Heating ventilation air conditioning system abnormal operation diagnosis method based on time sequence symbolization

Abstract

According to the method for diagnosing abnormal operation of the heating, ventilation and air conditioning system based on time series symbolization, time series symbolization representation is applied to abnormal operation diagnosis of the heating, ventilation and air conditioning system for the first time, and the characteristics of time series data can be highlighted in massive heating data by processing the heating data collected by an intelligent heating platform; the identification mode and trend analysis are more intuitive, and random noise in the time sequence can be smoothed by adopting a symbolization mode, so that small changes in the time sequence are less sensitive and have better robustness. The abnormal operation condition of the heating ventilation air-conditioning system is efficiently diagnosed, and the calculation burden is greatly reduced.

Description

Technical Field

[0001] The present invention relates to a method for diagnosing abnormal operation of a heating, ventilation and air conditioning system, and in particular to a method for diagnosing abnormal operation of a heating, ventilation and air conditioning system based on time series symbolization. The method uses multiple characteristic parameters of a time series as independent variables to diagnose abnormal operation conditions of an abnormal heating, ventilation and air conditioning system. Background Art

[0002] In modern heating, ventilation and air conditioning (HVAC) systems, with the widespread use of equipment and sensors, the system generates a large amount of data. These data include temperature, pressure, flow, heat and other parameters, which together affect the operating efficiency and comfort of the system. However, this huge amount of data makes system optimization and fault diagnosis complicated. In order to extract valuable information from this data, mathematical methods need to be applied for in-depth analysis. Mathematical methods can help identify patterns in the data, predict system performance, optimize operating strategies, and ultimately improve the overall efficiency and reliability of the system. By using statistics, machine learning and data mining techniques, these data can be processed more effectively, the operating effect of the HVAC system can be improved, and intelligent and automated management can be achieved.

[0003] At present, the abnormal operation diagnosis of HVAC systems is still limited to meeting the basic operation requirements of the system. When abnormal operating conditions occur, the existing diagnostic system often relies on the experience analysis of experts. This not only consumes a lot of manpower and material resources, but also leads to problems such as low maintenance efficiency, delayed response and insufficient diagnostic depth. In addition, this diagnostic method based on manual experience is difficult to effectively accumulate sufficient data on similar abnormal conditions in a short period of time, which is prone to misdiagnosis and missed diagnosis, reducing the accuracy of diagnosis and the portability of the system.

[0004] When the HVAC system has an operating fault, the relevant characteristic parameters such as water supply temperature, valve opening and other parameters will show obvious characteristics on the time series curve. The symbolic representation of the time series can quickly identify the time series data similar to the known fault conditions in the massive operating data by reducing the dimension of the time series curve into a symbolic representation form, and can realize the rapid and accurate diagnosis of the three types of fault conditions: heat source failure, thermal power station shutdown and electric control valve oscillation. Summary of the invention

[0005] In order to overcome the shortcomings of the above-mentioned prior art, the present invention proposes a method for diagnosing abnormal operation of a HVAC system based on time series symbolization. The method is based on time series symbolization and uses multiple characteristic parameters of the time series as independent variables for detection. It is suitable for diagnosing abnormal operation of HVAC system energy stations (such as air conditioning rooms or thermal power stations) and provides reliable guidance for the optimized operation of HVAC systems.

[0006] The technical solution adopted to achieve the purpose of the present invention is: A method for diagnosing abnormal operation of a heating, ventilation and air conditioning system based on time series symbolization, the method comprising the following steps: 1. Obtain an abnormal heat consumption data set, which includes normal operating condition data and abnormal operating condition data of various HVAC systems. Part of the abnormal operating condition data is marked with abnormal time periods and abnormal types; the marked data is divided into a one-step training set, a two-step training set, and a test set at a ratio of 0.07, 0.63, and 0.3; 2. Building a rule base Step 2.1: Smooth the time series data. The smoothed time series is recorded as X=(x1,x2,…, x n ), n is the number of data points in the smoothed time series; calculate the slopes of adjacent data points and get the slope sequence K=(k1, k2,…, k i ,…, k n-1 ), and according to the positive or negative slope, m is calculated according to formula (4). i To assign a value: (3) Among them, t i represents the i-th time point; x i Represents the time point t i Heating parameter value at t i+1 represents the i+1th time point; x i+1 Represents the time point t i+1 Heating parameter value at ; (4) In the formula, m i No practical meaning, similar to the sgn function; Step 2.2: Select time series segmentation points According to the slope sequence, we get the m sequence, m=(m1, m2, …, m i , …, m n-1 ), determine the segmentation points according to formula (6), (6) Then the sequence of segmentation points It is expressed as: (5) In the formula, represents the jth segmentation point in the time series X; J is the number of segmentation points, C j is the position of the current segmentation point in the time series X; Since the segmentation point is located between the time series data and does not include the starting point and end point of the time series data, C0 is used to represent the starting point of the time series data, and C J+1 is the end point of the time series data; Step 2.3: Divide the time series into trend segments Connect the time series data of different segmentation points to form a trend segment sequence T. Where, T j Indicates segmentation point and The trend segment formed by connecting the time series data between them; Step 2.4: Determine the angle threshold Calculate the average slope in each trend segment and the angle value corresponding to the average slope. The angle values ​​of all trend segments form an angle sequence. ; Obtain the maximum and minimum values ​​in the angle sequence under the same abnormal operating condition, and then obtain the upper and lower bounds of the angle threshold under different abnormal operating conditions; Step 2.5: Extract timing rules According to the upper and lower limits of the angle threshold, it is determined whether each trend segment belongs to the trend of rising, falling and stable change. If the angle value of a trend segment is not less than the upper limit, it is considered to be an rising trend, represented by R; if the angle value of a trend segment is not greater than the lower limit, it is considered to be a falling trend, represented by F; if the angle value of a trend segment is between the upper and lower limits, it is considered to be a stable change trend, represented by S; All trend segments are symbolized into three trends of R, F and S according to the angle threshold corresponding to the abnormal operating condition, forming the time sequence rule of the abnormal condition. The time sequence rule of the abnormal condition refers to the symbolic combination of the name of the corresponding abnormal condition and the abnormal time period corresponding to the characteristic parameters; The timing rules under all abnormal operating conditions constitute the rule base, and the construction of the rule base is completed; 3. Abnormal operation diagnosis After obtaining the time series of the corresponding characteristic parameters of the HVAC system to be diagnosed and extracting the timing rules, the timing rules of the HVAC system to be diagnosed are matched with the timing rules in the rule base. If the timing rules of a certain type of abnormal working condition in the rule base are met, it is diagnosed that the abnormal operation data of the current HVAC system to be diagnosed belongs to the same type of abnormal working condition.

[0007] Furthermore, the method of the present invention is applicable to abnormal operation diagnosis of an energy station of a HVAC system, wherein the energy station includes an air-conditioning room or a thermal power station.

[0008] Furthermore, the abnormal operating conditions include thermal power station shutdown conditions, electric control valve oscillation conditions, heat source failures, etc.

[0009] Compared with the prior art, the present invention has the following beneficial effects: The outstanding essential features of the present invention are: 1) High detection and diagnosis accuracy. The use of symbolic representation of heating data time series greatly improves the accuracy of diagnosis of abnormal operating conditions.

[0010] 2) There are various methods for data smoothing. In addition to the moving average method used in the embodiment, there are also a series of mathematical methods such as Kalman filtering, spline smoothing, local weighted regression, Fourier transform, autoregressive model, etc. Different smoothing methods can provide different balance points when processing different types of data and application scenarios, making the optimized data results closer to the actual results and with smaller errors, thereby helping to optimize the effect of data analysis.

[0011] 3) Convenient operation and strong practicality. The present invention can be applied to the detection of operating fault conditions of various HVAC systems, with fast operation speed and easy operation.

[0012] The significant improvements of the present invention are: 1) Simplify complexity. Data complexity can be simplified by converting continuous time series data into discrete symbolic representations.

[0013] 2) Reduce dimensionality. Symbolization methods can significantly reduce the dimensionality of data.

[0014] 3) Improve efficiency. Compared with the original continuous time series, symbolization can speed up data processing and analysis.

[0015] This application is the first to apply symbolic representation of time series to abnormal operation diagnosis of HVAC systems. By processing the heating data collected by the smart heating platform, the characteristics of time series data can be highlighted in the massive heating data, making the identification pattern and trend analysis more intuitive. The symbolic method can also smooth the random noise in the time series, making the small changes in the time series less sensitive and more robust. It not only efficiently diagnoses the abnormal operation of the HVAC system, but also greatly reduces the computational burden. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a schematic diagram of the changing trend of the time series (the first picture is the rising trend R, the second picture is the falling trend F, and the third picture is the stable trend S).

[0017] Figure 2 The present invention is a flow chart for diagnosing abnormal operating conditions of a thermal power station using the diagnostic method of the present invention for a certain section of heating time series data.

[0018] Figure 3 This is a diagram showing the changes in relevant parameters during the heat source failure period of thermal power station D1.

[0019] Figure 4 This is a diagram showing the changes in relevant parameters during the heat source failure period of thermal power station D1 after data smoothing.

[0020] Figure 5 This is a graph showing changes in relevant parameters during the shutdown period of thermal power station D2.

[0021] Figure 6 This is a graph showing changes in relevant parameters during the shutdown period of thermal power station D2 after data smoothing.

[0022] Figure 7 This is a diagram showing the changes in relevant parameters during the oscillation period of the D3 electric regulating valve in the thermal power station.

[0023] Figure 8 This is a diagram showing the changes in relevant parameters during the oscillation period of the D3 electric regulating valve in the thermal power station after data smoothing.

[0024] Fig. 9 This is a statistical information diagram of all heat source failure trend segments in the one-step training set.

[0025] Fig.10 It is the time series trend diagram of heat source failure condition.

[0026] Fig.11 This is a bar chart of the smoothed valve opening curve and slope assignment for thermal power station D1.

[0027] Fig.12 This is a bar chart of the smoothed valve opening curve and slope assignment for thermal power station D1.

[0028] Fig.13 This is a bar chart of the smoothed valve opening curve and slope assignment for thermal power station D3.

[0029] Fig.14 This is a bar chart of the smoothed instantaneous flow curve and slope assignment for thermal power station D3.

[0030] Fig.15 This is the confusion matrix for diagnosing the shutdown condition of thermal power station D6.

[0031] Fig.16 Changes in relevant parameters during abnormal periods at thermal power station D4. DETAILED DESCRIPTION

[0032] The present invention is described in detail below with reference to the accompanying drawings and embodiments, but the protection scope of the present invention is not limited thereto.

[0033] The present invention provides a method for diagnosing abnormal operation of a HVAC system based on time series symbolization, the method comprising the following steps: 1. Obtain an abnormal heat consumption data set, which includes normal operating condition data and abnormal operating condition data of various HVAC systems. Part of the abnormal operating condition data is marked with abnormal time periods and abnormal types; the marked data is divided into a one-step training set, a two-step training set, and a test set at a ratio of 0.07, 0.63, and 0.3; 2. Building a rule base 1): Data smoothing. This embodiment uses exponentially weighted moving average to smooth the time series data. The formula of EWMA can be expressed as: (1) (2) Where V(t) is the actual value at time t; a is a smoothing coefficient between 0 and 1; span is the amount of historical data considered by EWMA(t).

[0034] 2): Time series segmentation point selection. For the time series data X=(x1, x2,…,x n ). The slope of the time series data is calculated by comparing the changes in adjacent data points, and the slope sequence K=(k1, k2,…,k i ,…, k n-1 ), and assign a value according to the slope according to formula (4). i The calculation formula is shown in formula (3): (3) (4) Among them, t i represents the i-th time point; x i Represents the time point t i Heating parameter value at t i+1 represents the i+1th time point; x i+1 Represents the time point t i+1 The heating parameter value at that time; n is the number of data points in the time series, m i No real meaning. Similar to the sgn function.

[0035] If the product of two adjacent slopes is not greater than 0, and the two adjacent slopes are not equal, then the data xi corresponding to the latter slope is a segmentation point. If the product of two adjacent slopes is greater than 0, then the data xi corresponding to the latter slope is not a segmentation point. The sequence of segmentation points It is expressed as: (5) (6) In the formula, represents the jth segmentation point in the time series X; J is the number of segmentation points, C j The position of the current segment point in the time series.

[0036] Since the segmentation point is located between the time series data, it does not include the starting point and end point of the time series data. Therefore, C0 is used to represent the starting point of the time series data, and C J+1 is the end point of the time series data.

[0037] 3) Time series trend segment division. Connect the time series data of different segmentation points to form a trend segment sequence T (7) Where, T j Indicates segmentation point and The trend segment is formed by connecting the time series data between. That is: (8) 4) Determine the trend characteristics of each trend segment, calculate the average slope in each trend segment, calculate the angle value corresponding to the average slope, and obtain the angle sequence of the trend segment , after training with the training set, a suitable threshold is obtained. The calculation formulas for the average slope and angle value are shown in formula (9) and formula (10) respectively.

[0038] (9) (10) (11) In the formula, is the average slope within the jth trend segment; is the angle value corresponding to the jth trend segment.

[0039] 5) Determine the angle threshold. By statistically analyzing the trend information of different types of abnormal operating conditions during abnormal periods, the angle thresholds for different abnormal operating conditions are obtained. By statistically analyzing the statistical information of all heat source fault trend segments in the one-step training set, a box plot is drawn to find the maximum and minimum values.

[0040] This threshold is not fixed, but needs to be determined based on the statistical results of the trend segment in the training step. According to this angle threshold, it can be determined whether the trend segment is rising, falling or stable. Specifically, if the angle value of a trend segment is greater than or equal to the upper limit, it is judged as a rising trend (R); if it is less than or equal to the lower limit, it is judged as a falling trend (F); if the angle value is between the upper and lower limits, it is considered to be a stable change trend (S). The trend of a time series can usually be divided into three parts: rising trend (Rising), falling trend (Falling) and stable change (Stable), as shown in the attached figure. Figure 1 shown.

[0041] 6) Extraction of timing rules. All trend segments are symbolized into three trends, R, F and S, according to the angle threshold to form the timing rules of abnormal working conditions. The timing rules of abnormal working conditions refer to the symbolic combination of the name of the corresponding abnormal working condition and the abnormal time period corresponding to the characteristic parameters.

[0042] Based on the timing rules of different abnormal working conditions, the detected abnormal operation data is subjected to timing analysis, and the timing data is symbolized through SRTS (time series symbolization) RFS. The timing rules are verified and evaluated using the two-step training set and test set data. If the conditions are not met, the smoothing method can be adjusted so that the timing rules show obvious curve changes in different abnormal working conditions. The timing rules under all abnormal operating conditions constitute the rule base, and the construction of the rule base is completed.

[0043] 3. Abnormal operation diagnosis The time series of the corresponding characteristic parameters of the HVAC system to be diagnosed is obtained. After data smoothing, selecting time series segmentation points, dividing the time series trend segments, determining the angle threshold and extracting the timing rules, the timing rules of the HVAC system to be diagnosed are matched with the timing rules in the rule base. If the timing rules of a certain type of abnormal working condition in the rule base are met, it is diagnosed that the abnormal operation data of the current HVAC system to be diagnosed belongs to the same type of abnormal working condition.

[0044] The present invention uses the accuracy, precision, recall and F1 score in the confusion matrix to evaluate the accuracy of the diagnosis method.

[0045] The time series in the present invention includes at least one time variable and one characteristic parameter variable. The characteristic parameter can be outdoor temperature, indoor temperature, flow, heat, valve opening, etc. In machine learning, a semi-supervised learning method is used to label a part of the data of known faults, and expert knowledge is used to train the unlabeled data to find abnormal control data.

[0046] The diagnostic method of the present invention has the following characteristics: 1) Symbolic representation of time series: By reducing the dimension of the time series curve into a symbolic representation, when different feature parameters are selected, the extracted time series rules will also be different.

[0047] 2) There are many methods for data smoothing. You can choose the appropriate smoothing method according to the specific analysis objectives, which can more effectively highlight the key information in the data and improve the quality of the analysis results. Embodiment 1: This example takes a large-scale centralized heating system in a cold region as the research object. The system uses a thermal power plant as the heat source and has a total of 932 heating stations under its jurisdiction, covering an actual heating area of ​​about 26 million m 2 . All thermal power stations are equipped with data acquisition devices. The data used in the study comes from the smart heating system monitoring platform, including the operation data of the thermal power station during 2019-2020. The data set contains information such as the primary side supply and return water temperature, the secondary side supply and return water temperature, the valve opening, the instantaneous flow rate and the instantaneous heat. Then, the detection and diagnosis method of the present embodiment described below is applied to detect and diagnose the heat consumption of the heating system operation fault.

[0048] The abnormal operation diagnosis method of the HVAC system based on time series symbolization in this embodiment includes the following steps: Step S1: Use exponentially weighted moving average to smooth the time series data. The heat source failure condition occurred in the thermal power station D1 at 10:00 on December 27, 2019. The water supply temperature, valve opening and instantaneous heat data within 1 day before and after the abnormal data corresponding time point are selected to draw a line graph, see Figure 3 .right Figure 3 After performing exponential weighted moving average with a span index of 72 on the time series data of heat source failures in the model, it was found that the model data results were good, so the span index of 72 was used. The parameter change diagram after smoothing is shown in Figure 4 . Figure 4 The parameter changes after exponentially weighted moving average processing are shown, and it can be clearly observed that the characteristic parameters change during abnormal periods ( Figure 3 The trend of change in the middle grey area.

[0049] Thermal power station D2 was out of service at 10:00 on December 21, 2019. The water supply temperature, valve opening, and instantaneous heat data for a total of 1 day before and after the abnormal data are selected to draw a line graph, see Figure 5 The same is true for Figure 5 The time series data of the thermal power station shutdown in the above example is subjected to exponential weighted moving average with a span index of 72. The parameter change diagram after smoothing is shown in Figure 6 . Figure 6 The parameter changes after exponentially weighted moving average processing are shown, and the changing trend of the characteristic parameters can be clearly observed.

[0050] It was diagnosed that the electric regulating valve oscillation occurred in the thermal power station D3 between January 19 and January 21, 2020. A line graph was drawn for the valve opening and instantaneous flow during this period. Figure 7 .right Figure 7 The time series data in the example are subjected to exponential weighted moving average with a span index of 24. The parameter change diagram after smoothing is shown in Figure 8 . Figure 8 The parameter changes after exponentially weighted moving average processing are shown, and the changing trend of the parameters can be clearly observed.

[0051] Step S2: For the time series data X=(x1, x2,…,x n ), by comparing the changes in adjacent data points to calculate the slope of the time series data, and according to the positive and negative slope of m i Assignment. Get the slope sequence K=(k1,k2,…, k i ,…, k n-1 ). Slope k i The calculation formula is shown in formula (3): (3) (4) Where, t i represents a time point; x i Represents the time point t i The heating parameter value at that time.

[0052] The sequence of segmentation points is expressed as: (5) (6) In the formula, represents the jth segmentation point in the time series X; C j The position of the current segment point in the time series.

[0053] Since the segmentation point is located between the time series data, it does not include the starting point and end point of the time series data. Therefore, C0 is used to represent the starting point of the time series data, and C J+1 is the end point of the time series data.

[0054] Step S3: Time series trend segment division. Connect the time series data of different segmentation points obtained in step 2 to form a trend segment sequence T (7) Where, t i Indicates segmentation point and The trend segment is formed by connecting the time series data between. That is: (8) Step S4: Determine the trend characteristics of each trend segment, calculate the average slope in each trend segment, and calculate the angle value corresponding to the average slope, and obtain the angle sequence of the trend segment. The calculation formulas for the average slope and angle value are shown in formula (9) and formula (10) respectively. (9) (10) (11) In the formula, is the average slope within the jth trend segment; is the angle value corresponding to the jth trend segment.

[0055] The primary side water supply temperature data is smoothed by exponentially weighted moving average with a span index of 72, and then the slope k of each smoothed data point is calculated in turn. i , and assign the slope according to formula (4). By analyzing the slope assignment m i The data points are divided into different trend segments based on the changes in the function. A continuous period of function is regarded as a time period. When the function value changes, it is considered that a trend segment ends or the next trend segment begins. Fig.11 According to the turning point of the slope assignment change, the primary side water supply temperature data can be divided into 10 trend segments. The average slope is calculated for the data points in each trend segment. , and calculate the angle value arctan( ). The average slope and angle values ​​of each trend segment are summarized in Table 1.

[0056]

[0057] Similarly, for the valve opening data smoothed by the exponentially weighted moving average with a span index of 72, the slope k of each data point is calculated in turn. i , and the slope is assigned. According to the turning point of the slope assignment change, the valve opening data can be divided into 5 trend segments, see Fig.12 Calculate the average slope for each data point in the trend segment , and calculate the angle value arctan( ). The average slope and angle values ​​of each trend segment are shown in Table 2.

[0058]

[0059] The valve opening data is smoothed by exponentially weighted moving average with a span index of 24, which eliminates small fluctuations in the data and retains the short-term trend of the data. Then, the slope k of each smoothed data point is calculated in turn. i , and assign a slope value to each data point. According to the turning point of the slope assignment change, the valve opening can be divided into 32 trend segments, see Fig.13 Calculate the average slope for each data point in the trend segment , and calculate the angle value arctan( ). The average slope and angle values ​​of each trend segment are summarized in Table 3.

[0060]

[0061] For the instantaneous flow data smoothed by the exponentially weighted moving average with a span index of 24, the slope ki of each data point is calculated in turn, and the slope is assigned. According to the turning point of the slope assignment change, the instantaneous flow can be divided into 34 trend segments, see Fig.14 Calculate the average slope for each data point in the trend segment. , and calculate the angle value arctan( ). The average slope and angle values ​​of each trend segment are summarized in Table 4.

[0062]

[0063] Step S5: Determine the angle threshold. The data in this paper adopts a method based on semi-supervised learning. First, the abnormal heat consumption dataset is divided into a training set and a test set at a ratio of 7:3. In order to reduce the workload of manual labeling, the training set is further divided into a one-step training set and a two-step training set at a ratio of 1:9 in the training set. The two-step training set is used to verify the effect and performance of the timing rules of the one-step training set, and the test set is used to evaluate the generalization ability of the model. That is, the ratios of the one-step training set, the two-step training set and the test set are 0.07, 0.63 and 0.3, respectively. In the one-step training set, heat source fault conditions were marked in 48 thermal power stations, and the primary water supply temperature and valve opening curves during the fault period were divided into trend segments. All temperature curves are divided into 433 trend segments, and all valve curves are divided into 609 trend segments. The angle values ​​of all trend segments are counted to draw a box plot, see Fig. 9 .

[0064] like Fig. 9As shown in the figure, in the one-step training set, for the heat source failure condition, the upper and lower bounds of the angle value of the primary water supply temperature trend segment are 3.67° and -3.26° respectively; the upper and lower bounds of the angle value of the valve opening trend segment are 2.77° and -2.84° respectively. That is, for the heat source failure condition, if the angle value θ of a primary water supply temperature trend segment is greater than or equal to 3.67°, the trend segment is considered to be an upward trend (R); if θ is less than or equal to -3.26°, the trend segment is considered to be a downward trend (F); if θ is between -3.26° and 3.67°, it is considered to be a steady change trend (S). If the angle value θ of a valve opening trend segment is greater than or equal to 2.77°, the trend segment is considered to be an upward trend (R); if θ is less than or equal to -2.84°, the trend segment is considered to be a downward trend (F); if θ is between -2.84° and 2.77°, it is considered to be a steady change trend (S).

[0065] Step S6: Timing rule extraction. According to the angle threshold, all trend segments are symbolized into three trends: R, F and S, to form the timing rules of abnormal working conditions. Note: The timing rules are composed of different symbol sequences of R and F. Generally, there is no S trend in the timing rule combination of abnormal working conditions. Based on the timing rules of different abnormal working conditions, the detected abnormal heat consumption data is subjected to timing analysis, the timing data is symbolized by SRTS, and compared with the timing rules under different abnormal working conditions. If the timing rules of a certain type of abnormal working condition are met, it is diagnosed that the current abnormal heat consumption data belongs to the same type of abnormal working condition.

[0066] Based on the definition of angle threshold, the primary water supply temperature and valve opening curve of thermal power station D1 between 14:00 on December 26 and 14:00 on December 28 is taken as an example to perform symbolic representation of each trend segment, see Fig.10 .

[0067] Depend on Fig.10 It can be seen that the AB section of the curve is a section where the primary side water supply temperature drops sharply due to insufficient heat supply at the heat source, matching the timing rule F; the corresponding A'-B' section is a section where the valve opening increases sharply in response to insufficient heat supply from the heat source and the thermal power station obtains more heat, matching the timing rule R; the BC section of the curve is when the heat supply from the heat source gradually recovers and the primary side water supply temperature rises, matching the timing rule R; the corresponding B'-C' section is a section where the valve opening falls back to the normal operating range, matching the timing rule F. The timing rules for the condition of insufficient heat supply from the heat source are as follows: there is a timing rule combination of F + R on the primary water supply temperature time series curve, and there is a timing rule combination of R + F on the valve opening time series curve, as shown in Table 5.

[0068]

[0069] Step S7: Use the accuracy, precision, recall, and F1 score in the confusion matrix to evaluate the accuracy of the diagnostic method.

[0070] The data in the thermal power station D6 was diagnosed by symbolizing the time series, and a total of 22 thermal power station shutdown data were diagnosed. The confusion matrix is ​​as follows Fig.15 shown.

[0071] like Fig.15 As shown in Table 6, there are 20 data points marked as shutdown data points in thermal power station D6. Through time series symbolization, a total of 22 shutdown fault data points are diagnosed. The evaluation indicators for shutdown faults are shown in Table 6.

[0072]

[0073] As shown in Table 7, the time series symbolization also has a high diagnostic accuracy for the shutdown condition of the thermal power station, and the accuracy and precision rate are all above 91%. In the test set, 13,099 data were analyzed for the shutdown condition of the thermal power station, and a total of 922 data of the shutdown period of the thermal power station were diagnosed, accounting for 7.04% of the total data in the test set.

[0074] Randomly select the operating conditions diagnosed as thermal power station shutdown in the test set for analysis and display. Thermal power station D4 experienced a shutdown condition at 10:00 on December 21, 2019. Select the water supply temperature, valve opening, and instantaneous heat data within 1 day before and after the abnormal data corresponding time point to draw a line graph, as shown in the figure below: Fig.16 shown.

[0075] like Fig.16 As shown in the figure, during the shaded time period, the valve opening of the thermal power station D4 suddenly dropped to 0, and the instantaneous flow and instantaneous heat also dropped to 0, indicating that the valve was closed during this period, and the thermal power station was cut off, resulting in the cessation of heat supply. At the same time, the heat exchange of the thermal power station stopped, and the primary side water supply temperature dropped sharply, gradually cooling to the ambient temperature. After 22:00 on December 21, the thermal power station resumed normal operation, and all operating parameters quickly returned to normal operation. The fault condition characteristics are consistent with the operating fault condition characteristics caused by the thermal power station shutdown, verifying the diagnostic effect of the time series symbolic representation on the thermal power station shutdown condition.

[0076] Any matters not described in the present invention are applicable to the prior art.

Claims

1. A method for diagnosing abnormal operation of a HVAC system based on time series symbolization, the method comprising the following steps:

1. Obtain an abnormal heat consumption data set, which includes normal operating condition data and abnormal operating condition data of various HVAC systems. Part of the abnormal operating condition data is marked with abnormal time periods and abnormal types; the marked data is divided into a one-step training set, a two-step training set, and a test set at a ratio of 0.07, 0.63, and 0.3; 2. Building a rule base Step 2.1: Smooth the time series data. The smoothed time series is recorded as X=(x1, x2,…,x n ), n is the number of data points in the smoothed time series; calculate the slopes of adjacent data points and get the slope sequence K=(k1,k2, …, k i , …, k n-1 ), and according to the positive or negative slope, m is calculated according to formula (4). i To assign a value: （3） Among them, t i represents the i-th time point; x i Represents the time point t i Heating parameter value at t i+1 represents the i+1th time point; x i+1 Represents the time point t i+1 Heating parameter value at ; （4） In the formula, m i No practical meaning, similar to the sgn function; Step 2.2: Select time series segmentation points According to the slope sequence, we get the m sequence, m=(m1, m2, …, m i , …, m n-1 ), determine the segmentation points according to formula (6), （6） Then the sequence of segmentation points It is expressed as: （5） In the formula, represents the jth segmentation point in the time series X; J is the number of segmentation points, C j is the position of the current segmentation point in the time series X; Since the segmentation point is located between the time series data and does not include the starting point and end point of the time series data, C0 is used to represent the starting point of the time series data, and C J+1 is the end point of the time series data; Step 2.3: Divide the time series into trend segments Connect the time series data of different segmentation points to form a trend segment sequence T. Where, T j Indicates segmentation point and The trend segment formed by connecting the time series data between them; Step 2.4: Determine the angle threshold Calculate the average slope in each trend segment and the angle value corresponding to the average slope. The angle values ​​of all trend segments form an angle sequence. ; Obtain the maximum and minimum values ​​in the angle sequence under the same abnormal operating condition, and then obtain the upper and lower bounds of the angle threshold under different abnormal operating conditions; Step 2.5: Extract timing rules According to the upper and lower limits of the angle threshold, it is determined whether each trend segment belongs to the trend of rising, falling and stable change. If the angle value of a trend segment is not less than the upper limit, it is considered to be an rising trend, represented by R; if the angle value of a trend segment is not greater than the lower limit, it is considered to be a falling trend, represented by F; if the angle value of a trend segment is between the upper and lower limits, it is considered to be a stable change trend, represented by S; All trend segments are symbolized into three trends of R, F and S according to the angle threshold corresponding to the abnormal operating condition, forming the time sequence rule of the abnormal condition. The time sequence rule of the abnormal condition refers to the symbolic combination of the name of the corresponding abnormal condition and the abnormal time period corresponding to the characteristic parameters; The timing rules under all abnormal operating conditions constitute the rule base, and the construction of the rule base is completed; 3. Abnormal operation diagnosis After obtaining the time series of the corresponding characteristic parameters of the HVAC system to be diagnosed and extracting the timing rules, the timing rules of the HVAC system to be diagnosed are matched with the timing rules in the rule base. If the timing rules of a certain type of abnormal working condition in the rule base are met, it is diagnosed that the abnormal operation data of the current HVAC system to be diagnosed belongs to the same type of abnormal working condition.

2. The method according to claim 1, characterized in that The method is applicable to abnormal operation diagnosis of an energy station of a HVAC system, wherein the energy station includes an air conditioning room or a thermal power station.

3. The method according to claim 1, characterized in that The abnormal operating conditions include thermal power station shutdown conditions, electric control valve oscillation conditions, and heat source failures.

Images