An edge ai-based power consumption behavior profiling method and system

Abstract

The application relates to the field of power consumption behavior profiling, and in particular to a power consumption behavior profiling method and system based on edge AI. The method comprises the following steps: acquiring multi-dimensional data of an electrical equipment, performing multi-state level analysis based on the multi-dimensional data of the electrical equipment, determining a current life cycle state type of the electrical equipment, and generating a current life cycle state type of the electrical equipment; establishing a health benchmark according to a time axis and deducing a device health life cycle sequence based on the current life cycle state type of the electrical equipment, and generating a device health state duration predicted according to time; and performing health comprehensive evaluation based on the current life cycle state type of the electrical equipment and the device health state duration, and generating and outputting a user comprehensive portrait report for user decision-making. In the power consumption behavior profiling process, high-value electrical equipment information can be extracted, and unnecessary economic losses and safety risks can be reduced.

Description

Technical Field

[0001] This application relates to the field of electricity consumption behavior profiling, and in particular to a method and system for electricity consumption behavior profiling based on edge AI. Background Technology

[0002] In the field of intelligent power management and equipment health diagnosis, the refined monitoring and condition assessment of user-side electrical equipment is the core link to achieve energy conservation and consumption reduction, ensure power safety and realize predictive maintenance. The accuracy and foresight of its analysis results are directly related to user energy costs, equipment lifespan and timely elimination of safety hazards. It is a key technological cornerstone for building a new power system and digital energy ecosystem.

[0003] However, existing electrical equipment condition monitoring methods lack a mechanism for deeply integrating and collaboratively profiling equipment health status and user electricity consumption behavior patterns when faced with high-dimensional, asynchronous electricity consumption behavior data that reflects different physical processes. This not only makes it difficult to extract intuitive and high-value information to guide user decisions from massive amounts of data, but may also lead to delayed or misjudged maintenance decisions, resulting in unnecessary economic losses and safety risks. Summary of the Invention

[0004] This application provides a method and system for profiling electricity consumption behavior based on edge AI to solve the above-mentioned technical problems.

[0005] Firstly, this application provides a method for profiling electricity consumption behavior based on edge AI. The method includes: acquiring multi-dimensional data on the operation of electrical appliances; performing multi-state hierarchical analysis based on the multi-dimensional data to determine the current lifecycle state type of the electrical appliances and generate the current lifecycle state type of the appliances; establishing a health benchmark along a time axis and extrapolating the health lifecycle sequence of the appliances based on the current lifecycle state type of the appliances, generating a predicted health status duration of the appliances based on time patterns; and performing a comprehensive health assessment based on the current lifecycle state type of the appliances and the health status duration of the appliances, generating and outputting a comprehensive user profile report for user decision-making.

[0006] The above technical solution first deploys edge computing nodes in electrical equipment scenarios, collecting electrical parameters through smart meters and integrating them with operational logs read from the equipment's built-in interfaces to create multi-dimensional operational data for the electrical equipment. Next, it invokes a multi-state hierarchical analysis model of edge AI to determine the equipment's lifecycle state type, such as "sub-healthy state." Based on this type, and combined with historical and data from similar equipment models, a health benchmark is established. A time-series prediction model is used to extrapolate the equipment's health lifecycle sequence and the duration of its health status. Finally, it integrates multi-dimensional information to conduct a comprehensive health assessment, outputting a comprehensive user profile report to the user, including user type, equipment status, predictions, and suggestions. By monitoring equipment status in real time, it provides early warnings of faults, shortens repair response time, reduces inconvenience to daily life and production losses for businesses, and improves operational efficiency. It helps users take low-cost maintenance measures in the early stages of equipment performance degradation, extending equipment lifespan, identifying high-energy-consuming equipment and guiding energy-saving renovations, reducing usage and energy costs, and aligning with low-carbon requirements. The report is visually presented and adaptable to different user needs, lowering the decision-making threshold. Local data processing at edge nodes ensures privacy and adaptability to multiple scenarios.

[0007] Optionally, the step of performing multi-state hierarchical analysis based on the multi-dimensional operating data of the electrical equipment to determine the current life cycle state type of the electrical equipment and generate the current life cycle state type of the electrical equipment includes: the multi-dimensional operating data of the electrical equipment includes current harmonic spectrum, high-frequency current waveform, steady-state power consumption trajectory, and reference voltage phase signal; by tracking the temporal component ratio changes of the current harmonic spectrum, identifying the continuous degradation and drift of its harmonic structure, and generating a sub-health state determination result; based on the component degradation trend indicated by the sub-health state determination result, by extracting the transient phase of the start-up phase from the high-frequency current waveform... Waveform distortion characteristics are identified to determine abnormal oscillation decay or plateauing of rising edges, generating a pre-death state determination result. When the pre-death state determination result confirms that the device has entered the critical point of functional failure, the steep drop or surge of the current power consumption trajectory is analyzed using the steady-state power consumption trajectory as a standard, and the loss of synchronization or jump of the current voltage phase signal is analyzed using the reference voltage phase signal as a standard. Corresponding characteristics characterizing the termination of device function are identified, generating a death state determination result. Based on the sub-health state determination result, the pre-death state determination result, and the death state determination result, the current life cycle state type of the electrical appliance is comprehensively generated.

[0008] Optionally, the step of identifying the continuous degradation and drift of the harmonic structure by tracking the temporal component ratio changes of the current harmonic spectrum and generating a sub-health state judgment result includes: extracting the harmonic component distribution within a typical normal working cycle from the historical operating data of the electrical equipment itself, and establishing a dynamic reference baseline reflecting the proportional relationship of each harmonic amplitude; slicing the current harmonic spectrum by time window, comparing it proportionally with the dynamic reference baseline one by one, capturing abnormal increases or decreases in the amplitude of specific harmonics, and generating harmonic ratio offset events; performing trend analysis on the harmonic ratio offset events within a continuous time window, and determining that a continuous degradation and drift feature is formed when the proportion of a specific harmonic shows a unidirectional continuous deviation and an increasing amplitude; mapping the harmonic number and deviation direction corresponding to the continuous degradation and drift feature to a preset electrical component degradation knowledge base to generate a sub-health state judgment result containing the specific degradation component type and stage.

[0009] Optionally, the step of extracting waveform distortion features from the high-frequency current waveform during the startup transient phase, identifying abnormal oscillation attenuation or rising edge plateauing, and generating a pre-death state determination result includes: when the sub-health state determination result indicates that a specific power component or power mechanism has continuous degradation, at each startup instant of the electrical equipment, extracting the startup transient information of the high-frequency current waveform, analyzing the instantaneous dynamic behavior of its waveform profile; analyzing the attenuation process of the high-frequency current waveform oscillation after reaching its peak value, identifying whether there is a phenomenon where the oscillation amplitude fails to recover to the threshold level within the expected number of times or the number of oscillation cycles is significantly shorter than the standard startup mode, and generating an attenuation anomaly marker characterizing the loss of inertia of the mechanism; tracking the slope continuity during the current rising phase, and when the interruption of rising kinetic energy is detected, forming a plateau-like flat top or exhibiting a stepped hesitant climb, generating a rising edge struggle marker characterizing the obstruction of driving power; spatiotemporally associating the attenuation anomaly marker with the rising edge struggle marker, and if the two appear in the same startup event, it is determined that the core moving parts of the equipment or the drive circuit have functional failure, and a pre-death state determination result is generated.

[0010] Optionally, the step of analyzing the sharp drop or surge of the current power consumption trajectory using the steady-state power consumption trajectory as a standard and analyzing the loss of synchronization or jump of the current voltage phase signal using the reference voltage phase signal as a standard to identify the associated characteristics characterizing the termination of device function and generate a death state determination result includes: when the pre-death state determination result simultaneously contains the attenuation anomaly marker and the rising edge struggle marker, based on the steady-state power consumption trajectory, capturing the cliff-like drop or sudden surge that occurs after it leaves the normal operating range, generating an extreme power consumption anomaly event; and based on the reference voltage phase signal, identifying its continuous angular deviation or instantaneous angular jump relative to the grid standard phase. A phase-lock event is generated; the extreme power consumption anomaly event and the phase-lock event are time-axis aligned and causally correlated: when the steady-state power consumption trajectory experiences a precipitous drop, and simultaneously accompanied by a continuous angular shift in the reference voltage phase signal, a correlation feature is determined to form that the main circuit is open or the drive core is completely disabled; when the steady-state power consumption trajectory experiences a sudden surge, and simultaneously accompanied by an instantaneous angular jump in the reference voltage phase signal, a correlation feature is determined to form that the internal component is short-circuited or the power mechanism is mechanically locked; based on the establishment of any correlation feature, it is directly determined that the equipment has lost its basic operating function, and a death state judgment result representing the irreversible shutdown of the equipment is generated.

[0011] Optionally, the step of establishing a health benchmark and extrapolating the equipment health lifecycle sequence based on the current lifecycle state type of the electrical appliance, and generating the equipment health state duration predicted by time regularity, includes: taking the moment when the electrical appliance is first marked as the sub-healthy state determination result as the starting point of the lifecycle extrapolation sequence, integrating all multi-dimensional operational data of the electrical appliance recorded from this starting point to the current moment and their corresponding lifecycle state types to construct the equipment health lifecycle sequence; and based on the equipment health lifecycle sequence, fitting the trend slope of the duration of the evolution from the sub-healthy state to the pre-death state with the harmonic ratio offset event. A first lifespan decay model is constructed based on the rate of change. A second lifespan decay model is constructed by fitting the duration of evolution from the pre-death state to the death state with the frequency of occurrence of the decay anomaly marker and the rising edge struggle marker. The first lifespan decay model is used to characterize the predictive relationship of the duration of evolution from the sub-healthy state to the pre-death state, and the second lifespan decay model is used to characterize the predictive relationship of the duration of evolution from the pre-death state to the death state. Based on the first lifespan decay model and the second lifespan decay model, and by fusing the duration of the current state with the degradation rate, the remaining time until the function terminates is calculated, and the lifespan of the device's healthy state is generated.

[0012] Optionally, the step of constructing a first lifetime decay model by fitting the duration of the evolution from a sub-healthy state to a pre-death state with the trend slope of the harmonic ratio shift events includes: locating a first historical evolution duration from the sub-healthy state determination result to the pre-death state determination result in the device health life cycle sequence; analyzing all the harmonic ratio shift events within the first historical evolution duration interval, extracting the key harmonic components that lead to the state determination in each event, and calculating the average rate of change of their amplitudes from the dynamic reference baseline to generate a harmonic degradation trend slope; performing an evolution correlation analysis between the first historical evolution duration and the harmonic degradation trend slope: when the absolute value of the harmonic degradation trend slope exceeds a preset accelerated degradation threshold, the speed at which it evolves to the pre-death state will be significantly accelerated; and dynamically predicting the first lifetime decay model based on the evolution correlation analysis and the accelerated degradation threshold to determine the time required for the evolution from the current sub-healthy state to the pre-death state.

[0013] Optionally, the step of constructing a second lifetime decay model by fitting the duration of evolution from the pre-death state to the death state with the occurrence frequency of the decay anomaly marker and the rising edge struggle marker includes: locating the second historical duration from the first occurrence of the pre-death state determination result to the generation of the death state determination result in the device health life cycle sequence; analyzing all device startup events within this time interval, and respectively counting the number of occurrences and the frequency of occurrence of the decay anomaly marker and the rising edge struggle marker within a unit time window; superimposing the occurrence frequency of the decay anomaly marker and the occurrence frequency of the rising edge struggle marker to generate a joint frequency index characterizing the overall functional instability; performing instability correlation analysis between the second historical duration and the joint frequency index: identifying that when the joint frequency index exceeds the critical instability threshold, the duration of evolution from the pre-death state to the death state will be drastically shortened; and dynamically predicting the second lifetime decay model based on the critical instability threshold and the instability correlation analysis results to determine the time required for evolution from the current pre-death state to the death state.

[0014] Optionally, the step of conducting a comprehensive health assessment based on the current lifecycle state type of the electrical appliance and the health status duration of the equipment, and generating and outputting a comprehensive user profile report for user decision-making, includes: deriving and assigning user electricity consumption behavior profile tags based on the current lifecycle state type of the electrical appliance and its corresponding health status duration; if the equipment is in a sub-healthy state and the duration indicates a maintenance window, the user behavior profile is defined as a risk-averse user, indicating that the user has tolerance for potential equipment failure risks and tends to perform maintenance within the plan; if the equipment is in a pre-death state and the duration indicates an emergency risk, the user behavior profile is defined as a cost-sensitive user, indicating that the user is highly sensitive to losses caused by sudden equipment downtime and tends to take emergency loss-mitigation measures; based on the electricity consumption behavior profile tags, generating a structured comprehensive user profile report that matches the tags: for risk-averse users, the report focuses on providing preventative maintenance planning and risk warning suggestions based on the duration; for cost-sensitive users, the report focuses on providing decision-making suggestions based on the duration for emergency loss mitigation, spare parts preparation, or immediate replacement.

[0015] Secondly, this application provides an edge AI-based electricity consumption behavior profiling system, the system comprising: a state analysis module, used to acquire multi-dimensional operational data of electrical appliances, perform multi-state hierarchical analysis based on the multi-dimensional operational data of electrical appliances, determine the current life cycle state type of the electrical appliances, and generate the current life cycle state type of the electrical appliances; a state prediction module, used to establish a health benchmark along a time axis based on the current life cycle state type of the electrical appliances and deduce the health life cycle sequence of the equipment, and generate the equipment health status duration predicted by time patterns; and a comprehensive profiling module, used to perform a comprehensive health assessment based on the current life cycle state type of the electrical appliances and the equipment health status duration, and generate and output a comprehensive user profile report for user decision-making. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of an application scenario provided in an embodiment of this application;

[0018] Figure 2 A flowchart illustrating an edge AI-based electricity consumption behavior profiling method provided in an embodiment of this application;

[0019] Figure 3 This is a schematic diagram of the structure of an edge AI-based electricity consumption behavior profiling system provided in one embodiment of this application. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0021] Furthermore, the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article, unless otherwise specified, generally indicates that the preceding and following related objects have an "or" relationship.

[0022] The embodiments of this application will now be described in further detail with reference to the accompanying drawings.

[0023] Existing electrical equipment condition monitoring methods lack a mechanism for deeply integrating and collaboratively profiling equipment health status and user electricity consumption patterns when faced with high-dimensional, asynchronous electricity consumption behavior data that reflects different physical processes. This not only makes it difficult to extract intuitive and high-value information to guide user decisions from massive amounts of data, but may also lead to delayed or misjudged maintenance decisions, resulting in unnecessary economic losses and safety risks.

[0024] Based on this, this application provides a method and system for profiling electricity consumption behavior based on edge AI. First, edge computing nodes are deployed in the electrical equipment scenario to collect electrical parameters through smart meters and combine this with operational logs read from the equipment's built-in interface, integrating them into multi-dimensional operational data for the electrical equipment. Next, a multi-state hierarchical analysis model of edge AI is invoked to determine the equipment's lifecycle state type, such as "sub-healthy state." Based on this type, a health benchmark is established by combining historical data with data from similar equipment models. A time series prediction model is used to extrapolate the equipment's health lifecycle sequence and the duration of its health status. Finally, multi-dimensional information is integrated to conduct a comprehensive health assessment, outputting a comprehensive user profile report to the user, including user type, equipment status, predictions, and suggestions. By monitoring equipment status in real time, early warnings of faults are provided, shortening repair response time, reducing inconvenience to household life and production losses for businesses, and improving operational efficiency. This helps users take low-cost maintenance measures in the early stages of equipment performance degradation, extending equipment lifespan, identifying high-energy-consuming equipment and guiding energy-saving renovations, reducing usage and energy costs, and aligning with low-carbon requirements. The report is visually presented and adaptable to different user needs, lowering the decision-making threshold. Local data processing at edge nodes ensures privacy and adaptability to multiple scenarios.

[0025] Figure 1 This is a schematic diagram illustrating an application scenario provided by this application. In the process of profiling electricity consumption behavior, the method provided in this application can extract high-value information about electrical equipment, reducing unnecessary economic losses and safety risks.

[0026] Specifically, the method of this application is applied to any server that communicates with a smart meter to obtain multi-dimensional operational data of electrical appliances provided by the smart meter. First, edge computing nodes are deployed in the electrical appliance scenario to collect electrical parameters through the smart meter and combine them with the operational logs read from the device's built-in interface to integrate multi-dimensional operational data of the electrical appliance. Next, a multi-state hierarchical analysis model of edge AI is invoked to determine the device's lifecycle state type, such as "sub-healthy state." Based on this type, a health benchmark is established by combining historical data with data from devices of the same model. A time series prediction model is used to extrapolate the device's health lifecycle sequence and the duration of its health status. Finally, multi-dimensional information is integrated to conduct a comprehensive health assessment and output a comprehensive user profile report containing user type, device status, predictions, and suggestions to the user.

[0027] For specific implementation details, please refer to the following examples.

[0028] Figure 2 This is a flowchart illustrating a method for profiling electricity consumption behavior based on edge AI, provided as an embodiment of this application. The method of this embodiment can be applied to servers in the above scenarios. Figure 2 As shown, the method includes:

[0029] S201. Obtain multi-dimensional data on the operation of electrical equipment, perform multi-state hierarchical analysis based on the multi-dimensional data on the operation of electrical equipment, determine the current life cycle state type of electrical equipment, and generate the current life cycle state type of electrical equipment.

[0030] Multidimensional operational data for electrical equipment can be a set of multidimensional parameters generated during the operation of the equipment, including current harmonic spectrum, high-frequency current waveform, steady-state power consumption trajectory, and reference voltage phase signal. The data comes from smart meters. The current life cycle state type of an electrical appliance can be a classification result used to characterize the stage of the electrical appliance in its life cycle, such as "sub-healthy state", "pre-death state", and "death state".

[0031] Specifically, with the widespread adoption of smart homes and the Industrial Internet of Things (IIoT), the number of electrical appliances has surged. However, monitoring and maintaining their operational status often relies on periodic inspections or reactive repairs, lacking real-time and predictive capabilities. Current technologies often assess the condition of electrical appliances based on single parameters (such as operating time) or simple threshold alarms, making it difficult to accurately capture the complex degradation process of equipment. This leads to untimely maintenance, low energy efficiency, or unexpected downtime. For example, an air conditioner in a state of mild degradation may only exhibit a slight decrease in energy efficiency, but traditional methods cannot identify such subtle changes, intervening only after a malfunction occurs, resulting in energy waste and user inconvenience. This step utilizes edge AI technology to acquire multi-dimensional operational data of electrical appliances in real time and performs multi-state hierarchical analysis, enabling more precise capture of equipment status changes and determination of its lifecycle state type. By accurately determining the current lifecycle state type of electrical appliances, users and managers can clearly understand the "health status" of the equipment, avoiding over-maintenance or untimely maintenance due to subjective judgment bias.

[0032] S202. Based on the current life cycle status type of the electrical appliance, establish a health benchmark according to the time axis and deduce the equipment health life cycle sequence to generate the equipment health status duration predicted by time pattern.

[0033] The equipment health lifecycle sequence can be a sequence of changes in equipment health status deduced through time series analysis. The starting point for the lifecycle deduction is the moment when the electrical equipment is first marked as being in a sub-healthy state. It integrates all multi-dimensional operational data of the electrical equipment recorded from this starting point up to the current moment, along with their corresponding lifecycle state types. The equipment health status duration can be the predicted duration of the electrical equipment's current state, based on time-related patterns.

[0034] Specifically, in electrical equipment management, knowing only the current state is insufficient for long-term decision-making. Users need to understand the future health trends of the equipment to plan maintenance or replacement. Current technologies often rely on fixed models or statistical methods for health prediction, lacking adaptability to individual equipment differences and dynamic changes, leading to significant prediction bias. For example, a refrigerator may have different lifespans during its stable operation period due to varying usage frequencies, but traditional methods using general lifespan models cannot accurately predict its specific duration. This step establishes a dynamic health benchmark based on the current lifecycle state type of the appliance, along a timeline, and extrapolates the equipment's health lifecycle sequence, thereby generating the health status duration. The extrapolation of the equipment health status duration and health lifecycle sequence provides crucial support for proactive equipment management. Users can plan maintenance, spare parts preparation, and equipment replacement in advance, effectively reducing equipment failure rates, minimizing production and life disruptions caused by equipment downtime, and significantly improving equipment reliability and management efficiency.

[0035] S203. Based on the current life cycle status type of the electrical appliance and the duration of the health status, perform a comprehensive health assessment, generate and output a comprehensive user profile report for user decision-making.

[0036] A comprehensive user profile report can include information such as device health status, prediction results, and suggested measures, aiming to assist users in decision-making.

[0037] Specifically, different users have significantly different needs for managing electrical appliances: household users are more concerned about whether equipment failures affect daily life and whether repair costs are reasonable; enterprise users (such as manufacturing plants) are more concerned about the impact of equipment downtime on production and whether maintenance plans align with production schedules; equipment leasing companies focus on the remaining lifespan of the equipment and whether early replacement is necessary to avoid leasing disputes. Traditional monitoring solutions mostly output raw data or general alarm information, requiring users to interpret the data and make decisions themselves, which is a high barrier to entry and difficult to adapt to individual needs. This step, through a comprehensive health assessment, integrates the current status and future predictions into a holistic profile, generating and outputting a comprehensive user profile report for user decision-making. The comprehensive user profile report provides intuitive and comprehensive insights into equipment health, enabling users to quickly make maintenance or optimization decisions, improving equipment reliability and energy efficiency; personalized report output enhances user participation and supports preventative maintenance, reducing downtime and costs.

[0038] The method provided in this embodiment first deploys edge computing nodes in the electrical equipment scenario, collects electrical parameters through smart meters, and integrates them with the equipment's built-in interface to read operation logs, forming multi-dimensional data on the electrical equipment's operation. Next, it calls the edge AI's multi-state hierarchical analysis model to determine the equipment's lifecycle state type, such as "sub-healthy state." Based on this type, and combined with historical and similar equipment data, a health benchmark is established, and a time series prediction model is used to extrapolate the equipment's health lifecycle sequence and health status duration. Finally, it integrates multi-dimensional information to conduct a comprehensive health assessment, outputting a comprehensive user profile report to the user, including user type, equipment status, predictions, and suggestions. By monitoring equipment status in real time, it provides early warnings of faults, shortens repair response time, reduces inconvenience to family life and production downtime losses for businesses, and improves operational efficiency. It helps users take low-cost maintenance in the early stages of equipment performance degradation, extending equipment lifespan, identifying high-energy-consuming equipment and guiding energy-saving renovations, reducing usage and energy costs, and aligning with low-carbon requirements. The report is visually presented and adaptable to different user needs, lowering the decision-making threshold. Local data processing at edge nodes ensures privacy and adapts to multiple scenarios.

[0039] In some embodiments, the multidimensional data of electrical equipment operation includes current harmonic spectrum, high-frequency current waveform, steady-state power consumption trajectory, and reference voltage phase signal; by tracking the changes in the proportion of components of the current harmonic spectrum over time, the continuous degradation and drift of its harmonic structure is identified, generating a sub-health state determination result; based on the component degradation trend indicated by the sub-health state determination result, by extracting the waveform distortion characteristics of the transient phase of the high-frequency current waveform, its abnormal oscillation decay or rising edge plateauing struggle is identified, generating a pre-death state determination result; when the pre-death state determination result confirms that the equipment has entered the functional failure critical point, the steep drop or surge of the current power consumption trajectory is analyzed using the steady-state power consumption trajectory as the standard, and the loss of synchronization or jump of the current voltage phase signal is analyzed using the reference voltage phase signal as the standard, identifying the correlation characteristics characterizing the termination of equipment function, generating a death state determination result; based on the sub-health state determination result, the pre-death state determination result, and the death state determination result, the current life cycle state type of the electrical appliance is comprehensively generated.

[0040] The current harmonic spectrum can be a frequency and amplitude distribution map of each harmonic component in the current signal. A high-frequency current waveform can be an instantaneous curve showing the change of high-frequency current over time during equipment operation. A steady-state power consumption trajectory can be a continuous trajectory showing the change of power consumption per unit time during the stable operation phase of the equipment. A reference voltage phase signal can be a phase reference curve of the supply voltage during normal equipment operation, used to compare and determine whether the current voltage phase is abnormal. Persistent degradation drift can be a quantitative indicator used to characterize the gradual and irreversible degradation of the performance of key components inside electrical equipment. It specifically refers to the slow but continuous trend of change in the energy proportion or amplitude of each harmonic (such as the third, fifth, and seventh harmonics) in the current harmonic spectrum of the equipment over time, deviating from its healthy baseline state. A sub-health state judgment result can be an output used to characterize that electrical equipment is in an early stage of performance degradation, indicating that although the equipment can operate normally, potential fault signs have appeared (such as increased harmonic distortion). The component degradation trend can be analyzed based on the sub-health status assessment results, determining the direction and speed of performance degradation of core components of electrical equipment (such as compressor motor windings, capacitors, circuit board chips, etc.). High-frequency current waveform distortion characteristics can be abnormal features where the high-frequency current waveform deviates from the normal pattern when the core components of electrical equipment degrade. Specifically, this includes two typical characteristics: abnormal oscillation decay (e.g., normally oscillates to a stable value after 5 cycles, but abnormally oscillates for 10 cycles without stabilizing) and struggling with a plateauing rising edge (e.g., normally the rising edge reaches its peak within 0.1 seconds, but abnormally it experiences a 0.5-second plateau, failing to quickly reach the peak). The pre-death state assessment result can be an output used to characterize electrical equipment approaching the critical point of functional failure, indicating that the core components of the equipment (such as motors or power modules) have severely degraded and may fail at any time. The functional failure critical point can be the critical node where the core components of electrical equipment degrade to the point where they can no longer maintain basic functions and are about to enter a state of complete failure; it is the dividing point between the pre-death state and the death state. The death state assessment result can be an output used to characterize the complete termination of the function of electrical equipment, indicating that the equipment can no longer operate normally or has suffered permanent damage.

[0041] Specifically, in intelligent electricity consumption behavior profiling systems, accurately determining the life cycle status of electrical appliances is crucial for preventing malfunctions, optimizing maintenance, and improving energy efficiency. Traditional condition monitoring methods often rely on single parameters (such as total power consumption or running time), failing to capture subtle degradation processes within the equipment. This leads to delayed or misjudgments in condition assessment. For example, in a sub-healthy state, equipment may only exhibit slight harmonic changes, but traditional methods struggle to identify such early warnings, thus missing maintenance opportunities. In the pre-death stage, waveform distortion may be masked by noise, easily overlooked without specialized analysis until the equipment suddenly fails. Furthermore, traditional fault diagnosis for electrical appliances often relies on "post-fault alarms," ​​meaning that problems are only discovered through fault codes or manual inspection after the equipment has completely failed and ceased operation. This results in users facing the inconvenience of "sudden shutdowns." For instance, if a household refrigerator enters a functional decline stage due to compressor motor winding degradation, traditional methods can only diagnose the fault after the refrigerator completely stops cooling, by which time the food inside has already spoiled. Similarly, if a factory motor enters a pre-death state due to bearing wear, traditional methods require waiting until the motor jams and stops before discovering the problem, causing production line shutdowns. To address the above issues, this step first acquires multi-dimensional operational data of the electrical equipment in real time through the data acquisition module on the edge AI device, including the current harmonic spectrum, high-frequency current waveform, steady-state power consumption trajectory, and reference voltage phase signal, and performs preprocessing such as filtering, noise reduction, and feature alignment. Next, the system tracks the temporal component ratio changes of the current harmonic spectrum (e.g., the third harmonic gradually increases from 5% to 10%), identifies the continuous degradation and drift of its harmonic structure, and generates a sub-health state assessment result. Based on the component degradation trend indicated by this result, the system extracts waveform distortion features from the high-frequency current waveform during the initial transient phase (e.g., the oscillation decay time increases from the normal value of 10 milliseconds to 20 milliseconds, or a plateau appears on the rising edge). The system first analyzes the current power consumption trajectory using a steady-state power consumption trajectory as a standard to analyze whether there is a sharp drop (e.g., power consumption drops sharply from the rated value, such as 100 watts, to 20 watts) or a surge (e.g., power consumption suddenly increases to 150 watts). Simultaneously, it analyzes the current voltage phase signal using a reference voltage phase signal as a standard to analyze whether there is a loss of synchronization (e.g., phase deviation exceeds the standard value, such as 5 degrees) or a jump (e.g., instantaneous phase change, such as 10 degrees). This identifies the associated characteristics representing the termination of device function and generates a death status determination result. Finally, based on the sub-health, pre-death, and death status determination results, the system uses a rule engine or weighted fusion method to comprehensively generate the current life cycle state type of the electrical appliance.

[0042] The method provided in this embodiment, based on multi-state hierarchical analysis, can accurately capture the full-cycle state changes of electrical equipment from early degradation to functional failure, improving the accuracy and timeliness of state determination, helping users to identify potential faults in advance, and reducing unexpected downtime and maintenance costs. The hierarchical determination framework enhances the system's adaptability and robustness, providing reliable input for subsequent health prediction and decision support, thereby improving overall power management efficiency and equipment reliability.

[0043] In some embodiments, the distribution of harmonic components within a typical normal working cycle is extracted from the historical operating data of the electrical equipment itself, and a dynamic reference baseline reflecting the proportional relationship of the amplitude of each harmonic is established. The current harmonic spectrum is sliced ​​according to time windows, and each slice is compared with the dynamic reference baseline to capture abnormal increases or decreases in the amplitude of specific harmonics, generating harmonic ratio shift events. Trend analysis is performed on the harmonic ratio shift events within continuous time windows. When the proportion of a specific harmonic shows a unidirectional continuous deviation and the amplitude increases, it is determined that a continuous degradation drift feature has been formed. Based on the harmonic number and deviation direction corresponding to the continuous degradation drift feature, it is mapped to a preset electrical component degradation knowledge base to generate a sub-health status judgment result containing the specific degradation component type and stage.

[0044] A typical normal operating cycle refers to the time period during which an electrical device completes a full functional operation and remains stable. The typical normal operating cycle varies among different types of equipment. Harmonic component distribution refers to the proportion of the amplitude of each harmonic (e.g., the 3rd, 5th, and 7th harmonics) in the current harmonic spectrum to the total current amplitude within the typical normal operating cycle of the electrical device. It is a core indicator reflecting whether the electrical characteristics of the equipment are normal. A dynamic reference baseline is a harmonic ratio baseline constructed based on the harmonic component distribution of the electrical device's typical normal operating cycle, which can dynamically adapt to the long-term slight aging of the equipment. A time window slice is a continuous data segment divided into fixed time intervals from the real-time acquired current harmonic spectrum data, with each segment corresponding to an independent analysis unit. A harmonic ratio deviation event refers to an abnormal event where, after comparing the current harmonic spectrum within a certain time window with the dynamic reference baseline, the ratio deviation of a specific harmonic exceeds a preset threshold. A persistent degradation drift characteristic refers to the characteristic of "unidirectional continuous deviation + increasing amplitude" of the ratio deviation event of the same specific harmonic within multiple consecutive time windows (e.g., more than 3), which is a core indicator of the equipment entering a sub-healthy state. The preset electrical component degradation knowledge base can be a structured database that stores the correlation between the degradation of different electrical components (such as capacitors, inductors, and motor windings) and specific harmonic changes.

[0045] Specifically, traditional sub-health status assessments can only inform users that "the equipment is malfunctioning," but cannot pinpoint the specific deteriorating components. This forces repair personnel to disassemble the equipment and check each component individually, which is time-consuming and labor-intensive. For example, if a household refrigerator experiences harmonic anomalies, traditional methods only indicate a "sub-healthy state." Repair personnel must then check components such as the compressor, capacitors, and circuit boards in sequence, resulting in lengthy troubleshooting times. If users ignore the anomaly due to the inability to locate the components, it may lead to accelerated component degradation and ultimately cause the refrigerator to shut down. At the same time, traditional methods often determine the status based on a comparison of single harmonic data at a certain moment with a baseline. This is easily affected by interference factors such as power grid fluctuations and short-term load changes in equipment, leading to false anomalies. For example, if an industrial motor experiences a short-term fluctuation in power grid voltage, causing the proportion of the 5th harmonic to suddenly rise from 3% to 4.2% at a certain moment (exceeding a fixed threshold of 1%), traditional methods would immediately classify it as sub-healthy. However, if the harmonic proportion returns to normal after 10 minutes, the anomaly is clearly a false signal. To address the above issues, this step first involves the edge AI device acquiring historical operating data of the electrical appliance from local storage or the cloud. Data mining algorithms are then used to extract the current harmonic spectrum data of the device during a typical normal operating cycle. A dynamic reference baseline reflecting the stable proportional relationship of the amplitudes of each harmonic (e.g., 3rd, 5th, and 7th) is then established and loaded into memory. Next, the system slices the real-time acquired current harmonic spectrum streaming data into preset time windows (e.g., every 5 minutes). For each time window slice, the system compares the distribution of harmonic components within it with the dynamic reference baseline. By calculating the proportional differences, it captures abnormal increases or decreases in the amplitude of specific harmonics. When the proportional difference exceeds a certain threshold... When the preset tolerance is exceeded, a harmonic ratio offset event containing the harmonic number, offset direction, and offset amount is generated. Then, the system performs trend analysis on the offset event sequence of the same harmonic generated within a continuous time window. If the analysis shows that the harmonic ratio deviates in the same direction in multiple consecutive windows and the deviation magnitude shows an increasing trend, it is determined that a continuous degradation drift feature has been formed. Finally, the system matches and queries this feature (including the harmonic number and offset direction) with the electrical component degradation knowledge base pre-installed on the edge AI device. According to the built-in mapping rules, it generates and outputs a sub-health status judgment result containing the specific degradation component type (such as "power converter IGBT") and degradation stage (such as "early saturation characteristic degradation").

[0046] The method provided in this embodiment enables early, accurate, and diagnosable identification of sub-health conditions in electrical equipment, significantly improving the foresight and effectiveness of predictive maintenance. By establishing a personalized dynamic reference baseline and analyzing the continuous drift of harmonic ratios, it effectively overcomes the lag and misjudgment risk of traditional static threshold methods, allowing users to know potential faulty components and risk levels before equipment performance substantially declines, thus gaining valuable time for arranging preventive maintenance, avoiding unplanned downtime, and reducing maintenance costs.

[0047] In some embodiments, when the sub-health status determination result indicates that a specific power component or power mechanism is in continuous degradation, at the moment of each start-up of the electrical equipment, the start-up transient information of the high-frequency current waveform is extracted, and the instantaneous dynamic behavior of its waveform profile is analyzed: the decay process of the high-frequency current waveform oscillation after reaching the peak is analyzed, and it is identified whether there is a phenomenon that the oscillation amplitude does not recover to the threshold level within the expected number of times or the number of oscillation cycles is significantly shorter than the standard start-up mode, and a decay anomaly mark characterizing the loss of inertia of the mechanism is generated; the slope continuity is tracked during the current rise phase, and when the interruption of the rising kinetic energy is detected to form a plateau-like flat top or to present a stepped hesitant climb, a rising edge struggle mark characterizing the obstruction of driving power is generated; the decay anomaly mark and the rising edge struggle mark are spatiotemporally correlated, and if the two appear in the same start-up event, it is determined that the core moving parts of the equipment or the drive circuit have functional failure, and a pre-death status determination result is generated.

[0048] Startup transient information can be the brief dynamic changes in the high-frequency current waveform of electrical equipment after power is connected and before it reaches a stable operating state. Attenuation anomaly markers can be indicators used to quantitatively characterize abnormalities in the oscillation attenuation process of the high-frequency current waveform during the startup transient phase. Rising edge struggle markers can be indicators used to quantitatively characterize the resistance to power generation during the current rise process of the high-frequency current waveform during the startup transient phase.

[0049] Specifically, in the chain of progressively deteriorating health status of electrical equipment, identifying the critical point from "sub-health" to "functional failure" (i.e., the "pre-death" state) is the most challenging and critical early warning link in the entire predictive maintenance system. The sub-health state reveals a slow, cumulative trend of performance degradation, but the equipment can still operate basically normally. The pre-death state, on the other hand, indicates that this degradation has reached the bottom line of the equipment's core functions. Although the equipment may still be able to start or run occasionally, its failure risk has increased sharply and may lead to complete shutdown at any time. Traditional diagnostic methods have serious shortcomings in judging this critical point: First, they often rely on steady-state operating parameters (such as average power consumption and steady-state current), but these parameters may not change much before the equipment fails completely, making them unable to provide effective early warnings. Second, even if the starting current is analyzed, if only its overall amplitude or effective value is focused on, the earlier fault fingerprints contained in the transient details of the waveform will be missed, because the functional failure of the core components first affects the instantaneous dynamic characteristics of the current, rather than its macroscopic statistics. To address the above issues, this step first involves the edge AI system receiving a "sub-health status assessment result" for a specific electrical device, indicating continuous degradation of its specific power components or power mechanisms. The system then activates a high-frequency current waveform start-up transient monitoring mode for that device. At each subsequent start-up moment, the system captures data from the start-up process using a high-frequency current sensor and precisely extracts "start-up transient information" lasting only a very short time (e.g., within 500 milliseconds after start-up). Next, the system performs dual-path parallel analysis on this information: one path specifically analyzes the decay process of the current waveform oscillation after reaching its peak value, comparing it with the standard start-up decay pattern stored in the device's feature library. If it identifies that the oscillation amplitude has not recovered to the threshold level (e.g., 30% of the peak current) within the expected number of cycles (e.g., 3 cycles), or If the number of cycles in the entire oscillation process is significantly shorter than the standard mode (e.g., the standard is 5 cycles, but in reality it is only 2), an "abnormal decay marker" is generated. Another path specifically analyzes the rising phase of the current, tracking its continuity through high-precision slope calculation. When a clear plateau-like flat top (a section with a slope close to zero) or a discontinuous step-like climb is detected during the rising process, a "struggling rising edge marker" is generated. Finally, the system sends these two markers to the correlation analysis module for "spatiotemporal correlation" judgment: if it is determined that these two markers are generated within the same time period of the same startup event, the highest level alarm is triggered, determining that the power core of the device is at the critical point of functional failure, and a "pre-death state judgment result" is generated accordingly. This result will clearly point to the core component that may fail (such as the "compressor motor drive system").

[0050] The method provided in this embodiment effectively overcomes the lag of traditional steady-state parameter monitoring, and can identify clear signs of functional failure of the power system before the equipment experiences a catastrophic shutdown. This allows users to gain valuable emergency response time, arrange for component replacement or emergency repair in a timely manner, and effectively avoid production interruptions, data loss or service interruptions caused by sudden failure of core equipment, ensuring the continuity and safety of operations, while minimizing the economic losses and safety risks caused by the failure.

[0051] In some embodiments, when the pre-death state determination result includes both an attenuation anomaly marker and a rising edge struggle marker, based on the steady-state power consumption trajectory, the phenomenon of a precipitous drop or sudden surge after it leaves the normal operating range is captured, generating an extreme power consumption anomaly event; based on the reference voltage phase signal, the continuous angle shift or instantaneous angle jump relative to the grid standard phase is identified, generating a phase lockout event; the extreme power consumption anomaly event and the phase lockout event are time-axis aligned and causally correlated: when the steady-state power consumption trajectory experiences a precipitous drop, and is accompanied by a continuous angle shift in the reference voltage phase signal, it is determined that a correlation feature of main circuit open circuit or complete failure of drive core is formed; when the steady-state power consumption trajectory experiences a sudden surge, and is accompanied by an instantaneous angle jump in the reference voltage phase signal, it is determined that a correlation feature of internal component breakdown short circuit or mechanical lock-up of power mechanism is formed; based on the establishment of any correlation feature, it is directly determined that the equipment has lost its basic operating function, generating a death state determination result characterizing irreversible shutdown of the equipment.

[0052] Steady-state power consumption trajectory can be a continuous and smooth data trajectory of power consumption per unit time during the stable operation phase of an electrical device. It is a core indicator reflecting the energy consumption stability and normal operating status of the device. The normal operating range can be the normal fluctuation range of the steady-state power consumption trajectory defined based on the rated power consumption and historical normal operating data of the electrical device. It serves as a benchmark range for judging whether power consumption is abnormal. Extreme power consumption anomalies can be structured event information recording a precipitous drop or sudden surge in the steady-state power consumption trajectory, including information such as "anomaly type (drop / surge), anomaly occurrence time, power consumption change amplitude, and duration." Phase lockout events can be structured event information recording a continuous angular shift or instantaneous angular jump in the voltage phase signal, including information such as "lockout type (shift / jump), occurrence time, phase difference amplitude, and duration." Time axis alignment and causal logic association can be the analytical process for verifying the temporal synchronicity and logical causal relationship between extreme power consumption anomalies and phase lockout events.

[0053] Specifically, in the final stage of equipment health evolution, the determination from "pre-death" (the critical point of functional failure) to "death" (permanent termination of function) is a decisive step in the entire life cycle status profile. The pre-death state provides a serious warning, but the equipment may still have intermittent and unstable operating capabilities; while the determination of the death state means that the equipment has completely failed, and any maintenance investment will be meaningless. At this critical moment, it is crucial to avoid misjudgment: prematurely declaring "death" may lead to the wrong scrapping of equipment that can still be repaired, resulting in a waste of resources; while making a determination too late will cause users to rely on equipment that may completely shut down at any time, causing production interruptions or safety accidents. Traditional fault diagnosis methods are often inadequate in this final judgment because they are usually based on a single parameter (such as only power abnormality or only phase abnormality). A single abnormal signal may be caused by temporary line interference, sensor error, or failure of non-critical auxiliary components, which is insufficient to support the final conclusion of equipment "death". To address the above issues, this step begins by having the edge AI system generate a "pre-death state determination result" containing "attenuation anomaly markers" and "rising edge struggle markers." Then, the system activates the highest-level ultimate state monitoring mode. In this mode, the system continuously monitors the device's "steady-state power consumption trajectory" and "reference voltage phase signal." If the system detects a "cliff-like drop" or "sudden surge" in the "steady-state power consumption trajectory," it immediately generates a "power consumption extreme anomaly event" record. Simultaneously, if the system identifies a "persistent angle shift" or "instantaneous angle jump" in the "reference voltage phase signal," it immediately generates a "phase unlocking event" record. Subsequently, the system enters the correlation analysis phase: it... Align these two events on a precise timeline and check if they occur synchronously within an extremely short time window (e.g., milliseconds). Next, perform a causal logic correlation judgment: if the alignment reveals that "power consumption precipitous drop" and "phase persistent shift" are synchronous, then the correlation characteristic of "main circuit open circuit or complete failure of the drive core" is determined; if the alignment reveals that "power consumption sudden surge" and "phase instantaneous jump" are synchronous, then the correlation characteristic of "internal component breakdown short circuit or mechanical seizure of the power mechanism" is determined; finally, as long as any of the above correlation characteristics are true, the system will no longer wait for other evidence and will directly generate the final "death status judgment result," clearly declaring the end of the life cycle of the electrical equipment.

[0054] The method provided in this embodiment effectively avoids misjudgments or omissions caused by false alarms of a single parameter, making the diagnostic conclusions highly reliable. This ensures that users can obtain a clear and accurate final diagnosis as soon as the equipment completely fails, thereby decisively stopping unnecessary maintenance investment and immediately initiating the equipment replacement or scrapping process. This minimizes the risk of business interruption and related losses caused by sudden permanent equipment shutdown, achieving a perfect closed loop for the full life cycle management of equipment.

[0055] In some embodiments, the moment when an electrical appliance is first marked as being in a sub-healthy state is taken as the starting point of the lifespan extrapolation sequence. All multi-dimensional operational data of the electrical appliance recorded from this starting point up to the current moment, along with their corresponding lifespan state types, are integrated to construct an equipment health lifespan sequence. Based on this sequence, a first lifespan decay model is constructed by fitting the trend slope of the duration of the evolution from a sub-healthy state to a pre-death state with the harmonic ratio shift event. A second lifespan decay model is constructed by fitting the duration of the evolution from a pre-death state to a death state with the frequency of decay anomaly markers and rising edge struggle markers. The first lifespan decay model characterizes the predictive relationship of the duration of the evolution from a sub-healthy state to a pre-death state, and the second lifespan decay model characterizes the predictive relationship of the duration of the evolution from a pre-death state to a death state. Based on the first and second lifespan decay models, and by fusing the duration of the current state with the degradation rate, the remaining time until functional termination is calculated, generating the equipment health state duration.

[0056] The first lifespan decay model can be a mathematical model used to predict the duration required for an electrical appliance to evolve from a sub-healthy state to a pre-death state. Its core input parameters are the "actual duration from sub-health to pre-death" and the "trend slope of the harmonic ratio offset event." By fitting the correlation between these two parameters, accurate prediction of the evolution duration of similar devices at this stage can be achieved. The second lifespan decay model can also be a mathematical model used to predict the duration required for an electrical appliance to evolve from a pre-death state to a dead state. Its core input parameters are the "actual duration from pre-death to death" and the "occurrence frequency of decay anomaly markers and rising edge struggle markers." By fitting the negative correlation between these two parameters, prediction of the evolution duration of this stage can be achieved.

[0057] Specifically, traditional equipment maintenance relies heavily on manual experience (such as "replace the capacitor after 5 years of use for air conditioners" or "inspect the motor after 1000 hours of operation"), failing to consider individual equipment differences. This can easily lead to over-maintenance or under-maintenance. For example, two air conditioners of the same model might have different capacitors. One, due to low usage (2 hours per day), might still have a usable capacitor after 5 years, leading to forced replacement based on traditional experience and wasted costs. The other, due to high usage (10 hours per day), might experience capacitor degradation after only 3 years, which traditional experience doesn't cover, resulting in under-maintenance and malfunction. Furthermore, traditional methods record equipment status in a fragmented manner, failing to present a complete evolution from "sub-health - pre-death - death." This makes it difficult for maintenance personnel to trace the root cause of the malfunction and analyze the reasons for degradation. For instance, if a factory water pump suddenly enters a dead state, traditional methods only record "current shutdown," but cannot trace "when the sub-health occurred" or "what abnormalities occurred during the pre-death stage." Maintenance personnel would have to start from scratch, which is time-consuming and labor-intensive. If the equipment is under warranty, the lack of a complete status evolution record can easily lead to disputes between users and manufacturers regarding the cause of the malfunction. To address the above issues, this step begins by having the edge AI system construct a "device health lifecycle sequence" when the device is initially identified as being in a "sub-healthy state." This sequence continuously records all subsequent state changes (such as entering a pre-death state), the duration of each state, and corresponding key characteristic data (such as the trend slope of harmonic ratio shift and the frequency of start-up anomaly markers). When lifespan prediction is needed, the system invokes this sequence. Based on historical case data recorded in the sequence, the system uses regression analysis and other fitting methods to establish a "first lifespan decay model." This model describes how the duration from sub-healthy to pre-death changes with the trend slope of harmonic ratio shift. Simultaneously, a "second lifespan decay model" is established. The model describes how the duration from pre-death to death varies with the frequency of startup anomaly markers. Next, the system reads the current state of the device: if the device is currently in a sub-healthy state, it obtains the duration it has been in this state and the slope of the currently monitored harmonic offset trend, substitutes these values ​​into the "first lifetime decay model" to calculate the remaining state duration, and adds the typical duration from pre-death to death calculated by the "second lifetime decay model" to obtain the total "device health state duration." If the device is already in a pre-death state, it directly obtains the duration it has been in this state and the frequency of the current startup anomaly markers, substitutes these values ​​into the "second lifetime decay model" to calculate the remaining "device health state duration."

[0058] The method provided in this embodiment constructs a time series and decay model based on the equipment's own historical degradation data, effectively overcoming the shortcomings of traditional population-based statistical life prediction methods with insufficient accuracy, making the prediction results more consistent with the actual operating conditions of the equipment; the generated "equipment health status duration" provides users with a clear time expectation, enabling them to scientifically formulate maintenance strategies and accurately arrange spare parts procurement and equipment replacement plans based on the prediction results, thereby maximizing the utilization value of the equipment, minimizing the risk of unexpected downtime and maintenance costs, and realizing the refinement and intelligence of asset life cycle management.

[0059] In some embodiments, within the device health lifecycle sequence, the first historical evolution time from the sub-health state determination result to the pre-death state determination result is located; all harmonic ratio offset events within the first historical evolution time interval are analyzed, the key harmonic components leading to the state determination in each event are extracted, and the average rate of change of their amplitudes deviating from the dynamic reference baseline is calculated to generate a harmonic degradation trend slope; by performing an evolution correlation analysis between the first historical evolution time and the harmonic degradation trend slope: when the absolute value of the harmonic degradation trend slope exceeds a preset accelerated degradation threshold, the speed at which it evolves to the pre-death state will be significantly accelerated; based on the evolution correlation analysis and the accelerated degradation threshold, a first lifespan decay model is dynamically predicted to predict the time required for the current sub-health state to evolve to the pre-death state.

[0060] The first historical evolution duration can be the time span from when electrical equipment is first identified as being in a sub-healthy state to when it is first identified as being in a pre-death state within the equipment's health life cycle sequence. It is a core time parameter reflecting the speed at which the equipment evolves from a sub-healthy state to a pre-death state. The key harmonic component can be the core harmonic component that causes the equipment to evolve from a sub-healthy state to a pre-death state within the first historical evolution duration. It is usually a harmonic of a specific order (such as the 3rd or 5th harmonic), and the degree and rate at which its amplitude deviates from the dynamic reference baseline directly determines the state evolution process. The slope of the harmonic degradation trend can be the slope of a trend line formed by linear regression fitting with time as the horizontal axis and the amplitude deviation of the key harmonic component as the vertical axis. It is used to visually characterize the acceleration or deceleration trend of the key harmonic component's degradation. Evolutionary correlation analysis can be an analytical process to explore the causal relationship between the first historical evolution duration and the slope of the harmonic degradation trend. By quantifying the correlation between the two (e.g., for every 0.1% increase in the slope per day, the evolution duration shortens by 2 days), the influence of the harmonic degradation rate on the state evolution process is clarified, providing logical support for the first lifetime decay model. A preset accelerated degradation threshold is used as the critical slope value to determine whether harmonic degradation has entered the accelerated stage. When the absolute value of the harmonic degradation trend slope exceeds this threshold, the rate of evolution from sub-healthy to pre-death will significantly accelerate.

[0061] Specifically, in the predictive maintenance system based on equipment health status profiling, accurately predicting the time frame for the critical evolutionary stage from "sub-health" to "pre-death" is the core link in the entire method's transition from condition diagnosis to lifespan management. Sub-health reveals the beginning of equipment degradation, while pre-death indicates the critical point of functional failure. The time window between the two directly determines the urgency of maintenance decisions and the planning space. Traditional prediction methods face severe challenges here: First, they often rely on the overall mean time between failures (MTBF) of the equipment or fixed inspection cycles. These static indicators cannot reflect the unique degradation trajectory of individual equipment in actual operation. Second, even if they focus on changes in certain characteristic parameters, they are mostly isolated state point judgments, lacking continuous tracking and quantification of the key dynamic factor of the "rate" of parameter change. This results in the inability to capture the nonlinear acceleration phenomenon in the degradation process, thus creating a prediction blind spot when the equipment suddenly deteriorates, causing a lag in maintenance actions. To address the above issues, this step first involves the edge AI system precisely extracting two time points from the constructed "device health lifecycle sequence": the first time a "sub-health state determination result" is generated, and the first time a "pre-death state determination result" is generated. The difference between these two points is calculated to obtain the "first historical evolution duration." Subsequently, the system focuses on all "harmonic ratio offset events" recorded within this duration, analyzing each event to identify the dominant "key harmonic components" (e.g., the seventh harmonic pointing to power transistor aging) that are associated with specific component degradation by a pre-defined electrical component degradation knowledge base. The system calculates the average deviation of the amplitude of these key harmonic components from the "dynamic reference baseline" over the entire time interval. The system calculates the degradation rate to generate a comprehensive "harmonic degradation trend slope". Next, the system performs "evolution correlation analysis", which models the correlation between the "first historical evolution duration" of multiple devices or multiple time periods and the corresponding "harmonic degradation trend slope" and sets an "accelerated degradation threshold" based on the distribution of historical data. When the system makes a prediction for a device that is currently in a sub-healthy state, it compares the real-time calculated "harmonic degradation trend slope" with this threshold: if the absolute value of the slope does not exceed the threshold, the remaining time is predicted according to the conventional correlation; if it exceeds the threshold, the accelerated prediction mode is immediately activated to significantly shorten the predicted remaining evolution time. Finally, the system constructs a dynamic "first lifetime decay model" based on this complete logic.

[0062] The method provided in this embodiment enables dynamic, accurate, and adaptive prediction of the time it takes for equipment to evolve from a sub-healthy state to a pre-death state. By deeply analyzing the trend slope of harmonic degradation and establishing an accelerated discrimination mechanism, it effectively overcomes the shortcomings of traditional prediction methods in responding slowly to nonlinear degradation processes, allowing the life prediction to closely match the real-time health deterioration rate of the equipment. The generated "first life decay model" provides users with highly personalized and timely updated time warnings, enabling them to accurately plan maintenance windows, optimize the spare parts supply chain, and rationally allocate maintenance resources based on the actual degradation dynamics of the equipment.

[0063] In some embodiments, within the device health lifecycle sequence, the second historical duration from the first occurrence of the pre-death state determination result to the generation of the death state determination result is located; all device startup events within this time interval are analyzed, and the occurrence count and frequency of decay anomaly markers and rising edge struggle markers within a unit time window are statistically analyzed; the occurrence frequency of decay anomaly markers and rising edge struggle markers are superimposed to generate a joint frequency index characterizing the overall functional instability; by performing instability correlation analysis on the second historical duration and the joint frequency index, it is identified that when the joint frequency index exceeds the critical instability threshold, the duration of the device evolving from the pre-death state to the death state will be drastically shortened; based on the critical instability threshold and the instability correlation analysis results, a second lifespan decay model is dynamically predicted to predict the time required for the device to evolve from the current pre-death state to the death state.

[0064] The second historical duration can be the time span from when an electrical device is first determined to be in a pre-death state to when it is first determined to be in a dead state within the device's health life cycle sequence. It is a core time parameter for measuring the speed at which a device evolves from pre-death to death. The unit time window can be a time unit used to statistically analyze the frequency of decay anomaly markers and rising edge struggle markers. It is set according to the device startup frequency, ensuring that each window contains at least one device startup event. The joint frequency index can be a comprehensive index obtained by weighting the occurrence frequencies of decay anomaly markers and rising edge struggle markers. It is used to quantify the overall functional instability of the device. The weights are set according to the influence of the two types of markers on the device function (e.g., decay anomaly marker weight 0.6, rising edge struggle marker weight 0.4). Instability correlation analysis can be an analytical process to explore the causal relationship between the second historical duration and the joint frequency index. By quantifying the correlation between the two (e.g., for every 0.2 times / hour increase in the joint frequency index, the duration shortens by 1.5 days), the influence of the functional instability speed on the state evolution process is clarified, providing logical support for the second life decay model. The critical instability threshold can be a critical value of the joint frequency index for determining whether the equipment is in an accelerated stage of functional instability. When the joint frequency index exceeds this threshold, the rate of evolution of the equipment from pre-death to death will be shortened sharply.

[0065] Specifically, in the predictive maintenance system based on equipment health status profiling, accurately predicting the time from "pre-death" to "death"—the ultimate evolutionary stage—is the crucial step in moving from risk warning to final decision-making. The pre-death state reveals the critical point of equipment functional failure, while the death state declares the end of the equipment's life. The time window between the two is extremely short and highly uncertain, directly determining the urgency of maintenance actions and the effectiveness of the final disposal strategy. Traditional prediction methods face significant challenges here: First, they often treat the pre-death state as a uniform, linear deterioration process, ignoring the sudden and nonlinear characteristics of equipment functional instability near the critical point. Second, even if certain abnormal markers are considered, they are mostly isolated event records, lacking continuous tracking and comprehensive analysis of the key dynamic indicator of "occurrence frequency" of abnormal events. This results in the inability to capture the cumulative effect and accelerating trend of functional instability, leaving the system unprepared when equipment suddenly collapses, causing catastrophic downtime. To address the above issues, this step first involves the edge AI system precisely extracting two time points from the constructed "device health lifecycle sequence": the time from the first "pre-death state determination result" to the time of the "death state determination result." The difference between these two points is calculated to obtain the "second historical survival duration." Subsequently, the system focuses on all device startup events recorded within this duration, analyzing each event and counting the number of times the "decay anomaly marker" and "rising edge struggle marker" are triggered. The system calculates the frequency of these two markers within a set unit time window (e.g., every 24 hours or every 100 startups). Next, the system superimposes these two frequencies (e.g., through weighted summation or arithmetic average) to generate a comprehensive "joint frequency." The system first uses an "instability correlation analysis" to quantify the overall functional instability of the equipment during the pre-death phase. Next, it performs this analysis, which correlates the "second historical survival duration" of multiple devices or time periods with the corresponding "joint frequency index," analyzes the statistical patterns between them, and sets a "critical instability threshold" based on historical data distribution. When the system predicts the future of an equipment in a pre-death state, it compares the real-time calculated "joint frequency index" with this threshold. If the index does not exceed the threshold, the remaining time is predicted based on conventional correlations; if it exceeds the threshold, an emergency prediction mode is immediately activated, significantly shortening the predicted remaining evolution time. Finally, the system constructs a dynamic "second lifetime decay model" based on this complete logic.

[0066] By comprehensively analyzing the frequency of decay anomalies and rising edge struggle markers and establishing a critical instability discrimination mechanism, the traditional prediction method effectively overcomes the drawback of delayed response to the functional instability acceleration process, enabling the lifetime prediction to closely match the real-time functional deterioration of the equipment. The generated "second lifetime decay model" provides users with a highly personalized and timely updated ultimate time warning, enabling them to decisively take final measures such as emergency maintenance, shutdown replacement, or activation of backup equipment in the last window before the equipment's functional failure.

[0067] In some embodiments, user electricity consumption behavior profile labels are derived and assigned based on the current life cycle state type of electrical equipment and its corresponding equipment health status duration. If the equipment is in a sub-healthy state and the duration indicates a maintenance window, the user behavior profile is defined as a risk-averse user. This type indicates that the user has a tolerance for potential equipment failure risks and tends to perform maintenance within the plan. If the equipment is in a pre-death state and the duration indicates an emergency risk, the user behavior profile is defined as a cost-sensitive user. This type indicates that the user is highly sensitive to losses caused by sudden equipment downtime and tends to take emergency loss-mitigation measures. Based on the electricity consumption behavior profile labels, a matching comprehensive user profile report is generated: for risk-averse users, the report focuses on providing preventive maintenance planning and risk warning suggestions based on the duration; for cost-sensitive users, the report focuses on providing decision-making suggestions for emergency loss-mitigation, spare parts preparation, or immediate replacement based on the duration.

[0068] Electricity consumption behavior profile tags are classification labels derived from the current life cycle status and health status duration of electrical equipment. These tags characterize users' equipment maintenance decision preferences and are primarily divided into two categories: "risk-averse" and "cost-sensitive." Risk-averse users are those whose equipment is in a sub-healthy state but has a sufficient health status duration (at least a 15-day maintenance window). Their core characteristic is a certain tolerance for potential equipment failure risks, prioritizing scheduled maintenance (such as weekends or production breaks) to avoid disrupting daily routines or incurring additional costs due to emergency repairs. Cost-sensitive users are those whose equipment is in a near-death state and has an extremely short health status duration (less than 7 days, posing an emergency risk). Their core characteristic is a high sensitivity to economic losses caused by sudden equipment downtime (such as factory production line shutdowns or food spoilage in medical refrigerators), prioritizing emergency loss mitigation measures (such as repairs within 24 hours or immediate replacement of spare parts) to reduce downtime risks. A comprehensive user profile report is a structured report generated based on electricity consumption behavior profile tags, which includes equipment status, lifespan, and maintenance recommendations, and presents content differentiated by user type.

[0069] Specifically, in the scenario of comprehensive health assessment of electrical equipment and user decision support, traditional methods only output equipment status and simple maintenance suggestions, without considering differences in user decision preferences. This results in suggestions that are highly generalized but lack specificity, making it difficult for users to quickly formulate solutions tailored to their needs. To address these issues, this step first involves the edge AI system acquiring the current "current lifecycle status type of the electrical appliance" (e.g., "sub-healthy" or "pre-death") and the calculated "device health status duration" (e.g., "30 days remaining" or "3 days remaining"). The system's built-in decision logic engine performs rule matching based on these two core inputs. For example, if the status is "sub-healthy" and the duration is longer than a preset "emergency action threshold" (e.g., 15 days), the system "derives and assigns" the "risk-averse user" label to the user associated with the device. Conversely, if the status is "pre-death" or the duration is shorter than the threshold, the system assigns the "cost-sensitive user" label. Next, the system calls the report template corresponding to the profile label. For "risk-averse users," the system generates a "Comprehensive User Profile Report," whose content structure focuses on "preventive maintenance planning." For example, the report clearly identifies the user type at the beginning, and the main body details the maintenance window recommendations based on the remaining days, the risk and cost analysis of performing maintenance during different windows, and the long-term equipment health monitoring plan. For "cost-sensitive users," the system generates a report that focuses on "emergency decision support." For example, the report prominently warns of urgent risks at the beginning, the main body directly compares the potential losses of immediate action versus delayed action, provides spare parts supply chain information, emergency service contact information, and strongly recommends immediately developing a replacement or overhaul plan. Finally, this structured, personalized report, which can be directly used for decision-making, is output to the user.

[0070] The "Comprehensive User Profile Report" generated through the method provided in this embodiment greatly improves the efficiency of transforming technical information into management actions. It can guide users to make maintenance decisions that best suit their interests based on their own risk preferences and resource conditions, thereby achieving an optimal balance between maintenance costs and downtime risks while ensuring equipment reliability, and significantly improving the level of refinement and intelligence in equipment management.

[0071] Figure 3 A schematic diagram of the structure of an edge AI-based electricity consumption behavior profiling system provided in an embodiment of this application is shown below. Figure 3 As shown, the electricity consumption behavior profiling system 300 based on edge AI in this embodiment includes: a state analysis module 301, a state prediction module 302, and a comprehensive profiling module 303;

[0072] The status analysis module 301 is used to acquire multi-dimensional operating data of electrical equipment, perform multi-state hierarchical analysis based on the multi-dimensional operating data of electrical equipment, determine the current life cycle status type of electrical equipment, and generate the current life cycle status type of electrical equipment; the status prediction module 302 is used to establish a health benchmark according to the time axis based on the current life cycle status type of electrical equipment and deduce the health life cycle sequence of equipment, and generate the equipment health status duration predicted by time pattern; the comprehensive profile module 303 is used to perform a comprehensive health assessment based on the current life cycle status type of electrical equipment and the equipment health status duration, and generate and output a comprehensive user profile report for user decision-making.

[0073] Optionally, when the state analysis module 301 performs multi-state hierarchical analysis based on the multi-dimensional operating data of the electrical equipment, determines the current life cycle state type of the electrical equipment, and generates the current life cycle state type of the electrical equipment, it is specifically used for: the multi-dimensional operating data of the electrical equipment includes current harmonic spectrum, high-frequency current waveform, steady-state power consumption trajectory, and reference voltage phase signal; by tracking the changes in the component proportions of the current harmonic spectrum over time, it identifies the continuous degradation and drift of its harmonic structure and generates a sub-health state determination result; based on the component degradation trend indicated by the sub-health state determination result, it extracts the starting voltage from the high-frequency current waveform... The waveform distortion characteristics during the transient phase are identified to determine abnormal oscillation decay or plateauing of the rising edge, generating a pre-death state determination result. When the pre-death state determination result confirms that the device has entered the critical point of functional failure, the steep drop or surge of the current power consumption trajectory is analyzed using the steady-state power consumption trajectory as a standard, and the loss of synchronization or jump of the current voltage phase signal is analyzed using the reference voltage phase signal as a standard. The associated characteristics representing the termination of device function are identified, generating a death state determination result. Based on the sub-health state determination result, the pre-death state determination result, and the death state determination result, the current life cycle state type of the electrical appliance is comprehensively generated.

[0074] Optionally, when the state analysis module 301 identifies the continuous degradation drift of the harmonic structure by tracking the changes in the component ratio of the current harmonic spectrum over time and generates a sub-health state judgment result, it is specifically used for: extracting the distribution of harmonic components within a typical normal working cycle from the historical operating data of the electrical equipment itself, and establishing a dynamic reference baseline reflecting the proportional relationship of each harmonic amplitude; slicing the current harmonic spectrum by time window, comparing it proportionally with the dynamic reference baseline one by one, capturing abnormal increases or decreases in the amplitude of specific harmonics, and generating harmonic ratio offset events; performing trend analysis on the harmonic ratio offset events within a continuous time window, and determining that a continuous degradation drift feature is formed when the proportion of a specific harmonic shows a unidirectional continuous deviation and an increasing amplitude; mapping the harmonic number and deviation direction corresponding to the continuous degradation drift feature to a preset electrical component degradation knowledge base, and generating a sub-health state judgment result containing the specific degradation component type and stage.

[0075] Optionally, when the state analysis module 301 identifies abnormal oscillation decay or rising edge plateauing based on the waveform distortion characteristics extracted from the high-frequency current waveform during the startup transient phase and generates a pre-death state determination result, it is specifically used for: when the sub-health state determination result indicates that a specific power component or power mechanism has continuous degradation, at each startup instant of the electrical equipment, extracting the startup transient information of the high-frequency current waveform and analyzing the instantaneous dynamic behavior of its waveform profile; analyzing the decay process of the high-frequency current waveform oscillation after reaching its peak value, identifying whether there is a phenomenon that the oscillation amplitude has not recovered to the threshold level within the expected number of times or the number of oscillation cycles is significantly shorter than the standard startup mode, and generating a decay abnormality mark characterizing the loss of inertia of the mechanism; tracking the slope continuity during the current rising phase, and when the interruption of the rising kinetic energy is detected to form a plateau-like flat top or a stepped hesitant climb, generating a rising edge struggle mark characterizing the obstruction of driving power; spatiotemporally associating the decay abnormality mark with the rising edge struggle mark, if the two appear in the same startup event, it is determined that the core moving parts of the equipment or the drive circuit have functional failure, and a pre-death state determination result is generated.

[0076] Optionally, when the state analysis module 301 identifies the associated features characterizing the termination of device function based on the steep drop or surge phenomenon of the current power consumption trajectory analyzed using the steady-state power consumption trajectory as a standard and the loss of synchronization or jump phenomenon of the current voltage phase signal analyzed using the reference voltage phase signal as a standard, and generates a death state determination result, it is specifically used for: when the pre-death state determination result simultaneously includes the attenuation anomaly marker and the rising edge struggle marker, based on the steady-state power consumption trajectory, capturing the cliff drop or sudden surge phenomenon that occurs after it leaves the normal operating range, and generating an extreme power consumption anomaly event; based on the reference voltage phase signal, identifying the continuous angular deviation of its relative to the grid standard phase. A phase lock-out event is generated by shifting or instantaneously changing the angle. The extreme power consumption anomaly event and the phase lock-out event are aligned on the time axis and correlated by causal logic: when the steady-state power consumption trajectory drops sharply and is accompanied by a continuous angle shift in the reference voltage phase signal, it is determined that the main circuit is open or the drive core is completely disabled. When the steady-state power consumption trajectory spikes suddenly and is accompanied by an instantaneous angle shift in the reference voltage phase signal, it is determined that the internal component is short-circuited or the power mechanism is mechanically locked. Based on the establishment of any correlation feature, it is directly determined that the equipment has lost its basic operating function, and a death state judgment result representing irreversible shutdown of the equipment is generated.

[0077] Optionally, when the state prediction module 302 establishes a health benchmark and extrapolates the equipment health life cycle sequence based on the current life cycle state type of the electrical appliance, and generates the equipment health state duration predicted by time regularity, it is specifically used to: take the moment when the electrical appliance is first marked as the sub-healthy state determination result as the starting point of the life cycle extrapolation sequence, integrate all the multi-dimensional operating data of the electrical appliance recorded from this starting point to the current moment and their corresponding life cycle state types, and construct the equipment health life cycle sequence; based on the equipment health life cycle sequence, by fitting the duration of the evolution from the sub-healthy state to the pre-death state with the harmonic ratio deviation... A first lifetime decay model is constructed based on the trend slope of the migration event; a second lifetime decay model is constructed by fitting the duration of evolution from the pre-death state to the death state with the occurrence frequency of the decay anomaly marker and the rising edge struggle marker; the first lifetime decay model is used to characterize the predictive relationship of the duration of evolution from the sub-healthy state to the pre-death state, and the second lifetime decay model is used to characterize the predictive relationship of the duration of evolution from the pre-death state to the death state; based on the first lifetime decay model and the second lifetime decay model, and by fusing the duration of the current state with the degradation rate, the remaining time until the function terminates is calculated, and the lifespan of the device's health state is generated.

[0078] Optionally, when constructing a first lifetime decay model based on the duration of evolution from a sub-healthy state to a pre-death state by fitting the trend slope of the harmonic ratio offset events, the state prediction module 302 is specifically used for: locating the first historical evolution duration from the sub-healthy state determination result to the pre-death state determination result in the equipment health life cycle sequence; analyzing all the harmonic ratio offset events within the first historical evolution duration interval, extracting the key harmonic components that lead to the state determination in each event, and calculating the average rate of change of their amplitudes deviating from the dynamic reference baseline to generate a harmonic degradation trend slope; performing an evolution correlation analysis between the first historical evolution duration and the harmonic degradation trend slope: when the absolute value of the harmonic degradation trend slope exceeds a preset accelerated degradation threshold, the speed at which it evolves to the pre-death state will be significantly accelerated; and dynamically predicting the first lifetime decay model based on the evolution correlation analysis and the accelerated degradation threshold, using the time required for the evolution from the current sub-healthy state to the pre-death state.

[0079] Optionally, when constructing a second lifetime decay model based on the duration of evolution from the pre-death state to the death state by fitting the occurrence frequency of the decay anomaly marker and the rising edge struggle marker, the state prediction module 302 is specifically used for: locating the second historical duration from the first occurrence of the pre-death state determination result to the generation of the death state determination result in the device health life cycle sequence; analyzing all device startup events within this time interval, and respectively counting the number of occurrences and the frequency of occurrence of the decay anomaly marker and the rising edge struggle marker within a unit time window; superimposing the occurrence frequency of the decay anomaly marker and the occurrence frequency of the rising edge struggle marker to generate a joint frequency index characterizing the overall functional instability; performing instability correlation analysis between the second historical duration and the joint frequency index: identifying that when the joint frequency index exceeds the critical instability threshold, the duration of evolution from the pre-death state to the death state will be drastically shortened; and dynamically predicting the second lifetime decay model based on the critical instability threshold and the instability correlation analysis result, using the time required for evolution from the current pre-death state to the death state.

[0080] Optionally, the comprehensive profiling module 303 is specifically used for: deriving and assigning user electricity consumption behavior profile tags based on the current life cycle state type of the electrical equipment and its corresponding equipment health status duration; if the equipment is in a sub-healthy state and the duration indicates a maintenance window, the user behavior profile is defined as a risk-averse user, which indicates that the user has tolerance for potential equipment failure risks and tends to perform maintenance within the plan; if the equipment is in a pre-death state and the duration indicates an emergency risk, the user behavior profile is defined as a cost-sensitive user, which indicates that the user is highly sensitive to losses caused by sudden equipment downtime and tends to take emergency loss-mitigation measures; based on the electricity consumption behavior profile tags, generating a matching comprehensive user profile report: for the risk-averse user, the report focuses on providing preventive maintenance planning and risk warning suggestions based on the duration; for the cost-sensitive user, the report focuses on providing decision-making suggestions for emergency loss-mitigation, spare parts preparation, or immediate replacement based on the duration.

[0081] The system in this embodiment can be used to execute the methods of any of the above embodiments, and its implementation principle and technical effect are similar, so they will not be described again here.

Claims

1. A method for profiling electricity consumption behavior based on edge AI, characterized in that, include: Edge computing nodes are deployed in electrical equipment scenarios to acquire multi-dimensional operational data of the electrical equipment. Based on this multi-dimensional operational data, a multi-state hierarchical analysis model of edge AI is invoked to perform multi-state hierarchical analysis, determine the current lifecycle state type of the electrical equipment, and generate the current lifecycle state type of the electrical equipment, including: The multidimensional data of the electrical equipment operation includes current harmonic spectrum, high-frequency current waveform, steady-state power consumption trajectory and reference voltage phase signal; By tracking the temporal component ratio changes of the current harmonic spectrum, the continuous deterioration and drift of its harmonic structure are identified, and a sub-health state determination result is generated. Based on the component degradation trend indicated by the sub-health state determination result, by extracting the waveform distortion characteristics of the start-up transient phase in the high-frequency current waveform, its abnormal oscillation decay or rising edge plateauing struggle is identified, and a pre-death state determination result is generated. Once the pre-death state determination result confirms that the device has entered the critical point of functional failure, the steady-state power consumption trajectory is used as the standard to analyze the steep drop or surge of the current power consumption trajectory and the loss of step or jump of the current voltage phase signal is used as the standard to analyze the current voltage phase signal, identify the associated features that characterize the termination of device function, and generate the death state determination result. Based on the sub-health state determination result, the pre-death state determination result, and the death state determination result, the current life cycle state type of the electrical appliance is generated comprehensively; Based on the current life cycle status type of the electrical appliance, a health benchmark is established according to the time axis and the equipment health life cycle sequence is deduced to generate the equipment health status duration predicted by time pattern. Based on the current lifecycle status of the electrical appliance and the duration of its health status, a comprehensive health assessment is performed to generate and output a comprehensive user profile report for user decision-making, including: Based on the current life cycle status type of electrical equipment and its corresponding health status duration, user electricity consumption behavior profile labels are derived and assigned. If the equipment is in a sub-healthy state and its lifespan indicates a maintenance window, the user behavior profile is defined as a risk-averse user. This type indicates that the user has a tolerance for the potential failure risk of the equipment and tends to perform maintenance within the plan. If the device is in a pre-death state and its remaining life indicates an urgent risk, the user behavior profile is defined as a cost-sensitive user. This type of user is characterized by being highly sensitive to losses caused by sudden device downtime and tending to take emergency loss-mitigation measures. Based on the electricity consumption behavior profile tags, a matching comprehensive user profile report is generated: For the aforementioned risk-averse users, the report focuses on providing preventative maintenance planning and risk warning recommendations based on the lifespan of the device. For the aforementioned cost-sensitive users, the report focuses on providing decision-making recommendations based on lifespan, including emergency loss mitigation, spare parts preparation, or immediate replacement.

2. The method according to claim 1, characterized in that, The process of tracking the temporal component ratio changes of the current harmonic spectrum to identify the continuous degradation and drift of its harmonic structure, and generating a sub-health state assessment result, includes: From the historical operating data of electrical equipment, the distribution of harmonic components within a typical normal working cycle is extracted, and a dynamic reference baseline reflecting the proportional relationship of the amplitude of each harmonic is established. The current harmonic spectrum is sliced ​​according to a time window, and each slice is compared with the dynamic reference baseline to capture abnormal increases or decreases in the amplitude of specific harmonics and generate harmonic ratio offset events. Trend analysis is performed on the harmonic ratio shift events within a continuous time window. When the ratio of a specific harmonic shows a unidirectional continuous deviation and the amplitude increases, it is determined that a continuous degradation drift characteristic has been formed. Based on the harmonic order and deviation direction corresponding to the continuous degradation drift characteristics, the data is mapped to a preset electrical component degradation knowledge base to generate a sub-health status judgment result containing specific degradation component types and stages.

3. The method according to claim 2, characterized in that, The process involves extracting waveform distortion features from the high-frequency current waveform during the initial transient phase, identifying abnormal oscillation decay or plateauing on the rising edge, and generating a pre-death state determination result, including: When the sub-health state determination result indicates that a specific power component or power mechanism is experiencing continuous degradation, at the moment of each start-up of the electrical equipment, the transient information of the high-frequency current waveform is extracted, and the real-time dynamic behavior of its waveform profile is analyzed: Analyze the decay process of the high-frequency current waveform oscillation after reaching its peak value, identify whether there is a phenomenon that the oscillation amplitude does not recover to the threshold level within the expected number of times or the number of oscillation cycles is significantly shorter than the standard start-up mode, and generate a decay anomaly marker characterizing the loss of inertia of the mechanism. During the current rise phase, the slope continuity is tracked. When the interruption of the rising kinetic energy is detected, forming a plateau-shaped flat top or exhibiting a step-like hesitant climb, a rising edge struggle mark representing the obstruction of driving power is generated. The decay anomaly marker and the rising edge struggle marker are spatiotemporally correlated. If the two appear in the same startup event, it is determined that the core moving parts or drive circuit of the device have functional failure, and a pre-death state determination result is generated.

4. The method according to claim 3, characterized in that, The process of analyzing the sharp drop or surge of the current power consumption trajectory using the steady-state power consumption trajectory as a standard and analyzing the loss of synchronization or jump of the current voltage phase signal using the reference voltage phase signal as a standard, identifies the associated characteristics characterizing the termination of device function, and generates a death state determination result, including: When the pre-death state determination result contains both the decay anomaly marker and the rising edge struggle marker, based on the steady-state power consumption trajectory, capture the cliff drop or sudden surge phenomenon that occurs after it leaves the normal working range, and generate extreme power consumption anomaly events. Based on the reference voltage phase signal, identify its continuous angular offset or instantaneous angular jump relative to the standard phase of the power grid, and generate a phase loss event; The extreme power consumption anomaly event and the phase lock-out event are time-aligned and causally correlated: When the steady-state power consumption trajectory drops sharply and is accompanied by a continuous angular shift in the reference voltage phase signal, it is determined that the main circuit is open or the driving core is completely disabled. When the steady-state power consumption trajectory suddenly spikes, and the reference voltage phase signal simultaneously undergoes an instantaneous angle jump, it is determined that the associated characteristics of internal component breakdown short circuit or power mechanism mechanical lock-up are formed. Based on the establishment of any associated feature, it is directly determined that the equipment has lost its basic operating functions, and a death state determination result representing the irreversible shutdown of the equipment is generated.

5. The method according to claim 4, characterized in that, The process of establishing a health benchmark based on the current lifecycle state type of the electrical appliance, extrapolating the equipment health lifecycle sequence according to the time axis, and generating the predicted equipment health status duration based on time patterns includes: The moment when the electrical equipment is first marked as the sub-healthy state determination result is taken as the starting point of the lifespan extrapolation sequence. All multi-dimensional data of the operation of the electrical equipment and its corresponding life cycle state type recorded from this starting point to the current moment are integrated to construct the equipment health life cycle sequence. Based on the device health life cycle sequence, a first lifespan decay model is constructed by fitting the trend slope of the duration of the evolution from a sub-healthy state to a pre-death state with the harmonic ratio shift event. A second lifetime decay model is constructed by fitting the duration of the evolution from the pre-death state to the death state with the occurrence frequency of the decay anomaly marker and the rising edge struggle marker. The first lifespan decay model is used to characterize the predictive relationship of the duration of evolution from a sub-healthy state to a pre-death state, and the second lifespan decay model is used to characterize the predictive relationship of the duration of evolution from a pre-death state to a death state. Based on the first lifespan decay model and the second lifespan decay model, and by combining the duration of the current state with the degradation rate, the remaining time until the function terminates is calculated, and the health status duration of the device is generated.

6. The method according to claim 5, characterized in that, The first lifespan decline model is constructed by fitting the trend slope of the harmonic ratio shift event to the duration of the evolution from a sub-healthy state to a pre-death state, including: In the device health lifecycle sequence, the first historical evolution time from the sub-health state determination result to the pre-death state determination result is located; Analyze all the harmonic ratio offset events within the first historical evolution time interval, extract the key harmonic components that lead to the state determination in each event, and calculate the average rate of change of their amplitudes from the dynamic reference baseline to generate the harmonic degradation trend slope. By performing evolution correlation analysis between the first historical evolution duration and the slope of the harmonic degradation trend: when the absolute value of the slope of the harmonic degradation trend exceeds the preset accelerated degradation threshold, it is determined that the speed at which it evolves to the pre-death state will be significantly accelerated. The first lifespan decay model dynamically predicts the time required for the current sub-healthy state to evolve into a pre-death state based on the evolutionary correlation analysis and the accelerated degradation threshold.

7. The method according to claim 6, characterized in that, The second lifetime decay model is constructed by fitting the duration of the evolution from the pre-death state to the death state with the occurrence frequency of the decay anomaly marker and the rising edge struggle marker, including: In the device health lifecycle sequence, the second historical duration from the first occurrence of the pre-death state determination result to the generation of the death state determination result is located; Analyze all device startup events within this time interval, and count the number of times and frequency of occurrence of the decay anomaly marker and the rising edge struggle marker within a unit time window; The frequency of occurrence of the attenuation anomaly marker is superimposed with the frequency of occurrence of the rising edge struggle marker to generate a joint frequency index characterizing the overall degree of functional instability. By performing an instability correlation analysis between the second historical duration and the joint frequency index, it was identified that when the joint frequency index exceeds the critical instability threshold, the duration of the device evolving from a pre-death state to a dead state will be drastically shortened. Based on the critical instability threshold and the instability correlation analysis results, a second life decay model dynamically predicts the time required for the evolution from the current pre-death state to the death state.

8. A power consumption behavior profiling system based on edge AI, characterized in that, Applied to the method as described in any one of claims 1-7, comprising: The status analysis module is used to acquire multi-dimensional data on the operation of electrical equipment, perform multi-state level analysis based on the multi-dimensional data on the operation of electrical equipment, determine the current life cycle status type of electrical equipment, and generate the current life cycle status type of electrical equipment. The status prediction module is used to establish a health benchmark and deduce the equipment health life cycle sequence based on the current life cycle status type of the electrical appliance according to the time axis, and generate the equipment health status duration predicted by time pattern. The comprehensive profile module is used to conduct a comprehensive health assessment based on the current life cycle status type of the electrical appliance and the duration of the device's health status, and generate and output a comprehensive user profile report for user decision-making.

Images