Intelligent early warning method and system for heat dissipation of ultraviolet LED lamp driven by internet of things data

Abstract

The application provides an Internet of Things data-driven ultraviolet LED lamp heat dissipation intelligent early warning method and system, relates to the technical field of heat dissipation early warning, and the method comprises the following steps: collecting multi-source heterogeneous time sequence data flow; determining the length of a sliding window; in a defined time sequence data interval, jointly detecting anomalies by using an improved isolated forest algorithm; taking a thermal anomaly feature set as an input sequence, and using a long short-term memory network model to perform rolling prediction on the temperature change trend of the lamp tube; in combination with the current running time and cumulative service history of the ultraviolet LED lamp, dynamically calculating an early warning trigger threshold, and performing early warning trigger judgment on the temperature prediction value sequence. The application solves the technical problem that the heat dissipation performance is implicitly degraded due to ultraviolet radiation in the long-term service process, causing a thermal failure risk and affecting the long-term operation reliability of the ultraviolet LED lamp, and accurately issues an early warning before the thermal failure occurs, thereby improving the long-term operation reliability of the ultraviolet LED lamp.

Description

Technical Field

[0001] This application relates to the field of heat dissipation early warning technology, specifically to an intelligent early warning method and system for heat dissipation of ultraviolet LED lamps driven by IoT data. Background Technology

[0002] Currently, heat dissipation management for UV LED lamps mainly employs physical heat dissipation solutions such as air cooling, liquid cooling, or heat pipes, supplemented by thermistors for single-point temperature monitoring and fixed temperature thresholds for over-limit alarms. However, with the accumulation of service time, the internal material system of UV LED lamps undergoes irreversible and gradual degradation under long-term UV irradiation. This degradation accumulates invisibly through a slow increase in thermal resistance and a gradual decline in the efficiency of the heat conduction path, leading to a continuous deterioration of the lamp's heat dissipation performance without being noticed. Existing temperature monitoring methods based on fixed thresholds cannot detect the invisible degradation trend of heat dissipation performance. The fixed thresholds cannot adaptively adjust with changes in the lamp's service age and heat dissipation status. As a result, even when the actual chip junction temperature has quietly climbed into the dangerous range in the middle and late stages of the equipment's lifespan, the monitoring system still judges it as normal according to the standards of a new lamp, leading to missed thermal failures. Temperature monitoring cannot distinguish between temperature anomalies caused by substantial deterioration of the heat dissipation system and transient interference caused by fluctuations in the drive circuit or environmental disturbances, resulting in low reliability of warnings and frequent false alarms.

[0003] In summary, existing technologies suffer from the risk of thermal failure due to the latent degradation of heat dissipation performance caused by ultraviolet irradiation during long-term service, which further affects the long-term operational reliability of ultraviolet LED lamps. Summary of the Invention

[0004] The purpose of this application is to provide an IoT data-driven intelligent early warning method and system for heat dissipation of ultraviolet LED lamps, in order to solve the technical problem in the prior art where heat dissipation performance is implicitly degraded due to ultraviolet irradiation during long-term service, causing the risk of thermal failure and further affecting the long-term operational reliability of ultraviolet LED lamps.

[0005] To achieve the above objectives, this application provides an IoT data-driven intelligent early warning method and system for heat dissipation of ultraviolet LED lamps.

[0006] In a first aspect, this application provides an IoT data-driven intelligent early warning method for heat dissipation of ultraviolet LED lamps. This method is implemented through an IoT data-driven intelligent early warning system for heat dissipation of ultraviolet LED lamps. The method includes: constructing a multi-dimensional sensor network covering the entire thermal field of the lamp tube; collecting multi-source heterogeneous time-series data streams containing lamp tube surface temperature, driving current value, and ambient humidity value; and determining the length of a sliding window based on the dynamic change amplitude of the current operating condition of the ultraviolet LED lamp according to the multi-source heterogeneous time-series data streams. Within the time-series data interval defined by the sliding window, an improved isolated forest algorithm is used to perform joint anomaly detection on the multi-source heterogeneous data within the window, extracting a set of thermal anomaly features reflecting abnormal heat dissipation status. Using the set of thermal anomaly features as the input sequence, a pre-trained long short-term memory network model is used to perform rolling prediction of the lamp temperature change trend over multiple future time steps, generating a sequence of predicted temperature values ​​over multiple future time steps. Combining the current operating time and cumulative service history of the ultraviolet LED lamp, a warning trigger threshold adapted to the current operating conditions is dynamically calculated, and a warning trigger judgment is made on the temperature prediction value sequence.

[0007] Optionally, the multidimensional sensing network includes temperature sensors, current sensors, and ambient humidity sensors deployed on the surface of the ultraviolet LED tube body, in the preset air region of the tube, and at the location of the lamp driver circuit board.

[0008] Optionally, an initial sliding window is obtained; based on the multi-source heterogeneous time-series data stream, the degree of temperature change drasticness of the temperature value sequence within the initial sliding window is calculated; when the degree of temperature change drasticness exceeds a preset first degree of drasticness threshold, the time domain length of the sliding window is set to a first time domain length value; when the degree of temperature change drasticness is lower than a preset second degree of drasticness threshold, the time domain length of the sliding window is set to a second time domain length value greater than the first time domain length value; when the degree of temperature change drasticness index is between the first degree of drasticness threshold and the second degree of drasticness threshold, the time domain length of the initial sliding window remains unchanged.

[0009] Optionally, the average of the absolute values ​​of the first-order differences of the temperature value sequence within the initial sliding window can be used as a quantitative indicator of the severity of temperature changes.

[0010] Optionally, the temperature, current, and humidity data within the time-series data interval defined by the sliding window are arranged in chronological order to obtain a multi-dimensional time-series data matrix. For each sampling point in the multi-dimensional time-series data matrix, a time weight coefficient is calculated based on the time interval between the sampling point's timestamp and the current time. The time weight coefficients of each sampling point are embedded into the path length calculation process of the isolated forest algorithm to obtain an initial thermal anomaly feature set. The temperature and current data are input into a preset physical coupling discriminator, which removes interference events from the initial thermal anomaly feature set based on the ratio between the change in current value and the change in temperature value, thus obtaining the thermal anomaly feature set.

[0011] Optionally, the physical coupling discriminator determines whether the current temperature anomaly is accompanied by a current anomaly of the same origin based on the ratio between the change in current value and the change in temperature value; if the temperature anomaly is accompanied by a current anomaly of the same origin, the corresponding temperature anomaly is marked as an interference event caused by a drive circuit anomaly and deleted from the initial thermal anomaly feature set.

[0012] Optionally, each isolated tree of the isolated forest is constructed based on the multidimensional time-series data matrix; after calculating the original path length of each sampling point in each isolated tree of the isolated forest, the original path length is weighted and corrected using the time weight coefficient of the sampling point as a multiplicative adjustment factor; anomaly identification is performed based on each corrected isolated tree to obtain the initial thermal anomaly feature set.

[0013] Optionally, the thermal anomaly feature set is used as input samples, and after normalization, they are arranged into a fixed-length input tensor in chronological order. The input tensor is then fed into a long short-term memory network model for forward computation, and the predicted temperature values ​​for multiple future time steps are output sequentially to form the temperature prediction value sequence.

[0014] Optionally, the initial heat dissipation performance of the UV LED lamp at the time of manufacture is used as the benchmark threshold; historical heat dissipation characteristics are collected and the heat dissipation performance decay curve with respect to the cumulative running time is analyzed; based on the heat dissipation performance decay curve, the benchmark threshold is corrected for aging compensation with the current cumulative running time, and at the same time, the environmental impact is corrected by matching the current ambient humidity value with an environmental correction coefficient to obtain the warning trigger threshold.

[0015] Secondly, this application also provides an IoT data-driven intelligent early warning system for heat dissipation of ultraviolet LED lamps, used to execute the IoT data-driven intelligent early warning method for heat dissipation of ultraviolet LED lamps as described in the first aspect. The IoT data-driven intelligent early warning system for heat dissipation of ultraviolet LED lamps includes: a time-series data acquisition module, used to construct a multi-dimensional sensor network covering the entire thermal field of the lamp tube, and to acquire multi-source heterogeneous time-series data streams containing lamp tube surface temperature, driving current value, and ambient humidity value; a sliding window determination module, used to determine the length of the sliding window based on the multi-source heterogeneous time-series data stream and the dynamic change amplitude of the current operating condition of the ultraviolet LED lamp; and a joint anomaly detection module, used to... Within the time-series data interval defined by the sliding window, an improved isolated forest algorithm is used to perform joint anomaly detection on the multi-source heterogeneous data within the window, extracting a set of thermal anomaly features reflecting abnormal heat dissipation status. A rolling temperature prediction module uses the thermal anomaly feature set as input sequence and a pre-trained long short-term memory network model to perform rolling predictions of lamp temperature change trends over multiple future time steps, generating a sequence of predicted temperature values ​​for those multiple time steps. A warning trigger judgment module combines the current operating time and cumulative service history of the UV LED lamp to dynamically calculate a warning trigger threshold adapted to the current operating conditions, and performs warning trigger judgment on the temperature prediction value sequence.

[0016] One or more technical solutions provided in this application have at least the following technical effects or advantages: By constructing a multi-dimensional sensing network covering the entire thermal field of the UV LED lamp, a multi-source heterogeneous time-series data stream containing lamp surface temperature, driving current value, and ambient humidity value is collected. Based on the multi-source heterogeneous time-series data stream, the length of the sliding window is determined according to the dynamic change amplitude of the current operating condition of the UV LED lamp. Within the time-series data interval defined by the sliding window, an improved isolated forest algorithm is used to perform joint anomaly detection on the multi-source heterogeneous data within the window, extracting a set of thermal anomaly features reflecting abnormal heat dissipation status. Using the set of thermal anomaly features as the input sequence, a pre-trained long short-term memory network model is used to perform rolling prediction of the lamp temperature change trend in multiple future time steps, generating a temperature prediction value sequence for multiple future time steps. Combining the current operating time and cumulative service history of the UV LED lamp, a warning trigger threshold adapted to the current operating condition is dynamically calculated, and a warning trigger judgment is made on the temperature prediction value sequence. In other words, by constructing a multi-dimensional sensor network to collect multi-source data and adaptively adjusting the sliding window length according to dynamic changes in operating conditions to match the analysis granularity, pure thermal anomaly features are accurately extracted, driving the long short-term memory network model to perform multi-step rolling predictions. Finally, by combining the equipment aging and decay curve with the environmental humidity coefficient, the early warning trigger threshold is dynamically calculated, and an early warning is accurately issued before thermal failure occurs, fundamentally ensuring the operational reliability of the ultraviolet LED lamp throughout its entire life cycle.

[0017] The above description is merely an overview of the technical solution of this application. To better understand the technical means of this application and to facilitate its implementation according to the description, and to make the above and other objects, features, and advantages of this application more apparent, specific embodiments of this application are described below. It should be understood that the content described in this section is not intended to identify key or important features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent through the following description. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in 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 merely exemplary. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the IoT data-driven intelligent early warning method for heat dissipation of ultraviolet LED lamps, as described in this application.

[0020] Figure 2 This is a schematic diagram of the structure of the IoT data-driven intelligent early warning system for heat dissipation of ultraviolet LED lamps in this application.

[0021] Figure labeling: 11 Time series data acquisition module, 12 Sliding window determination module, 13 Joint anomaly detection module, 14 Temperature rolling prediction module, 15 Early warning trigger judgment module. Detailed Implementation

[0022] This application provides an IoT-driven intelligent early warning method and system for ultraviolet (UV) LED lamp heat dissipation. It addresses the technical problem in existing technologies where UV irradiation during long-term service leads to subtle degradation of heat dissipation performance, causing thermal failure risks and further impacting the long-term operational reliability of UV LED lamps. By constructing a multi-dimensional sensor network to collect multi-source data and adaptively adjusting the sliding window length according to dynamic changes in operating conditions to match the analysis granularity, it accurately extracts pure thermal anomaly features. This drives a long short-term memory (LSTM) network model to perform multi-step rolling predictions. Finally, by combining the equipment aging degradation curve and the environmental humidity coefficient, it dynamically calculates the early warning trigger threshold, accurately issuing warnings before thermal failure occurs, fundamentally ensuring the operational reliability of UV LED lamps throughout their entire lifecycle.

[0023] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. It should be understood that this application is not limited to the exemplary embodiments described herein. 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. It should also be noted that, for ease of description, only the parts related to this application are shown in the accompanying drawings, not all of them.

[0024] Example 1, please refer to the appendix. Figure 1 This application provides an IoT data-driven intelligent early warning method for heat dissipation of ultraviolet LED lamps. The IoT data-driven intelligent early warning method for heat dissipation of ultraviolet LED lamps is applied to an IoT data-driven intelligent early warning system for heat dissipation of ultraviolet LED lamps. The specific steps of the IoT data-driven intelligent early warning method for heat dissipation of ultraviolet LED lamps are as follows: A multi-dimensional sensor network covering the entire thermal field of the lamp tube is constructed to collect multi-source heterogeneous time-series data streams containing lamp tube surface temperature, drive current value and ambient humidity value.

[0025] Furthermore, this application also includes the following steps: the multidimensional sensing network includes temperature sensors, current sensors, and ambient humidity sensors deployed on the surface of the ultraviolet LED tube body, in the preset air domain of the tube, and at the location of the lamp driver circuit board.

[0026] Specifically, based on the physical structure and heat dissipation path of the UV LED lamp, the sensor deployment locations are determined, including temperature sensors, current sensors, and ambient humidity sensors. Three surface-mount K-type thermocouples are tightly bonded to the metal surface using high thermal conductivity silicone grease at the central area on the back of the lamp housing, the sidewall near the light-emitting surface of the lamp, and the edge of the contact surface between the heat sink base and the lamp. Temperature sensors are then secured with high-temperature resistant tape to ensure good thermal contact between the thermocouple's sensing end and the metal surface, thereby accurately collecting the housing temperature at different locations.

[0027] An integrated temperature and humidity sensor is fixed in place using a bracket at a position 10cm away from the surface of the lamp tube housing, at a 45° angle above the lamp tube and avoiding the direct path of the light emitted from the lamp tube. It measures both the ambient temperature and relative humidity, but the temperature data is mainly used for temperature compensation of the humidity.

[0028] On the positive power supply bus of the drive circuit, the primary wire of the Hall effect current sensor is passed through its central hole, so that the entire drive current passes through the sensor. The secondary output of the sensor is a voltage signal proportional to the current. At the same time, a PT100 platinum resistance temperature sensor is mounted on the circuit board near the current sampling resistor but far away from the high-power switching transistor to monitor the local temperature of the circuit board.

[0029] A multidimensional sensor network is a three-dimensional sensing system composed of multiple physical quantity sensors (temperature, current, humidity) working together at multiple spatial locations. The multidimensionality is manifested in the multi-dimensional sensing of physical quantities (heat, electricity, environment) and the multi-dimensional spatial layout (lamp body, air space, circuit board). The lamp body surface is the outer shell surface of the UV LED lamp tube, typically including a metal heat sink or the exposed surface of a ceramic substrate. This location is closest to the internal chip, and its temperature is the most direct external representation of the chip's junction temperature. The lamp's preset air space is the surrounding air space region at a certain distance from the lamp surface, such as 5cm to 15cm, used to collect environmental thermal state and air humidity. This location is not excessively affected by direct heat radiation from the lamp tube but still reflects the true state of the microenvironment in which the lamp is located. The lamp driver circuit board is an electronic circuit board that provides a stable driving current for the UV LED chips. It integrates power switching transistors, transformers, rectifier filters, and current sampling resistors, among other heat-generating components. It is both a heat source and a direct measurement point for electrical parameters. Temperature sensors are devices that convert the physical quantity of temperature into a measurable electrical signal, specifically referring to surface-mount thermocouples or NTC thermistors. Current sensors are devices that convert the current flowing through a drive circuit into a measurable electrical signal, specifically referring to Hall effect or resistance sampling current sensors. Ambient humidity sensors are devices used to measure the relative humidity of the surrounding air, typically using capacitive or resistive humidity-sensitive elements.

[0030] All sensor signal output lines use shielded twisted-pair cables, converging to a single multi-channel data acquisition terminal. This terminal contains multiple independent analog-to-digital converters to synchronously sample the weak voltage signals from thermocouples, the voltage signals from current sensors, and the digital or analog signals from humidity sensors. The microcontroller in the data acquisition terminal converts the sampled values ​​from each channel into engineering quantities—temperature, current, and humidity—according to their respective calibration curves, and adds a millisecond-level timestamp to each data set using a high-precision real-time clock. The converted data is then encapsulated into structured data frames according to a preset communication protocol and uploaded to an edge computing server via an RS485 bus at a frequency of once per second, forming a continuously arriving, multi-source heterogeneous time-series data stream aligned in both time and spatial dimensions.

[0031] Based on the multi-source heterogeneous time-series data stream, the length of the sliding window is determined according to the dynamic change range of the current operating conditions of the ultraviolet LED lamp.

[0032] Furthermore, this application also includes the following steps: obtaining an initial sliding window; calculating the degree of temperature change drasticness of the temperature value sequence within the initial sliding window based on the multi-source heterogeneous time-series data stream; when the degree of temperature change drasticness exceeds a preset first drasticness threshold, setting the time domain length of the sliding window to a first time domain length value; when the degree of temperature change drasticness is lower than a preset second drasticness threshold, setting the time domain length of the sliding window to a second time domain length value greater than the first time domain length value; when the degree of temperature change drasticness index is between the first drasticness threshold and the second drasticness threshold, keeping the time domain length of the initial sliding window unchanged.

[0033] Furthermore, this application also includes the following steps: using the average of the absolute values ​​of the first-order differences of the temperature value sequence within the initial sliding window as a quantitative indicator of the severity of temperature changes.

[0034] Specifically, based on the drastic changes in heat load or heat dissipation conditions during the operation of the UV LED lamp, a default initial sliding window length, such as 300 seconds, is set. Real-time data is continuously received. Whenever a new data frame arrives, all temperature values ​​prior to the current moment, with a time domain length equal to the current window length, are extracted from the multi-source heterogeneous time-series data stream, forming a temperature value sequence. The first-order difference is calculated point-by-point on the temperature value sequence according to the sampling time sequence; that is, each temperature value is subtracted from the previous temperature value, resulting in a difference sequence. Since the temperature may rise or fall, the difference values ​​can be positive or negative. The arithmetic mean of the absolute values ​​of all difference values ​​is then calculated, which is the quantitative indicator of the drastic temperature change.

[0035] The quantification of the drastic temperature change is compared with two pre-stored threshold values ​​in a configuration file. The first threshold is higher, and the second threshold is lower, with the first threshold being higher than the second. If the quantification of the drastic temperature change exceeds the first threshold, it indicates that the temperature is fluctuating rapidly. The sliding window's time domain length is switched from its current value to a shorter first time domain length, such as 120s, to quickly track temperature changes and avoid response lag due to an excessively long window. If the quantification of the drastic temperature change is lower than the second threshold, it indicates that the temperature is in a stable state. The window length is switched to a longer second time domain length, such as 600s, to incorporate more historical data and improve the statistical stability and noise resistance of subsequent anomaly detection. If the quantification of the drastic temperature change is exactly between the first and second thresholds, it indicates that the temperature is in a moderately fluctuating state. The current window length is maintained, and no switching operation is performed. The above calculation, comparison, and switching logic is repeated in each sampling period, so that the length of the sliding window is dynamically adjusted in real time as the temperature changes, always matching the dynamic change range of the current operating condition.

[0036] For example, assume the sliding window length is set to 300 seconds. During a certain continuous monitoring period, the system reads the temperature sequence within the current window from the database, in °C, with a sampling interval of 1 second, listing only representative data points: 62.3, 62.5, 62.7, 62.4, 61.8, 61.2, 60.5, 61.0, 62.3, 63.1, 64.2, 65.5, 66.0, 65.8, 65.2. The first-order difference is calculated point by point, i.e., subtracting the previous value from the next value, resulting in the difference sequence (in °C): +0.2, +0.2, -0.3, -0.6, -0.6, -0.7, +0.5, +1.3, +0.8, +1.1, +1.3, +0.5, -0.2, -0.6. Taking the absolute values ​​yields 0.2, 0.2, 0.3, 0.6, 0.6, 0.7, 0.5, 1.3, 0.8, 1.1, 1.3, 0.5, 0.2, and 0.6. The arithmetic mean of these absolute values ​​is 0.64℃ / s. Assume a pre-defined first severity threshold of 0.5℃ / s and a second severity threshold of 0.15℃ / s. Since the current calculated severity (0.64 > 0.5) exceeds the first threshold, the temperature is determined to be in a state of drastic fluctuation, and the sliding window length is immediately switched from 300s to the first time domain length value of 120s. After the switch is completed, the data within the 120s window is used to continue the calculation in the next sampling period. If the subsequent temperature sequence tends to be stable, such as the temperature values ​​within the window becoming 66.5, 66.4, 66.3, 66.5, 66.6, 66.4, 66.5, 66.3, 66.4, 66.5, with the absolute values ​​of the first difference being 0.1, 0.1, 0.2, 0.1, 0.2, 0.1, 0.2, 0.1, 0.1, and the average value being 0.13℃ / s, which is lower than the second threshold of 0.15℃ / s, the window length is switched from 120s back to the second time domain length value of 600s to accumulate more historical data of stable states for subsequent analysis.

[0037] The sliding window length is dynamically and adaptively adjusted based on the degree of temperature change, achieving automatic optimization and matching of the analysis granularity. When the temperature changes drastically, the short window ensures that the system can respond quickly and avoids missed detections caused by anomaly signals being submerged in historical data due to an excessively long window. When the temperature is stable, the long window provides a sufficient amount of statistical sample, enhancing the noise resistance and statistical significance of subsequent anomaly detection algorithms.

[0038] Within the time-series data interval defined by the sliding window, an improved isolated forest algorithm is used to perform joint anomaly detection on the multi-source heterogeneous data within the window, and extract a set of thermal anomaly features reflecting abnormal heat dissipation status.

[0039] Furthermore, this application also includes the following steps: arranging the temperature data, current data, and humidity data within the time-series data interval defined by the sliding window in chronological order to obtain a multi-dimensional time-series data matrix; calculating a time weight coefficient for each sampling point in the multi-dimensional time-series data matrix based on the time interval between the sampling point's timestamp and the current time; embedding the time weight coefficients of each sampling point into the path length calculation process of the isolated forest algorithm to obtain an initial thermal anomaly feature set; inputting the temperature data and current data into a preset physical coupling discriminator, which then performs interference event deletion on the initial thermal anomaly feature set based on the ratio between the change amplitude of the current value and the change amplitude of the temperature value, thereby obtaining the thermal anomaly feature set.

[0040] Furthermore, this application also includes the following steps: the physical coupling discriminator determines whether the current temperature anomaly is accompanied by a current anomaly of the same origin based on the ratio between the change in current value and the change in temperature value; if the temperature anomaly is accompanied by a current anomaly of the same origin, the corresponding temperature anomaly is marked as an interference event caused by a drive circuit anomaly and deleted from the initial thermal anomaly feature set.

[0041] Furthermore, this application also includes the following steps: constructing isolated trees of the isolated forest based on the multidimensional time-series data matrix; calculating the original path length of each sampling point in each isolated tree of the isolated forest, and then using the time weight coefficient of the sampling point as a multiplicative adjustment factor to weight and correct the original path length; and performing anomaly identification based on the corrected isolated trees to obtain the initial thermal anomaly feature set.

[0042] Specifically, all temperature, current, and humidity data within the current sliding window are extracted. These three types of data are then arranged and aligned according to the timestamps of the sampling times, forming an N-row, 3-column multidimensional time-series data matrix, where N equals the total number of sampling points within the window. Each row of the multidimensional time-series data matrix contains three values, corresponding to the temperature, current, and humidity values ​​at that given moment.

[0043] For each row in the multidimensional time-series data matrix, i.e., each sampling point, calculate the time interval difference between its timestamp and the current time. Calculate a time weighting coefficient based on the time interval between the sampling point's timestamp and the current time, ensuring that sampling points with shorter time intervals receive a larger weighting coefficient. The weighting coefficient ranges from 0 to 1. If a linear decay function is used, w... i =1-(tt i ) / (t-t0), where t is the current time, t i Let t0 be the timestamp of the i-th sampling point, t0 be the timestamp of the earliest sampling point within the window, and w be the timestamp of the earliest sampling point within the window. iLet be the weight coefficient for the i-th sampling point.

[0044] The multidimensional time-series data matrix is ​​used as the input dataset for the Isolation Forest algorithm. The total number of isolated trees in the Isolation Forest is set, typically 100 or 200. For each isolated tree, a subset of samples is randomly drawn with replacement from the multidimensional time-series data matrix. The size of the subset is usually set as a proportion of the total data volume, such as 256 points. A binary tree is constructed: In each split operation, one dimension is randomly selected from three feature dimensions: temperature, current, and humidity. A split value is randomly generated between the maximum and minimum values ​​of that dimension. Data points in the current node whose value in that dimension is less than the split value are assigned to the left child node, and the remaining data points are assigned to the right child node. The same split operation is then recursively performed on the left and right child nodes until each leaf node contains only one data point or the tree depth reaches a preset maximum depth, typically log2(subsample size). After all isolated trees are constructed, a complete Isolation Forest is formed. An isolated tree is the basic building block of an Isolation Forest and is a random binary tree. During construction, each tree independently and randomly selects feature dimensions and split values, recursively dividing the data until each leaf node contains only one data point or reaches the preset maximum tree depth.

[0045] Each isolated tree is traversed, and the original path length of each sampling point on that tree is calculated. Specifically, starting from the root node, based on the feature dimensions and segmentation values ​​stored in each node, the sampling point is passed down layer by layer, with the count value incremented by 1 each time it is passed, until a leaf node is reached. The accumulated count value is the original path length of that sampling point on that tree. After the original path length of each sampling point on all isolated trees has been calculated, the time weight coefficient of each sampling point is used as a multiplicative adjustment factor to weight and correct its original path length on each tree. That is, the corrected path length is equal to the original path length multiplied by the time weight coefficient of that sampling point. Since the time weight coefficient is a positive number not greater than 1, the weighted corrected path length is compressed to varying degrees compared to the original path length. This ensures that the sampling point closer to the current time is compressed less than the original path length after correction, thus allowing recent data to receive a higher contribution ratio in the final determination of outlier scores.

[0046] For each sampling point, the arithmetic mean of its corrected path lengths across all isolated trees is taken to obtain the average corrected path length. The average corrected path length for each sampling point is then substituted into the standard outlier score calculation formula for isolated forests: Score=2. (-E(path) / c(N))Where E(path) is the average corrected path length for that sampling point, and c(N) is the expected path length under the current data size N, which depends only on the number of data points N and can be calculated through theoretical formulas or by looking up tables. The final anomaly score for each sampling point is calculated. Because the time weight coefficient of new data is larger, its corrected path length is compressed less, meaning the corrected path length is relatively longer. In the anomaly score formula for isolated forests, the smaller the path length E(path), the higher the exponent 2. (-E(path) / c(N)) The larger the value, the higher the anomaly score. After correction, the new data has a shorter path length relative to the old data due to the smaller path length compression, thus contributing a higher percentage to the final anomaly score.

[0047] All sampling points are sorted from highest to lowest according to their anomaly scores. Sampling points with scores exceeding a preset anomaly threshold are marked as anomalies. The temperature, current, and humidity data corresponding to these anomalies are extracted to form an initial thermal anomaly feature set. Each sampling point in the initial thermal anomaly feature set is marked as a potential anomaly, but at this time it is not clear whether the root cause of the anomaly is a substantial deterioration of the heat dissipation system or a fluctuation interference in the drive circuit.

[0048] The temperature and current data of each anomalous sampling point in the initial thermal anomaly feature set are extracted, along with the average temperature and current values ​​of a previous stable operating period, and input into the physical coupling discriminator. The physical coupling discriminator is a discrimination module based on physical logic relationships. It utilizes the causal coupling relationship between current and temperature—that changes in current lead to corresponding changes in temperature—to perform secondary screening on each anomalous point in the initial thermal anomaly feature set, distinguishing whether its root cause is a problem with the heat dissipation system or a problem with the drive circuit.

[0049] The physical coupling discriminator calculates the temperature change amplitude of the current temperature at the anomaly point relative to the average temperature of the previous stable period, i.e., the absolute value of the current temperature minus the average temperature. Simultaneously, it calculates the current change amplitude of the current at the anomaly point relative to the average current of the previous stable period, i.e., the absolute value of the current current minus the average current. The ratio between the temperature change amplitude and the current change amplitude is calculated and compared to a pre-calibrated normal coupling ratio range. The normal coupling ratio range is an empirical interval representing the proportional relationship between the temperature change amplitude and the current change amplitude under normal operating conditions. This range is obtained through statistical calibration using a large amount of historical normal operating data. If the temperature-current change ratio of the current anomaly point falls within this range, it indicates a normal physical coupling relationship between the temperature and current changes; that is, the temperature change is driven by the current change, and it is an interference event caused by a drive circuit malfunction. Under normal operating conditions, small fluctuations in the drive current will correspondingly cause small synchronous changes in temperature. There is a relatively stable proportional relationship between the temperature change amplitude and the current change amplitude; the statistical distribution interval of this proportion is the normal coupling ratio range. If the ratio of the current anomaly falls within the preset normal coupling ratio range, it indicates that the ratio of the temperature change amplitude to the current change amplitude at the anomaly point is consistent with the physical coupling relationship under normal operating conditions. That is, the temperature change is synchronously driven by the current fluctuation. Therefore, the root cause of this temperature anomaly is the abnormal fluctuation of the output current of the drive circuit, rather than the performance degradation of the heat dissipation system itself. The physical coupling discriminator marks this type of temperature anomaly as an interference event caused by a drive circuit malfunction and removes it from the initial thermal anomaly feature set.

[0050] If the ratio between the change in current value and the change in temperature value falls outside the normal coupling ratio range, that is, if the ratio between the change in current value and the change in temperature value deviates significantly from the normal coupling ratio range, it indicates that the proportional relationship between the change in temperature and the change in current has deviated from the normal physical coupling law. In other words, the temperature has increased excessively in proportion to the current fluctuation. This usually manifests as a temperature rise that far exceeds the magnitude that the current fluctuation can explain, indicating that there is a substantial performance degradation in the heat dissipation system, such as increased thermal resistance and decreased heat dissipation efficiency. Such anomalies are identified as true heat dissipation abnormalities and are retained instead of being deleted.

[0051] Interference events caused by abnormal drive circuits are non-heat-dissipation-related events where fluctuations or drifts in the output current of the drive circuit lead to changes in heat generation, which in turn cause temperature changes. After the physical coupling discriminator identifies and removes abnormal points from all initial thermal anomaly feature sets one by one, the remaining abnormal points in the initial thermal anomaly feature sets constitute the pure thermal anomaly feature set that truly reflects the abnormal state of the heat dissipation system.

[0052] Through two core innovations of the improved Isolation Forest algorithm—temporal weighting embedding and physical coupling discrimination—high-precision and high-reliability joint detection of anomalies in the heat dissipation status of UV LEDs is achieved. Temporal weighting embedding focuses on recent thermal state changes, avoiding the dilution effect of historical normal data on current anomalies and improving detection timeliness. Physical coupling discrimination utilizes the physical causal relationship between current and temperature to effectively eliminate temperature interference events caused by fluctuations in the drive circuit, ensuring that the final set of retained thermal anomaly features only contains anomalies truly caused by the performance degradation of the heat dissipation system. Joint anomaly detection simultaneously inputs three-dimensional data of temperature, current, and humidity, identifying anomaly patterns that cannot be detected by single temperature monitoring within the feature space of the coordinated evolution of multiple physical quantities.

[0053] Using the thermal anomaly feature set as the input sequence, a pre-trained long short-term memory network model is used to perform rolling predictions on the temperature change trend of the lamps over multiple future time steps, generating a sequence of predicted temperature values ​​over multiple future time steps.

[0054] Furthermore, this application also includes the following steps: using the thermal anomaly feature set as input samples, performing normalization processing, and arranging them into a fixed-length input tensor in chronological order; feeding the input tensor into a long short-term memory network model for forward computation, and sequentially outputting temperature prediction values ​​for multiple future time steps to form the temperature prediction value sequence.

[0055] Specifically, each sample point in the thermal anomaly feature set contains three feature values: temperature, current, and humidity. Since the original numerical ranges of temperature, current, and humidity vary considerably, directly inputting them into the LSTM model would cause features with larger values ​​to dominate the loss function calculation, affecting the model's sensitivity to smaller numerical features. Therefore, all samples in the thermal anomaly feature set are normalized. The pre-calculated global minimum and maximum values ​​of each feature are read from the system configuration file and obtained through statistical analysis of a large amount of historical data during the offline training phase. A linear normalization formula is applied to each feature value of each sample, i.e., (feature value - minimum value) / (maximum value - minimum value) to obtain the normalized value, mapping all feature values ​​to the [0,1] interval, thus eliminating the influence of dimensional differences on model calculation.

[0056] After normalization, the thermal anomaly feature set is arranged into a fixed-length input tensor according to the chronological order of timestamps. The tensor has a shape of [T×3], where T is the number of sample points in the current thermal anomaly feature set, and 3 corresponds to the three feature dimensions of temperature, current, and humidity, which are the inputs to the LSTM model.

[0057] The Long Short-Term Memory (LSTM) network model employs a single-layer LSTM structure with 64 hidden layer neurons. Its input layer accepts a tensor of shape [T×3], where T is the number of sampling points in the thermal anomaly feature set within the current sliding window, and 3 corresponds to the three feature dimensions: temperature, current, and humidity. The output layer is a fully connected layer that maps the 64-dimensional hidden state of the last time step to a 1-dimensional temperature prediction value. At each time step t, the model receives the current input vector xt, a 3-dimensional vector comprising normalized temperature, normalized current, normalized humidity values, and the hidden state h{t-1} (64-dimensional) and cell state C{t-1} (64-dimensional) from the previous time step. Four gating units collaboratively perform selective forgetting and selective writing of information. The forgetting gate analyzes the current input features (temperature, current, humidity) and the hidden state from the previous time step, generating a retention ratio between 0 and 1 for each dimension of the cell state from the previous time step. 0 indicates complete forgetting of the historical information for that dimension, and 1 indicates complete retention, thus determining how much historical temperature trend information is discarded. The input gate generates two types of information: a writing ratio between 0 and 1, and a new candidate state between -1 and 1, representing the direction and magnitude of the temperature change increment at the current moment. These two values ​​are multiplied element-wise to obtain the new information to be written to the cell state at the current moment. The cell state update then adds the historical information filtered by the forget gate to the new information filtered by the input gate element-wise, completing the rolling update of the cell state. The output gate extracts the most relevant temperature features from the updated cell state at the current moment. An output ratio between 0 and 1 controls which information is output as the current hidden state. This hidden state serves both as the output of the current time step, passed to the fully connected layer for temperature prediction, and as the input for the next time step. After processing at the last time step, the fully connected layer linearly maps the final 64-dimensional hidden state to a 1-dimensional normalized temperature prediction value. All the weight matrices mentioned above, including those for the forget gate, input gate, output gate, candidate state generation, and output layer, as well as the bias vectors, are parameters automatically optimized and determined by the model during the training phase using the backpropagation algorithm.

[0058] During the offline training phase, a large number of continuous monitoring data segments were extracted from the historical operation database of the same model of UV LED lamps using a sliding window approach. The multidimensional features of the first Tin time steps of each segment, where Tin corresponds to the actual number of anomalies within the sliding window (e.g., 5), were used as input samples. The temperature values ​​of the last Tout time steps were used as prediction labels, where Tout represents the number of future steps to be predicted (e.g., 10 steps). Training, validation, and test sets were constructed, and the parameters were iteratively optimized using mean squared error as the loss function and the Adam optimizer (initial learning rate lr=0.001, β1=0.9, β2=0.999). The batch size was 32, and the maximum number of training epochs was 200. An early stopping strategy was also introduced: the training was terminated if the validation set loss did not decrease for 10 consecutive epochs to prevent overfitting. After training, the model parameters with the minimum loss on the validation set were saved. Electric current is the electrical source of heat, humidity affects the air convection heat transfer coefficient and thus changes the heat dissipation efficiency, and temperature is the final thermal characterization of the heat dissipation system state. These three constitute a complete causal chain of input from the electric heat source → environmental modulation → thermal state response. The three dimensions of temperature, current, and humidity in the input tensor of the Long Short-Term Memory Network model are precisely the digital mapping of this causal chain in time sequence, while the output is the temperature prediction value at multiple future moments, realizing end-to-end learning and prediction from multi-dimensional historical states to future thermal trends. In terms of the correspondence between the gating mechanism and the physical scene, the forget gate learns to judge the persistence of temperature trends to identify the implicit degradation of heat dissipation performance. For example, when the thermal resistance slowly increases and the temperature rises for a long time, the forget gate tends to retain this historical trend. The input gate learns to respond to disturbances such as sudden changes in current and writes new information into the cell state, while the output gate selects the most relevant features for prediction based on the current operating conditions.

[0059] The input tensor is fed into a pre-trained LSTM network model for forward inference computation. The internal structure of the LSTM network includes an input gate, a forget gate, an output gate, and a cell state update unit. At each time step, the model receives three feature values ​​from the tensor corresponding to that time step, combines them with the hidden state and cell state passed from the previous time step, uses the forget gate to decide which historical information to discard, and the input gate to decide which new information to write into the cell state. After updating the cell state, the output gate is used to calculate the output value for the current time step and the hidden state for the next time step. The model processes all time steps in the input tensor sequentially. After the last time step, the model maps the current hidden state to the temperature prediction value for the first future time step through a fully connected layer (output layer).

[0060] The predicted temperature value for the first future time step is concatenated with the current and humidity values ​​from the last input time step to form a new 3D input vector [predicted temperature, current, humidity]. This vector is then used as the new input, along with the current hidden state and cell state, and fed into the LSTM model for the next forward computation, outputting the predicted temperature value for the second future time step. This iterative process is repeated until a predetermined number of predicted temperature values ​​for future time steps (e.g., 10) are output. All predicted values ​​are arranged in chronological order to form a complete sequence of predicted temperature values.

[0061] By using an LSTM model to perform time-series prediction on a set of thermal anomaly features, we have achieved advanced perception of future temperature change trends in UV LED lamps. The gating mechanism of LSTM (forget gate, input gate, and output gate) can effectively capture the long-term dependencies and cumulative effects of temperature changes, so that the prediction results are not only based on the current temperature value, but also on implicit trend information such as the rate and acceleration of temperature change.

[0062] By combining the current operating time and cumulative service history of the ultraviolet LED lamp, a warning trigger threshold adapted to the current operating conditions is dynamically calculated, and the warning trigger judgment is made on the temperature prediction value sequence.

[0063] Furthermore, this application also includes the following steps: using the initial heat dissipation performance of the ultraviolet LED lamp at the time of manufacture as a benchmark threshold; collecting historical heat dissipation characteristic time series and analyzing the heat dissipation performance decay curve with respect to the cumulative running time; based on the heat dissipation performance decay curve, performing aging compensation correction on the benchmark threshold with the current cumulative running time, and simultaneously, performing environmental impact correction by matching the current ambient humidity value with an environmental correction coefficient, to obtain the warning trigger threshold.

[0064] Specifically, the upper limit of the rated case temperature of the UV LED lamp in a brand new state is read from the factory configuration parameters. This is the initial benchmark warning trigger threshold, which is the safe operating temperature boundary obtained through thermal simulation and actual measurement under standard test conditions during the lamp design stage.

[0065] The system retrieves the total cumulative service time of the luminaire from the database, i.e., the total number of hours it has been lit since its manufacture. It then retrieves the pre-stored heat dissipation performance degradation curve for this model of luminaire. During the luminaire's development phase, several samples of the same model are selected and subjected to accelerated aging tests under standard environmental conditions. At regular intervals, such as 500 hours, the equivalent thermal resistance or the case temperature under standard current is measured. The discrete measurement points are then subjected to curve fitting to obtain a degradation curve function with the cumulative service time as the x-axis and the temperature correction offset as the y-axis. The degradation curve is typically established using polynomial fitting or piecewise linear fitting and is permanently stored in the system configuration file.

[0066] Using the current cumulative runtime as an index, interpolation is performed on the heat dissipation performance degradation curve to obtain the equivalent temperature correction offset for that cumulative runtime. The equivalent temperature correction offset is the increase in shell temperature relative to the brand-new state under the same heat generation conditions due to heat dissipation performance degradation. Subtracting the equivalent temperature correction offset from the initial baseline threshold yields the aging-compensated threshold. As the luminaire ages, its heat dissipation capacity decreases, and the warning threshold must be lowered accordingly to maintain the same safety margin. For example, if the initial threshold is 85.0℃ and the aging offset is +4.2℃, then the aging-compensated threshold is 85.0 - 4.2 = 80.8℃.

[0067] The system reads the relative humidity value collected by the ambient humidity sensor. It has a pre-calibrated humidity-heat dissipation efficiency correction table, obtained through extensive thermal testing under varying humidity conditions, providing correction coefficients for different humidity levels. The correction coefficient typically ranges from 0.90 to 1, representing the comprehensive impact of humidity on the warning trigger threshold. At standard humidity, a correction coefficient of 1 indicates no adjustment is needed; at high humidity, a correction coefficient less than 1 indicates the threshold needs to be lowered to compensate for the adverse effects of humidity. The warning trigger threshold is obtained by multiplying the aging-compensated threshold by this humidity correction coefficient. If the current humidity is within the standard range and below the set humidity threshold, the correction coefficient is 1, and the warning trigger threshold remains unchanged; if the current humidity is above the set humidity threshold, the correction coefficient is less than 1, and the warning trigger threshold is lowered accordingly.

[0068] Each predicted value in the temperature prediction sequence over multiple future time steps is compared one by one with the currently calculated warning trigger threshold. If any predicted value exceeds the warning trigger threshold, a warning signal is triggered, and a tiered warning is issued via audible and visual alarms, app push notifications, or SMS notifications. If all predicted values ​​are below the warning trigger threshold, normal monitoring is maintained, and no warning is triggered. After the warning trigger judgment is completed, the next sampling cycle begins, data is updated continuously, and the entire process is repeated. The warning trigger threshold is a dynamic warning temperature threshold value obtained after aging compensation and environmental impact correction, which is dynamically adjusted according to the aging degree of the lighting fixtures and changes in environmental conditions.

[0069] By introducing the concept of full lifecycle management, the inapplicability of fixed thresholds to ultraviolet LEDs during long-term operation is solved. Based on aging compensation correction according to the heat dissipation performance decay curve, the warning trigger threshold can be dynamically adjusted according to the actual health status of the lamp. This avoids frequent false alarms due to normal temperature rise caused by lamp aging, and also avoids underestimating real risks by ignoring the increase in thermal resistance. This ensures the accuracy and reliability of the warning system throughout the entire product lifecycle, balancing equipment safety and operational continuity.

[0070] In summary, the IoT data-driven intelligent early warning method for heat dissipation of ultraviolet LED lamps provided in this application has the following technical effects: A multi-dimensional sensor network covering the entire thermal field of the lamp tube is constructed to collect multi-source heterogeneous time-series data streams containing lamp tube surface temperature, driving current value, and ambient humidity value; based on the multi-source heterogeneous time-series data streams, the length of the sliding window is determined according to the dynamic change amplitude of the current operating condition of the ultraviolet LED lamp; within the time-series data interval defined by the sliding window, an improved isolated forest algorithm is used to perform joint anomaly detection on the multi-source heterogeneous data within the window, extracting a set of thermal anomaly features reflecting abnormal heat dissipation status; using the set of thermal anomaly features as the input sequence, a pre-trained long short-term memory network model is used to perform rolling prediction of the lamp tube temperature change trend over multiple future time steps, generating a temperature prediction value sequence over multiple future time steps; combining the current operating time and cumulative service history of the ultraviolet LED lamp, an early warning trigger threshold adapted to the current operating condition is dynamically calculated, and an early warning trigger judgment is made on the temperature prediction value sequence. In other words, by constructing a multi-dimensional sensor network to collect multi-source data and adaptively adjusting the sliding window length according to dynamic changes in operating conditions to match the analysis granularity, pure thermal anomaly features are accurately extracted, driving the long short-term memory network model to perform multi-step rolling predictions. Finally, by combining the equipment aging and decay curve with the environmental humidity coefficient, the early warning trigger threshold is dynamically calculated, and an early warning is accurately issued before thermal failure occurs, fundamentally ensuring the operational reliability of the ultraviolet LED lamp throughout its entire life cycle.

[0071] Example 2: Based on the same inventive concept as the IoT data-driven intelligent early warning method for heat dissipation of ultraviolet LED lamps in Example 1, this application also provides an IoT data-driven intelligent early warning system for heat dissipation of ultraviolet LED lamps. Please refer to the appendix. Figure 2The IoT data-driven intelligent early warning system for ultraviolet LED lamp heat dissipation includes: a time-series data acquisition module 11, used to construct a multi-dimensional sensor network covering the entire thermal field of the lamp tube, and to acquire multi-source heterogeneous time-series data streams containing lamp tube surface temperature, driving current value, and ambient humidity value; a sliding window determination module 12, used to determine the length of the sliding window based on the multi-source heterogeneous time-series data stream and the dynamic change amplitude of the current operating condition of the ultraviolet LED lamp; and a joint anomaly detection module 13, used to perform multi-source anomaly detection within the time-series data interval defined by the sliding window using an improved isolated forest algorithm. Heterogeneous data are used for joint anomaly detection to extract a set of thermal anomaly features reflecting abnormal heat dissipation. The temperature rolling prediction module 14 is used to use the set of thermal anomaly features as an input sequence and a pre-trained long short-term memory network model to make rolling predictions on the temperature change trend of the lamp tube in multiple future time steps, generating a sequence of predicted temperature values ​​in multiple future time steps. The early warning trigger judgment module 15 is used to combine the current running time and cumulative service history of the ultraviolet LED lamp to dynamically calculate an early warning trigger threshold adapted to the current operating conditions and to make an early warning trigger judgment on the temperature prediction value sequence.

[0072] Furthermore, the timing data acquisition module 11 in the IoT data-driven intelligent early warning system for heat dissipation of ultraviolet LED lamps is also used for: the multi-dimensional sensing network includes temperature sensors, current sensors, and ambient humidity sensors deployed on the surface of the ultraviolet LED lamp body, in the preset air domain of the lamp tube, and at the location of the lamp driving circuit board.

[0073] Furthermore, the sliding window determination module 12 in the IoT data-driven intelligent early warning system for ultraviolet LED lamp heat dissipation is also used to: acquire an initial sliding window; calculate the degree of temperature change drasticness of the temperature value sequence within the initial sliding window based on the multi-source heterogeneous time-series data stream; when the degree of temperature change drasticness exceeds a preset first degree of drasticness threshold, set the time domain length of the sliding window to a first time domain length value; when the degree of temperature change drasticness is lower than a preset second degree of drasticness threshold, set the time domain length of the sliding window to a second time domain length value greater than the first time domain length value; when the degree of temperature change drasticness index is between the first degree of drasticness threshold and the second degree of drasticness threshold, keep the time domain length of the initial sliding window unchanged.

[0074] Furthermore, the sliding window determination module 12 in the IoT data-driven intelligent early warning system for ultraviolet LED lamp heat dissipation is also used to: use the average value of the absolute value of the first difference of the temperature value sequence within the initial sliding window as a quantitative indicator of the severity of temperature change.

[0075] Furthermore, the joint anomaly detection module 13 in the IoT data-driven intelligent early warning system for ultraviolet LED lamp heat dissipation is also used to: arrange the temperature data, current data, and humidity data within the time-series data interval defined by the sliding window in chronological order to obtain a multi-dimensional time-series data matrix; for each sampling point in the multi-dimensional time-series data matrix, calculate the time weight coefficient based on the time interval between the timestamp of the sampling point and the current time; embed the time weight coefficient of each sampling point into the path length calculation process of the isolated forest algorithm to obtain an initial thermal anomaly feature set; input the temperature data and current data into a preset physical coupling discriminator, and have the physical coupling discriminator delete interference events from the initial thermal anomaly feature set based on the ratio between the change amplitude of the current value and the change amplitude of the temperature value to obtain the thermal anomaly feature set.

[0076] Furthermore, the joint anomaly detection module 13 in the IoT data-driven intelligent early warning system for ultraviolet LED lamp heat dissipation is also used for: the physical coupling discriminator to determine whether the current temperature anomaly is accompanied by a current anomaly of the same source based on the ratio between the change amplitude of the current value and the change amplitude of the temperature value; if the temperature anomaly is accompanied by a current anomaly of the same source, the corresponding temperature anomaly is marked as an interference event caused by a driving circuit anomaly and deleted from the initial thermal anomaly feature set.

[0077] Furthermore, the joint anomaly detection module 13 in the IoT data-driven intelligent early warning system for ultraviolet LED lamp heat dissipation is also used to: construct isolated trees of the isolated forest based on the multidimensional time-series data matrix; calculate the original path length of each sampling point in each isolated tree of the isolated forest, and then use the time weight coefficient of the sampling point as a multiplicative adjustment factor to perform weighted correction on the original path length; and perform anomaly identification based on the corrected isolated trees to obtain the initial thermal anomaly feature set.

[0078] Furthermore, the temperature rolling prediction module 14 in the IoT data-driven intelligent early warning system for ultraviolet LED lamp heat dissipation is also used to: take the thermal anomaly feature set as input sample, perform normalization processing, and arrange it into a fixed-length input tensor in chronological order; send the input tensor into a long short-term memory network model for forward calculation, and output the temperature prediction values ​​for multiple future time steps in sequence to form the temperature prediction value sequence.

[0079] Furthermore, the warning trigger judgment module 15 in the IoT data-driven intelligent early warning system for ultraviolet LED lamp heat dissipation is also used to: take the initial heat dissipation performance of the ultraviolet LED lamp at the time of leaving the factory as the benchmark threshold; collect historical heat dissipation characteristic time sequence and analyze the heat dissipation performance decay curve with respect to the cumulative running time; based on the heat dissipation performance decay curve, perform aging compensation correction on the benchmark threshold with the current cumulative running time, and at the same time, perform environmental impact correction by matching the current ambient humidity value with the environmental correction coefficient to obtain the warning trigger threshold.

[0080] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The IoT data-driven intelligent early warning method and specific examples for heat dissipation of ultraviolet LED lamps in the foregoing embodiment one are also applicable to the IoT data-driven intelligent early warning system for heat dissipation of ultraviolet LED lamps in this embodiment. Through the foregoing detailed description of the IoT data-driven intelligent early warning method for heat dissipation of ultraviolet LED lamps, those skilled in the art can clearly understand the IoT data-driven intelligent early warning system for heat dissipation of ultraviolet LED lamps in this embodiment. Therefore, for the sake of brevity, it will not be described in detail here.

[0081] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0082] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application also intends to include such modifications and variations.

Claims

1. A smart early warning method for heat dissipation of ultraviolet LED lamps driven by IoT data, characterized in that, include: Construct a multi-dimensional sensor network covering the entire thermal field of the lamp tube, and collect multi-source heterogeneous time-series data streams including lamp tube surface temperature, driving current value and ambient humidity value; Based on the multi-source heterogeneous time-series data stream, the length of the sliding window is determined according to the dynamic change range of the current operating conditions of the ultraviolet LED lamp; Within the time-series data interval defined by the sliding window, an improved isolated forest algorithm is used to perform joint anomaly detection on multi-source heterogeneous data within the window, and extract a set of thermal anomaly features reflecting abnormal heat dissipation status. Using the thermal anomaly feature set as the input sequence, a pre-trained long short-term memory network model is used to make rolling predictions on the temperature change trend of the lamp tube in multiple future time steps, generating a sequence of predicted temperature values ​​in multiple future time steps. By combining the current operating time and cumulative service history of the ultraviolet LED lamp, a warning trigger threshold adapted to the current operating conditions is dynamically calculated, and the warning trigger judgment is made on the temperature prediction value sequence.

2. The IoT data-driven intelligent early warning method for heat dissipation of ultraviolet LED lamps as described in claim 1, characterized in that, Based on the aforementioned multi-source heterogeneous time-series data stream, the length of the sliding window is determined according to the dynamic change amplitude of the current operating condition of the ultraviolet LED lamp, including: Get the initial sliding window; Based on the multi-source heterogeneous time-series data stream, the degree of temperature change drasticness of the temperature value sequence within the initial sliding window is calculated; When the temperature change exceeds the preset first severity threshold, the time domain length of the sliding window is set to the first time domain length value. When the degree of temperature change is lower than the preset second degree threshold, the time domain length of the sliding window is set to a second time domain length value that is greater than the first time domain length value; When the intensity of temperature change is between the first intensity threshold and the second intensity threshold, the time domain length of the initial sliding window remains unchanged.

3. The IoT data-driven intelligent early warning method for heat dissipation of ultraviolet LED lamps as described in claim 2, characterized in that, The average of the absolute values ​​of the first-order differences of the temperature value sequence within the initial sliding window is used as a quantitative indicator of the severity of temperature changes.

4. The IoT data-driven intelligent early warning method for heat dissipation of ultraviolet LED lamps as described in claim 1, characterized in that, Within the time-series data interval defined by the sliding window, an improved isolated forest algorithm is used to perform joint anomaly detection on the multi-source heterogeneous data within the window, extracting a set of thermal anomaly features reflecting abnormal heat dissipation status, including: The temperature data, current data, and humidity data within the time-series data interval defined by the sliding window are arranged in chronological order to obtain a multidimensional time-series data matrix. For each sampling point in the multidimensional time series data matrix, a time weighting coefficient is calculated based on the time interval between the timestamp of the sampling point and the current time. The time weight coefficients of each sampling point are embedded into the path length calculation process of the isolated forest algorithm to obtain the initial set of thermal anomaly features; Temperature and current data are input into a preset physical coupling discriminator. The physical coupling discriminator removes interference events from the initial thermal anomaly feature set based on the ratio between the change in current value and the change in temperature value, thus obtaining the thermal anomaly feature set.

5. The IoT data-driven intelligent early warning method for heat dissipation of ultraviolet LED lamps as described in claim 4, characterized in that, The initial thermal anomaly feature set is subjected to interference event removal, including: The physical coupling discriminator determines whether the current temperature anomaly is accompanied by a current anomaly of the same origin based on the ratio between the change in current value and the change in temperature value. If a temperature anomaly is accompanied by a current anomaly of the same origin, the corresponding temperature anomaly is marked as an interference event caused by a drive circuit anomaly and removed from the initial thermal anomaly feature set.

6. The IoT data-driven intelligent early warning method for heat dissipation of ultraviolet LED lamps as described in claim 5, characterized in that, By embedding the time weight coefficients of each sampling point into the path length calculation process of the isolated forest algorithm, an initial set of hot anomaly features is obtained, including: The isolated trees of the isolated forest are constructed based on the multidimensional time-series data matrix. After calculating the original path length of each sampling point in each isolated tree of the isolated forest, the original path length is weighted and corrected using the time weight coefficient of the sampling point as a multiplicative adjustment factor. Anomaly identification is performed on each corrected isolated tree to obtain the initial thermal anomaly feature set.

7. The IoT data-driven intelligent early warning method for heat dissipation of ultraviolet LED lamps as described in claim 1, characterized in that, Using the aforementioned thermal anomaly feature set as the input sequence, a pre-trained long short-term memory network model is used to perform rolling predictions of the lamp temperature change trend over multiple future time steps, generating a sequence of predicted temperature values ​​over multiple future time steps, including: The thermal anomaly feature set is used as input samples, and after normalization, they are arranged into a fixed-length input tensor in chronological order. The input tensor is fed into a long short-term memory network model for forward computation, and the predicted temperature values ​​for multiple future time steps are output sequentially to form the temperature prediction value sequence.

8. The IoT data-driven intelligent early warning method for heat dissipation of ultraviolet LED lamps as described in claim 1, characterized in that, Based on the current operating time and cumulative service history of the UV LED lamp, dynamically calculate the warning trigger threshold adapted to the current operating conditions, including: The initial heat dissipation performance of the UV LED lamp at the time of manufacture is used as the benchmark threshold. Collect historical heat dissipation characteristics time series and analyze the heat dissipation performance degradation curve with respect to cumulative running time; Based on the heat dissipation performance decay curve, the baseline threshold is corrected for aging compensation using the current cumulative running time. At the same time, the environmental impact is corrected by matching the current ambient humidity value with an environmental correction coefficient, thus obtaining the warning trigger threshold.

9. The IoT data-driven intelligent early warning method for heat dissipation of ultraviolet LED lamps as described in claim 1, characterized in that, The multidimensional sensing network includes temperature sensors, current sensors, and ambient humidity sensors deployed on the surface of the ultraviolet LED tube, in the preset air zone of the tube, and at the location of the lamp driver circuit board.

10. An IoT data-driven intelligent early warning system for heat dissipation of ultraviolet LED lamps, characterized in that, The steps for implementing the IoT data-driven intelligent early warning method for heat dissipation of ultraviolet LED lamps according to any one of claims 1 to 9, wherein the IoT data-driven intelligent early warning system for heat dissipation of ultraviolet LED lamps comprises: The timing data acquisition module is used to build a multi-dimensional sensor network covering the entire thermal field of the lamp tube, and to collect multi-source heterogeneous timing data streams including lamp tube surface temperature, driving current value and ambient humidity value. The sliding window determination module is used to determine the length of the sliding window based on the multi-source heterogeneous time-series data stream and the dynamic change amplitude of the current operating condition of the ultraviolet LED lamp. The joint anomaly detection module is used to perform joint anomaly detection on multi-source heterogeneous data within the time-series data interval defined by the sliding window using an improved isolated forest algorithm, and extract a set of thermal anomaly features reflecting abnormal heat dissipation status. The temperature rolling prediction module is used to take the thermal anomaly feature set as the input sequence and use a pre-trained long short-term memory network model to make rolling predictions on the temperature change trend of the lamp tube in multiple future time steps, generating a sequence of predicted temperature values ​​in multiple future time steps. The early warning trigger judgment module is used to dynamically calculate an early warning trigger threshold adapted to the current operating conditions by combining the current running time and cumulative service history of the ultraviolet LED lamp, and to make early warning trigger judgment on the temperature prediction value sequence.

Images