UPS system intelligent monitoring and self-adaptive power supply method and system based on Internet of Things and AI

Abstract

The invention relates to the field of artificial intelligence and electric power, and discloses an intelligent monitoring and self-adaptive power supply method and system for a UPS system based on Internet of Things and AI, and the method comprises the following steps: collecting key parameters of the UPS system through the Internet of Things, evaluating the health degree of equipment through an AI algorithm, and dynamically adjusting a power supply mode in combination with an optimal control and reinforcement learning algorithm. Intelligent optimization and closed-loop control of a power supply strategy are realized, and the stability and efficiency of the system are improved; the system comprises a sensor acquisition module, an Internet of Things sensing network module, a health degree evaluation module, a self-adaptive power supply decision module, a reinforcement learning optimization module and an execution and feedback module. Through data preprocessing, health degree evaluation, reinforcement learning optimization and closed-loop control technologies based on the Internet of Things and edge calculation, intelligent and dynamic adjustment and optimization of a UPS system power supply mode are realized, and the data accuracy, the response speed, the system adaptability and the overall operation efficiency are improved.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and power, specifically to a method and system for intelligent monitoring and adaptive power supply of UPS systems based on the Internet of Things and AI. Background Technology

[0002] In existing UPS (Uninterruptible Power Supply) systems, the selection and adjustment of power supply modes typically rely on static preset algorithms and fixed control strategies. Traditional UPS systems determine the operating status of equipment through simple status monitoring and manually set thresholds, and select the power supply mode accordingly. However, this approach has significant limitations. Because traditional solutions lack dynamic adjustment mechanisms, they may not be able to make timely adaptive adjustments when encountering load fluctuations or equipment failures, leading to decreased system efficiency or potential safety hazards.

[0003] Furthermore, most existing sensor data acquisition technologies directly transmit raw data to the central processing unit for analysis, which increases data transmission latency and makes the data susceptible to noise interference. Especially in complex power environments, the accuracy and real-time performance of the data are difficult to guarantee. Without an effective preprocessing mechanism, the operating status of the equipment cannot be accurately reflected, thus affecting the decision-making and execution of power supply strategies, resulting in a less intelligent and efficient system.

[0004] Furthermore, existing optimization algorithms typically rely on rules or models for adjustment, lacking adaptive capabilities. This means that UPS systems often employ fixed power supply strategies, failing to adapt flexibly to real-time data changes, resulting in an inability to quickly adjust when load demands shift. Even hardware upgrades enhancing computing power cannot resolve the latency issue of strategy updates, hindering optimal system performance under varying operating environments. Therefore, this invention proposes an intelligent monitoring and adaptive power supply method and system for UPS systems based on IoT and AI to address the shortcomings of existing technologies. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method and system for intelligent monitoring and adaptive power supply of UPS systems based on the Internet of Things and AI, which solves the problems of slow response and inflexible power supply strategy adjustment in traditional UPS systems in terms of dynamic load changes and data accuracy.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a UPS system intelligent monitoring and adaptive power supply method based on the Internet of Things and AI, comprising the following steps: S1. Real-time data collection of input power, output power, conversion efficiency, harmonic distortion rate, and equipment temperature parameters of the uninterruptible power supply system via the Internet of Things (IoT) sensing network. S2. Based on the collected parameters, the health status of the device is assessed using AI algorithms, and a health score is generated. S3. Based on the health score, combined with the load type, equipment status and current environmental conditions, the optimal control algorithm is used to adjust the power supply mode of the UPS system; S4. Dynamically select the optimal power supply mode under different load conditions through reinforcement learning algorithm; S5. Real-time adjustment of power supply mode, and feedback of adjustment results to control system for closed-loop optimization.

[0007] This invention also provides an intelligent monitoring and adaptive power supply system for UPS systems based on the Internet of Things and AI, including: The sensor acquisition module includes multiple types of sensors such as voltage, current, temperature, harmonics, power quality, and battery capacity, used to acquire UPS operating parameters in real time; The Internet of Things (IoT) sensing network module is used to transmit data collected by sensors to the central processing unit via wireless or wired communication. The health assessment module is used to assess the status of UPS equipment based on real-time data; An adaptive power supply decision module is used to adjust the power supply strategy based on optimization algorithms and health assessment results; The reinforcement learning optimization module is used to optimize the power supply strategy online through reinforcement learning algorithms. The execution and feedback module is used to control the power supply mode and send feedback data back to the sensor acquisition module, forming a closed-loop regulation mechanism.

[0008] This invention provides a method and system for intelligent monitoring and adaptive power supply of UPS systems based on the Internet of Things and AI. It has the following beneficial effects: 1. This invention employs a sensor data preprocessing technology solution based on the Internet of Things and edge computing, achieving the technical effect of improving data acquisition accuracy and response speed. Compared with existing technologies that typically rely on direct data acquisition by a central processing unit, traditional solutions often face problems such as data transmission delays and noise interference, failing to reflect the health status of devices in a timely and accurate manner. This invention significantly improves the real-time performance and stability of the system by performing data filtering and noise reduction at edge nodes, avoiding power supply decision errors caused by data lag.

[0009] 2. This invention introduces a health assessment module and combines real-time data to intelligently evaluate UPS equipment, achieving the technical effect of maintaining efficient power supply under load changes and environmental fluctuations. Compared to existing health monitoring schemes that rely solely on statically set thresholds, these schemes often cannot adapt to dynamic changes under different load environments. This invention dynamically assesses the operating status of UPS equipment, enabling real-time detection of equipment health and helping the system adjust its power supply strategy in a timely manner, ensuring efficient system operation under various working conditions.

[0010] 3. This invention achieves automated online optimization of power supply strategies through a reinforcement learning optimization module, realizing the technical effect of dynamically adjusting the power supply mode based on real-time feedback. Unlike the relatively fixed power supply mode selection mechanism in existing technologies, which typically operate based on preset rules and cannot flexibly adjust according to changes in actual load and equipment status, this invention continuously learns and optimizes power supply strategies through reinforcement learning algorithms. It can find the optimal power supply solution under different operating conditions, improving the system's intelligence and adaptability.

[0011] 4. This invention employs an execution and feedback closed-loop control technology, achieving the technical effect of rapid adjustment and continuous optimization of the power supply mode. Compared to the unidirectional control command method in existing technologies, which lack timely feedback and easily lead to a mismatch between the power supply strategy and the actual situation, this invention ensures that the system can promptly correct its operating strategy based on feedback through real-time data transmission from the execution and feedback module, maximizing system operating efficiency and enabling rapid response and recovery in the event of anomalies. Attached Figure Description

[0012] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a system architecture diagram of the present invention. Detailed Implementation

[0013] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0014] Please see Figure 1 This invention provides a method for intelligent monitoring and adaptive power supply of a UPS system based on the Internet of Things and AI, comprising the following steps: S1. Real-time data collection of input power, output power, conversion efficiency, harmonic distortion rate, and equipment temperature parameters of the uninterruptible power supply system via the Internet of Things (IoT) sensing network. In the IoT and AI-based intelligent monitoring and adaptive power supply method for UPS systems described in this invention, step S1 undertakes the tasks of real-time perception and raw data modeling of the system, and belongs to the basic logical link of the entire control process. The data collected in step S1 not only directly determines the accuracy of the health score, but also determines the state construction of reinforcement learning and the boundary determination of the control model. Therefore, its construction method must have high temporal resolution, high accuracy and synchronization.

[0015] In this embodiment, the UPS system is equipped with a group of multi-type sensor units, which cover the following data types: Input power ( ) and output power ( System conversion efficiency ( Total Harmonic Distortion (THD) Equipment temperature ( ) and ambient temperature ( Remaining capacity of the battery pack ( ) and charging / discharging current ( (If applicable); UPS operating frequency, voltage fluctuation rate and other auxiliary signals.

[0016] Generally, input and output power are acquired based on a combination of voltage and current sensor outputs. In one specific implementation, real-time power is calculated using the following formula: ; in, The active power at the current moment is expressed in watts (W). This is the instantaneous effective voltage, measured in volts (V), and is obtained in real time by a voltage sensor. The instantaneous effective current is measured in amperes (A) and is obtained in real time by a current sensor. This is the phase difference between the voltage and current signals, measured in radians, and can be estimated through correlation analysis of the sampled signals.

[0017] As an optional approach, to reduce power fluctuation errors, the system employs a centroid-weighted sliding window method for average power estimation. The window width is typically set to 1 second, and the sampling frequency is 2kHz. The real-time calculation of the system conversion efficiency is performed in the following form: ; in, For conversion efficiency, there are no units, and the value is between 0 and 1; This refers to the instantaneous power output of the UPS system, measured in watts (W). The instantaneous power at the input terminal is expressed in watts (W). The system has a tolerance mechanism in place, when When the value is less than the threshold, efficiency evaluation is automatically paused to avoid division by zero anomalies.

[0018] In terms of harmonic distortion detection, this embodiment uses a Fast Fourier Transform (FFT) algorithm combined with a high-frequency sampling module to extract the spectral structure of the output voltage or current, and calculates the total harmonic distortion rate accordingly. The calculation method is as follows: ; in, Total harmonic distortion (THD) is expressed as a percentage (%). This represents the amplitude of the fundamental component, expressed in volts (V). For the first The amplitude of the first harmonic component, in volts (V); The recommended harmonic cutoff order is 25. The sampling window is one complete power grid cycle (20ms or 16.7ms), and the frequency resolution is no less than 50Hz.

[0019] For temperature monitoring, a surface-mount digital temperature sensor is used, employing multi-point parallel sampling via I2C or 1-Wire protocol. The following areas are specifically monitored: inverter module heatsink, rectifier filter capacitor, core battery cells, and the top area inside the casing. Temperature data is used for temperature rise analysis in the subsequent health assessment module, with temperature drift compensation performed when necessary.

[0020] The temperature parameters mainly include the following three values: This is the real-time internal temperature of the UPS equipment, in degrees Celsius (°C). The ambient reference temperature is expressed in degrees Celsius (°C). This is the set critical temperature of the equipment, in degrees Celsius (°C). This value is generally set to 70°C or is set according to the heat resistance rating of the components.

[0021] In the health assessment, the system normalizes and calculates the temperature dimension as follows: ; in, This is a health factor in the temperature dimension, without units, and ranging from 0 to 1.

[0022] To ensure data time consistency, this embodiment introduces a timestamp synchronization mechanism. Each sensor module is connected to a unified clock source (such as a local GPS module or an edge time server), and the collected data has a timestamp accurate to the millisecond level, which is used for alignment during subsequent data fusion and modeling input.

[0023] In some embodiments, the system also introduces a weighted priority data buffering mechanism. When data transmission is abnormal or bandwidth is limited, key indicators such as power, temperature, and THD are prioritized for storage; relatively minor information (such as frequency and voltage fluctuations) can be compressed for storage or transmitted at intervals.

[0024] To improve data acquisition accuracy, the system also incorporates a multi-point calibration mechanism. During the initial deployment phase, calibration is performed using a known standard power source and voltage regulator to establish a baseline curve. Subsequently, the system automatically performs a "zero-drift calibration" process every 48 hours to eliminate baseline offsets that may be caused by temperature drift, electromagnetic interference, or interface aging.

[0025] Finally, all the collected raw data will be encoded into data frames in a unified format (such as JSON, Protobuf, etc.) and reported to the edge gateway via MQTT or CoAP protocol. The edge gateway will then perform preliminary cleaning, fusion and forwarding to provide a clear and structured data input interface for the subsequent AI inference module.

[0026] Through the above data acquisition process, the system obtains UPS operating status parameters covering multiple dimensions such as electrical, thermal, and spectrum, ensuring effective input for subsequent AI inference models, control algorithms, and health scoring mechanisms, thereby supporting the closed-loop operation of the entire intelligent monitoring and adaptive power supply control logic.

[0027] S2. Based on the collected parameters, the health status of the device is assessed using AI algorithms, and a health score is generated. In step S1, the UPS system has acquired multi-dimensional data, including input power, output power, conversion efficiency, harmonic distortion rate, and equipment temperature, through the sensing network. These parameters constitute the basic input features for equipment health status assessment. The core of step S2 is to convert these raw physical parameters into structured and quantitative equipment operating status scores, which serve as the basis for subsequent control and maintenance decisions.

[0028] To achieve this goal, this invention proposes a health scoring mechanism based on fusion indicators, and combines an AI self-learning model for dynamic evaluation and historical tracking modeling, ultimately outputting a health score value ranging from 0 to 1. .

[0029] In this embodiment, the health score calculation formula is as follows: ; in, The system health score is dimensionless and ranges from 0 to 1. A higher score indicates a more stable system. The weight coefficients for the three rating dimensions satisfy the following conditions: This value can be configured according to the device type; the recommended default values ​​are 0.4, 0.3, and 0.3. The current real-time energy conversion efficiency of the UPS system is expressed as a percentage (%), which is calculated from the input and output power in step S1. This represents the theoretical optimal efficiency of the system under normal operation, expressed as a percentage (%). It is determined based on the system design specifications and is generally taken as 92% to 95%. The real-time total harmonic distortion rate of the UPS system output voltage is expressed as a percentage (%) and is calculated using Fourier transform. This is the maximum permissible harmonic distortion rate threshold for the equipment, expressed as a percentage (%), typically set to 4%. The temperature of the critical components of the UPS at the current moment is expressed in degrees Celsius (°C) and is obtained through a temperature sensor. The current ambient temperature of the system is expressed in degrees Celsius (°C), which can be collected from the top of the cabinet or the surrounding area of ​​the equipment casing. The upper limit of the safe temperature set for the equipment, in degrees Celsius (°C), is generally taken as [value missing] depending on the component grade. C to 75℃.

[0030] Alternatively, in this embodiment, the health score result is not directly used for threshold judgment in the control logic. Instead, it is further smoothed by an exponential sliding window filter to reduce false judgments caused by sampling noise. The calculation method is as follows: ; in, This is the filtered health score value; This is a smoothing factor, ranging from 0 to 1, with a recommended setting of 0.3 to 0.5; This is the score at the current moment; This is the score from the previous period.

[0031] Normally, the score is calculated every 30 seconds; under high load or high temperature operating conditions, the cycle can be shortened to 10 seconds. If the score falls below the set threshold (e.g., 0.6) three times consecutively, the system will issue an internal health warning signal.

[0032] In one possible implementation, the system also defines a set of scoring and grading strategies, and uses the results to control logic switching and strategy reconfiguration. The grading logic is as follows: Excellent condition and stable operation; : In good condition and can operate sustainably; The condition is moderate; it is recommended to enter the observation period. The condition is abnormal and requires early intervention. Severe degradation requires immediate repair or shutdown.

[0033] To adapt to different UPS equipment models and operating scenarios, this invention further introduces a dynamic scoring correction model based on a neural network structure. This model adopts a 3-layer feedforward structure, with the input layer containing 6 nodes, namely: Real-time conversion efficiency ; Total Harmonic Distortion ; Temperature difference ; Load level labels (such as constant load, light load, non-linear load, etc.); Remaining battery capacity (if connected to DC power); Current power supply mode label (e.g., Standard, ECO, Compensation, Learning Mode); The hidden layer consists of two hidden layers, each with 16 and 8 nodes respectively, and uses ReLU activation function; the output layer has 1 node, representing the predicted corrected score. The model can run on an edge server or be inferred on a local microcontroller in the form of a quantized model.

[0034] In some embodiments, the system supports online learning, meaning that during operation, the model weights are continuously adjusted based on the user's actual scenario, and the threshold settings are updated synchronously, making the scoring more closely reflect the on-site environment. The synchronization mechanism provides dynamic feedback correction based on the deviation between the scoring results and actual fault data.

[0035] To enhance system robustness, this invention further provides a health score backtracking analysis module. This module performs a quadratic fitting based on a 24-hour historical score sequence and, combined with load change trends, calculates the health degradation rate. ; in, The rate of health decline per unit of time, expressed in minutes per hour; This is the rating from 24 hours ago; This is the current rating value.

[0036] When the rate exceeds the set threshold (e.g., 0.02 / h), the system will determine that the device is in a rapid degradation process and automatically enter the policy protection state in advance.

[0037] S3. Based on the health score, combined with the load type, equipment status and current environmental conditions, the optimal control algorithm is used to adjust the power supply mode of the UPS system; In step S2 above, the UPS system has completed a health status assessment based on multi-dimensional collected parameters and output a health score reflecting the current operational stability and risk level of the system. This score, as a crucial indicator of the quantified state, provides a fundamental input for subsequent power supply strategy optimization. In step S3, the system further combines the current load type, equipment operating status, and environmental conditions, comprehensively utilizing the optimal control algorithm to complete the adaptive switching of the power supply mode and the optimal adjustment of the output power. This step not only forms a direct data flow connection with S2 but also serves as a benchmark for the subsequent reinforcement learning strategy selection (step S4), providing support for the system to build intelligent, hierarchical power supply scheduling.

[0038] In this embodiment, the UPS system uses a sensing network composed of multiple types of sensors to collect the following key operating parameters in real time and transmit them synchronously to edge computing nodes or cloud servers: Voltage sensor: Used to detect the AC or DC voltage at the input and output terminals of the UPS, measured in volts (V). Current sensor: Used to monitor changes in the operating current at the UPS load end, measured in amperes (A). Temperature sensor: Used to monitor the temperature of the power devices, battery modules and chassis environment inside the UPS, in degrees Celsius (°C). Harmonic detection sensor: used to extract spectral distortion signals from the output waveform and calculate the total harmonic distortion rate, in percentage (%). Power quality sensors: used to evaluate indicators such as the integrity of the power supply waveform, frequency deviation, and voltage stability; Remaining battery capacity detection sensor: used to obtain the remaining state of charge (SOC) of the UPS battery, in percentage (%). Generally, the above data is uploaded via LoRaWAN protocol or industrial Ethernet, and after being locally cached, standardized and preliminarily cleaned by edge computing devices, it is finally stored in a database that supports time-series structures (such as InfluxDB or TSDB) to provide data support for subsequent health score calculation and control model solving.

[0039] In this invention, the adjustment objective of the power supply strategy can be expressed as an optimization problem, with a performance objective function... Defined as follows: ; in, The objective function value represents a weighted index of power supply efficiency and load tracking capability per unit time. For UPS systems at all times The real-time conversion efficiency is defined as No unit; This refers to the UPS output power, measured in watts (W), which is set by the system controller. The current load power requirement is expressed in watts (W) and is calculated from current and voltage samples. Let be the weight factors in the objective function, satisfying This is used to balance efficiency maximization and power stability; a recommended value is [value missing]. .

[0040] To dynamically solve this optimization objective, this invention constructs a control path based on the Pontryagin maximum principle and introduces the following state equations and Hamiltonian function structure: ; in, This is a vector of system state variables, including temperature, voltage, current, load change rate, environmental fluctuations, etc. To control the input variable vector, including output power setting, power supply mode selection bit, switching frequency, etc.; Let be a vector of costate variables, describing the impact of the future optimization objective on the current decision; The system state space matrix is ​​set based on the UPS topology and control modeling. This is the Hamiltonian function, used to construct optimality conditions; The objective is to make satisfy: ; In one possible implementation, the control system will base its decisions on health score values. Load type (identified as) Ambient temperature trend Information such as [list of information] collectively determines the current power supply strategy mode. Combining this with the preset strategy mapping relationship in the table below, the system completes mode determination and parameter distribution: model Applicable Scenarios Control Target Standard mode Normal load Maintain stable output ECO mode Light load (<30%) Reduce energy consumption and improve efficiency Compensation model Nonlinear load Harmonic control improves waveform Emergency mode Low battery / high temperature Limit output to extend battery life Learning mode New load connection Modeling, feature acquisition, optimization iteration In this embodiment, mode switching is accomplished by the controller issuing a "mode command register word." After parsing, the UPS control unit executes the corresponding control path, such as changing the inverter modulation mode, switching the output filter channel, or adjusting the charging strategy. For complex operating conditions (such as rapid load changes + critical scoring state), the system will generate a "mode suggestion score" with reference to the fuzzy control matrix and select the final mode through a voting mechanism to avoid frequent oscillation switching.

[0041] To ensure the real-time performance of control calculations, this invention employs a rolling horizontal optimization strategy, that is, every control cycle... (e.g., 0.5 seconds), re-evaluate the state variables and solve for the optimal output trajectory, applying only the first step of the control action.

[0042] Furthermore, the power supply mode and power adjustment results output in step S3 will also serve as one of the environmental feedback signals for the reinforcement learning algorithm in step S4, participating in subsequent policy value update calculations, thereby realizing continuous iteration and dynamic learning of the system policy.

[0043] S4. Dynamically select the optimal power supply mode under different load conditions through reinforcement learning algorithm; In step S3, the UPS system has combined health scores, load status, and environmental factors to achieve the strategy output under the static objective function through the optimal control algorithm. However, with changes in load type, system state drift, and operating environment, the fixed control strategy is difficult to maintain global optimum continuously. Therefore, in step S4, this invention introduces a reinforcement learning algorithm mechanism based on Q-learning to construct a dynamic power supply mode selection system to achieve long-term adaptive adjustment and performance optimization of the UPS power supply strategy.

[0044] In this embodiment, the system adopts the Q-learning discrete state reinforcement learning framework, using the running state of the UPS as the state space. Power supply mode selection as the action space It uses energy efficiency, waveform quality, and equipment health as inputs for multi-dimensional rewards, and updates the state-action value function in real time. Ultimately, the drive system automatically selects the optimal power supply mode under different load conditions.

[0045] In general, the system defines a state vector. Includes the following dimensional variables: The current load rate, expressed as a percentage (%), is defined as the ratio of current power to rated power. The current health score is given, with a value range of 0-1, and is derived from the output of step S2; The real-time operating temperature of the equipment is expressed in degrees Celsius (°C). The ambient temperature is expressed in degrees Celsius (°C). The load change rate is defined as: The unit is percentage (% / cycle); The current power supply mode number is a discrete value taken from the set. This corresponds to the five standard power supply modes: ECO, compensation, emergency, and learning.

[0046] In one possible implementation, the system discretizes the aforementioned continuous variables into intervals, such as the load rate. Divided into "low" (<30%) and "medium" (30%) The 0.2 interval is divided into five levels for discretizing state mapping. Action space. Defined as: ; in, For standard mode; ECO mode; For compensation mode; Emergency mode; This is the learning mode.

[0047] In each iteration cycle, the system starts from the current state. Choose an action This corresponds to the power supply mode switching command and executes the action. After execution, the system extracts a reward from the results. and perceive new states. The action value function is then corrected using the following Q-learning core update formula: ; in, The learning rate, ranging from 0 to 1, represents the degree of influence of newly sampled information on the Q-value update. It is recommended to initially set it to 0.2. This is a discount factor, ranging from 0 to 1, reflecting the degree of contribution to future rewards; a value of 0.9 is recommended. This represents the maximum expected value in subsequent states. In the state Take action below The current Q value.

[0048] reward function The weighted combination function for multi-objective evaluation is expressed as follows: ; in, The weight coefficient for the reward item satisfies... It is recommended to set the default values ​​to 0.4, 0.3, and 0.3. For efficiency scoring items, the definition is: ; in, The system's current real-time conversion efficiency is expressed as a percentage (%). The theoretical optimal efficiency set for the system, expressed as a percentage (%), typically 92%-95%.

[0049] The harmonic control scoring item is defined as follows: ; in, The total harmonic distortion rate of the system output at the current moment is expressed as a percentage (%). The maximum THD threshold allowed by the system, expressed as a percentage (%), typically 4%.

[0050] The temperature rating item is defined as follows: ; in, This represents the current operating temperature of the core components of the equipment, expressed in degrees Celsius (°C). The current ambient temperature is expressed in degrees Celsius (°C). This is the set temperature threshold, in degrees Celsius (°C), typically set to 70°C~75°C.

[0051] In actual operation, to balance the exploration and utilization strategies, this embodiment adopts... - A greedy strategy is used to select actions. Specifically: With probability Select the power supply mode with the highest current Q value; With probability Randomly select other actions for strategy exploration; The initial value is set to 0.3, and it gradually decreases to 0.05 as the running time increases, using a linear or exponential decay function for adjustment.

[0052] As an option, the system also supports persistent storage of the Q-table every 12 hours to avoid losing learning results in case of power failure; at the same time, it is used in conjunction with the "experience replay pool" to store the most recent 1000 state-action-reward triplets for offline retraining.

[0053] In some embodiments, to improve convergence efficiency, this invention introduces a strategy pruning mechanism. If a power supply mode fails to obtain a positive reward after being executed 5 times consecutively in a certain state, the system will mark that state-action path as "inefficient" and reduce its subsequent selection probability.

[0054] Furthermore, to further enhance generalization ability, the system can be extended to a Q-network based on function approximation (such as DQN), where: The state input dimension can be increased to more than 8 dimensions, supporting continuous variable modeling; The network structure adopts a two-layer fully connected feedforward neural network, with 32 and 16 nodes in each layer, respectively; The activation function is ReLU, and the output is a Q-value vector of action space size (5); The network update uses gradient descent to optimize the loss function. ; in, These are the current main network parameters; The target network parameters are updated and synchronized at fixed intervals.

[0055] S5. Real-time adjustment of power supply mode, and feedback of adjustment results to control system for closed-loop optimization; In step S4, the system has constructed a strategy mapping between the UPS power supply mode and the operating state using the Q-learning algorithm, and outputs the optimal power supply mode decision result under the current state. In order to transmit this strategy from the theoretical layer to the physical layer, step S5 is required to complete the issuance of strategy instructions, adjustment of power paths, and real-time response of equipment behavior. Based on the feedback of the execution effect, the effectiveness of the strategy is verified and the model is corrected, thereby realizing the closed-loop self-optimization of the control system.

[0056] In this embodiment, the execution of the power supply mode is completed by the execution control module set in the control system. The control module is equipped with a multi-level control buffer structure, a decision instruction buffer area and a feedback channel mapping table to support the synchronous execution of multi-mode switching and state return flow.

[0057] Generally, switching power supply modes involves two types of commands: Static mode switching commands, such as standard mode (stable output), ECO mode (light load), compensation mode (high harmonics), emergency mode (low battery), and learning mode (new load introduced). Dynamic power parameter commands, such as output power setting Bus voltage regulation Upper limit of charging and discharging current Parallel control coefficient wait.

[0058] In one possible implementation, the power supply mode control signal is encapsulated in the form of "mode + parameter packet", and the distribution structure is as follows: ; in, This is a pattern instruction structure; The mode number is an integer that represents the currently selected mode, such as 1 for standard, 2 for ECO, etc. This is the output power setting, in watts (W), with a recommended accuracy of 0.1%. This is the reference value for the DC bus voltage of the inverter bridge, in volts (V). This is the maximum battery discharge current, measured in amperes (A). This is the parallel control coefficient, used for adjusting the synchronization ratio when coordinating multiple UPS units. It is dimensionless and ranges from 0 to 1. This is the delay period for mode switching (i.e., the time for the policy to take effect), measured in milliseconds (ms). It is used to buffer hard switching jitter and is typically set to 200-500ms.

[0059] Alternatively, command execution is accomplished through a digital controller (such as a DSP, ARM Cortex core MCU, etc.). The controller communicates with the power module via CAN-FD, Modbus RTU, or SPI protocols, and performs command polling and status feedback every cycle.

[0060] In practical applications, after the system completes the execution of the power supply mode, it will collect the following real-time information through the feedback path to form a feedback evaluation data set: Actual output power The unit is watts (W); actual conversion efficiency The unit is percentage (%), from Calculate the actual total harmonic distortion rate. The unit is percentage (%); actual temperature rise status. The unit is degrees Celsius (°C).

[0061] In this invention, a feedback error function is introduced to quantify the degree of deviation of the power supply strategy at the execution layer. The definition is as follows: ; in, This is the power supply execution error function, expressed in composite index values. Let be the error weighting coefficient, satisfying The recommended values ​​are 0.4, 0.3, and 0.3. The target output power is set in watts (W). The actual measured output power is expressed in watts (W). For reference conversion efficiency, the unit is percentage (%), derived from the strategy template; To measure conversion efficiency in real time, the unit is percentage (%); For reference harmonic targets, the unit is percentage (%), corresponding to the mode; This represents the actual harmonic distortion rate, expressed as a percentage (%).

[0062] Feedback error Once the set tolerance threshold is exceeded The system will automatically mark this execution strategy as "non-ideal execution" and trigger the following response chain: If two consecutive non-ideal executions occur, the sampling of the current action in the Q table will be temporarily frozen; If it occurs three times in a row, the "waitlist mode" judgment logic will be automatically triggered to select an alternative strategy from the waitlist strategy pool. Simultaneously, the error value and mode number are recorded in the feedback history table and used for Q-value penalty calculation.

[0063] In some embodiments, the system constructs a feedback stability function. This is used to determine the immunity of the current power supply mode under dynamic load, and is defined as follows: ; in, This is a stability index; the smaller the value, the more stable the output. The unit is W / s. The evaluation period is measured in seconds (s), typically between 5 and 10 seconds. This represents the rate of change of output power, expressed in W / s.

[0064] As an option, after each power mode switch is completed and feedback data is received, the system will use the current power mode as the basis for its operation. , and rate of change in health scores As input, construct evaluation triples: ; This is used to evaluate the effectiveness of the current power supply strategy and serves as a reward feedback source for the reinforcement learning module (see step S4). Wherein: This indicates the trend of health score changes; like This indicates that the device status deteriorated after the policy was executed, and policy rollback or switching should be considered. like If this occurs, then this execution is recorded as positive policy feedback.

[0065] To improve feedback speed and response efficiency, this embodiment deploys the feedback computing module on edge nodes, keeping feedback latency below 200ms. All feedback data is recorded in a "Time-Series Feedback Table," which can be used for strategy tracing, strategy version comparison, and historical tracking of power supply strategy optimization.

[0066] Please see Figure 2 The UPS system intelligent monitoring and adaptive power supply system based on IoT and AI includes: The sensor acquisition module contains various sensors used to acquire key operating parameters of the UPS (Uninterruptible Power Supply) equipment in real time. Specific sensors include, but are not limited to, voltage sensors, current sensors, temperature sensors, harmonic sensors, power quality sensors, and battery capacity sensors. These sensors accurately monitor the equipment's operating status, providing reliable foundational data for subsequent health assessments and power supply strategy optimization. To ensure compatibility and effective fusion of data from multiple sensors, all sensor-acquired data is encoded using a unified data format. Before transmission to the central processing unit, the data undergoes preliminary filtering, noise reduction, and data fusion processing at the edge computing device. These processing steps significantly improve data accuracy and response speed, reduce the risk of misjudgments due to noise or sensor errors, and ensure data transmission stability in high-frequency changing environments. The IoT sensing network module enables real-time communication between sensor-collected data and the central processing unit. Whether via wireless (e.g., Wi-Fi, Zigbee, LoRa) or wired (e.g., Ethernet, RS485), the sensing network module ensures timely data transmission and system-wide information synchronization. This module's efficient communication capabilities allow the system to respond promptly to environmental changes, ensuring the stability of the UPS equipment under load fluctuations or sudden emergencies. The health assessment module is responsible for evaluating the health status of the UPS equipment based on real-time data collected from various sensors. By analyzing key parameters such as voltage, current, temperature, and battery capacity, this module can monitor the UPS's operational health status in real time and identify potential faults or efficiency degradation issues. The health scoring model comprehensively considers this real-time data and, through set thresholds and rules, provides the system with an assessment of the current equipment's health status, helping the system adjust its operating strategies in a timely manner. The adaptive power supply decision module automatically adjusts the UPS power supply strategy based on health assessment results and optimization algorithms. This module dynamically selects the optimal power supply mode (such as standard mode, energy-saving mode, standby mode, etc.) according to the UPS's current operating environment, load demand, and health score to ensure efficient equipment operation. By considering load changes, efficiency optimization, and equipment lifespan balance, this module can significantly improve the overall performance of the UPS system. The reinforcement learning optimization module is the core of the UPS system's adaptive power supply decision-making. This module continuously optimizes the power supply strategy based on historical data and real-time feedback using reinforcement learning algorithms. The system learns the optimal power supply strategy through interaction with the environment (e.g., load fluctuations, fault detection) and gradually adjusts and updates the decision rules. The reinforcement learning algorithm optimizes the system's control strategy through a trial-and-error process, continuously learning and adjusting to maximize the efficiency and reliability of the UPS system. The execution and feedback module is the execution unit in the system, responsible for adjusting the UPS's operating mode according to the instructions from the decision-making module. During execution, this module monitors the UPS equipment's operating status in real time and sends the results back to the sensor acquisition module. This closed-loop feedback mechanism ensures real-time correction and optimization of the power supply strategy. If the system detects abnormal equipment status or substandard strategy execution, the feedback module will immediately activate the adjustment mechanism, restoring normal equipment operation by updating the control strategy or changing the power supply mode.

[0067] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent monitoring and adaptive power supply of UPS systems based on IoT and AI, characterized in that, Includes the following steps: S1. Real-time data collection of input power, output power, conversion efficiency, harmonic distortion rate, and equipment temperature parameters of the uninterruptible power supply system via an Internet of Things (IoT) sensing network. S2. Based on the collected parameters, the health status of the device is assessed using AI algorithms, and a health score is generated. S3. Based on the health score, combined with the load type, equipment status and current environmental conditions, the optimal control algorithm is used to adjust the power supply mode of the UPS system; S4. Dynamically select the optimal power supply mode under different load conditions through reinforcement learning algorithm; S5. Real-time adjustment of power supply mode, and feedback of adjustment results to control system for closed-loop optimization.

2. The intelligent monitoring and adaptive power supply method for UPS systems based on IoT and AI according to claim 1, characterized in that, The health rating assessment includes calculating the health value of the equipment based on the difference between real-time conversion efficiency and optimal efficiency, the ratio of real-time harmonic distortion rate to the maximum allowable value, and the ratio of real-time equipment temperature to critical temperature.

3. The intelligent monitoring and adaptive power supply method for UPS systems based on IoT and AI according to claim 1, characterized in that, The health score is calculated using the following formula: ； in, Rate your health. These are the weighting coefficients. For real-time conversion efficiency, For optimal efficiency. For real-time harmonic distortion rate, The maximum harmonic distortion rate, For the real-time temperature of the equipment, For ambient temperature, This refers to the critical temperature of the equipment.

4. The intelligent monitoring and adaptive power supply method for UPS systems based on IoT and AI according to claim 1, characterized in that, The step of adjusting the power supply mode of the UPS system using the optimal control algorithm is to dynamically adjust the output power by using the optimal control algorithm to maximize the ratio between the output power and the input power.

5. The intelligent monitoring and adaptive power supply method for UPS systems based on IoT and AI according to claim 1, characterized in that, The optimal control algorithm is based on the Pontryagin maximum principle and adjusts the control input through real-time state updates and feedback to enable the system to reach the optimal state during operation.

6. The intelligent monitoring and adaptive power supply method for UPS systems based on IoT and AI according to claim 1, characterized in that, The UPS system collects the following parameters in real time through a sensing network composed of multiple types of sensors and transmits them to edge computing nodes or cloud servers for processing: Voltage sensors are used to detect the voltage at the input and output terminals of the UPS; Current sensor, used to monitor changes in load current; Temperature sensors are used to monitor the temperature of UPS circuit components and the ambient temperature; Harmonic detection sensors are used to measure the total harmonic distortion rate in output current or voltage. Power quality sensors are used to assess the integrity and stability of power waveforms. The remaining battery capacity detection sensor is used to acquire real-time battery power data; Data is transmitted via LoRaWAN or Ethernet protocols and processed and stored by edge computing units or cloud time-series databases for subsequent health assessment and AI decision-making modules.

7. The intelligent monitoring and adaptive power supply method for UPS systems based on IoT and AI according to claim 1, characterized in that, The reinforcement learning algorithm selects the optimal power supply mode in real time through the Q-learning algorithm, and the Q value is updated based on the feedback results of the system state and the power supply mode.

8. The intelligent monitoring and adaptive power supply method for UPS systems based on IoT and AI according to claim 7, characterized in that, The state space of the Q-learning algorithm includes load type, device health, temperature, and current power supply mode. The reward function is evaluated based on energy efficiency, harmonic distortion rate, and device lifespan.

9. A UPS system intelligent monitoring and adaptive power supply system based on IoT and AI, applied to the UPS system intelligent monitoring and adaptive power supply method based on IoT and AI as described in any one of claims 1-8, characterized in that, include: The sensor acquisition module includes multiple types of sensors such as voltage, current, temperature, harmonics, power quality, and battery capacity, used to acquire UPS operating parameters in real time; The Internet of Things (IoT) sensing network module is used to transmit data collected by sensors to the central processing unit via wireless or wired communication. The health assessment module is used to assess the status of UPS equipment based on real-time data; An adaptive power supply decision module is used to adjust the power supply strategy based on optimization algorithms and health assessment results; The reinforcement learning optimization module is used to optimize the power supply strategy online through reinforcement learning algorithms. The execution and feedback module is used to control the power supply mode and send feedback data back to the sensor acquisition module, forming a closed-loop regulation mechanism.

10. The UPS system intelligent monitoring and adaptive power supply system based on IoT and AI according to claim 9, characterized in that, The various sensors in the sensor acquisition module are encoded using a unified data format, and preliminary filtering, noise reduction, and data fusion processing are performed in the edge computing device to improve data accuracy and response rate.

Images