0028] In one possible embodiment, the step of fitting the second-order response curve includes:

[0029] The consumption data y(t) and corresponding data and the set control input data u(t) in each valid sample of the homogenizer are obtained, and an initial differential equation model describing the dynamic relationship between output and input using system parameters is constructed; the consumption data y(t) is natural gas consumption data or heating section data;

[0030] The consumed data y(t) and the control input data u(t) are processed by a preset low-pass filter to obtain the filtered estimates of their derivatives of each order.

[0031] Using the filtered estimate, the initial differential equation model is transformed into a standard linear regression equation with system parameters as variables;

[0032] The linear regression equation is solved using the least squares method to obtain estimated values ​​of the system parameters.

[0033] The estimated system parameters are substituted into a pre-defined second-order transfer function structure to generate a transfer function model, the formula of which is as follows:

[0034]

[0035] Where A represents the control input data u(t), which is the steady-state value of natural gas or electricity consumption data in each production process; a1 and a2 represent output coefficients, and b1, b2, c1, and c2 represent input coefficients;

[0036] For the transfer function model identified from multiple valid samples based on the homogenizing furnace, the mean values ​​of the input and output coefficients are calculated respectively. The obtained average input and output coefficients are then substituted into the preset second-order transfer function structure to obtain the standard transfer function model.

[0037] The standard control curve is obtained based on the standard transfer function model.

[0038] The above scheme uses the least squares method to estimate the numerator and denominator coefficients of the second-order transfer function, so that the model output is as close as possible to the actual data, thereby finding the best matching transfer function model.

[0039] In one possible embodiment, after obtaining the standard control curve, the standard control curve is further smoothed using Makima interpolation, the steps of which include:

[0040] Obtain consumption data and calculate the slope between two adjacent consumption data points. The calculation formula is as follows:

[0041]

[0042] Where x i Indicates time, y i This represents the consumption data at the corresponding time point;

[0043] Calculate the weighting factor w i The calculation formula is as follows:

[0044]

[0045] Based on the weighting factor and the slope between the two energy consumption time series data, the data point x is calculated using the following formula. i slope k i :

[0046]

[0047] In every two adjacent data points x i and x i+1 Between these, a cubic polynomial is constructed based on Makima interpolation, and its formula is:

[0048]

[0049] Among them, a i b i c i d i The coefficients of the polynomial are given by the difference condition y(x) and the slope k. i Determine x as the data point x i and x i+1 Any data between;

[0050] By concatenating the cubic polynomials over all data intervals, a smooth standard control curve is obtained.

[0051] Since the sampling frequency of the raw energy consumption data is low, while the sampling frequency of the control system fitting results is high, in order to meet the high real-time detection requirements of the system, the above scheme improves the virtual sampling frequency by interpolating the raw natural gas consumption data, thereby improving the real-time performance of anomaly monitoring.

[0052] To achieve the above objectives, a third aspect of the present invention provides a data-driven abnormal state detection system for an aluminum alloy homogenizing furnace, comprising: a data storage module, a data acquisition module, a segmentation module, a preliminary judgment module for the heating section, a preliminary judgment module for natural gas, and a verification module;

[0053] The data storage module has built-in heating power standard control curve and natural gas standard control curve. The heating power standard control curve is obtained by fitting the heating section data of the aluminum alloy homogenizing furnace under normal conditions with the time domain response curve of the second-order control system under a step signal. The natural gas standard control curve is obtained by fitting the natural gas consumption data of the aluminum alloy homogenizing furnace under normal conditions with the time domain response curve of the second-order control system under a step signal.

[0054] The data acquisition module is used to acquire natural gas consumption data and electricity consumption data for each production process of the aluminum alloy homogenizing furnace under test.

[0055] The segmentation module is used to segment the power consumption data to obtain the heating segment data;

[0056] The heating section preliminary judgment module is used to call the heating power standard control curve in the data storage module, calculate the cumulative error between the heating section data and the preset heating power standard control curve through a fixed sliding window, and preliminarily determine that the next production process is abnormal when the cumulative error is greater than the set first cumulative error threshold.

[0057] The natural gas preliminary judgment module is used to call the natural gas standard control curve in the data storage module, calculate the cumulative error between the natural gas consumption data and the preset natural gas standard control curve through a fixed sliding window, and preliminarily determine that the production process is abnormal when the cumulative error is greater than the set second cumulative error threshold.

[0058] The verification module is used to verify and issue alarms for the natural gas consumption data and electricity consumption data during this production process using the isolated forest anomaly detection model.

[0059] To achieve the above objectives, a third aspect of the present invention provides a computer device, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus.

[0060] A memory is used to store computer programs; a processor is used to execute the programs stored in the memory to implement the steps of the data-driven aluminum alloy homogenizing furnace abnormal state detection method described in the first aspect above.

[0061] To achieve the above objectives, a fourth aspect of the present invention provides a computer-readable storage medium having instructions stored thereon that, when executed by one or more processors, cause the processors to perform the steps of the data-driven aluminum alloy homogenizing furnace abnormal state detection method described in the first aspect above.

[0062] The beneficial effects of this invention are as follows:

[0063] 1. Using energy consumption data from the homogenizing furnace, a two-stage detection system is employed to determine whether the aluminum alloy homogenizing furnace is in an ideal and healthy state, thereby improving the timeliness, accuracy, and automation level of anomaly detection. The first stage, based on an anomaly detection method using the control system response curve and cumulative error, detects natural gas and electricity consumption data during each production process of the aluminum alloy homogenizing furnace, achieving high real-time detection of anomalies. The second stage uses an anomaly detection model based on isolated forests to detect natural gas and electricity consumption data, achieving high-precision anomaly detection of anomalies and providing auxiliary judgment and verification. Both stages are executed sequentially, jointly ensuring rapid, accurate, and automated detection.

[0064] 2. By introducing Euclidean distance metric, the distance from abnormal data points to the center of normal sample clusters in the feature visualization space is calculated and compared with the preset alarm threshold. This allows for intuitive quantification of the degree of abnormality, alleviating the problems of false alarms and missed alarms, and significantly improving the reliability and effectiveness of alarm signals. Attached Figure Description

[0065] Figure 1 This embodiment provides data on natural gas and electricity consumption in September for a homogenizing furnace production line in the aluminum alloy smelting section of a certain enterprise.

[0066] Figure 2 This is a schematic diagram of the implementation process of the anomaly detection method provided in this embodiment;

[0067] Figure 3 This is a graph showing the natural gas and electricity consumption during the 20th and 39th production processes in this embodiment.

[0068] Figure 4 This is a schematic diagram of the segmented power data provided in this embodiment;

[0069] Figure 5 This is a schematic diagram showing the fitting results of natural gas energy consumption data during some production processes provided in this embodiment;

[0070] Figure 6This is a schematic diagram of a standard natural gas curve provided in this embodiment;

[0071] Figure 7 This is a schematic diagram showing the fitting results of energy consumption data during the electric heating stage of a portion of the production process provided in this embodiment;

[0072] Figure 8 This is a schematic diagram of the standard electric heating stage curve provided in this embodiment;

[0073] Figure 9 This is a schematic diagram of the anomaly detection results of the 39th production process provided in this embodiment;

[0074] Figure 10 This is a schematic diagram of the anomaly detection results of the 20th production process provided in this embodiment;

[0075] Figure 11 This is a schematic diagram of sensor fault output provided in this embodiment;

[0076] Figure 12 A schematic diagram illustrating the feature visualization results of the isolated forest state classification and recognition device provided in this embodiment;

[0077] Figure 13 A schematic diagram showing the natural gas and electricity consumption data for the second and thirteenth production processes provided in this embodiment;

[0078] Figure 14 This is a schematic diagram of the abnormality detection results of the second production process provided in this embodiment;

[0079] Figure 15 This is a schematic diagram of the abnormality detection results of the 34th production process provided in this embodiment;

[0080] Figure 16 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0081] The technical solution of the present invention will be further described in detail below through specific embodiments.

[0082] refer to Figure 2 First, a baseline curve model is constructed and an adaptive threshold is set to perform preliminary anomaly screening of the input energy consumption data. Second, an anomaly detection model based on the isolated forest algorithm is built to find anomalies in the pre-screened data. Finally, a decision analysis module based on metric learning is introduced to calculate the distance of anomalous data in the feature space, perform a final comprehensive decision and confidence assessment, and determine whether it is a false anomaly. This system ensures that the detection method has both high real-time performance and high accuracy, providing reliable technical support for ensuring the stable operation and safe production of the aluminum alloy homogenization process.

[0083] Example 1

[0084] This embodiment provides a data-driven method for detecting abnormal states in aluminum alloy homogenizing furnaces, such as... Figure 2 As shown, the method includes the following steps:

[0085] Step S0: Fit the heating section data of the aluminum alloy homogenizing furnace under normal conditions with the time domain response curve of the second-order control system under a step signal to obtain the standard control curve of heating power.

[0086] Natural gas consumption data under normal conditions of aluminum alloy homogenizer were pre-fitted with the time-domain response curve of the second-order control system under a step signal to obtain the natural gas standard control curve.

[0087] The specific steps are as follows:

[0088] Step S01: Obtain the data on natural gas and electricity consumed by the aluminum alloy homogenizing furnace under normal conditions, and separate the natural gas data and electricity data to retain valid data samples.

[0089] It should be noted that after obtaining the natural gas and electricity consumption data of the aluminum alloy homogenizing furnace under normal conditions, the natural gas consumption data and electricity consumption data are divided into multiple complete production processes according to the production cycle; or the shutdown data is replaced with 0, and samples are divided based on the 0 value. That is, any non-zero consecutive data combination will be considered as a sample. It can be understood that one production process can complete the identification of two adjacent production processes. To avoid the situation of short-term equipment operation, a valid sample is recorded as having at least 5 consecutive points with values, which can retain valid data samples.

[0090] It is understandable that after sample segmentation, the natural gas and electricity consumption in each of the 20 production processes is obtained, such as... Figure 3 As shown in (a).

[0091] Step S02: The electrical energy consumption data of the aluminum alloy homogenizing furnace under normal conditions is segmented, and the heating segment data in the segmented electrical energy data is selected and fitted by the standard response curve model of the second-order control system to obtain the standard control curve of heating electrical energy.

[0092] It should be noted that the smelting of aluminum alloys requires three stages: heating, holding, and cooling. By analyzing the natural gas and electricity consumption data during the smelting process, it was found that the natural gas data sample itself had a good fit and did not need to be segmented, while the electricity data required segmented fitting.

[0093] In this embodiment, adaptive segmentation based on cumulative error is used to segment the electrical energy data, and the control system response curve is used to fit the electrical energy heating stage.

[0094] Specifically, extract heating section data from the electricity consumption data, including:

[0095] A theoretical curve based on process standards is selected as the benchmark energy consumption curve;

[0096] Set a third cumulative error threshold;

[0097] For each complete production process, starting from the starting point of the power consumption data, the continuous power consumption data is included in a sliding window, and the cumulative error between all power consumption data in the sliding window and the corresponding benchmark power consumption curve is calculated in real time.

[0098] When the cumulative error value exceeds the third cumulative error threshold, the end point of the current window is marked as a segment point; the cumulative error value is reset, and the next window is started from this point as the new starting point, until the power consumption data is traversed, and the multiple segments of power consumption data are divided into heating segment - heat preservation segment - cooling segment.

[0099] like Figure 4 As shown, by analyzing the electrical energy data samples using the above segmentation method, the boundary point between heating and heat preservation can be easily obtained. Furthermore, the potential starting point during the heating stage fitting is approximately at the 2nd or 3rd data point. This segment of electrical energy data is then fitted using the standard response curve of a second-order control system. The segmented electrical energy data samples are shown below. Figure 4 As shown.

[0100] The steps for obtaining the standard control curve for heating electrical energy include:

[0101] Obtain the heating section data y(t) and corresponding data and set control input data u1(t) in each valid sample of the homogenizer, and construct an initial differential equation model that describes the dynamic relationship between output and input using system parameters;

[0102] The formula for the initial differential equation is as follows:

[0103] (1)

[0104] Among them, y (n) (t) represents the nth derivative of the heating section data y(t) with respect to time, y (n-1) (t) represents the (n-1)th derivative of the heating section data y(t) with respect to time, u (m-1) (t) represents the m-1th derivative of the heating section data y(t) with respect to time, where a1,…,a n The coefficients of the output terms are represented as b1, ..., b2. m Indicates the coefficients of the input terms;

[0105] The heating section data y(t) and control input data u1(t) are processed by a preset low-pass filter L(s) to obtain the filtered estimates z of their derivatives. k (t) and w k (t), the formula is as follows:

[0106] (2)

[0107] (3)

[0108] Using the filtered estimate, the initial differential equation model is transformed into a standard linear regression equation with system parameters as variables;

[0109] In this embodiment, L(s) is applied to equation (1), and combined with equations (2) and (3), the filtered equation is transformed into the following linear regression equation:

[0110] (4)

[0111] Rearranging equation (4) yields equation (5) as follows:

[0112] (5)

[0113] Construct the regression matrix φ(t)=[-z n-1 (t),…, -z0(t),w m-1 [(t),…,w0(t)] and parameter matrix θ=[a1,…,a n ,b1,…,b m ] T ,y f (t)=z n (t), Equation (5) can be written in linear regression form:

[0114] y f (t) = φ(t) T θ

[0115] The linear regression equation is solved using the least squares method to obtain estimated values ​​of the system parameters.

[0116] The estimated system parameters are substituted into a pre-defined second-order transfer function structure to generate a transfer function model, the formula of which is as follows:

[0117]

[0118] Where A represents the control input data u1(t), which is determined by the steady-state value of the power consumption data in each production process; a1 and a2 represent the output coefficients, and b1, b2, c1, and c2 represent the input coefficients.

[0119] It should be noted that in this embodiment, only the calculation method of the input term coefficient b is given; the calculation method of the input term coefficient b has the same form as that of c. Because the transfer function fitted to the data of the electric heating stage contains zeros, numerator and denominator coefficients are introduced to facilitate the direct establishment of a benchmark model, and the numerator and denominator coefficients are calculated for the electric energy data.

[0120] The measured electrical energy data were compared with the simulated response to obtain the fitting rate. The fitting results for energy consumption data in the heating stage of some production processes are as follows: Figure 5 As shown in Table 1, after removing sample data with excessively low fit, the numerator and denominator coefficients of the transfer function of the control system with a high data fit rate in the electric heating stage are shown in Table 2. The numerator and denominator coefficients are processed to obtain the average value and standard deviation of the numerator and denominator coefficients for the electric heating stage data fit, as shown in Table 2. Substituting the average value of the numerator and denominator data, the transfer function of the established standard control system for the electric heating stage is obtained as G1(S):

[0121]

[0122] Analysis of the energy consumption data fitting results for the electric heating stage revealed that the control system obtained from this data fitting showed a high fitting rate for most of the production process when the input step function u1(t) = 2.5. The time-domain response curve obtained under this input is the standard electric heating stage curve, as shown below. Figure 6 As shown.

[0123] For the transfer function model identified from multiple valid samples based on the homogenizing furnace, the mean values ​​of the input and output coefficients are calculated respectively. The obtained average input and output coefficients are then substituted into the preset second-order transfer function structure to obtain the standard transfer function model.

[0124] The standard control curve for heating electrical energy is obtained based on the standard transfer function model.

[0125] Table I. Numerator and denominator coefficients of the transfer function of control systems with high fitting rate for energy consumption data in the electric heating stage.

[0126] Serial Number a1 b1 c1 a2 b2 c2 1 1 0.0397 0.0011 0 0.0606 0.0018 2 1 0.0576 0.0009 0 0.1086 0.0013 3 1 0.0455 0.0011 0 0.0733 0.0018 4 1 0.0474 0.0009 0 0.0790 0.0014 5 1 0.0559 0.0007 0 0.1034 0.0010 6 1 0.0496 0.0010 0 0.0833 0.0016 7 1 0.0670 0.0011 0 0.1183 0.0017 8 1 0.0630 0.0011 0 0.1107 0.0017 9 1 0.0662 0.0009 0 0.1233 0.0013 10 1 0.0572 0.0008 0 0.1021 0.0012 11 1 0.0708 0.0009 0 0.1255 0.0014 12 1 0.0581 0.0010 0 0.0996 0.0016 13 1 0.0504 0.0011 0 0.0752 0.0018 14 1 0.0517 0.0014 0 0.0833 0.0022 15 1 0.0801 0.0014 0 0.1390 0.0023 16 1 0.0767 0.0012 0 0.1398 0.0019 17 1 0.0720 0.0013 0 0.1316 0.0019 18 1 0.0637 0.0012 0 0.1077 0.0020 19 1 0.0587 0.0009 0 0.1026 0.0015 20 1 0.0652 0.0012 0 0.1072 0.0018 21 1 0.0379 0.0009 0 0.0604 0.0014 22 1 0.0717 0.0010 0 0.1344 0.0016 23 1 0.0545 0.0006 0 0.0982 0.0009 24 1 0.0603 0.0009 0 0.1113 0.0015 25 1 0.0563 0.0010 0 0.1011 0.0016 26 1 0.0739 0.0012 0 0.1386 0.0018 27 1 0.0596 0.0007 0 0.1133 0.0010 28 1 0.0584 0.0013 0 0.1059 0.0020 29 1 0.0586 0.0011 0 0.1081 0.0017 30 1 0.0560 0.0012 0 0.0951 0.0020

[0127] Table II shows the mean and standard deviation of the numerator and denominator coefficients of the transfer function of control systems with high data fitting rates during the electric heating stage.

[0128] Identification results a1 b1 c1 a2 b2 c2 average value 1 0.0595 0.0010 0 0.1047 0.0016 Standard deviation 0 0.0102 0.0002 0 0.0219 0.0003

[0129] Table V. Degree of Abnormalities in the Production Process

[0130] Abnormal production process number Distance index Ld Abnormal percentage in conclusion 2 1078.5753 927.8% abnormal 34 198.9082 171.0% False Alarm

[0131] In one possible embodiment, after obtaining the standard control curve, the standard control curve is further smoothed using Makima interpolation, the steps of which include:

[0132] Obtain heating section data, where x i Indicates time, y i This represents the energy consumption value of the heating section at the corresponding moment;

[0133] The slope between two adjacent heating segment data points is calculated using the following formula:

[0134]

[0135] Calculate the weighting factor w i The calculation formula is as follows:

[0136]

[0137] The data point x is calculated using the following formula, based on the weighting factor and the slope between the two consumption data points. i slope k i :

[0138]

[0139] At data boundary point x i Between x and x, several cubic polynomials are constructed based on Makima interpolation, with the following formula:

[0140]

[0141] Among them, a i b i c i d i The coefficients of the polynomial are given by the difference condition y(x) and the slope k. i Sure;

[0142] By concatenating the cubic polynomials over all data intervals, a smooth interpolation curve is obtained.

[0143] Among them, the interpolation condition y(x) i )=y i =a i y(x) i+1 )=y i+1 ;y`(x i )=m i =b i y`(x i+1 )=mi+1 c i= 3S i -2m i -m i+1 / h i ;d i= m i+1 +m i -2S i / h i 2 ;

[0144] It should be noted that since the energy consumption data is obtained by the sensor sampling every 15 minutes, while the standard sample has a sampling period of 1 second, data length alignment is required first to compare the measured data with the standard sample. After interpolating the measured data using the above method to generate high-resolution interpolated data, to ensure that the lengths of the interpolated data and the simulated response data are consistent, the minimum length of the two is taken, and the first few data points of the minimum length of both the interpolated data and the simulated response data are truncated to ensure that they are compared within the same time period.

[0145] Step S03: The natural gas consumption data of the aluminum alloy homogenizing furnace under normal conditions is fitted by the standard response curve model of the second-order control system to obtain the natural gas standard control curve.

[0146] The specific fitting steps can be found in step S02. The only difference is that in this step, the natural gas data y(t) and the corresponding data and the set control input data u2(t) in each effective sample of the homogenizer are obtained, and an initial differential equation model describing the dynamic relationship between the output and the input is constructed using system parameters. The natural gas data y(t) and the control input data u2(t) are processed by a preset low-pass filter to obtain the filtered estimates of their derivatives of each order.

[0147] In this embodiment, the measured natural gas data is compared with the simulated response to obtain the fitting rate. The fitting results for natural gas energy consumption data in some production processes are as follows: Figure 7 As shown in Table III, after removing sample data with excessively low fit, the numerator and denominator coefficients of the transfer function of the control system with a high natural gas data fit rate are obtained. The numerator and denominator coefficients are then processed to obtain the average value and standard deviation of the numerator and denominator coefficients for the natural gas data fit, as shown in Table IV. Substituting the average value of the numerator and denominator data, the transfer function of the established natural gas standard control system is obtained as G2(S):

[0148]

[0149] Analysis of the fitting results for natural gas energy consumption data revealed that the control system obtained from this natural gas data fitting showed a high fitting rate for most production processes when the input step function u2(t) = 2.5. The time-domain response curve obtained under this input is a standard natural gas curve, such as... Figure 8 As shown.

[0150] Table III. Numerator and denominator coefficients of transfer functions for control systems with high natural gas data fit rates.

[0151] Serial Number <![CDATA[a1]]> <![CDATA[b1]]> <![CDATA[c1]]> <![CDATA[a2]]> <![CDATA[b2]]> <![CDATA[c2]]> 1 1 0.3554 0.0485 0 9.8918 0.0517 2 1 0.3161 0.0434 0 9.026 0.0549 3 1 0.3248 0.0423 0 9.2617 0.0532 4 1 0.3806 0.0444 0 10.7193 0.0562 5 1 0.3268 0.0424 0 9.2622 0.0517 6 1 0.4761 0.0651 0 13.5004 0.0625 7 1 0.5667 0.0615 0 14.0034 0.0495 8 1 0.4924 0.0552 0 12.5412 0.0514 9 1 0.6432 0.0785 0 15.8246 0.0654 10 1 0.5003 0.0560 0 12.6946 0.0488 11 1 0.4977 0.0530 0 12.1791 0.0454 12 1 0.5029 0.0599 0 13.0632 0.0535 13 1 0.4043 0.0479 0 10.2350 0.0493 14 1 0.3384 0.0399 0 8.1564 0.0461 15 1 0.4778 0.0598 0 12.1842 0.0620 16 1 0.4617 0.0559 0 11.3981 0.0591 17 1 0.4169 0.0536 0 10.8904 0.0562 18 1 0.5766 0.0699 0 16.1237 0.0739 19 1 0.4067 0.0551 0 12.3837 0.0574 20 1 0.4031 0.0523 0 11.0408 0.0705 21 1 0.4396 0.0508 0 10.7872 0.0497 22 1 0.5338 0.0600 0 13.9959 0.0510 23 1 0.3668 0.0364 0 8.5251 0.0448 24 1 0.3925 0.0392 0 10.1578 0.0524 25 1 0.4129 0.0481 0 11.4306 0.0537 26 1 0.3299 0.0379 0 8.2972 0.0471 27 1 0.5096 0.0666 0 14.9034 0.0728 28 1 0.4587 0.0649 0 14.2694 0.0795 29 1 0.4767 0.0627 0 14.1476 0.0723 30 1 0.3597 0.0523 0 10.9377 0.0654

[0152] Table IV. Mean and standard deviation of numerator and denominator coefficients of transfer functions for control systems with high natural gas data fit.

[0153] Identification results <![CDATA[a1]]> <![CDATA[b1]]> <![CDATA[c1]]> <![CDATA[a2]]> <![CDATA[b2]]> <![CDATA[c2]]> average value 1 0.4383 0.0534 0 11.7277 0.0569 Standard deviation 0 0.0834 0.0103 0 2.2159 0.0094

[0154] Step S1, refer to Figure 1 The natural gas consumption and electricity consumption data of the aluminum alloy homogenizing furnace under test were obtained during each production process.

[0155] Similarly, the real-time data on natural gas and electricity consumption of the aluminum alloy homogenizing furnace is acquired, and the natural gas and electricity data are separated to retain valid data samples for reference. Figure 4 ;

[0156] It should be noted that after acquiring the natural gas consumption data and electricity consumption data of the aluminum alloy homogenizing furnace under test in real time, the natural gas consumption data and electricity consumption data are divided into multiple complete production processes according to the production cycle; or a combination of at least 5 consecutive and non-zero data is determined as a complete production process; each complete production process is used as a valid sample for detecting abnormal conditions of the aluminum alloy homogenizing furnace.

[0157] It is understandable that after sample segmentation, normal production processes may occur, such as the natural gas and electricity consumption during the 20th production process. Figure 3 As shown in (a), abnormal production processes may also occur. In these processes, the natural gas and electricity consumption data differ significantly from most other processes. For example, in the 39th production process, where both natural gas and electricity consumption were abnormal, a sudden increase occurred during the initial decrease in natural gas and electricity consumption data. These anomalies were caused by secondary heating in the homogenizer and are considered abnormal production processes. Figure 3 As shown in (b).

[0158] Step S2: Extract the heating section data from the power consumption data, calculate the cumulative error between the heating section data and the preset heating power standard control curve through a fixed sliding window, and preliminarily determine that the current production process is abnormal when the cumulative error is greater than the set first cumulative error threshold.

[0159] And / or, calculate the cumulative error between natural gas consumption data and the preset natural gas standard control curve through a fixed sliding window, and preliminarily determine that the current production process is abnormal when the cumulative error is greater than the set second cumulative error threshold.

[0160] In one possible embodiment, the formula for calculating the first cumulative error threshold or the second cumulative error threshold is: T=μ+kσ; where μ is the cumulative mean error of the heating section data sample under normal conditions of the aluminum alloy homogenizing furnace, or the cumulative mean error of the natural gas consumption data sample; σ is the cumulative standard deviation of the cumulative error of the heating section data sample under normal conditions of the aluminum alloy homogenizing furnace, or the cumulative standard deviation of the cumulative error of the natural gas consumption data sample; and k is a preset safety factor.

[0161] It should be noted that when calculating the error threshold, the set sampling frequency f is used. s The length L of the fixed sliding window for calculating cumulative error is determined. The formula for calculating the fixed sliding window is as follows:

[0162] L∈[f s / 10, f s / 2]

[0163] Within a sliding window of length L, calculate the cumulative error between the electricity consumption data, natural gas consumption data, and their corresponding baseline model predictions.

[0164] In this embodiment, error thresholds of 13 and 5 are set for the data from the natural gas and electric heating stages, respectively. Each interval consists of 10 data points and lasts for 90 seconds. The system checks whether the cumulative error within each interval exceeds the threshold. Since the actual sampling frequency is low, initial sampling deviations may cause false alarms. Therefore, this method selects to start checking for error points exceeding the threshold from the 10th interval, which does not affect the accuracy of the judgment. If an error point exceeding the threshold exists within an interval, an alarm message is output, including the interval number, time range, and the threshold exceeded, such as... Figure 9 The results of the 39th anomaly detection during the production process show that, from Figure 9 (a) It can be seen that the natural gas error curve of this production process exceeded the threshold of 13 in the initial interval. Figure 9 (b) It can be seen that the error curve of the electric heating stage of this production process exceeds the threshold of 5 from the 24th interval, reflecting the real-time nature of anomaly monitoring. If the error in all intervals does not exceed the threshold, the prompt message "All interval errors are within the threshold range" will be output. Figure 10 The results of the 20th production process anomaly detection are shown.

[0165] Step S3: Use the isolated forest anomaly detection model to verify and alarm the natural gas consumption data and electricity consumption data in this production process.

[0166] In one possible embodiment, an isolated forest anomaly detection model is used to review and issue alarms for the natural gas consumption data and electricity consumption data during the current production process, including:

[0167] Energy consumption index features and data distribution index features are extracted from the natural gas consumption data and electricity consumption data of this production process as feature vectors, and then fed into the trained isolated forest anomaly detection model to identify normal data points and abnormal data points.

[0168] The energy consumption index features include total natural gas energy consumption, total electricity energy consumption, and the ratio of total natural gas energy consumption to total electricity energy consumption; the data distribution index features include the Gamma distribution, skewness, and kurtosis of natural gas consumption data and electricity consumption data.

[0169] in, , ;

[0170] ;

[0171] Total energy consumption of natural gas and total energy consumption of electricity can be used to detect abnormalities in total energy consumption and identify sudden increases or decreases in energy consumption.

[0172] The ratio of total natural gas energy consumption to total electricity energy consumption can detect whether the energy consumption ratio deviates significantly from the historical average. This ratio can also indicate whether there are any disconnections in the electricity or natural gas data acquisition sensors.

[0173] The Gamma distribution can be used to extract features about the shape and scale of data, making it easier to understand the data distribution; the probability density function of the Gamma distribution is:

[0174]

[0175] Where x is the data sample; k is the shape parameter, which controls the left or right skewness of the data distribution; θ is the scale parameter, which controls the degree of diffusion of the data distribution. It is the Gamma function.

[0176] Skewness is used to measure the asymmetry of the data distribution; kurtosis is used to measure the sharpness of the data distribution.

[0177]

[0178]

[0179] Where x is the data sample, μ is the mean, σ is the standard deviation, and E represents the expected value.

[0180] The training steps for the isolated forest anomaly detection model include:

[0181] Historical natural gas consumption and electricity consumption data for aluminum alloy homogenizing furnaces were obtained as the training set; the proportion of outlier samples to total samples was required to be 6%.

[0182] Extract the energy consumption index features and data distribution index features of each sample in the training set as feature vectors;

[0183] The feature vectors are used as input to train an isolated forest anomaly detection model.

[0184] In this embodiment, based on the characteristics of electricity and natural gas data, an isolated forest algorithm is used for unsupervised learning to detect abnormal data points, identify whether there are faults in the electricity or natural gas data acquisition sensors, and display the time period of the fault. The requirement is that the proportion of abnormal sample points to the total number of sample points is 6%.

[0185] If the total electricity consumption is less than 0.5 kWh and the total natural gas consumption is greater than 0.5 m³ / h... 3 The problem is identified as a malfunction in the natural gas acquisition sensor. If the total natural gas energy consumption is less than 0.5m³... 3 Furthermore, if the total power consumption exceeds 0.5 kWh, the power acquisition sensor is determined to be faulty. The sensor fault output is as follows: Figure 11 As shown, on-site investigation confirmed that there was indeed an abnormality of the power acquisition sensor going offline during the aforementioned time period.

[0186] To differentiate between different samples and observe the characteristics of total natural gas and electricity consumption in the aluminum alloy smelting process, total natural gas consumption, total electricity consumption, and the ratio of total natural gas consumption to total electricity consumption were selected as visualization feature dimensions. Normal and abnormal data points were projected onto the visualization space. The feature visualization results... Figure 12 As shown.

[0187] The top left corner shows the second and 39th production processes. The energy consumption data for natural gas and electricity in the second production process is as follows: Figure 13 As shown in (a) and 13(b); the natural gas and electricity consumption during the 39th production process is as follows: Figure 3 As shown in (b). To compare the abnormal and normal states, the natural gas and electricity consumption data for the 13th production process are as follows: Figure 13 As shown in (c) and 13(d), it is evident that the energy consumption level during abnormal production is significantly lower than that during normal production, and the shape parameters also differ considerably from those in the normal production process. Therefore, in Figure 12In the process, the abnormal production process (square) and the normal production process (circle) have huge differences in the three coordinate axis scales and the three characteristic vectors (total energy consumption of natural gas, total energy consumption of electricity, and the ratio of total energy consumption of natural gas to total energy consumption of electricity).

[0188] In this embodiment, the feature visualization 3D view based on the isolated forest algorithm shows that when the abnormal production process is set to 6% of all production processes, there are two abnormal production processes: the 2nd production process and the 39th production process. However, the 2nd production process is detected in the anomaly detection model based on the prediction and cumulative error of the control system, and the output result is as follows. Figure 14 As shown in (a) and 14(b), the natural gas anomaly detection results show that alarms were generated in all sections, and the anomaly detection results for the electric heating stage show that alarms were generated from section 13 onwards, diagnosing the second production process anomaly.

[0189] However, when detecting the 34th production process using the anomaly detection model based on control system prediction and cumulative error, it was found that the natural gas anomaly detection results showed an alarm generated in interval 109 to interval 114, such as... Figure 15 As shown in (a), the abnormal detection results of its electric heating stage show that all interval errors are within the threshold range, such as Figure 15 As shown in (b), in the feature visualization 3D view based on the isolated forest algorithm, the 34th production process is a normal production process. Comparing the simulated response and the measured data, it was found that although the measured data exceeded the threshold in these intervals, the sum was normal, that is, a false alarm occurred at this time.

[0190] Furthermore, in order to address the false alarm problem, this embodiment utilizes a decision fusion model based on metric learning to conduct a re-inspection.

[0191] Calculate the centroids of all normal data points in the visualization space;

[0192] Calculate the Euclidean distance between the centroids of abnormal data points and normal data points. If the Euclidean distance is greater than the alarm threshold, the abnormal data point is determined to be a real anomaly, and an alarm is triggered. If the Euclidean distance is less than or equal to the alarm threshold, the abnormal data point is determined to be a false anomaly, and no alarm is triggered.

[0193] Specifically, the alarm threshold is the average distance of all data points to the centroid in the visualization space.

[0194] It is understandable that, in one embodiment, the average distance from all points to the center coordinates of the normal point can be calculated using Euclidean distance, and then the percentage of the distance from any point to the center coordinates of the normal point relative to the average distance can be expressed as a measure of the degree of anomaly. Furthermore, no dimensional standardization is performed in this operation because the difference between anomaly points and normal points is not significant after standardization of the three feature vectors used. Therefore, a direct comparison in the original space is chosen. The average distance from the anomaly point to the center coordinates of the normal point is 116.31, and the re-inspection threshold is set to 232.6. As shown in Table V, the distance from the second production process to the center coordinates of the normal point is 1078.5753 > 232.6, with an anomaly percentage of 927.8%. The distance from the 34th production process to the center coordinates of the normal point is 198.9082 < 232.6, with an anomaly percentage of 171%. The 34th production process is determined to be a normal production process.

[0195] After on-site investigation, it was found that the aluminum ingot samples in the above production cycle did indeed have product quality problems, and some samples developed micro-cracks during subsequent processing.

[0196] Example 2

[0197] Based on the same inventive concept, this application also provides a data-driven aluminum alloy homogenizing furnace abnormal state detection system for implementing the above-mentioned system. The solution provided by this system is similar to the implementation scheme described in the above method. Therefore, the specific limitations of the data-driven aluminum alloy homogenizing furnace abnormal state detection system described below can be found in the limitations of the data-driven aluminum alloy homogenizing furnace abnormal state detection method described above, and will not be repeated here.

[0198] Specifically, a data-driven abnormal state detection system for an aluminum alloy homogenizing furnace includes: a data storage module, a data acquisition module, a segmentation module, a preliminary judgment module for the heating section, a preliminary judgment module for natural gas, and a verification module;

[0199] The data storage module has built-in heating power standard control curve and natural gas standard control curve. The heating power standard control curve is obtained by fitting the heating section data of the aluminum alloy homogenizing furnace under normal conditions with the time domain response curve of the second-order control system under a step signal. The natural gas standard control curve is obtained by fitting the natural gas consumption data of the aluminum alloy homogenizing furnace under normal conditions with the time domain response curve of the second-order control system under a step signal.

[0200] The data acquisition module is used to acquire natural gas consumption data and electricity consumption data for each production process of the aluminum alloy homogenizing furnace under test.

[0201] The segmentation module is used to segment the power consumption data to obtain the heating segment data;

[0202] The heating section preliminary judgment module is used to call the heating power standard control curve in the data storage module, calculate the cumulative error between the heating section data and the preset heating power standard control curve through a fixed sliding window, and preliminarily determine that the next production process is abnormal when the cumulative error is greater than the set first cumulative error threshold.

[0203] The natural gas preliminary judgment module is used to call the natural gas standard control curve in the data storage module, calculate the cumulative error between the natural gas consumption data and the preset natural gas standard control curve through a fixed sliding window, and preliminarily determine that the production process is abnormal when the cumulative error is greater than the set second cumulative error threshold.

[0204] The verification module is used to verify and issue alarms for the natural gas consumption data and electricity consumption data during this production process using the isolated forest anomaly detection model.

[0205] Example 3

[0206] This application provides an embodiment of a computer device. The computer device may be a terminal, and its internal structure diagram may be as follows: Figure 16 As shown, the computer device includes a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. When executed by the processor, the computer program implements a data-driven method for detecting abnormal states in an aluminum alloy homogenizing furnace. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0207] Those skilled in the art will understand that Figure 16 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0208] Example 4

[0209] Based on the above embodiments, this embodiment provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the data-driven aluminum alloy homogenizing furnace abnormal state detection method described in Embodiment 1.

[0210] The above embodiments are merely illustrative of the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. Any modifications made to the technical solutions based on the technical concept proposed in this invention shall fall within the scope of protection of this invention.

Claims

1. A data-driven method for detecting abnormal states in an aluminum alloy homogenizing furnace, characterized in that, The method includes the following steps: The heating section data of the aluminum alloy homogenizing furnace under normal conditions was pre-fitted with the time domain response curve of the second-order control system under a step signal to obtain the standard control curve of heating power. Natural gas consumption data under normal conditions of aluminum alloy homogenizer were pre-fitted with the time-domain response curve of second-order control system under step signal to obtain natural gas standard control curve; Obtain natural gas consumption and electricity consumption data for each production process of the aluminum alloy homogenizing furnace under test; Extract heating section data from power consumption data, calculate the cumulative error between heating section data and preset heating power standard control curve through a fixed sliding window, and preliminarily determine that the current production process is abnormal when the cumulative error is greater than the set first cumulative error threshold. And / or, calculate the cumulative error between natural gas consumption data and the preset natural gas standard control curve through a fixed sliding window, and preliminarily determine that the current production process is abnormal when the cumulative error is greater than the set second cumulative error threshold; An isolated forest anomaly detection model was used to verify and issue alarms for natural gas consumption and electricity consumption data during this production process.

2. The data-driven method for detecting abnormal states in an aluminum alloy homogenizing furnace according to claim 1, characterized in that, The calculation formula for the first cumulative error threshold or the second cumulative error threshold is: T=μ+kσ; where μ is the cumulative mean of the heating section data sample under normal conditions of the aluminum alloy homogenizing furnace, or the cumulative mean of the natural gas consumption data sample; σ is the cumulative standard deviation of the heating section data sample under normal conditions of the aluminum alloy homogenizing furnace or the cumulative standard deviation of the natural gas consumption data sample; and k is the preset safety factor. The formula for calculating the length of a fixed sliding window is L∈[f s / 10, f s / 2];where f s This refers to the data sampling frequency.

3. A data-driven method for detecting abnormal states in an aluminum alloy homogenizing furnace according to claim 1 or 2, characterized in that, The steps for obtaining natural gas consumption data and electricity consumption data for each production process of the aluminum alloy homogenizing furnace under test include: After acquiring real-time data on natural gas consumption and electricity consumption during the production process of the aluminum alloy homogenizing furnace to be tested, the data is divided into multiple complete production processes according to the production cycle; or a combination of at least 5 consecutive and non-zero data points is determined as a complete production process; each complete production process is used as a valid sample for detecting abnormal conditions of the aluminum alloy homogenizing furnace.

4. The data-driven method for detecting abnormal states in an aluminum alloy homogenizing furnace according to claim 3, characterized in that, An isolated forest anomaly detection model was used to verify and issue alarms for natural gas consumption and electricity consumption data during this production process, including: Energy consumption index features and data distribution index features are extracted from the natural gas consumption data and electricity consumption data of this production process as feature vectors, and then fed into the trained isolated forest anomaly detection model to identify normal data points and abnormal data points. The energy consumption index features include total natural gas energy consumption, total electricity energy consumption, and the ratio of total natural gas energy consumption to total electricity energy consumption; the data distribution index features include the Gamma distribution, skewness, and kurtosis of natural gas consumption data and electricity consumption data. The total energy consumption of natural gas, the total energy consumption of electricity, and the ratio of total energy consumption of natural gas to total energy consumption of electricity are selected as visualization feature dimensions, and normal data points and abnormal data points are projected onto the visualization space. Calculate the centroids of all normal data points in the visualization space; Calculate the Euclidean distance between the centroids of abnormal data points and normal data points. If the Euclidean distance is greater than the alarm threshold, the abnormal data point is determined to be a real anomaly, and an alarm is triggered. If the Euclidean distance is less than or equal to the alarm threshold, the abnormal data point is determined to be a false anomaly, and no alarm is triggered. The training steps for the isolated forest anomaly detection model include: Historical natural gas consumption and electricity consumption data of aluminum alloy homogenizing furnaces were obtained as a training set. Extract the energy consumption index features and data distribution index features of each sample in the training set as feature vectors; The feature vectors are used as input to train an isolated forest anomaly detection model.

5. The data-driven method for detecting abnormal states in an aluminum alloy homogenizing furnace according to claim 4, characterized in that, The steps for obtaining the standard control curve include: The consumption data y(t) and corresponding data and the set control input data u(t) in each valid sample of the homogenizer are obtained, and an initial differential equation model describing the dynamic relationship between output and input using system parameters is constructed; the consumption data y(t) is natural gas consumption data or heating section data; The consumed data y(t) and the control input data u(t) are processed by a preset low-pass filter to obtain the filtered estimates of their derivatives of each order. Using the filtered estimate, the initial differential equation model is transformed into a standard linear regression equation with system parameters as variables; The linear regression equation is solved using the least squares method to obtain estimated values ​​of the system parameters. The estimated system parameters are substituted into a pre-defined second-order transfer function structure to generate a transfer function model, the formula of which is as follows: Where A represents the control input data u(t), which is the steady-state value of natural gas or electricity consumption data in each production process; a1 and a2 represent output coefficients, and b1, b2, c1, and c2 represent input coefficients; For the transfer function model identified from multiple valid samples based on the homogenizing furnace, the mean values ​​of the input and output coefficients are calculated respectively. The obtained average input and output coefficients are then substituted into the preset second-order transfer function structure to obtain the standard transfer function model. The standard control curve is obtained based on the standard transfer function model.

6. The data-driven method for detecting abnormal states in an aluminum alloy homogenizing furnace according to claim 4, characterized in that, Extract heating section data from the electricity consumption data, including: A theoretical curve based on process standards is selected as the benchmark energy consumption curve; Set a third cumulative error threshold; For each complete production process, starting from the starting point of the power consumption data, the continuous power consumption data is included in a sliding window, and the cumulative error between all power consumption data in the sliding window and the corresponding benchmark power consumption curve is calculated in real time. When the cumulative error value exceeds the third cumulative error threshold, the end point of the current window is marked as a segment point; the cumulative error value is reset, and the next window is started from this point as the new starting point, until the power consumption data is traversed, and the multiple segments of power consumption data are divided into heating segment - heat preservation segment - cooling segment.

7. The data-driven method for detecting abnormal states in an aluminum alloy homogenizing furnace according to claim 4, characterized in that, After obtaining the standard control curve, the Makima interpolation method is used to smooth the standard control curve. The steps include: Obtain consumption data and calculate the slope between two adjacent consumption data points. The calculation formula is as follows: Where x i Indicates time, y i This represents the consumption data at the corresponding time point; Calculate the weighting factor w i The calculation formula is as follows: The data point x is calculated using the following formula, based on the weighting factor and the slope between the two consumption data points. i slope k i : At data boundary point x i Between x and x, a cubic polynomial is constructed based on Makima interpolation, and its formula is: Among them, a i b i c i d i The coefficients of the polynomial are given by the difference condition y(x) and the slope k. i Determined; x is the data point x i and x i+1 Any data between; By concatenating the cubic polynomials over all data intervals, a smooth standard control curve is obtained.

8. A data-driven abnormal state detection system for an aluminum alloy homogenizing furnace, comprising: Data storage module, data acquisition module, segmentation module, preliminary judgment module for heating section, preliminary judgment module for natural gas, and verification module; The data storage module has built-in heating power standard control curve and natural gas standard control curve. The heating power standard control curve is obtained by fitting the heating section data of the aluminum alloy homogenizing furnace under normal conditions with the time domain response curve of the second-order control system under a step signal. The natural gas standard control curve is obtained by fitting the natural gas consumption data of the aluminum alloy homogenizing furnace under normal conditions with the time domain response curve of the second-order control system under a step signal. The data acquisition module is used to acquire natural gas consumption data and electricity consumption data for each production process of the aluminum alloy homogenizing furnace under test. The segmentation module is used to segment the power consumption data to obtain the heating segment data; The heating section preliminary judgment module is used to call the heating power standard control curve in the data storage module, calculate the cumulative error between the heating section data and the preset heating power standard control curve through a fixed sliding window, and preliminarily determine that the next production process is abnormal when the cumulative error is greater than the set first cumulative error threshold. The natural gas preliminary judgment module is used to call the natural gas standard control curve in the data storage module, calculate the cumulative error between the natural gas consumption data and the preset natural gas standard control curve through a fixed sliding window, and preliminarily determine that the production process is abnormal when the cumulative error is greater than the set second cumulative error threshold. The verification module is used to verify and issue alarms for the natural gas consumption data and electricity consumption data during this production process using the isolated forest anomaly detection model.

9. A computer device, characterized in that: It includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; A memory is used to store computer programs; a processor is used to execute the programs stored in the memory to implement the data-driven method for detecting abnormal states of aluminum alloy homogenizing furnaces as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the data-driven method for detecting abnormal states of aluminum alloy homogenizing furnaces as described in any one of claims 1 to 7.

Images