A fault detection method, device and storage medium of a heating device

Abstract

The application provides a fault detection method and device of a heating device, and a storage medium. The heating device comprises a thermistor. The method comprises: obtaining first resistance value time sequence data of the thermistor in a first operation period of the heating device; performing feature extraction on the first resistance value time sequence data to obtain a first target feature vector; each component in the first target feature vector represents a feature value of the first resistance value time sequence data in a feature dimension; inputting a second target feature vector into a target fault detection model to obtain a fault detection result of the heating device; and the second target feature vector is determined based on the first target feature vector. The application extracts features from the resistance value time sequence data of the thermistor to obtain a feature vector for representing characteristics of the resistance value time sequence data, and determines a fault detection result based on the feature vector by using a fault detection model, which is conducive to avoiding interference caused by complex operation conditions of the heating device, thereby improving the fault detection accuracy of the heating device.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fault detection, in particular to a fault detection method, device and storage medium of a heating device. BACKGROUND

[0002] The heating device is a core component of a temperature control system of a new energy vehicle, and is mainly used for engine preheating and cabin heating in winter. Since the thermistor in the heating device has the characteristic that the resistance value changes with temperature, it is easy to age and fail in the long-term working process, thereby inducing the heating device failure. Therefore, it is very important to carry out fault detection of the heating device to ensure driving safety.

[0003] The current fault detection method mainly sets a static threshold of the current or temperature of the heating device, and determines the fault by judging whether the real-time sampling value exceeds the threshold. However, the working condition of the heating device is complex and changeable, and the current fault detection method is difficult to adapt, resulting in poor accuracy of the fault detection of the heating device. SUMMARY

[0004] The main purpose of the embodiments of the present application is to provide a fault detection method, device and storage medium of a heating device, which aims to improve the accuracy of the fault detection of the heating device.

[0005] The embodiments of the present application provide a fault detection method of a heating device, the heating device comprising a thermistor, the method comprising: acquiring first resistance value time sequence data of the thermistor in a first running period of the heating device; performing feature extraction on the first resistance value time sequence data to obtain a first target feature vector; each component in the first target feature vector represents a feature value of the first resistance value time sequence data in a feature dimension; inputting a second target feature vector into a target fault detection model to obtain a fault detection result of the heating device; the second target feature vector is determined based on the first target feature vector.

[0006] In an embodiment, before the second target feature vector is input into the target fault detection model to obtain the fault detection result of the heating device, the method further includes: obtaining a plurality of first sample feature vectors; each of the first sample feature vectors is obtained based on feature extraction on a second resistance value time series data of the thermistor; each of the second resistance value time series data corresponds to a second running period of the heating device; the thermistor is in a normal working state during each of the second running periods; a kernel density estimation algorithm is used to perform smoothing superposition on kernel functions centered on each of the first sample feature vectors to obtain a continuous first probability density function; a second probability density function is determined based on the first probability density function and a target value of a bandwidth; the bandwidth in the first probability density function is a to-be-determined coefficient; and the target fault detection model is constructed based on the second probability density function.

[0007] In an embodiment, the obtaining of the plurality of first sample feature vectors includes: obtaining a target covariance matrix of a first sample feature matrix; the first sample feature matrix includes a plurality of second sample feature vectors; each of the second sample feature vectors is obtained based on feature extraction on one of the second resistance value time series data; the target covariance matrix is determined based on covariances between feature dimension vectors in the first sample feature matrix; each component in a same feature dimension vector corresponds to a feature value of a same feature dimension of each of the second resistance value time series data; eigenvalue decomposition is performed on the target covariance matrix to obtain a plurality of target eigenvalues and third sample feature vectors corresponding to each of the target eigenvalues; cumulative variance contribution rates of the first k target eigenvalues are determined in a descending order of the target eigenvalues; k is a positive integer and is less than a total number of the target eigenvalues; and the third sample feature vectors corresponding to the first k target eigenvalues are determined as the first sample feature vectors in response to the cumulative variance contribution rates of the first k target eigenvalues being greater than a cumulative variance contribution rate threshold.

[0008] In an embodiment, the obtaining of the target covariance matrix of the first sample feature matrix includes: obtaining a mean vector and a standard deviation vector of the first sample feature matrix; each component in the mean vector represents an average value of a feature dimension vector in the first sample feature matrix; each component in the standard deviation vector represents a standard deviation of a feature dimension vector in the first sample feature matrix; the first sample feature matrix is standardized based on the mean vector and the standard deviation vector; and the target covariance matrix is determined based on covariances between feature dimension vectors in the standardized first sample feature matrix.

[0009] In an embodiment, before the second probability density function is determined based on the first probability density function and the target value of the bandwidth, the method further comprises: obtaining a plurality of candidate values of the bandwidth; for each candidate value of the bandwidth, performing the following steps to obtain a log-likelihood value corresponding to the candidate value: determining a probability density function corresponding to the candidate value based on the second probability density function and the candidate value of the bandwidth; determining a probability density value of each of the first sample feature vectors respectively using the probability density function corresponding to the candidate value; determining a log-likelihood value corresponding to the candidate value based on the probability density values of the plurality of first sample feature vectors; and selecting a candidate value with the maximum log-likelihood value from the plurality of candidate values as the target value.

[0010] In an embodiment, the target fault detection model is constructed based on the second probability density function, comprising: obtaining a target probability density threshold; the target probability density threshold is used to determine whether a probability density value determined based on the second probability density function represents a fault of the heating device; and constructing the target fault detection model based on the second probability density function and the target probability density threshold.

[0011] In an embodiment, the target probability density threshold is obtained by: obtaining a plurality of fourth sample feature vectors and a plurality of candidate probability density thresholds; each of the fourth sample feature vectors is obtained by feature extraction on a third resistance value time series data of the thermistor; each of the third resistance value time series data corresponds to a third operation period of the heating device; the thermistor is in an abnormal working state during at least part of the third operation period; for each of the fourth sample feature vectors, a probability density value of the fourth sample feature vector is determined based on the second probability density function; for each of the candidate probability density thresholds, a fault detection accuracy of the candidate probability density threshold is determined based on a comparison result of the candidate probability density threshold and the probability density values of the plurality of fourth sample feature vectors, and actual working states of the thermistor corresponding to the plurality of fourth sample feature vectors; and a candidate probability density threshold with the highest fault detection accuracy is selected from the plurality of candidate probability density thresholds as the target probability density threshold.

[0012] In an embodiment, before the second target feature vector is input into the target fault detection model to obtain the fault detection result of the heating device, the method further comprises: obtaining a second sample feature matrix; the second sample feature matrix comprises third sample feature vectors corresponding to the first k target eigenvalues of the target covariance matrix respectively; the cumulative variance contribution rate of the first k target eigenvalues is greater than a cumulative variance contribution rate threshold; and the second target feature vector is determined based on a product of the first target feature vector and the second sample feature matrix.

[0013] In an embodiment, the determining the second target feature vector based on the product of the first target feature vector and the second sample feature matrix comprises: normalizing the first target feature vector based on a mean vector and a standard deviation vector of the first sample feature matrix; and determining the second target feature vector based on the product of the normalized first target feature vector and the second sample feature matrix.

[0014] In an embodiment, after the inputting the second target feature vector into the target fault detection model to obtain the fault detection result of the heating device, the method further comprises: updating the target fault detection model based on the first resistance value time series data and the actual working state of the thermistor in the first operation period.

[0015] In an embodiment, the heating device is configured with a plurality of heating gears; the first resistance value time series data comprises resistance value time series data of the thermistor under each of the heating gears; the first target feature vector comprises at least one first component; and the extracting features from the first resistance value time series data to obtain the first target feature vector comprises: extracting at least one kind of cooperative feature between the resistance value time series data under the plurality of heating gears to obtain the at least one first component; wherein the cooperative feature is used to represent the consistency of changes between the resistance value time series data under the plurality of heating gears; and each of the first components represents a feature value of the resistance value time series data under the plurality of heating gears in one cooperative feature dimension.

[0016] In one embodiment, the first target feature vector includes at least one of the following: at least one second component, at least one third component, and at least one fourth component; the feature extraction of the first resistance time series data to obtain the first target feature vector includes at least one of the following: extracting at least one distribution statistical feature of the first resistance time series data to obtain the at least one second component; wherein the distribution statistical feature is used to characterize the distribution pattern of the first resistance time series data; each second component characterizes the feature value of the first resistance time series data in a distribution statistical feature dimension; extracting at least one trend evolution feature of the first resistance time series data to obtain the at least one third component; wherein the trend evolution feature is used to characterize the trend change pattern of the first resistance time series data; each third component characterizes the feature value of the first resistance time series data in a trend evolution feature dimension; extracting at least one time-dependent feature of the first resistance time series data to obtain the at least one fourth component; wherein the time-dependent feature is used to characterize the correlation between resistance values ​​at adjacent time points in the first resistance time series data; each fourth component characterizes the feature value of the first resistance time series data in a time-dependent feature dimension.

[0017] This application embodiment also provides a fault detection device for a heating device, the device including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the above-described fault detection method for the heating device.

[0018] This application embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements the above-described fault detection method for the heating device.

[0019] This application provides a fault detection method, device, and storage medium for a heating device. The method includes: acquiring first resistance time-series data of a thermistor during a first operating period of the heating device; extracting features from the first resistance time-series data to obtain a first target feature vector, where each component of the first target feature vector represents a feature value of the first resistance time-series data in one feature dimension; and inputting a second target feature vector determined based on the first target feature vector into a target fault detection model to obtain a fault detection result for the heating device. The technical solution of this application, by extracting features from the resistance time-series data of the thermistor to obtain a feature vector characterizing the resistance time-series data, and using a fault detection model to determine the fault detection result based on the feature vector, helps to avoid interference caused by the complex operating conditions of the heating device, thereby improving the accuracy of fault detection for the heating device. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating the fault detection method provided in the embodiments of this application.

[0021] Figure 2 This is a schematic diagram illustrating the construction of the fault detection model provided in the embodiments of this application.

[0022] Figure 3 This is a schematic diagram of the specific process of the fault detection method provided in the embodiments of this application.

[0023] Figure 4 This is a schematic diagram of the structure of the fault detection device provided in the embodiments of this application. Detailed Implementation

[0024] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0025] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, a first object can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the digit " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0026] The fault detection method for a heating device provided in this application can be applied to a fault detection device for a heating device or the software of such a device. The fault detection device can be a terminal, a server, or a vehicle equipped with the heating device. In some embodiments, the terminal can be a smartphone, tablet, laptop, desktop computer, etc.; the server can be configured as an independent physical server, or as a server cluster or distributed system composed of multiple physical servers. The software can be an application that implements the fault detection method for the heating device, but is not limited to the above forms.

[0027] The fault detection method for the heating device provided in this application will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0028] Please see Figure 1This application provides a fault detection method for a heating device, the heating device including a thermistor, and the fault detection method may include: Step S101: Obtain the first resistance value timing data of the thermistor during the first operating period of the heating device; Optionally, the first operating period may include multiple first time points; the first resistance value timing data may include the resistance value of the thermistor at each time point.

[0029] Optionally, the heating device can be configured with multiple heating levels, each with a different heating power. The heating levels can be displayed as power selection levels or temperature setting levels. The first resistance timing data can include the resistance timing data of the thermistor at each heating level.

[0030] Step S102: Perform feature extraction on the first resistance time series data to obtain the first target feature vector; each component in the first target feature vector represents the feature value of the first resistance time series data in one feature dimension. In practical implementation, at least one of the following features can be extracted from the first resistance value time series data to obtain the first target feature vector: distribution statistics, trend evolution, and time-series dependence. Distribution statistics can be used to characterize the distribution pattern of the first resistance value time series data. Trend evolution can be used to characterize the trend change pattern of the first resistance value time series data. Time-series dependence can be used to characterize the correlation between resistance values ​​at adjacent time points in the first resistance value time series data.

[0031] When the heating device is equipped with multiple heating levels, the collaborative features of the first resistance value time series data under multiple heating levels can be extracted to obtain the first target feature vector. The collaborative features can be used to characterize the consistency of changes among the resistance value time series data under multiple heating levels. The above-mentioned distribution statistical features, trend evolution features, and time series dependency features can characterize the corresponding features of the first resistance value time series data under each heating level.

[0032] Step S103: Input the second target feature vector into the target fault detection model to obtain the fault detection result of the heating device; the second target feature vector is determined based on the first target feature vector.

[0033] Optionally, the target fault detection model can be a model that has learned the mapping relationship between the feature vector of the resistance time series data and the operating state of the thermistor. When the target fault detection model determines that the thermistor is in an abnormal operating state based on the second target feature vector, it can output a fault detection result to indicate that the heating device has failed.

[0034] In practice, the first target feature vector can be directly determined as the second target feature vector, or the first target feature vector can be subjected to at least one of the following processes: dimensionality reduction and standardization, to obtain the second target feature vector.

[0035] In this way, by extracting features from the resistance time series data of the thermistor, a feature vector is obtained to characterize the resistance time series data. The fault detection model is then used to determine the fault detection result based on the feature vector, which helps to avoid interference caused by the complex operating conditions of the heating device, thereby improving the accuracy of fault detection.

[0036] Considering that the timing relationship between the resistance values ​​of a thermistor at multiple heating levels will become abnormal under abnormal operating conditions compared to normal operation, in practical implementation, the coordination characteristics between the timing data of the resistance values ​​at multiple heating levels can be used to determine whether the heating device has malfunctioned.

[0037] Optionally, the first target feature vector includes at least one first component. In one embodiment, the feature extraction of the first resistance timing data in step S102 above to obtain the first target feature vector includes: At least one collaborative feature is extracted from the resistance time series data under multiple heating levels to obtain at least one first component; wherein, the collaborative feature is used to characterize the consistency of change among the resistance time series data under multiple heating levels; each first component characterizes the feature value of the resistance time series data under multiple heating levels in a collaborative feature dimension.

[0038] Optionally, the above-mentioned at least one synergistic feature may include at least one of the following: the degree of linear correlation of resistance changes, the degree of fluctuation of the resistance difference sequence, the degree of fluctuation of the resistance ratio sequence, and the difference in the slope of the linear trend of resistance. The resistance difference sequence can include the resistance difference between the thermistors at multiple heating levels at the same time point; the resistance ratio sequence can include the ratio between the resistance of the thermistors at multiple heating levels at the same time point. When the number of heating levels is greater than 2, the above difference can be the average of the resistance difference under all pairwise combinations of heating levels; the above ratio can be the average of the resistance ratio under all pairwise combinations of heating levels.

[0039] In practice, the degree of linear correlation of the above resistance changes can be characterized by the Pearson correlation coefficient. Under normal operating conditions, the resistance values ​​of multiple ranges are highly positively correlated, and the correlation coefficient can be close to 1; however, under abnormal operating conditions, an abnormal resistance value in one range will cause the correlation coefficient to decrease.

[0040] The degree of fluctuation in the above resistance difference sequence can be characterized by the standard deviation. Under normal operating conditions, the resistance differences of multiple ranges are stable, and the standard deviation of the resistance difference sequence is small; however, under abnormal operating conditions, abnormal fluctuations in a certain range of resistance will increase the standard deviation of the resistance difference sequence.

[0041] The degree of fluctuation in the above resistance ratio sequence can be characterized by the coefficient of variation of the resistance ratio sequence. This coefficient of variation can be the ratio of the standard deviation to the mean of the resistance ratio sequence. Under normal operating conditions, the resistance ratios of multiple ranges are stable, and the coefficient of variation of the resistance ratio sequence is low; however, under abnormal operating conditions, the fluctuations of the resistance ratios of multiple ranges intensify, and the coefficient of variation of the resistance ratio sequence increases.

[0042] The difference in the slope of the linear trend of resistance values ​​can be characterized by the absolute difference in the slope of the linear trend of resistance values. For each heating level, the slope of the linear trend of resistance values ​​can be obtained by linearly fitting the resistance time-series data of the thermistor under a single heating level. Under normal operating conditions, the slopes of the linear trend of resistance values ​​across multiple levels are consistent, and the difference in slope is close to zero; however, under abnormal operating conditions, the slope of the linear trend of resistance values ​​in a certain level is abnormal, which will increase the difference in slope.

[0043] In this way, by fully utilizing the characteristic that the cooperative relationship between the resistance value timing data of the thermistor in multiple heating levels will be abnormal under abnormal working conditions, and by extracting the first target feature vector based on at least one cooperative feature between the resistance value timing data of multiple heating levels, the accuracy of fault detection of heating equipment can be improved.

[0044] Optionally, the first target feature vector includes at least one of the following: at least one second component, at least one third component, and at least one fourth component. In one embodiment, the feature extraction of the first resistance timing data in step S102 above to obtain the first target feature vector includes at least one of the following: At least one distribution statistical feature of the first resistance value time series data is extracted to obtain at least one second component; wherein, the distribution statistical feature is used to characterize the distribution pattern of the first resistance value time series data; each second component characterizes the feature value of the first resistance value time series data in one dimension of the distribution statistical feature; At least one trend evolution feature is extracted from the first resistance value time series data to obtain at least one third component; wherein, the trend evolution feature is used to characterize the trend change law of the first resistance value time series data; each third component characterizes the feature value of the first resistance value time series data in a trend evolution feature dimension; At least one time-dependent feature of the first resistance value time series data is extracted to obtain at least one fourth component; wherein, the time-dependent feature is used to characterize the correlation between the resistance values ​​at adjacent time points in the first resistance value time series data; each fourth component characterizes the feature value of the first resistance value time series data in a time-dependent feature dimension.

[0045] Considering that the resistance time-series data of a thermistor exhibits a stable distribution pattern under normal operating conditions, but this distribution pattern changes significantly under abnormal operating conditions, in practical applications, the statistical characteristics of the thermistor's resistance time-series data can be used to determine whether the heating device has malfunctioned.

[0046] The above-mentioned at least one distribution statistical characteristic may include at least one of the following: overall resistance level, degree of dispersion of resistance fluctuation, degree of relative resistance fluctuation, regularity of resistance change, symmetry of resistance distribution, sharpness of resistance distribution, and degree of dispersion of resistance distribution.

[0047] In practice, the overall resistance level can be represented by the average value. Under normal operating conditions, the average value of the resistance timing data should be stable within the rated range, while under abnormal operating conditions (especially in the early stages of a fault), the average value of the resistance timing data will drift.

[0048] The dispersion of the aforementioned resistance fluctuations can be characterized by variance. Variance is the square of the standard deviation and is more sensitive to large fluctuations in resistance time-series data. Compared to the standard deviation, variance amplifies the impact weight of large fluctuations, making abnormal fluctuations more prominent. Using variance to characterize the dispersion of resistance fluctuations can enhance the ability to identify fault modes with severe fluctuations.

[0049] The relative fluctuation of the resistance value can be characterized by the resistance coefficient of variation. This coefficient of variation is the ratio of the standard deviation to the mean of the resistance value time series data. Under normal operating conditions, the resistance coefficient of variation is low; however, under abnormal operating conditions, the resistance fluctuation intensifies, and the resistance coefficient of variation increases significantly.

[0050] The regularity of the resistance changes mentioned above can be characterized by information entropy. Under normal operating conditions, the resistance changes in the resistance time series data show strong regularity, and the corresponding information entropy value is low; while under abnormal operating conditions, the resistance changes in the resistance time series data are disordered, and the corresponding information entropy value will increase.

[0051] The symmetry of the resistance distribution described above can be characterized by skewness. Under normal operating conditions, the resistance values ​​in the resistance time series data are approximately symmetrically distributed, and the corresponding skewness is close to 0; however, under abnormal operating conditions, the resistance values ​​in the resistance time series data may exhibit a left-skewed or right-skewed distribution, and the absolute value of the corresponding skewness will increase.

[0052] The sharpness of the resistance distribution can be characterized by kurtosis. Under normal operating conditions, the resistance values ​​in the resistance time series data are close to a normal distribution, and the corresponding kurtosis is close to 0; while under abnormal operating conditions, the resistance values ​​in the resistance time series data may exhibit a peaked or flat distribution, and the corresponding kurtosis will increase or decrease.

[0053] The dispersion of the aforementioned resistance distribution can be characterized by the interquartile range (ICM). This ICM can be the difference between the upper and lower quartiles of the resistance time series data. Compared to the range, the ICM is more robust to outliers and more accurately reflects the stability of the resistance distribution. The ICM will increase under abnormal operating conditions compared to normal operating conditions.

[0054] Considering that the resistance trend of a thermistor changes under abnormal operating conditions compared to normal operation, in practice, the trend evolution characteristics of the thermistor's resistance time-series data can be used to determine whether the heating device has malfunctioned.

[0055] The aforementioned trend evolution characteristics may include at least one of the following: linear trend slope characteristics of resistance value, trend stability of resistance value change, degree of curvature of resistance value curve change, and degree of horizontal drift of resistance value.

[0056] In practice, the aforementioned resistance linear trend slope characteristic can be characterized by the resistance linear trend slope. This resistance linear trend slope can be obtained by linearly fitting the resistance time series data using the least squares method. Under normal operating conditions, the resistance linear trend slope should be close to 0; however, under abnormal operating conditions (especially in the early stages of a fault), the resistance linear trend slope may increase or decrease.

[0057] The stability of the aforementioned resistance change trend can be characterized by the monotonicity index. The monotonicity index can be obtained by statistically analyzing the consistency ratio of the resistance change direction at adjacent time points in the resistance time series data. Under normal operating conditions, the monotonicity index is low (mainly fluctuating); while under abnormal operating conditions, continuous monotonic changes may occur, and the monotonicity index will rise abnormally.

[0058] The degree of curvature variation in the resistance curve can be characterized by the standard deviation of curvature. In practice, the curvature value can be calculated point by point for the resistance time series data to obtain a curvature sequence; then, the standard deviation of the curvature sequence can be calculated to obtain the standard deviation of curvature. Compared to the normal operating state, the resistance curve corresponding to the resistance time series data may exhibit severe curvature fluctuations under abnormal operating conditions, and its standard deviation of curvature will increase significantly.

[0059] The severity of the aforementioned resistance level drift can be characterized by the sliding window mean drift. In practice, the resistance time series data can be divided into continuous sliding windows, and the difference between the means of adjacent sliding windows can be calculated to obtain a difference sequence. Then, the standard deviation of the difference sequence can be taken as the sliding window mean drift. Compared to normal operating conditions, the resistance level of the resistance time series data may fluctuate abnormally under abnormal operating conditions, and the sliding window mean drift will increase.

[0060] Considering the time-dependent nature of the thermistor's resistance timing data, and the difference in timing dependence between normal and abnormal operating states, in practical implementation, the timing dependence characteristics of the thermistor's resistance timing data can be used to determine whether the heating device has malfunctioned.

[0061] The aforementioned time-dependent feature may include at least one of the following: resistance autocorrelation feature, average level of local resistance fluctuations, stability of local resistance fluctuations, frequency of resistance fluctuations, and duration of sustained resistance stability.

[0062] In practice, the aforementioned resistance autocorrelation characteristics can be characterized by the first-order autocorrelation coefficient. The first-order autocorrelation coefficient measures the degree of linear correlation between the resistance value at each time point in the resistance time series data and the resistance value at the previous time point. Under normal operating conditions, the resistance value changes continuously in the resistance time series data, resulting in a high autocorrelation coefficient; however, under abnormal operating conditions, the resistance value in the resistance time series data may undergo abrupt changes, leading to a decrease in the autocorrelation coefficient.

[0063] The average level of the aforementioned local resistance fluctuations can be characterized by the mean of the sliding window standard deviation sequence. In practice, the resistance time series data can be divided into continuous sliding windows, and the standard deviation of the resistance within each sliding window can be calculated to obtain a standard deviation sequence. Then, the mean of the standard deviation sequence can be taken as the mean of the sliding window standard deviation sequence. Compared to the normal operating state, the local resistance fluctuations are aggravated under abnormal operating conditions, and the mean of the sliding window standard deviation sequence will increase.

[0064] The frequency of the aforementioned resistance fluctuations can be characterized by the standard deviation of the sliding window standard deviation sequence. In practice, the resistance time series data can be divided into continuous sliding windows, and the standard deviation of the resistance within each sliding window can be calculated to obtain a standard deviation sequence. Then, the standard deviation of the standard deviation sequence can be calculated to obtain the standard deviation of the sliding window standard deviation sequence. Compared to the normal operating state, the local fluctuations of the resistance are unstable under abnormal operating conditions, and the standard deviation of the sliding window standard deviation sequence will increase.

[0065] The frequency of the aforementioned resistance fluctuations can be characterized by local extremum density. In practical implementation, the local extremum density can be obtained by counting the number of local extrema occurring in the first resistance time series data per unit time. Compared to normal operating conditions, frequent resistance fluctuation events may occur under abnormal operating conditions (especially in the early stages of a fault), and the local extremum density will increase.

[0066] The duration of sustained resistance stability can be characterized by the maximum sustained stability time of the resistance value. In practical implementation, the maximum duration for which the resistance value remains continuously within its mean ± δ range (δ can be determined based on the resistance value variation coefficient mentioned above) in the resistance time series data can be statistically analyzed to obtain the sustained stability time of the resistance value. Compared to normal operating conditions, the sustained stability time of the resistance value will be shorter under abnormal operating conditions.

[0067] In this way, by making full use of the characteristic that the resistance distribution, resistance trend and / or resistance time dependence of thermistors will change under abnormal working conditions compared with normal working conditions, and by using the first target feature vector obtained by extracting at least one distribution statistical feature, at least one trend evolution feature and / or at least one time dependence feature from the first resistance time series data, the accuracy of fault detection of heating equipment can be improved.

[0068] Optionally, the aforementioned target fault detection model can be an unsupervised fault detection model constructed based on the kernel density estimation (KDE) algorithm.

[0069] In one embodiment, before inputting the second target feature vector into the target fault detection model in step S103 above to obtain the fault detection result of the heating device, the fault detection method provided in this application embodiment further includes: Multiple first sample feature vectors are obtained; each first sample feature vector is obtained by feature extraction based on a second resistance value time series data of a thermistor; each second resistance value time series data corresponds to a second operating period of the heating device; the thermistor is in normal operating state in each second operating period; Using the kernel density estimation algorithm, the kernel functions centered on each first sample feature vector are smoothly superimposed to obtain a continuous first probability density function; Based on the first probability density function and the target value of the bandwidth, the second probability density function is determined; in the first probability density function, the bandwidth is an undetermined coefficient. A target fault detection model is constructed based on the second probability density function.

[0070] In practical implementation, the timing data of the second resistance value of the thermistor during multiple second operating periods of the heating device can be acquired. These multiple second operating periods can be different operating periods of equal duration. For each second operating period, feature extraction can be performed on the timing data of the second resistance value to obtain a second sample feature vector. The aforementioned first sample feature vector can be either the second sample feature vector or a sample feature vector obtained after dimensionality reduction of the second sample feature vector.

[0071] In this design, each component of the second sample feature vector represents a feature value of the second resistance time series data in one feature dimension. The process of feature extraction for the second resistance time series data can be referred to the process of feature extraction for the first resistance time series data described above, and will not be repeated here.

[0072] Next, for each first sample feature vector, a kernel function can be constructed with the first sample feature vector as the center; and the kernel density estimation algorithm can be used to smoothly superimpose the kernel functions corresponding to multiple first sample feature vectors to obtain a continuous first probability density function.

[0073] Then, the target value of the bandwidth can be substituted into the first probability density function to obtain the second probability density function. Optionally, the target value of the bandwidth can be a preset value, or it can be obtained by filtering from multiple candidate values ​​of the bandwidth based on the probability density values ​​corresponding to multiple first sample feature vectors and candidate values ​​of the bandwidth.

[0074] Finally, a target fault detection model can be constructed based on the second probability density function. Optionally, the target fault detection model can be constructed solely based on the second probability density function. In actual implementation, the target fault detection model can output the probability density value corresponding to the input second target feature vector; furthermore, the user or fault detection device can determine the fault detection result of the heating device based on the comparison result between the probability density threshold and the probability density value output by the target fault detection model.

[0075] Optionally, the target fault detection model can also be constructed based on the second probability density function and the probability density threshold. In actual implementation, the target fault detection model can determine the probability density value corresponding to the second target feature vector based on the input second target feature vector, and can output the fault detection result of the heating device based on the comparison result of the probability density value and the probability density threshold.

[0076] Thus, by using the kernel density estimation algorithm to construct an unsupervised fault detection model, the target fault detection model can be trained using only the resistance time series data of the thermistors during multiple normal working periods as training samples, without any fault labels. In addition, the kernel density estimation algorithm can adaptively fit the true distribution of normal samples, which can endow the model with strong noise resistance and working condition adaptability, thereby improving the robustness of the target fault detection model and enabling the fault detection model to improve the fault detection accuracy of heating devices.

[0077] Considering that extracting multiple features from the thermistor's resistance time-series data results in high-dimensional feature vectors, with potential information overlap or redundancy among some features, and that high-dimensional features increase the computational complexity of subsequent fault detection models, this paper proposes a method to further improve the robustness of the target fault detection model. This method involves dimensionality reduction of the extracted feature vectors during the model construction phase, and then using the reduced feature vectors to build the target fault detection model.

[0078] In one embodiment, obtaining multiple first sample feature vectors includes: Obtain the target covariance matrix of the first sample feature matrix; wherein, the first sample feature matrix includes multiple second sample feature vectors; each second sample feature vector is obtained based on feature extraction of a second resistance time series data; the target covariance matrix is ​​determined based on the covariance between each feature dimension vector in the first sample feature matrix; each component in the same feature dimension vector corresponds one-to-one to represent the feature value of each second resistance time series data in the same feature dimension; Eigenvalues ​​are decomposed into the target covariance matrix to obtain multiple target eigenvalues ​​and a third sample feature vector corresponding to each target eigenvalue. The cumulative variance contribution rate of the first k target feature values ​​is determined sequentially in descending order of their target feature values; k is a positive integer and is less than the total number of target feature values. In response to the cumulative variance contribution rate of the first k target feature values ​​being greater than the cumulative variance contribution rate threshold, the third sample feature vectors corresponding to the first k target feature values ​​are respectively determined as the first sample feature vectors.

[0079] In practice, multiple second-sample feature vectors obtained from feature extraction can be combined into a first-sample feature matrix. Then, the covariance between every two feature dimension vectors in the first-sample feature matrix can be calculated to obtain the target covariance matrix.

[0080] Then, eigenvalue decomposition can be performed on the target covariance matrix. If the cumulative variance contribution rate of the first k target eigenvalues ​​is greater than a cumulative variance contribution rate threshold, then the third sample eigenvectors corresponding to the first k target eigenvalues ​​can be determined as the first sample eigenvectors.

[0081] Thus, by using eigenvalue decomposition (also known as orthogonal transformation) of the target covariance matrix, the original high-dimensional features can be mapped to a low-dimensional space, thereby eliminating linear correlations between some features and avoiding the impact of multicollinearity on fault detection results. Simultaneously, based on the screening criterion that the cumulative variance contribution rate of the first k target feature values ​​is greater than a set threshold, low-variance components can be discarded, achieving denoising and improving feature quality. Furthermore, the principal component directions are retained according to the cumulative variance contribution rate, ensuring that the dimensionality-reduced feature vectors can still effectively represent fault-sensitive information. Therefore, the target fault detection model built based on the dimensionality-reduced feature vectors can have stronger robustness.

[0082] In one embodiment, obtaining the target covariance matrix of the first sample feature matrix includes: Obtain the mean vector and standard deviation vector of the first sample feature matrix; where each component of the mean vector represents the average value of a feature dimension vector in the first sample feature matrix; and each component of the standard deviation vector represents the standard deviation of a feature dimension vector in the first sample feature matrix. The feature matrix of the first sample is standardized based on the mean vector and standard deviation vector. The target covariance matrix is ​​determined based on the covariance between the feature dimension vectors in the standardized first sample feature matrix.

[0083] In practice, for each feature dimension vector in the first sample feature matrix, the average value of each component in the feature dimension vector can be calculated; the average values ​​corresponding to each feature dimension vector can be combined to obtain the above mean vector.

[0084] Similarly, for each feature dimension vector in the first sample feature matrix, the standard deviation of each component in the feature dimension vector can be calculated; the standard deviations corresponding to each feature dimension vector can be combined to obtain the above standard deviation vector.

[0085] Next, for each feature dimension vector in the first sample feature matrix, the corresponding components of the mean vector and the standard deviation vector mentioned above can be used to standardize the feature dimension vector, resulting in a standardized feature dimension vector. The standardized first sample feature matrix can include each standardized feature dimension vector.

[0086] Then, the covariance between every two feature dimension vectors in the standardized first sample feature matrix can be calculated to obtain the target covariance matrix.

[0087] In this way, each feature dimension vector can be mapped to a standard distribution with zero mean and unit variance, thereby eliminating dimensional differences between different vehicles and operating conditions, and ensuring the comparability of each feature dimension in covariance calculation. Dimensionality reduction of the feature vectors based on this can improve the quality of the reduced feature vectors, making the target fault detection model built based on the reduced feature vectors more robust.

[0088] In one embodiment, before determining the second probability density function based on the first probability density function and the target value of the bandwidth, the fault detection method provided in this application further includes: Obtain multiple candidate values ​​for bandwidth; For each candidate value of bandwidth, the following steps are performed to obtain the log-likelihood value corresponding to the candidate value: based on the second probability density function and the candidate value of bandwidth, determine the probability density function corresponding to the candidate value; using the probability density function corresponding to the candidate value, determine the probability density value of each first sample feature vector respectively; based on the probability density values ​​of multiple first sample feature vectors, determine the log-likelihood value corresponding to the candidate value. Select the candidate value with the largest log-likelihood value from multiple candidate values ​​as the target value.

[0089] Optionally, the bandwidth can be a smoothing parameter in the probability density function, used to control the degree of broadening of the probability density function.

[0090] In actual implementation, multiple discrete bandwidth candidate values ​​can be obtained by uniformly sampling within a preset bandwidth range at a preset step size; alternatively, multiple bandwidth empirical values ​​can be set as bandwidth candidate values ​​according to the actual application scenario.

[0091] Next, for each candidate value of the bandwidth, a probability density function corresponding to the candidate value can be constructed based on the second probability density function and the candidate value of the bandwidth; and the probability density value of each first sample feature vector can be determined by using the probability density function corresponding to the candidate value; then the sum of the natural logarithmic values ​​of the probability density values ​​of each first sample feature vector can be determined as the log-likelihood value corresponding to the candidate value.

[0092] Finally, the candidate value with the largest log-likelihood value can be selected from multiple candidate values ​​as the target value for bandwidth.

[0093] Thus, by maximizing the log-likelihood value to achieve the adaptive selection of the optimal bandwidth, it can be ensured that the constructed target fault detection model can accurately characterize the probability density distribution of normal resistance time series data, making it easier for abnormal resistance time series data to fall into the low probability density region, thereby improving the fault detection accuracy of the heating device.

[0094] In one embodiment, the above-mentioned construction of the target fault detection model based on the second probability density function includes: Obtain the target probability density threshold; the target probability density threshold is used to determine whether the probability density value determined based on the second probability density function indicates a malfunction in the heating device; A target fault detection model is constructed based on the second probability density function and the target probability density threshold.

[0095] Optionally, the target probability density threshold can be a preset value, or it can be selected from multiple candidate probability density thresholds based on the fault detection accuracy of multiple candidate probability density thresholds.

[0096] When the target fault detection model is constructed based on the second probability density function and the target probability density threshold, after the above-mentioned second target feature vector is input into the target fault detection model, the target fault model can use the second probability density function to determine the probability density value of the second target feature vector.

[0097] The target fault detection model can then compare the probability density value with the target probability density threshold. If the probability density value is lower than the target probability density threshold, it can output a fault detection result to characterize the failure of the heating device.

[0098] Thus, by introducing a target probability density threshold, the target fault detection model built based on the second probability density function can have fault discrimination function. It can automatically output fault detection results based on the input feature vector without external intervention, which can improve the fault detection efficiency of heating devices.

[0099] In one embodiment, obtaining the target probability density threshold includes: Multiple fourth sample feature vectors and multiple candidate probability density thresholds are obtained; wherein, each fourth sample feature vector is obtained by feature extraction based on a third resistance value time series data of a thermistor; each third resistance value time series data corresponds to a third operating period of the heating device; the thermistor is in an abnormal working state in at least part of the third operating period; For each fourth sample feature vector, the probability density value of the fourth sample feature vector is determined based on the second probability density function; For each candidate probability density threshold, the fault detection accuracy of the candidate probability density threshold is determined based on the comparison results between the candidate probability density threshold and the probability density values ​​of multiple fourth sample feature vectors, as well as the actual working state of the thermistor corresponding to the multiple fourth sample feature vectors. The candidate probability density threshold with the highest fault detection accuracy is selected from multiple candidate probability density thresholds as the target probability density threshold.

[0100] In practical implementation, the timing data of the third resistance value of the thermistor during multiple third operating periods of the heating device can be acquired. For each third operating period, feature extraction can be performed on the timing data of the third resistance value to obtain a fourth sample feature vector. Optionally, the multiple third operating periods can be different operating periods of equal time length. Each component in the fourth sample feature vector can represent the feature value of the third resistance value timing data in one feature dimension. Then, for each fourth sample feature vector, the probability density value corresponding to that fourth sample feature vector can be determined based on the aforementioned second probability density function.

[0101] Then, for each fourth sample feature vector, its probability density value can be compared with the current candidate probability density threshold: if the probability density value is less than the threshold, the thermistor and the corresponding fourth sample feature vector can be judged as an abnormal working state; if the probability density value is greater than or equal to the threshold, the thermistor and the corresponding fourth sample feature vector can be judged as a normal working state. Furthermore, the working state determination result can be compared with the actual working state corresponding to the fourth sample feature vector, and the percentage of all fourth sample feature vectors whose working state determination results match the actual working state can be calculated to obtain the fault detection accuracy of the candidate probability density threshold.

[0102] Finally, the candidate probability density threshold with the highest fault detection accuracy can be selected from multiple candidate probability density thresholds as the target probability density threshold.

[0103] Thus, by traversing multiple candidate probability density thresholds and selecting the threshold with the highest fault detection accuracy as the target probability density threshold, it can be ensured that the selected target probability density threshold has the optimal fault discrimination capability. This enables the constructed target fault detection model to accurately distinguish between the normal working state and the abnormal working state of the thermistor, and effectively reduce the false alarm rate and false negative rate of the heating device fault, thereby improving the fault detection accuracy of the heating device.

[0104] In one embodiment, before inputting the second target feature vector into the target fault detection model in step S103 above to obtain the fault detection result of the heating device, the fault detection method provided in this application embodiment further includes: Obtain the second sample feature matrix; the second sample feature matrix includes the third sample feature vectors corresponding to the first k target feature values ​​of the target covariance matrix; the cumulative variance contribution rate of the first k target feature values ​​is greater than the cumulative variance contribution rate threshold; The second target feature vector is determined based on the product of the first target feature vector and the second sample feature matrix.

[0105] In practice, during the inference phase after the target fault detection model is built, the dimensionality-reduced feature vector determined during the target fault detection model building phase—the third sample feature vector corresponding to the first k target feature values ​​whose cumulative variance contribution rate is greater than the cumulative variance contribution rate threshold—can be used to reduce the dimensionality of the first target feature vector to obtain the second target feature vector.

[0106] In this way, by extracting multiple features from the time series data of the first resistance value of the thermistor, it is possible to eliminate the information overlap or redundancy that may exist between some features, and avoid the increase of the computational complexity of the target fault detection model by high-dimensional features, thereby improving the fault detection accuracy and efficiency of the heating device.

[0107] In one embodiment, determining the second target feature vector based on the product of the first target feature vector and the second sample feature matrix includes: The first target feature vector is standardized based on the mean vector and standard deviation vector of the first sample feature matrix; The second target feature vector is determined by multiplying the standardized first target feature vector with the second sample feature matrix.

[0108] In practice, before using the dimensionality-reduced feature vector to reduce the dimensionality of the first target feature vector, the mean vector and standard deviation vector of the first sample feature matrix determined in the target fault detection model construction stage can be used to standardize the first target feature vector. For specific implementation, please refer to the above-mentioned process of standardizing the first sample feature matrix.

[0109] In this way, the first target feature vector can be mapped to a standard distribution with zero mean and unit variance, thereby eliminating the dimensional differences between different vehicles and operating conditions, and ensuring the accuracy of fault detection using the target fault detection model.

[0110] In one embodiment, after inputting the second target feature vector into the target fault detection model in step S103 above to obtain the fault detection result of the heating device, the fault detection method provided in this application embodiment further includes: The target fault detection model is updated based on the timing data of the first resistance value and the actual working state of the thermistor during the first operating period.

[0111] In practice, the aforementioned multiple first sample feature vectors can be updated based on the timing data of the first resistance value, and the updated multiple first sample feature vectors can be used to update the aforementioned first probability density function. For the specific update process, please refer to the determination process of the first probability density function, which will not be repeated here.

[0112] If the thermistor is in normal operating condition during the first operating phase, the target value of the bandwidth can be updated using the updated first sample feature vectors. Then, the second probability density threshold can be updated based on the updated target value of the bandwidth and the first probability density threshold. For details on the update process, please refer to the process for determining the target value of the bandwidth and the second probability density threshold, which will not be repeated here.

[0113] Alternatively, the aforementioned multiple fourth sample feature vectors can be updated based on the timing data of the first resistance value. Furthermore, the updated fourth sample feature vectors and the actual operating state of the thermistor during the first operating period can be used to update the aforementioned target probability density threshold. For details on the update process, please refer to the process for determining the target probability density threshold, which will not be elaborated here.

[0114] Thus, a closed-loop iterative optimization mechanism for the target fault detection model was established: the resistance time series data and actual working status of the thermistor were continuously collected, and parameters such as the probability density function, the target value of the bandwidth, and the target probability density threshold in the target fault detection model were periodically fed back to optimize them, forming a data-driven continuous improvement closed loop. This enabled the target fault detection model to adapt to the long-term operating conditions of the heating device, thereby further improving the robustness of the target fault detection model.

[0115] In one embodiment, step S101 above: obtaining the first resistance value timing data of the thermistor during the first operating period of the heating device, may include: Obtain the original resistance timing data of the thermistor during the first operating phase; The original resistance timing data is preprocessed to obtain the first resistance timing data.

[0116] Considering the complex and variable operating conditions of heating devices, the original resistance timing data may suffer from problems such as noise interference, missing data, time alignment deviations, and inconsistent dimensions. In actual implementation, at least one of the following preprocessing steps can be performed on the original resistance timing data: Abnormal transition points and invalid data segments in the original resistance timing data are removed, and a sliding window mean filter is applied to the original resistance timing data; this can effectively eliminate noise interference in the resistance timing data.

[0117] Linear interpolation is used to fill in missing data: Linear interpolation is performed using valid resistance data before and after the missing time period; in this way, the missing resistance time series data caused by communication interruption, abnormal acquisition equipment, etc. can be filled, ensuring the continuity and integrity of the resistance time series data.

[0118] Align the resistance timing data of multiple heating levels by timestamp; this ensures the synchronization of resistance values ​​at the same time point and facilitates the extraction of collaborative features between the resistance timing data of multiple heating levels in the future.

[0119] The original resistance time series data is standardized by using the mean and standard deviation of the original resistance time series data. This maps the original resistance time series data to a standard distribution with zero mean and unit variance, thereby eliminating the dimensional differences between different vehicles and operating conditions and improving the consistency of subsequent feature extraction.

[0120] The determination of the aforementioned abnormal transition points can employ a neighborhood consistency check, eliminating only isolated single-point pulse noise outside the normal neighborhood to avoid falsely clearing fault signals. The elimination of invalid data segments can be based on a safety boundary set according to the physical measurement limits of the data acquisition system, eliminating only abnormal sensor values ​​exceeding this boundary to prevent falsely clearing of extreme resistance changes caused by faults. The sliding window width can be set based on the minimum duration of the fault characteristic to ensure that the trend of resistance changes caused by the fault is not smoothly eliminated.

[0121] Similarly, the aforementioned second and third resistance value timing data can also be obtained by performing at least one of the aforementioned preprocessing steps on the corresponding original resistance value timing data. Optionally, the aforementioned first, second, and third running segments can be different running segments with equal time lengths.

[0122] Optionally, the heating device described above can be installed in a vehicle, and the heating device can be configured with two heating levels. The thermistor described above can be a positive temperature coefficient (PTC) resistor. Please refer to [link to relevant documentation]. Figure 3 The fault detection method provided in this application embodiment may further include the following steps: 1. Data Acquisition and Preprocessing (1) According to the preset sampling period, the resistance data of the PTC resistor at the two heating levels during the second running period of the heating device are collected respectively to obtain the timing data of the PTC resistance value at the two levels during the second running period. (2) Perform the following preprocessing on the timing data of the two PTC resistance values ​​to obtain the timing data of the second resistance value of the PTC resistor during the second operating period: Data cleaning: Remove abnormal jump points and invalid data segments, and use sliding window mean filtering to eliminate high-frequency noise interference.

[0123] Data alignment: The timing data of the two PTC resistance values ​​are aligned according to the timestamp to ensure the synchronization of the two resistance values ​​at the same time point, which facilitates subsequent collaborative feature extraction.

[0124] Standardization processing: The Z-score standardization method is used to map the time series data of each resistance level to a standard distribution with zero mean and unit variance, eliminating the dimensional differences between different vehicles and different operating conditions, and ensuring the consistency of subsequent feature extraction.

[0125] The aforementioned second resistance timing data includes preprocessed timing data for two PTC resistance values. By repeatedly executing the above data acquisition and preprocessing steps, multiple second resistance timing data sets can be obtained. The second running segments corresponding to these multiple second resistance timing data sets can be different running segments of equal duration.

[0126] 2. Multidimensional feature extraction and dimensionality reduction (1) For each second resistance value time series data, 20 features of the second resistance value time series data are extracted to obtain the second sample feature vector.

[0127] The 20 features mentioned above may include mean, variance, resistance coefficient of variation, information entropy, skewness, kurtosis, interquartile range, linear trend slope, monotonicity index, standard deviation of curvature, mean drift of sliding window, first-order autocorrelation coefficient, mean of sliding window standard deviation sequence, standard deviation of sliding window standard deviation sequence, local extremum density, duration of resistance stability, Pearson correlation coefficient, standard deviation of resistance difference sequence, coefficient of variation of resistance ratio sequence, and trend slope difference.

[0128] Among them, the Pearson correlation coefficient, the standard deviation of the difference sequence, the coefficient of variation of the ratio sequence, and the difference in trend slope are four features that are collaborative features of the two PTC resistance value time series data in the second resistance value time series data. The remaining 16 features need to be extracted separately for the two PTC resistance value time series data in the second resistance value time series data. Thus, the resulting second sample feature vector can be 36. A 1-dimensional vector.

[0129] It is worth mentioning that by introducing curve shape features such as linear trend slope and curvature standard deviation, the nonlinear law of PTC resistance changing with temperature can be specifically characterized. By introducing two-level synergistic features and analyzing the synchronicity, difference, and proportional relationship between the two resistance levels, the reliability of fault detection can be improved. By introducing the resistance stability duration feature, the stability characteristics of PTC resistance within a specific range can be specifically characterized. By introducing information entropy features, the information complexity of resistance time series data can be specifically characterized, enabling the analysis of fault evolution laws from an information theory perspective.

[0130] Therefore, a complete feature engineering scheme was designed to address the nonlinear characteristics of PTC resistance changing with temperature. This feature engineering scheme is specifically designed for the nonlinear characteristics of PTC resistance, such as temperature sensitivity, step response characteristics, and long-term aging trend. It can effectively characterize the inherent laws of PTC resistance change and fully explore the fault evolution information contained in the PTC resistance time series data.

[0131] (2) Principal Component Analysis (PCA) was used to reduce the dimensionality of the second sample feature vector.

[0132] The multiple second sample feature vectors obtained by extracting 20 features from multiple second resistance time series data can constitute 36 The first sample feature matrix is ​​n-dimensional; n is the total number of multiple second sample feature vectors.

[0133] Z-score standardization is performed on the first sample feature matrix using the mean vector and standard deviation vector to eliminate dimensional differences. Specifically, the first sample feature matrix has a value of 36. In the case of n dimensions, it includes 36 feature dimension vectors, each of which includes n components; its mean vector and standard deviation vector also include 36 components, and each component in the mean vector / standard deviation vector represents the mean / standard deviation of the n components in a feature dimension vector.

[0134] Calculate the target covariance matrix of the standardized first sample feature matrix (36) (36 dimensions), and perform eigenvalue decomposition on the target covariance matrix to obtain 36 target eigenvalues ​​and the third sample feature vector corresponding to each target eigenvalue; Based on the order of multiple target feature values ​​from largest to smallest, determine the first k target feature values ​​whose cumulative variance contribution rate is greater than or equal to the cumulative variance contribution rate threshold; k can be understood as the number of principal components, where k is a positive integer and its value range can be 1 < k < 36. The first sample feature matrix carrying the original features is projected onto the principal component space corresponding to the first k target feature values ​​to obtain the dimensionality-reduced feature vectors, which are the third sample feature vectors corresponding to the first k target feature values. Each third sample feature vector includes 36 components.

[0135] Thus, principal component analysis (PCA) is used to reduce the dimensionality of high-dimensional features, eliminating redundant features and noise interference, and retaining the core feature components that are sensitive to faults. PCA dimensionality reduction reduces the feature dimension, decreasing the computational complexity of the subsequent KDE detection model; it also achieves feature denoising, improving the robustness of the KDE detection model and providing high-quality feature input for subsequent fault detection.

[0136] 3. KDE detection model construction and parameter optimization (1) Obtain multiple first sample feature vectors, that is, the third sample feature vectors corresponding to the first k target feature values; (2) Using the kernel density estimation algorithm, the kernel functions centered on each of the first sample feature vectors are smoothly superimposed to obtain a continuous first probability density function; the first probability density function contains unknown coefficient bandwidth; (3) A dual-grid search strategy is adopted to determine the optimal bandwidth value and the optimal probability density threshold: Bandwidth grid search: Traverse different bandwidth values, calculate the goodness of fit of each bandwidth value to multiple first sample feature vectors, and select the bandwidth value with the best goodness of fit (i.e. the target value of the bandwidth mentioned above); optionally, the goodness of fit can be characterized by the log-likelihood value. Threshold grid search: Traverse different probability density thresholds, verify the fault detection accuracy of each probability density threshold on multiple fourth sample feature vectors, and select the probability density threshold with the best fault detection accuracy (i.e. the target probability density threshold mentioned above).

[0137] (4) Determine the second probability density function based on the first probability density function and the optimal bandwidth value; construct the KDE detection model based on the second probability density function and the optimal probability density threshold, i.e. the above target fault detection model.

[0138] Thus, an unsupervised fault detection model can be constructed using the kernel density estimation (KDE) algorithm, allowing for model training without fault labels. By optimizing the bandwidth and probability density threshold of the KDE detection model through dual-grid search, adaptive calibration of the model parameters is achieved, enabling rapid deployment across vehicle models and component batches.

[0139] The specific implementation process of this embodiment can be found in the relevant descriptions in the above embodiments, and will not be repeated here.

[0140] Please see Figure 4In the inference phase after the KDE detection model is constructed, the fault detection method provided in this application embodiment may further include the following steps: 1. Load the completed KDE detection model; 2. Extract the above 20 features from the first resistance value time series data of the PTC resistor to obtain the first target feature vector (36). (1-dimensional); The first resistance value timing data can be obtained by preprocessing the original resistance value timing data of the thermistor during the first operating period; 3. Standardize the first target feature vector using the mean vector and standard deviation vector of the first sample feature matrix; 4. Using the second sample feature matrix (36) k-dimensional) on the standardized first target feature vector (36) Dimensionality reduction is performed on the 1-dimensional vector to obtain the second target feature vector (k). 1-dimensional); the second sample feature matrix may include the third sample feature vector (36) corresponding to the first k target feature values ​​mentioned above. 1D); 5. Input the second target feature vector into the KDE detection model to obtain the fault detection result; (1) The KDE detection model determines the probability density value corresponding to the second target feature vector based on the second probability density function, and compares the probability density value with the target probability density threshold; (2) If the probability density value is less than the target probability density threshold, the KED detection model outputs a fault detection result including fault warning and feature record. The feature record can be used for fault tracing. If the probability density value is greater than or equal to the target probability density threshold, the KED detection model determines that the thermistor is in normal operating condition. In this case, the fault detection result may not be output, or the fault detection result may be output to indicate that the thermistor is in normal operating condition.

[0141] 6. Continuously collect the resistance value timing data and actual working status of PTC resistors, and regularly provide feedback to optimize the parameters of the KDE detection model, forming a data-driven continuous improvement closed loop.

[0142] The specific implementation process of this embodiment can be found in the relevant descriptions in the above embodiments, and will not be repeated here.

[0143] The fault detection method for heating devices provided in this application utilizes deep feature engineering based on the resistance characteristics of PTC (Potentially Transmitted Chloride) cells to fully exploit the fault evolution patterns in PTC resistance time-series data, achieving high-precision fault detection with minimal data requirements. Furthermore, the use of kernel density estimation (KDE) to construct an unsupervised fault detection model effectively avoids the model training difficulties caused by scarce fault labels and overcomes the limitations of model generalization ability resulting from parameter calibration. In addition, a closed-loop iterative feedback mechanism continuously optimizes the fault detection model parameters. Compared to existing technologies, the technical solution of this application achieves significant improvements in feature depth, model generalization ability, and engineering deployment feasibility, making it more suitable for practical application scenarios of fault detection in heating devices for new energy vehicles.

[0144] This application also provides a fault detection device for a heating device, including a processor and a memory. The memory stores a program or instructions that can be run on the processor. When the program or instructions are executed by the processor, they implement the various steps of the above-described fault detection method for the heating device and achieve the same technical effect. To avoid repetition, they will not be described again here.

[0145] Figure 4 To illustrate the hardware structure of the fault detection device for the heating device according to the embodiments of this application, the fault detection device for the heating device may include: The processor 401 can be implemented using a general-purpose central processing unit (CPU), microprocessor, application specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application. The memory 402 can be implemented as a read-only memory (ROM), static storage device, dynamic storage device, or random access memory (RAM). The memory 402 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 402 and is called and executed by the processor 401 to execute the fault detection method of the heating device in the embodiments of this application. Input / output interface 403 is used to implement information input and output; The communication interface 404 is used to enable communication between the fault detection device of this heating device and other devices. Communication can be achieved through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.). Bus 405 transmits information between various components of the fault detection device for the heating device (e.g., processor 401, memory 402, input / output interface 403, and communication interface 404); The processor 401, memory 402, input / output interface 403 and communication interface 404 are connected to each other via bus 405 within the fault detection device of the heating device.

[0146] The fault detection device for the heating device provided in this application embodiment can implement all the steps of the above-described fault detection method embodiment for the heating device and achieve the same technical effect. To avoid repetition, it will not be described again here.

[0147] This application also provides a storage medium storing a program or instructions. When the program or instructions are executed by a processor, they implement the various steps of the above-described fault detection method for the heating device and achieve the same technical effect. To avoid repetition, they will not be described again here.

[0148] The processor can be the processor in the fault detection device of the heating device in the above embodiment. The storage medium can be implemented using computer read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk, etc.

[0149] This application embodiment also provides a chip, which includes a processor and a communication interface. The communication interface and the processor are coupled. The processor is used to run programs or instructions to implement the various steps of the above-described embodiment of the fault detection method for the heating device, and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0150] It should be understood that the chip mentioned in the embodiments of this application may also be referred to as a system-on-a-chip, system chip, chip system, or system-on-a-chip, etc.

[0151] This application provides a computer program product, which is stored in a storage medium and executed by at least one processor to implement the various steps of the above-described embodiment of the fault detection method for heating device, and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0152] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not delete other identical elements present in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of this application is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0153] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a computer software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods of the various embodiments of this application.

[0154] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A method for detecting faults in a heating device, characterized in that, The heating device includes a thermistor; the method includes: Acquire the timing data of the first resistance value of the thermistor during the first operating period of the heating device; Feature extraction is performed on the first resistance time series data to obtain a first target feature vector; each component in the first target feature vector represents the feature value of the first resistance time series data in one feature dimension. The second target feature vector is input into the target fault detection model to obtain the fault detection result of the heating device; the second target feature vector is determined based on the first target feature vector.

2. The method as described in claim 1, characterized in that, Before inputting the second target feature vector into the target fault detection model to obtain the fault detection result of the heating device, the method further includes: Multiple first sample feature vectors are obtained; wherein each first sample feature vector is obtained by feature extraction based on a second resistance value time series data of the thermistor; each second resistance value time series data corresponds to a second operating period of the heating device; the thermistor is in normal working state in each second operating period. Using a kernel density estimation algorithm, the kernel functions centered on each of the first sample feature vectors are smoothly superimposed to obtain a continuous first probability density function; Based on the first probability density function and the target value of the bandwidth, a second probability density function is determined; in the first probability density function, the bandwidth is an undetermined coefficient. The target fault detection model is constructed based on the second probability density function.

3. The method as described in claim 2, characterized in that, The process of obtaining multiple first sample feature vectors includes: Obtain the target covariance matrix of the first sample feature matrix; wherein, the first sample feature matrix includes multiple second sample feature vectors; each second sample feature vector is obtained based on feature extraction of a second resistance time series data; the target covariance matrix is ​​determined based on the covariance between each feature dimension vector in the first sample feature matrix; each component in the same feature dimension vector corresponds to and represents the feature value of each second resistance time series data in the same feature dimension; The target covariance matrix is ​​subjected to eigenvalue decomposition to obtain multiple target eigenvalues ​​and a third sample feature vector corresponding to each target eigenvalue; The cumulative variance contribution rate of the first k target feature values ​​is determined sequentially in descending order of the multiple target feature values; k is a positive integer and is less than the total number of the multiple target feature values. In response to the cumulative variance contribution rate of the first k target feature values ​​being greater than the cumulative variance contribution rate threshold, the third sample feature vectors corresponding to the first k target feature values ​​are respectively determined as the first sample feature vectors.

4. The method as described in claim 3, characterized in that, The step of obtaining the target covariance matrix of the first sample feature matrix includes: Obtain the mean vector and standard deviation vector of the first sample feature matrix; wherein, each component of the mean vector represents the average value of a feature dimension vector in the first sample feature matrix; and each component of the standard deviation vector represents the standard deviation of a feature dimension vector in the first sample feature matrix. The first sample feature matrix is ​​standardized based on the mean vector and the standard deviation vector; The target covariance matrix is ​​determined based on the covariance between the feature dimension vectors in the standardized first sample feature matrix.

5. The method as described in claim 2, characterized in that, Before determining the second probability density function based on the first probability density function and the target value of the bandwidth, the method further includes: Obtain multiple candidate values ​​for the bandwidth; For each candidate value of the bandwidth, the following steps are performed to obtain the log-likelihood value corresponding to the candidate value: based on the second probability density function and the candidate value of the bandwidth, determine the probability density function corresponding to the candidate value; using the probability density function corresponding to the candidate value, determine the probability density value of each first sample feature vector; based on the probability density values ​​of the plurality of first sample feature vectors, determine the log-likelihood value corresponding to the candidate value. The candidate value with the largest log-likelihood value is selected from the plurality of candidate values ​​as the target value.

6. The method as described in claim 2, characterized in that, The construction of the target fault detection model based on the second probability density function includes: Obtain a target probability density threshold; the target probability density threshold is used to determine whether the probability density value determined based on the second probability density function indicates that the heating device has malfunctioned; The target fault detection model is constructed based on the second probability density function and the target probability density threshold.

7. The method as described in claim 6, characterized in that, The process of obtaining the target probability density threshold includes: Multiple fourth sample feature vectors and multiple candidate probability density thresholds are obtained; wherein each fourth sample feature vector is obtained by feature extraction based on a third resistance value time series data of the thermistor; each third resistance value time series data corresponds to a third operating period of the heating device; the thermistor is in an abnormal working state during at least part of the third operating period; For each of the fourth sample feature vectors, the probability density value of the fourth sample feature vector is determined based on the second probability density function; For each candidate probability density threshold, the fault detection accuracy of the candidate probability density threshold is determined based on the comparison results between the candidate probability density threshold and the probability density values ​​of the plurality of fourth sample feature vectors, and the actual working state of the thermistor corresponding to the plurality of fourth sample feature vectors. The candidate probability density threshold with the highest fault detection accuracy is selected from the plurality of candidate probability density thresholds as the target probability density threshold.

8. The method as described in claim 1, characterized in that, Before inputting the second target feature vector into the target fault detection model to obtain the fault detection result of the heating device, the method further includes: Obtain the second sample feature matrix; the second sample feature matrix includes the third sample feature vectors corresponding to the first k target feature values ​​of the target covariance matrix; the cumulative variance contribution rate of the first k target feature values ​​is greater than the cumulative variance contribution rate threshold; The second target feature vector is determined based on the product of the first target feature vector and the second sample feature matrix.

9. The method as described in claim 8, characterized in that, Determining the second target feature vector based on the product of the first target feature vector and the second sample feature matrix includes: The first target feature vector is standardized based on the mean vector and standard deviation vector of the first sample feature matrix; The second target feature vector is determined by multiplying the standardized first target feature vector with the second sample feature matrix.

10. The method as described in claim 1, characterized in that, After inputting the second target feature vector into the target fault detection model to obtain the fault detection result of the heating device, the method further includes: The target fault detection model is updated based on the first resistance value timing data and the actual working state of the thermistor during the first operating period.

11. The method as described in claim 1, characterized in that, The heating device is configured with multiple heating levels; the first resistance timing data includes the resistance timing data of the thermistor at each heating level; the first target feature vector includes at least one first component; The step of extracting features from the first resistance time-series data to obtain a first target feature vector includes: At least one collaborative feature is extracted from the resistance time-series data under the multiple heating levels to obtain at least one first component; wherein, the collaborative feature is used to characterize the consistency of change among the resistance time-series data under the multiple heating levels; each first component characterizes the feature value of the resistance time-series data under the multiple heating levels in a collaborative feature dimension.

12. The method as described in claim 1, characterized in that, The first target feature vector includes at least one of the following: at least one second component, at least one third component, and at least one fourth component; the feature extraction of the first resistance time series data to obtain the first target feature vector includes at least one of the following: At least one distribution statistical feature of the first resistance time series data is extracted to obtain at least one second component; wherein, the distribution statistical feature is used to characterize the distribution pattern of the first resistance time series data; each second component characterizes the feature value of the first resistance time series data in one dimension of the distribution statistical feature. At least one trend evolution feature is extracted from the first resistance time series data to obtain at least one third component; wherein, the trend evolution feature is used to characterize the trend change pattern of the first resistance time series data; each third component represents the feature value of the first resistance time series data in one trend evolution feature dimension; At least one time-dependent feature is extracted from the first resistance time-series data to obtain at least one fourth component; wherein, the time-dependent feature is used to characterize the correlation between resistance values ​​at adjacent time points in the first resistance time-series data; each fourth component characterizes the feature value of the first resistance time-series data in a time-dependent feature dimension.

13. A fault detection device for a heating apparatus, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program, and the computer program, when executed by the processor, implements the method as described in any one of claims 1 to 12.

14. A storage medium, characterized in that, The storage medium stores a computer program that, when executed by a processor, implements the method as described in any one of claims 1 to 12.

Images