An intelligent device start-stop prediction method, a computer device and a storage medium

Abstract

The application belongs to the technical field of equipment steelmaking, and discloses an equipment start-stop intelligent prediction method, a computer device and a storage medium. The method comprises the following steps: acquiring standard time sequence data of each operating parameter of the equipment within a preset sampling period; determining the working state of the equipment at each sampling time within the preset sampling period based on the standard time sequence data and a preset state determination condition, and generating state sequence data; determining the structured statistical feature data of the equipment based on the standard time sequence data and the state sequence data; generating an equipment operation summary text according to the structured statistical feature data and a preset natural language template, and inputting the equipment operation summary text into a large language model to determine a semantic embedding vector; determining a structured feature vector according to the structured statistical feature data, and fusing the structured feature vector with the semantic embedding vector to obtain a fusion vector; and determining the future state change of the equipment based on the semantic embedding vector and a first prediction model. The application can improve the accuracy of predicting the start-stop of an electric furnace.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of equipment steelmaking, in particular to an equipment start-stop intelligent prediction method, a computer device and a storage medium. BACKGROUND

[0002] In continuous production such as equipment steelmaking, the accuracy of start-stop prediction is directly related to energy consumption control, furnace lining life, dust removal and feeding coordination, personnel and equipment scheduling, and safety boundary control. The existing technology is restricted by many factors: the equipment operation cycle is affected by multi-source disturbances such as material type ratio, power distribution strategy, steel grade switching, temperature control target, shift and seasonal differences, upstream feeding and downstream tapping cycle, etc., showing strong nonlinearity and non-stationary characteristics; the electric arc working condition has high-frequency fluctuations and short-time instantaneous interruption, which is easy to be confused with the real stop in signal form; the state boundaries of "stop, standby, low load, process suspension" are different in different field areas, causing inconsistent historical labeling; field data often accompanied by noise, missing, sensor drift, delay and timestamp misplacement, asynchronous sampling makes it difficult to align cross-signal correlation; key stop events are naturally sparse, and the class distribution is extremely unbalanced, making it difficult to cover all working conditions, which is easy to cause overfitting and insufficient generalization; the parameters system is obviously different after the cross-line, cross-equipment and new and old production capacity reconstruction, resulting in significant decay of prediction accuracy in migration application; the maintenance cost of fixed threshold and experience rule is high, and the adaptability is poor, which is easy to produce false alarm and miss alarm; the caliber and governance level are inconsistent between different systems, which also brings the result difference of "different judgment in the same furnace". Based on the above objective limitations, the existing technology is still difficult to meet the requirements of energy consumption optimization, equipment protection and cycle coordination in predicting the start-stop of the equipment in terms of stability and accuracy. SUMMARY

[0003] The purpose of the embodiments of the present application is to provide an equipment start-stop intelligent prediction method, a computer device and a storage medium, to solve the problem of low accuracy of existing technology in predicting the start-stop of the electric furnace.

[0004] To achieve the above purpose, the first aspect of the present application provides an equipment start-stop intelligent prediction method, which comprises:

[0005] acquiring standard time series data of each operating parameter of the target equipment within a preset sampling period;

[0006] determining the working state of the target equipment at each sampling time within the preset sampling period based on the standard time series data and the preset state determination condition, and generating state sequence data;

[0007] determining the structured statistical feature data of the target equipment based on the standard time series data and the state sequence data;

[0008] According to the structured statistical feature data and the preset natural language template, a device operation summary text is generated, and the device operation summary text is input into a large language model to determine a semantic embedding vector;

[0009] According to the structured statistical feature data, a structured feature vector is determined, and the semantic embedding vector and the structured feature vector are fused to obtain a fusion vector;

[0010] Based on the semantic embedding vector and the first prediction model, a future state change of the target device is determined, and based on the fusion vector and the second prediction model, a future abnormal shutdown probability of the target device is determined.

[0011] In the embodiments of the present application, the target device is an electric furnace, the electric furnace includes a dust removal fan and an electric furnace body; the operating parameters include the current value of the dust removal fan and the electric furnace inclination angle value of the electric furnace body; based on the standard time sequence data and the preset state determination condition, the working state of the target device at each sampling time in the preset sampling period is determined, and the state sequence data is generated, including: determining the current value of the dust removal fan at each sampling time; determining the electric furnace inclination angle value of the electric furnace body at each sampling time; determining the working state of the dust removal fan according to the current value at each sampling time and the corresponding preset state determination condition; determining the working state of the electric furnace body according to the electric furnace inclination angle value at each sampling time and the corresponding preset state determination condition; in the case that the dust removal fan is in the state of starting or the electric furnace body is in the state of starting, it is determined that the target device is in the state of starting at the sampling time; in the case that the dust removal fan is in the state of starting and the electric furnace body is in the state of stopping, it is determined that the target device is in the state of stopping at the sampling time; the state sequence data is generated based on the sampling time of the standard time sequence data, the current value at each sampling time, the electric furnace inclination angle value and the working state of the target device at the sampling time.

[0012] In the embodiments of the present application, the preset state determination condition corresponding to the dust removal fan satisfies formula (1):

[0013] (1)

[0014] wherein, is the working state of the dust removal fan; is the current value corresponding to the dust removal fan at the sampling time t; is the current threshold value corresponding to the dust removal fan in the state of starting; is the current threshold value corresponding to the dust removal fan in the state of stopping; is the minimum duration required for the dust removal fan to maintain the state of starting; is the minimum duration required for the dust removal fan to maintain the state of stopping; is the duration for the dust removal fan to maintain the working state.

[0015] In the embodiments of the present application, the preset state judgment condition corresponding to the electric furnace body also satisfies formula (2):

[0016] (2)

[0017] wherein, is the working state of the electric furnace body; is the electric furnace inclination value corresponding to the electric furnace body at the sampling time t; is the electric furnace inclination threshold value corresponding to the electric furnace body in the start-up state; is the electric furnace inclination threshold value corresponding to the electric furnace body in the shutdown state; is the minimum duration required for the electric furnace body to maintain the start-up state; is the minimum duration required for the electric furnace body to maintain the shutdown state; is the duration for the electric furnace body to maintain the working state.

[0018] In the embodiments of the present application, determining the structured statistical feature data based on the standard time sequence data and the state sequence data includes: determining the feature data of the dust removal fan based on the standard time sequence data of the current value of the dust removal fan; determining the feature data of the electric furnace body based on the standard time sequence data of the electric furnace inclination value of the electric furnace body; determining the number of state switches of the target device based on the state sequence data; and determining the structured statistical feature data based on the feature data of the dust removal fan, the feature data of the electric furnace body, and the number of state switches of the target device.

[0019] In the embodiments of the present application, determining the feature data of the dust removal fan based on the standard time sequence data of the current value of the dust removal fan includes: determining the average current value of the dust removal fan based on the standard time sequence data of the current value of the dust removal fan; obtaining the maximum current value of the dust removal fan in the standard time sequence data; determining the standard deviation of the current fluctuation of the dust removal fan based on the standard time sequence data; screening out the current value exceeding the preset rated upper limit in the standard time sequence data, taking the time point corresponding to the current value as an overload time point, and counting the number of overload time points as the number of overloads; and taking the average current value, the maximum current value, the standard deviation of the current fluctuation, and the number of overloads as the feature data of the dust removal fan.

[0020] In the embodiments of the present application, determining the feature data of the electric furnace body based on the standard time sequence data of the electric furnace inclination value of the electric furnace body includes: determining the average electric furnace inclination value of the electric furnace body based on the standard time sequence data of the electric furnace inclination value of the electric furnace body; obtaining the maximum electric furnace inclination value of the electric furnace body in the standard time sequence data; determining the number of inclination changes of the electric furnace body according to the change amplitude between the electric furnace inclination values in the adjacent two time points in the standard time sequence data and the preset amplitude change range; and taking the average electric furnace inclination value, the maximum electric furnace inclination value, and the number of inclination changes as the feature data of the electric furnace body.

[0021] In the embodiments of the present application, the structured feature vector satisfies formula (3):

[0022] (3)

[0023] wherein, is the structured feature vector; indicates that the structured feature vector is 9-dimensional, including the average current value, the maximum current value, the current fluctuation standard deviation, the overload frequency, the average furnace inclination value, the maximum furnace inclination value, the inclination change frequency, the switching frequency and the continuous running time length of the target device, and the continuous running time length is obtained based on the standard timing data.

[0024] The second aspect of the present application provides a computer device, comprising:

[0025] a memory configured to store instructions;

[0026] and a processor configured to call the instructions from the memory and capable of implementing the above-mentioned method when executing the instructions.

[0027] The third aspect of the present application provides a machine-readable storage medium having instructions stored thereon, the instructions being used to cause a machine to execute the above-mentioned method.

[0028] Through the above technical solution, the target device running parameters are acquired and processed under a unified time reference, the state sequence data is generated in combination with state determination, the structured statistical feature data quantitatively representing the device working condition is extracted, the device running summary text is generated based on a preset natural language template, the semantic embedding vector is obtained through a large language model, and then the structured feature vector is fused to form a fusion vector, and finally different prediction models are used to predict the future state change situation and the future abnormal shutdown probability of the target device, so that the complementary advantages of numerical features and semantic features are retained, the forward-looking judgment ability of the target device running trend and abnormal risk is significantly improved, and the accuracy of state recognition and the reliability of prediction are improved.

[0029] Other features and advantages of the embodiments of the present application will be described in detail in the following specific embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0030] The accompanying drawings are used to provide a further understanding of the embodiments of the present application, and constitute a part of the specification, and are used to explain the embodiments of the present application together with the following specific embodiments, but do not constitute a limitation on the embodiments of the present application. In the drawings:

[0031] Figure 1 a flowchart of a device start-stop intelligent prediction method according to the embodiments of the present application is schematically shown;

[0032] Figure 2 Fig. 1 schematically shows an architecture diagram of an intelligent equipment start-stop prediction system according to an embodiment of the present application;

[0033] Figure 3 Fig. 2 schematically shows a structural diagram of a computer device according to an embodiment of the present application. DETAILED DESCRIPTION

[0034] In order to make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. It should be understood that the specific embodiments described herein are merely used to explain and illustrate the embodiments of the present application and should not be used to limit the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative effort fall within the scope of protection of the present application.

[0035] It should be noted that if the embodiments of the present application involve directional indications (such as up, down, left, right, front, back, etc.), the directional indications are merely used to explain the relative positional relationship, movement condition, etc. between components in a certain posture (as shown in the drawings), and if the certain posture changes, the directional indications also change accordingly.

[0036] In addition, if the embodiments of the present application involve descriptions of “first”, “second”, etc., the descriptions of “first”, “second”, etc. are merely for description purposes and should not be understood as indicating or implying the relative importance of the technical features indicated or the number of the technical features indicated. Therefore, the features limited by “first”, “second” can explicitly or implicitly include at least one of the features. In addition, the technical solutions of the various embodiments can be combined with each other, but it must be based on the fact that a person of ordinary skill in the art can realize the combination, and when the combination of the technical solutions contradicts each other or cannot be realized, it should be considered that the combination of the technical solutions does not exist and is not within the scope of protection claimed by the present application.

[0037] Figure 1 Fig. 1 schematically shows a flowchart of an intelligent equipment start-stop prediction method according to an embodiment of the present application. As shown in Fig. 1, the present application provides an intelligent equipment start-stop prediction method, which can include the following steps. Figure 1

[0038] Step 101: Obtain standard time sequence data of each operating parameter of a target equipment in a preset sampling period.

[0039] ​In the embodiments of the present application, the target device refers to a specific device or a logical object composed of multiple physical units in a production system that is monitored and evaluated; the operating parameter refers to a measurable quantity and a discrete state quantity that can represent the working condition and state of the object, such as current, power, inclination, temperature, pressure, rotating speed, valve position, on-off quantity, etc.; the preset sampling period can refer to a fixed time interval that is set in advance according to the field beat, signal change rate and resource constraint, and is used as the step length for unified sampling or resampling; the standard time series data refers to a structured data set that is indexed by a time stamp, arranged at equal intervals according to the preset sampling period, and normalized in unit, field name, device identification and time reference, which is data that has been cleaned.

[0040] Step 102: Based on the standard time series data and the preset state determination condition, the working state of the target device at each sampling time within the preset sampling period is determined, and state sequence data is generated.

[0041] In the embodiments of the present application, the preset state determination condition can refer to a set of state recognition rules set for the target device, which can include threshold interval, hysteresis upper and lower limits, minimum duration and logical relationship between multiple signals, etc., for maintaining consistency and noise resistance of determination under different working conditions and signal quality conditions; the working state refers to a discrete category that can represent the running mode of the target device, for reflecting the working condition recognition result of the device at the sampling time, such as running, shutdown, standby or undetermined; the state sequence data refers to a structured data sequence formed by the working state label and its time stamp arranged in time sequence, which can be accompanied by derived markers such as state holding time and state switching position, to support subsequent quantitative analysis of continuous running time, state switching times and out-of-limit events.

[0042] Step 103: Based on the standard time series data and the state sequence data, the structured statistical feature data of the target device is determined.

[0043] In the embodiments of the present application, the structured statistical feature data refers to a fixed-length numerical vector formed by aggregating, counting and deriving the operating parameters and state information within the monitoring period according to the agreed fields, units and calculation scope, which includes not only parameter statistics (such as mean, extreme value, standard deviation, quantile, change rate, fluctuation amplitude, etc.) and state statistics (such as continuous running time, shutdown time, state proportion, state switching times and switching positions, etc.), but also event statistics and threshold-related counting items (such as out-of-limit times, out-of-limit duration, mutation times). Under the consistent time reference of the preset sampling period, the standard time series data and the state sequence data are aggregated and measured according to the unified scope, to generate fixed-length, comparable and traceable structured statistical feature data of the target device, so that the dispersed parameter records and state results are converted into a structured representation that can be directly called by subsequent processing.

[0044] Step 104: generating a device operation summary text according to the structured statistical feature data and the preset natural language template, inputting the device operation summary text into a large language model to determine a semantic embedding vector.

[0045] In the embodiments of the present application, the preset natural language template refers to a set of text generation rules pre-set according to a specific field order, expression manner and unit format, which is used to ensure the consistency of the device operation summary texts generated by different time periods or different target devices in terms of semantics and structure; the device operation summary text refers to a natural language description generated by the preset natural language template based on the structured statistical feature data, which is used to express quantitative statistical information in the form of words; the large language model refers to a model with natural language processing capability, which can receive the device operation summary text and output a corresponding semantic embedding vector; the semantic embedding vector refers to a fixed-length numerical vector generated by the large language model based on the device operation summary text, which is used to represent the semantic information contained in the text in a multi-dimensional space, so as to facilitate subsequent calculation, comparison and analysis.

[0046] Step 105: determining a structured feature vector according to the structured statistical feature data, and fusing the semantic embedding vector and the structured feature vector to obtain a fused vector.

[0047] In the embodiments of the present application, the structured feature vector refers to a fixed-length vector arranged and combined by the structured statistical feature data in a fixed field order and a given numerical form, which is used to directly reflect the quantitative characteristics of the target device operation; the fused vector refers to a comprehensive numerical vector formed by merging the structured feature vector and the semantic embedding vector through feature splicing or the like, which is fixed in dimension and simultaneously has the characteristics of being computable, comparable and inputtable into a prediction model in terms of semantic expression and numerical characterization.

[0048] Step 106: determining the future state change of the target device based on the semantic embedding vector and the first prediction model, and determining the future abnormal shutdown probability of the target device based on the fused vector and the second prediction model.

[0049] In the embodiments of the present application, the first prediction model can refer to a model capable of predicting the future state of the target device based on the semantic embedding vector, and the prediction result is the state change at each sampling time in the future; the fusion vector refers to a comprehensive numerical vector obtained by combining the structured feature vector and the semantic embedding vector according to a fixed rule, which has numerical features and semantic features; the second prediction model refers to a model capable of calculating the future abnormal shutdown probability based on the fusion vector, and the output is a probability value representing the risk degree of abnormal shutdown of the target device in the prediction period. Preferably, the first prediction model is an LSTM model (Long Short Term Memory model); and the second prediction model is an XGBoost model (a boosting enhancement strategy based additive model).

[0050] Through the above technical solutions, the target device operating parameters are acquired and processed under a unified time reference, the state sequence data is generated in combination with the state determination, the structured statistical feature data quantitatively representing the device working condition is extracted, the device operation summary text is generated based on the preset natural language template, the semantic embedding vector is obtained through the large language model, the fusion vector is formed by fusing the structured feature vector, and finally the different prediction models are used to respectively predict the future state change of the target device and the future abnormal shutdown probability, so that the complementary advantages of numerical features and semantic features are retained, the forward-looking judgment ability of the target device operating trend and abnormal risk is significantly improved, and the accuracy of state recognition and the reliability of prediction are improved.

[0051] In the embodiments of the present application, the method can further include: determining the health score and / or the abnormal probability of the target device based on the fusion vector and a third prediction model.

[0052] In the embodiments of the present application, preferably, the third prediction model can be an MLP model (Multi-Layer Perceptron model). The health score is a quantitative evaluation value of the comprehensive operating state of the device in the current or prediction period, which is used to reflect the degree of device health; and the abnormal probability is the possibility of the device in the prediction period to be in an abnormal operating state or to fail, which is used to represent the risk level.

[0053] In the embodiment of the present application, the target device can be an electric furnace, the electric furnace comprising a dust removal fan and an electric furnace body; the operating parameters comprise a current value of the dust removal fan and an electric furnace inclination value of the electric furnace body; based on the standard time sequence data and the preset state determination condition, the working state of the target device at each sampling time in a preset sampling period is determined, and the state sequence data is generated, which can comprise: determining the current value of the dust removal fan at each sampling time; determining the electric furnace inclination value of the electric furnace body at each sampling time; determining the working state of the dust removal fan according to the current value at each sampling time and the corresponding preset state determination condition; determining the working state of the electric furnace body according to the electric furnace inclination value at each sampling time and the corresponding preset state determination condition; in the case that the dust removal fan is in the start-up state or the electric furnace body is in the start-up state, it is determined that the target device is in the start-up state at the sampling time; in the case that the dust removal fan is in the start-up state and the electric furnace body is in the stop state, it is determined that the target device is in the stop state at the sampling time; the state sequence data is generated based on the sampling time of the standard time sequence data, the current value at each sampling time, the electric furnace inclination value and the working state of the target device at the sampling time.

[0054] In the embodiment of the present application, when the target device is an electric furnace, it is composed of a dust removal fan and an electric furnace body; the operating parameters comprise a current value of the dust removal fan and an electric furnace inclination value of the electric furnace body, the current value being a real-time value capable of representing the operating load condition of the dust removal fan, and the electric furnace inclination value being a real-time value capable of reflecting the inclination angle and working posture of the electric furnace body; the preset state determination condition is a rule set according to the equipment operating characteristics and engineering experience, which is used to determine the start-up or stop state in the case that different operating parameters reach or are lower than a specific threshold value and continue for a certain time; the working state is a discrete identification of the operating mode of the equipment at a certain sampling time, the start-up state indicating that the equipment is in the running operation, and the stop state indicating that the equipment is in the non-running state; the state sequence data is a structured data set containing the sampling time, the current value, the electric furnace inclination value and the corresponding working state recorded in sequence according to the sampling time, which is used to completely depict the running state change process of the target device in the monitoring period.

[0055] Taking the electric furnace as the target device, the key operating parameters thereof are collected and the state is determined in a unified sampling period, the working states of the dust removal fan and the electric furnace body at each sampling time are determined according to the preset state determination condition by combining the current value of the dust removal fan and the electric furnace inclination value of the electric furnace body, and on this basis, the state sequence data capable of accurately reflecting the overall running condition of the electric furnace is generated, thereby providing the basis data with consistent time sequence and unified caliber for subsequent statistical feature extraction, running trend analysis and prediction model input.

[0056] In the embodiment of the present application, the preset state determination condition corresponding to the dust removal fan satisfies formula (1):

[0057] (1)

[0058] wherein, is the working state of the dust removal fan; is the current value corresponding to the dust removal fan at the sampling time t; is the current threshold value corresponding to the dust removal fan in the start state; is the current threshold value corresponding to the dust removal fan in the stop state; is the minimum duration required for the dust removal fan to maintain the start state; is the minimum duration required for the dust removal fan to maintain the stop state; is the duration for the dust removal fan to maintain the working state.

[0059] In the embodiments of the present application, the preset state determination condition corresponding to the electric furnace body also satisfies formula (2):

[0060] (2)

[0061] wherein, is the working state of the electric furnace body; is the electric furnace inclination value corresponding to the electric furnace body at the sampling time t; is the electric furnace inclination threshold value corresponding to the electric furnace body in the start state; is the electric furnace inclination threshold value corresponding to the electric furnace body in the stop state; is the minimum duration required for the electric furnace body to maintain the start state; is the minimum duration required for the electric furnace body to maintain the stop state; is the duration for the electric furnace body to maintain the working state.

[0062] In the embodiments of the present application, determining the structured statistical feature data based on the standard time sequence data and the state sequence data includes: determining the feature data of the dust removal fan according to the standard time sequence data of the current value of the dust removal fan; determining the feature data of the electric furnace body according to the standard time sequence data of the electric furnace inclination value of the electric furnace body; determining the number of state switches of the target device according to the state sequence data; determining the structured statistical feature data based on the feature data of the dust removal fan, the feature data of the electric furnace body, and the number of state switches of the target device.

[0063] In the embodiments of the present application, the feature data of the dust removal fan refers to a numerical set obtained by statistical calculation of the operation parameters of the dust removal fan based on the standard time series data, and is used to reflect the operation level and fluctuation characteristics of the dust removal fan in the monitoring period; the feature data of the electric furnace body refers to a numerical set obtained by statistical calculation of the operation parameters of the electric furnace body based on the standard time series data, and is used to reflect the working posture and change rule of the electric furnace body in the monitoring period; the state switching times refer to the cumulative number of times that the target device is converted from one working state to another working state in the state sequence data, and are used to measure the frequency of the change of the operation state of the device; the structured statistical feature data refers to a fixed-length numerical vector obtained by combining the feature data of the dust removal fan, the feature data of the electric furnace body and the state switching times of the target device according to a unified field order and a fixed numerical format, and is used to quantitatively describe the operation characteristics of the device and as the standardized input of the subsequent processing link.

[0064] Under the unified time reference and data caliber, the standard time series data and the state sequence data are combined and processed, the feature data of the dust removal fan and the feature data of the electric furnace body are extracted respectively, and the state switching times reflecting the frequency of the change of the operation state of the target device in the state sequence data are combined to comprehensively form the structured statistical feature data capable of fully representing the operation characteristics of the target device, thereby providing fixed-length, comparable and traceable feature input for subsequent analysis, modeling and prediction.

[0065] In the embodiments of the present application, the feature data of the dust removal fan is determined according to the standard time series data of the current value of the dust removal fan, including: determining the average current value of the dust removal fan based on the standard time series data of the current value of the dust removal fan; obtaining the maximum current value of the dust removal fan in the standard time series data; determining the current fluctuation standard deviation of the dust removal fan based on the standard time series data; screening out the current value exceeding the preset rated upper limit in the standard time series data, taking the time point corresponding to the current value as the overload time point, and counting the number of overload time points as the overload times; taking the average current value, the maximum current value, the current fluctuation standard deviation and the overload times as the feature data of the dust removal fan.

[0066] In the embodiments of the present application, the current value of the dust removal fan refers to the running current value of the dust removal fan at each sampling time recorded in the standard time sequence data, which is used to represent the load level and running condition thereof; the average current value refers to the arithmetic mean of the current value of the dust removal fan in the monitoring period, which is used to reflect the average level of the overall running load; the maximum current value refers to the maximum value of the current value of the dust removal fan in the sampling time, which is used to reflect the peak load condition occurring in the running process; the current fluctuation standard deviation refers to the standard deviation of the current value of the dust removal fan in the monitoring period, which is used to measure the fluctuation degree and stability of the running current; the preset rated upper limit current value refers to the current threshold value preset according to the rated parameters or running safety standards of the equipment, which is used to determine the overload state; the overload time point refers to the sampling time at which the current value exceeds the preset rated upper limit current value in the standard time sequence data; the overload times refers to the cumulative number of the above overload time points, which is used to reflect the frequency of the overload running of the equipment in the monitoring period; the feature data of the dust removal fan refers to the numerical set obtained by combining the average current value, the maximum current value, the current fluctuation standard deviation and the overload times in the order of the established fields, which is used to quantitatively describe the running characteristics of the dust removal fan and as the input of the subsequent data processing and prediction model.

[0067] In the embodiments of the present application, the feature data of the electric furnace body is determined according to the standard time sequence data of the electric furnace body. The method comprises the following steps: determining the average electric furnace inclination value of the electric furnace body based on the standard time sequence data of the electric furnace inclination value of the electric furnace body; obtaining the maximum electric furnace inclination value of the electric furnace body in the standard time sequence data; determining the inclination change times of the electric furnace body according to the change amplitude between the electric furnace inclination values in the adjacent two time points in the standard time sequence data and the preset amplitude change range; and taking the average electric furnace inclination value, the maximum electric furnace inclination value and the inclination change times as the feature data of the electric furnace body.

[0068] In the embodiments of the present application, the electric furnace inclination value refers to the inclination angle value of the electric furnace body relative to the reference position at each sampling time recorded in the standard time sequence data, which is used to represent the working posture of the electric furnace in the production process; the average electric furnace inclination value refers to the arithmetic mean of the electric furnace inclination values in the monitoring period, which is used to reflect the overall level of the electric furnace posture in the period; the maximum electric furnace inclination value refers to the maximum value of the electric furnace inclination value in the monitoring period, which is used to represent the extreme posture change in the running process; the preset amplitude change range refers to the electric furnace inclination change threshold value set according to the process requirements or operation experience, which is used to determine whether the posture change reaches the effective amplitude in the statistical sense; the inclination change frequency refers to the cumulative number of times that the electric furnace inclination value changes by more than the preset amplitude change range between adjacent two sampling times, which is used to reflect the frequency of electric furnace posture adjustment or process operation; the feature data of the electric furnace body refers to a numerical set composed of the average electric furnace inclination value, the maximum electric furnace inclination value and the inclination change frequency in the order of unified fields, which is used to quantitatively describe the running posture characteristics of the electric furnace body and as the input of subsequent data processing and prediction model.

[0069] In the embodiments of the present application, the structured feature vector satisfies formula (3):

[0070] (3)

[0071] wherein, is the structured feature vector; indicates that the structured feature vector is 9-dimensional, including the average current value, the maximum current value, the current fluctuation standard deviation, the overload frequency, the average electric furnace inclination value, the maximum electric furnace inclination value, the inclination change frequency, the switching frequency and the continuous running time of the target device, and the continuous running time is obtained based on the standard time sequence data.

[0072] Through the above technical solution, the collection and standardization processing of the running parameters of the target device are realized under the unified sampling period, the state sequence data which can accurately reflect the running condition of the device is generated combined with the preset state judgment condition, and on this basis, the feature data of the dust removal fan and the electric furnace body and the state switching frequency of the target device are extracted to generate structured statistical feature data; further, the structured statistical feature data is converted into device running summary text by the preset natural language template and input into the large language model to generate semantic embedding vectors, which are fused with the structured feature vectors to form fusion vectors; finally, different prediction models are used to predict and evaluate the future state change, the future abnormal shutdown probability, the health score and / or the abnormal probability of the target device based on the semantic embedding vectors and the fusion vectors, so as to significantly improve the accuracy of the target device running state recognition and future risk prediction on the basis of the complementarity of numerical features and semantic features.

[0073] Figure 2An architecture diagram of an equipment start-stop intelligent prediction system according to an embodiment of the present application is schematically shown. As shown in the figure, Figure 2 In the embodiment of the present application, the system is divided into a data acquisition layer, a data processing and storage layer, a model analysis layer, and a visualization management layer. Data and results are gradually transmitted between each link, and a closed-loop prediction and management mechanism is finally formed.

[0074] In the data acquisition layer, the running parameters of the target equipment within the preset sampling period are collected through PLC, DSC, SCADA, instruments and meters, sensors, and manual input, etc. The running parameters include but are not limited to the current value of the dust removal fan and the electric furnace inclination angle value of the electric furnace body. After these raw data are preprocessed by the data processing and storage layer, such as missing value filling, noise filtering, abnormal value marking, time alignment, dynamic deduplication, etc., standard time series data with unified standards, time synchronization, and traceability are formed, laying a foundation for subsequent analysis.

[0075] In the data processing and storage layer, the raw time series data are processed by missing value filling, noise filtering, abnormal value marking, time alignment, dynamic deduplication, etc. to remove incomplete or unreliable records, and are stored and accessed at high speed through distributed storage. The role of this layer is to convert raw data into standard time series data, ensuring consistent time reference, unified data standards, traceable source, and the ability to handle large-scale data storage and concurrent access, providing stable data input for state determination and feature extraction in the upper layer.

[0076] In the model analysis layer, first, the working states of the dust removal fan and the electric furnace body at each sampling time are determined based on the standard time series data and the preset state determination conditions, and the working state of the target equipment as a whole is generated through joint logic to form state sequence data. Then, the average current value, maximum current value, current fluctuation standard deviation, and overload times of the dust removal fan, and the average electric furnace inclination angle value, maximum electric furnace inclination angle value, and inclination angle change times of the electric furnace body are calculated according to the standard time series data, and combined with the state switching times in the state sequence data, complete structured statistical feature data are generated, and a structured feature vector is constructed. Next, the structured statistical feature data is converted into an equipment operation summary text through a preset natural language template, and is input into a locally deployed large language model to generate a semantic embedding vector. The semantic embedding vector and the structured feature vector are fused to obtain a fusion vector.

[0077] In the prediction stage, the semantic embedding vector is output to a first prediction model (such as LSTM) to predict the future state change of the target device; the fusion vector is input to a second prediction model (such as XGBoost) to calculate the probability of future abnormal shutdown; and the fusion vector is input to a third prediction model (such as MLP) to output a health score and / or an abnormal probability. These prediction results, together with historical and real-time data, enter the visualization management layer for model evaluation, trend analysis, state display, and can trigger alarm notifications and generate maintenance work orders through a rule engine to realize closed-loop management of prediction results and operation and maintenance execution.

[0078] In the visualization management layer, the prediction results and real-time data are used for model evaluation, trend analysis, state display, alarm notification, and maintenance work order generation. The model evaluation module evaluates the prediction accuracy and stability, the trend analysis module provides visualization of the time evolution of the running state and risk, the state display module intuitively presents the current and predicted device state, the alarm notification module triggers alarm information push when the risk exceeds the threshold, the maintenance work order module converts the prediction results into operation and maintenance tasks, and the rule engine automatically executes event triggering and management actions according to the prediction results and on-site strategies. The combination of model output and operation and maintenance management realizes full-link closed-loop management from prediction to execution, improving operation and maintenance response efficiency and decision-making scientificity.

[0079] Through the above technical solutions, the whole process from multi-source collection, standardized processing, state determination, feature extraction, semantic description, feature fusion to multi-model prediction and result management is realized, taking into account the complementarity of numerical features and semantic features, greatly improving the accuracy of target device running state recognition and the reliability of future risk prediction.

[0080] The following is an embodiment:

[0081] 1. Collecting original time series data: Collecting multi-variable time series data of target devices from edge collection devices, PLCs or SCADA systems to form an initial raw data table, which includes collection timestamps, device indicators (such as furnace inclination, dust removal fan current) and running states.

[0082] 2. Cleaning and structuring time series data: Standardizing the original data and outputting processed standard time series data, including steps such as outlier filtering, outlier detection, and duplicate data processing to ensure data integrity and consistency. Taking a certain furnace as an example:

[0083] a. Outlier filtering (based on rule judgment):

[0084]

[0085] Effective range: such as fan current ∈ [0A, 200A], furnace inclination ∈ [-20°, +20°];

[0086] Effect: Eliminate invalid collection points, such as out of range, missing, device offline, etc.

[0087] b, Outlier detection (Z-Score):

[0088]

[0089] μ: historical mean; σ: historical standard deviation; : Outlier; if >3 is abnormal;

[0090] Effect: Detect extreme deviation data points (such as sudden abnormal current of the device);

[0091] c, Repeat data processing: If the data points with the same timestamp have different values, take the average value;

[0092] d, Time stamp alignment and interpolation: Linear interpolation is used for low-frequency data.

[0093]

[0094] 3, Start and stop state determination: In industrial field, the start and stop state of the device cannot be simply determined by the Boolean signal, and the stable and reliable multi-condition determination logic needs to be constructed based on the operating parameters of the target device and other characteristic variables, combined with threshold value judgment, time window and state retention mechanism.

[0095] a, Determination function of dust removal fan current:

[0096] Define the working state function S(t) of the dust removal fan to represent the running state of the dust removal fan at time t:

[0097]

[0098] Wherein, is the working state of the dust removal fan; is the current value corresponding to the dust removal fan at the sampling time t; is the current threshold value corresponding to the dust removal fan in the start state; is the current threshold value corresponding to the dust removal fan in the stop state; is the minimum duration required for the dust removal fan to maintain the start state; is the minimum duration required for the dust removal fan to maintain the stop state; is the duration for the dust removal fan to maintain the working state; if it does not meet the start or stop state, it is considered that the dust removal fan is in the transition state. Preferably, the current threshold value corresponding to the dust removal fan in the start state can be 10A; the current threshold value corresponding to the dust removal fan in the stop state can be 5A; the minimum duration of the start state and the stop state can be 10s.

[0099] b、The electric furnace body's electric furnace inclination angle determination function:

[0100] State function Indicates the running state of the electric furnace body at time t:

[0101]

[0102] Where, is the working state of the electric furnace body; is the electric furnace inclination angle value corresponding to the electric furnace body at the sampling time t; is the electric furnace inclination angle threshold value corresponding to the electric furnace body in the start state; is the electric furnace inclination angle threshold value corresponding to the electric furnace body in the stop state; is the minimum duration required for the electric furnace body to maintain the start state; is the minimum duration required for the electric furnace body to maintain the stop state; is the duration for which the electric furnace body maintains this working state; if the start or stop state is not met, it is considered that the electric furnace body is in a transition state.

[0103] c、Combined multi-condition determination:

[0104] The working state of the dust removal fan and the working state of the electric furnace body are jointly logically determined to form the final start / stop state determination function of the target device :

[0105]

[0106] As long as either the dust removal fan or the electric furnace body determines "start", it is considered that the target device is in the start state; only when the dust removal fan and the electric furnace body are both "stop", it is determined that the target device is in the stop state; the rest is considered standby or unknown state. As shown in Table 1.

[0107] Table 1

[0108]

[0109] 4、Natural language feature generation: convert multi-dimensional structured time series data (such as electric furnace inclination and dust removal fan current) into natural language running summary, which can be used as subsequent semantic embedding input large language model. Feature extraction example:

[0110] a、Electric furnace inclination value:

[0111] Let the sequence of electric furnace inclination values be , then:

[0112] Average electric furnace tilt angle Used to measure the overall tilt trend of the equipment.

[0113]

[0114] Maximum electric furnace tilt angle This reflects the maximum degree of tilt and helps determine whether there is a risk of exceeding the limit.

[0115]

[0116] Number of tilt angle changes The tilt angle change count represents the number of times the tilt angle change exceeds ±3° between adjacent moments, reflecting the degree of attitude instability during equipment operation. A threshold is set. =3.

[0117]

[0118] b. Current value of the dust collector fan:

[0119] Let the current sequence be ,but:

[0120] Average current value : Measures the overall workload level of the equipment.

[0121]

[0122] Maximum current value This reflects the equipment's short-term maximum workload.

[0123]

[0124] Standard deviation of current fluctuation : Measures the amplitude of current fluctuations and reflects the stability of the system.

[0125]

[0126] Overload times : Represents a current sequence I In Exceeding the rated upper limit current The number of times (e.g., 50A) is used to identify potential faults or overload risks.

[0127]

[0128] c. Target device status switching:

[0129] Suppose the equipment operating status record sequence is as follows: ;

[0130] State transition count : represents the frequency of changes between the states of startup ↔ stop / standby, reflecting the frequent startup and stop or fluctuating state of the device:

[0131]

[0132] d. Natural language construction logic:

[0133] Natural language construction logic: According to the extracted statistical feature algorithm, the preset language template is applied to automatically generate the running summary text. The generated language template is as follows: "In the past 72 hours, the average inclination of the electric furnace is 11.3 degrees, the maximum is 18.1 degrees, and there are 14 times of large inclination; the average current of the dust removal fan is 42.5 amperes, the maximum reaches 56.3 amperes, the current fluctuation is moderate, and 9 times of exceeding the rated upper limit are detected during the period; the device state has been switched 6 times, and the latest continuous running time is 14 hours."

[0134] 5. Semantic feature embedding extraction: mainly includes three types of features: semantic embedding vector, structured feature vector, and fusion vector of embedding vector + structured feature vector.

[0135] a. Semantic embedding vector:

[0136] The semantic embedding vector is a high-dimensional numerical vector generated by the existing large language model after understanding a piece of natural language. It compresses the information such as semantics, context, trend, and implicit relationship in the text into a set of numbers for further modeling.

[0137] Input: natural language summary text (from structured time series data); processing: processed by large model embedding mode; output: fixed-dimensional vector (such as 4096 dimensions).

[0138] Example: Natural language feature generation: "In the past 72 hours, the average inclination of the electric furnace is 11.3 degrees, the maximum is 18.1 degrees, and there are 14 times of large inclination; the average current of the dust removal fan is 42.5 amperes, the maximum reaches 56.3 amperes, the current fluctuation is moderate, and 9 times of exceeding the rated upper limit are detected during the period; the device state has been switched 6 times, and the latest continuous running time is 14 hours."

[0139] The generated vector after semantic embedding can be represented as:

[0140] [0.028, 0.046, 0.079, 0.145, 0.231, 0.406, 0.512, 0.649, 0.724, 0.855, 0.921, 0.781, 0.563, 0.433, 0.301, 0.198, 0.157, 0.093,..., 4096 dimensions].

[0141] b. Structured statistical feature data: Structured statistical feature data is a numerical statistical indicator calculated from standard time series data, such as maximum value, minimum value, average value, standard deviation, number of mutations, etc., used to quantitatively describe the operation of the device.

[0142] Input: Standard time series data (such as temperature, current, inclination, running state, etc.); Processing: processed by large model embedding mode; Output: fixed-dimensional vector (such as 9-dimensional).

[0143] As shown in Table 2:

[0144] Table 2

[0145]

[0146] c. Fusion of embedding vector + structured statistical feature vector: The final input received by the model is the fusion vector after splicing the two vectors. As shown in Table 3.

[0147] Table 3

[0148]

[0149] 6. Multi-model prediction algorithm: Load multiple prediction models for multi-dimensional prediction of the target device state.

[0150] a. First prediction model (LSTM long short-term memory model), predict the future state sequence of the target device:

[0151] Objective: Based on semantic embedding vectors, predict the working state (on / off / abnormal / stop) of the target device in the future (such as 6 hours);

[0152] Input: Input 4096-dimensional semantic embedding vector ;

[0153] Output: Future state sequence prediction ;

[0154] Output example: [running, running, abnormal, stop,...] Predicted state every hour in the next 6 hours.

[0155] b. Second prediction model (XGBoost model, additive model based on boosting enhancement strategy), predict whether the target device will stop in the future:

[0156] Objective: Determine whether the target device is likely to have an unplanned stop within the next 12 hours under the current state;

[0157] Input: Input 4105 fusion vector, , Current time semantic embedding vector (4096 dimensions), Structured feature vector (9 dimensions, such as: average current, maximum furnace inclination value, continuous running time, etc.);

[0158] Output: shutdown probability ;

[0159] Output schematic: -> High risk (possible shutdown).

[0160] c. Third prediction model (MLP model, multi-layer perception model), generating health score or abnormal probability:

[0161] Objective: to evaluate the health degree of the equipment (range 0~1) or abnormal risk based on the current state;

[0162] Input: input fusion vector, ;

[0163] Output health score: ;

[0164] is an activation function, used to compress the linear output of the model, i.e. the health score, to the interval [0, 1]; is a weight matrix, whose dimensions match the input fusion vector x, used to determine the influence degree of different input features in calculating the health score; is a bias term.

[0165] Output schematic: health score = 0.74.

[0166] 7. Result output and intelligent linkage response: the final integrated prediction result is output in a structured form, including future shutdown probability distribution, abnormal score, state evolution trend, etc., and the system links the prediction result with the visualization platform, alarm system, maintenance work order system and production control strategy, which can automatically trigger early warning, generate diagnosis suggestions, recommend operation and maintenance measures, and realize closed-loop management from data collection to intelligent decision-making.

[0167] Figure 3 An illustrative structural diagram of a computer device according to an embodiment of the present application is shown. As shown in Figure 3 the present application provides a computer device, which can include:

[0168] a memory configured to store instructions;

[0169] and a processor configured to call instructions from the memory and capable of implementing the above method when executing the instructions.

[0170] The embodiment of the present application further provides a machine readable storage medium, which has instructions stored thereon, and the instructions are used to cause a machine to execute the method described above.

[0171] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system or a computer program product. Therefore, the present application can adopt a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware aspects. Moreover, the present application can adopt a computer program product embodied on one or more computer usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer usable program code.

[0172] The present application is described with reference to flowcharts and / or block diagrams of the method, device (system), computer program product according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of the flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device produce a device implemented in the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 an apparatus that performs the functions specified in the flow(s) or block(s).

[0173] These computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce a manufactured product including instruction apparatus, which implements the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 an apparatus that performs the functions specified in the flow(s) or block(s).

[0174] These computer program instructions can also be loaded into a computer or other programmable data processing device, so that a series of operation steps are performed on the computer or other programmable device to produce a computer implemented process, so that the instructions executed on the computer or other programmable device provide a process for implementing the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 an apparatus that performs the functions specified in the flow(s) or block(s).

[0175] In a typical configuration, the computing device includes one or more processors (CPU), input / output interface, network interface and memory.

[0176] Memory can include non-persistent memory, Random Access Memory (RAM), and / or non-volatile memory, such as Read Only Memory (ROM) or flash memory, in a computer readable medium. Memory is an example of computer readable media.

[0177] Computer readable media includes permanent and non-permanent, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read only memory (ROM), electrically erasable programmable read only memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), digital versatile disc (DVD), or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information accessible to a computing device. According to the definition herein, computer readable media does not include transitory media, such as modulated data signals and carrier waves.

[0178] It should also be noted that the terms "comprising", "containing", or any other variant thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article or apparatus that comprises a list of elements does not include only those elements in the list, but can also include other elements not expressly listed or inherent to such process, method, article or apparatus. Without further limitation, an element defined by the statement "comprising a" does not exclude the presence of additional identical elements in the process, method, article or apparatus that includes the element.

[0179] The above merely provides an example of the present application and is not intended to limit the present application. The present application can have various modifications and changes for those skilled in the art. Any modification, equivalent replacement, improvement, etc. within the spirit and principle of the present application shall be included in the scope of claims of the present application.

Claims

1. A method for intelligent prediction of plant start-up and shut-down, characterized in that, The method comprises: acquiring standard time sequence data of each operating parameter of a target device in a preset sampling period; the target device is an electric furnace, the electric furnace comprises a dust removal fan and an electric furnace body; the operating parameters comprise a current value of the dust removal fan and an electric furnace inclination value of the electric furnace body; determining a working state of the target device at each sampling time in the preset sampling period based on the standard time sequence data and a preset state determination condition, and generating state sequence data; the working state comprises a working state of the dust removal fan and a working state of the electric furnace body; determining feature data of the dust removal fan according to the standard time sequence data of the current value of the dust removal fan; determining feature data of the electric furnace body according to the standard time sequence data of the electric furnace inclination value of the electric furnace body; determining the number of state switches of the target device according to the state sequence data; determining structured statistical feature data based on the feature data of the dust removal fan, the feature data of the electric furnace body and the number of state switches of the target device; generating a device operation summary text according to the structured statistical feature data and a preset natural language template, inputting the device operation summary text into a large language model to determine a semantic embedding vector; determining a structured feature vector according to the structured statistical feature data, and fusing the semantic embedding vector and the structured feature vector to obtain a fusion vector; determining a future state change of the target device based on the semantic embedding vector and a first prediction model, and determining a future abnormal shutdown probability of the target device based on the fusion vector and a second prediction model.

2. The method of claim 1, wherein, The determination of the working state of the target device at each sampling time in the preset sampling period based on the standard time sequence data and the preset state determination condition, and the generation of the state sequence data comprise: determining the current value of the dust removal fan at each sampling time; determining the electric furnace inclination value of the electric furnace body at each sampling time; determining the working state of the dust removal fan according to the current value at each sampling time and the corresponding preset state determination condition; determining the working state of the electric furnace body according to the electric furnace inclination value at each sampling time and the corresponding preset state determination condition; in the case that the dust removal fan is in a start-up state or the electric furnace body is in a start-up state, determining that the target device is in a start-up state at this sampling time; in the case that the dust removal fan is in a start-up state and the electric furnace body is in a shutdown state, determining that the target device is in a shutdown state at this sampling time; generating the state sequence data based on the sampling time of the standard time sequence data, the current value at each sampling time, the electric furnace inclination value and the working state of the target device at this sampling time.

3. The method of claim 2, wherein, The preset state determination condition corresponding to the dust removal fan satisfies the following formula: wherein, is the working state of the dust removal fan; is the current value corresponding to the dust removal fan at the sampling time t; is the current threshold value corresponding to the dust removal fan in the start-up state; is the current threshold value corresponding to the dust removal fan in the shutdown state; is the minimum duration required for the dust removal fan to remain in the start-up state; is the minimum duration required for the dust removal fan to remain in the shutdown state; is the duration for the dust removal fan to remain in the working state.

4. The method of claim 3, wherein, The preset state determination condition corresponding to the electric furnace body further satisfies the following formula: wherein, is the working state of the electric furnace body; is the electric furnace inclination value corresponding to the electric furnace body at the sampling time t; is the electric furnace inclination threshold value corresponding to the electric furnace body in the start-up state; is the electric furnace inclination threshold value corresponding to the electric furnace body in the shut-down state; is the minimum duration required for the electric furnace body to remain in the start-up state; is the minimum duration required for the electric furnace body to remain in the shut-down state; is the duration for the electric furnace body to remain in the working state.

5. The method according to any one of claims 1 to 4, characterized in that, The determination of the feature data of the dust removal fan according to the standard time sequence data of the current value of the dust removal fan comprises: determining the average current value of the dust removal fan based on the standard time sequence data of the current value of the dust removal fan; obtaining a maximum current value of the dust removal fan in the standard time sequence data; determining a current fluctuation standard deviation of the dust removal fan based on the standard time sequence data; screening out a current value exceeding a preset upper limit in the standard time sequence data, taking a time point corresponding to the current value as an overload time point, and counting a number of overload time points as an overload frequency; taking the average current value, the maximum current value, the current fluctuation standard deviation, and the overload frequency as feature data of the dust removal fan.

6. The method of claim 5, wherein, the feature data of the electric furnace body according to the standard time sequence data of the electric furnace body includes: determining an average electric furnace inclination value of the electric furnace body based on the standard time sequence data of the electric furnace body; obtaining a maximum electric furnace inclination value of the electric furnace body in the standard time sequence data; determining an inclination change frequency of the electric furnace body according to a change range between electric furnace inclination values in adjacent two time points in the standard time sequence data and a preset range of amplitude change; taking the average electric furnace inclination value, the maximum electric furnace inclination value, and the inclination change frequency as the feature data of the electric furnace body.

7. The method of claim 6, wherein, the structured feature vector satisfies the following formula: wherein, is the structured feature vector; indicates that the structured feature vector is 9-dimensional, including the average current value, the maximum current value, the current fluctuation standard deviation, the overload times, the average furnace tilt value, the maximum furnace tilt value, the tilt change times, the switching times, and the continuous running duration of the target device, which is obtained based on the standard timing data.

8. A computer device, comprising: including: a memory configured to store instructions; and a processor configured to call the instructions from the memory and enable the method according to any one of claims 1 to 7 to be implemented when the instructions are executed.

9. A machine-readable storage medium, characterized in that, The machine readable storage medium has instructions stored thereon for causing a machine to perform the method according to any one of claims 1 to 7.

Images