Transformer comprehensive monitoring and evaluation system based on big data analysis

Abstract

The present application relates to the technical field of transformer monitoring and evaluation, in particular to a transformer comprehensive monitoring and evaluation system based on big data analysis, comprising a data acquisition module, a model construction module and a prediction analysis module. The model construction module constructs a transformer normal operation model. After determining the evaluation features, the prediction analysis module uses the data collected by the data acquisition module to form an evaluation sequence, and simultaneously obtains a prediction sequence through the transformer normal operation model. Then, the similarity between the evaluation sequence and the prediction sequence is compared, and the similarity is converted into a percentage as the evaluation score of the transformer. The system analyzes the recent data to obtain a score reflecting the performance of the transformer, providing a basis for managers to make judgments. This method can greatly save computing power and improve the performance, efficiency and accuracy of the evaluation.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of transformer monitoring and evaluation, in particular to a transformer comprehensive monitoring and evaluation system based on big data analysis. BACKGROUND

[0002] The transformer is very important power equipment in the power system and is an important hub for realizing conversion of power transmission, so keeping a good transformer state is of great significance to the safety of the power system. The power system operation and power supply quality can be ensured, and personal and property losses caused by faults can be reduced.

[0003] At present, the detection of the transformer mainly concentrates on the early warning of the faults of the transformer. The transformer has many components and assemblies, and each assembly and each link will cause the transformer to malfunction. For example, in the fault early warning process, the abnormal oil temperature needs to be early warned and judged, the partial discharge needs to be judged, and the temperature of the iron core, the grounding current and the like need to be early warned and judged one by one. The early warning mode is mostly based on big data analysis and judgment. The data to be collected is also relatively comprehensive. However, the transformer is in a normal operating state most of the time, so real-time and continuous calling of data for prediction not only occupies unnecessary computing power, but also causes loss of energy consumption. Therefore, it is an urgent requirement to design a transformer comprehensive monitoring and evaluation system reflecting the performance state of the transformer. SUMMARY

[0004] The technical problem to be solved by the application is to provide a transformer comprehensive monitoring and evaluation system reflecting the performance state of the transformer based on big data analysis.

[0005] The technical solution of the technical problem to be solved by the application is a transformer comprehensive monitoring and evaluation system based on big data analysis, which comprises a data acquisition module, a model construction module and a prediction analysis module. The model construction module constructs a transformer normal operation model by applying a recurrent neural network algorithm based on normal operation data obtained by the data acquisition module. After determining the evaluation features, the prediction analysis module applies the data collected by the data acquisition module to form an evaluation sample. The evaluation sequence is extracted through the evaluation sample, and a prediction sequence is obtained through the transformer normal operation model. Then, the similarity of the evaluation sequence and the prediction sequence is compared, and the similarity is converted into a percentage as an evaluation score of the transformer.

[0006] Better, in the data acquisition process: first, the interval time and the starting time point of the time span are determined to determine the collection points of the sequence data; then, the sample data is collected according to the determined collection points within a day.

[0007] Better, the time span comprises at least one detection period.

[0008] More preferably, the time span comprises a stable period and a fluctuation period, and the method for obtaining the stable period is:

[0009] Step1, divide the time span of one day into m segments and sequentially number them;

[0010] Step2, collect j data b1, b2, b3…bj in the a segment, where a=1, 2, 3…m; b1, b2, b3…bj are data collected at equal time intervals in this time span;

[0011] Step3, calculate the mean And the variance b:

[0012]

[0013] Form a sample data

[0014] Step4, convert the sample data Into And form a sample data set, then use clustering algorithm to cluster the data set to obtain clusters reflecting different fluctuation states and average loads;

[0015] Step5, among the sample data of one cluster, obtain the number a of sample data B according to the original sample data; then find whether there is a number that increases one by one or decreases one by one in the sample data of this cluster;

[0016] If so, the time span with consecutive numbers is a stable period; if not, select another cluster and repeat step 5;

[0017] The method for obtaining the fluctuation period is to select a time span within 8-18 hours as the fluctuation period.

[0018] More preferably, in step Step1: each segment is 5-15 minutes.

[0019] More preferably, b1, b2, b3…bj are the values of an electrical parameter at different time points.

[0020] More preferably, at least three evaluation features are set, and the average of the evaluation scores corresponding to all evaluation features is taken as the evaluation score of the transformer performance.

[0021] More preferably, at least three evaluation features are set, and the lowest evaluation score is taken as the evaluation score of the transformer.

[0022] More preferably, the time span comprises a stable period and a fluctuation period, and the method for obtaining the stable period is:

[0023] Step1, divide the time length of a day into m segments and sequentially number them;

[0024] Step2, collect j data b1, b2, b3……bj in the a segment, wherein a=1, 2, 3……m; b1, b2, b3……bj are data collected at equal time intervals in the segment;

[0025] Step3, calculate the average number and the variance b to form sample data

[0026] Step4, convert the sample data to and form a sample data set, and then use a clustering algorithm to cluster the data set to obtain clusters reflecting different fluctuation states and average loads;

[0027] Step5, the sample data of one of the clusters, according to the original sample data, obtain the number a of sample data B; then find whether there is a sequentially increasing or sequentially decreasing number in the sample data of the cluster;

[0028] If there is, then the time segments with consecutive numbers form a stable time period; if not, select another cluster and repeat the description

[0029] Step 5;

[0030] The method for obtaining the fluctuation period is:

[0031] Step6, convert the sample data to (b) and form a sample data set, and then use a clustering algorithm to cluster the data set to obtain clusters reflecting different fluctuation states;

[0032] Step7, extract sample data reflecting a cluster with larger fluctuations, extract a sample with number a, obtain the average value of a group of time segments centered on a, and if the average values of at least two time segments are different, select the group of time segments to form a fluctuation period.

[0033] Better, if the number a in the selected group of time segments belongs to the cluster label in Step5, repeat step7 and extract a number for re-determination.

[0034] The beneficial effects of the present application are:

[0035] Compared with the prior art, the system can evaluate the performance of the transformer by calling part of data of the day after model training, obtain a score reflecting the performance of the transformer, and provide a basis for judgment for the management personnel. The method can greatly improve the performance, efficiency and accuracy of evaluation. When the performance score is too low, the fault detection and warning system of the transformer can be started to determine the fault. Compared with directly applying big data to detect and warn the transformer, the method can greatly save computing power and improve efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0036] Figure 1 is a system composition schematic diagram of an embodiment of the application.

[0037] Figure 2 is a flowchart of an embodiment of the application.

[0038] In the figure: 300, prediction analysis module; 200, model construction module; 100, data acquisition module; DETAILED DESCRIPTION

[0039] To make the technical solutions and beneficial effects of the application clearer, the embodiments of the application are further explained in detail below.

[0040] The transformer comprehensive monitoring and evaluation system based on big data analysis comprises a data acquisition module 100, a model construction module 200 and a prediction analysis module 300. The data acquisition module 100 is used to acquire operation data of the transformer. The model construction module 200 applies a recurrent neural network algorithm to construct a normal operation model of the transformer based on the data acquired by the data acquisition module 100. The normal operation model of the transformer reflects the mutual correlation and relevance between various parameters of the transformer in the normal operation state of the transformer. The prediction analysis module 300 also applies new data acquired by the data acquisition module 100 to make a prediction on some characteristic parameters of the transformer through the model construction module 200, and then obtains a performance score of the operation state of the transformer by comparing the predicted data with the actual data. The system is used to evaluate the performance of the transformer and realize numerical representation of the state of the transformer. The maintenance personnel or management personnel can determine the state of the transformer through the numerical value of the state of the transformer. If the numerical value is too low, it indicates that the transformer is in a poor operation state, and some components of the transformer are damaged and need to be repaired.

[0041] The data acquisition module 100 can acquire the operation data of the transformer from the database of the dispatching system or the monitoring system of the power system by calling historical data. Alternatively, the system is integrated in the monitoring system of the transformer substation as a separate system, and the operation data of the transformer is acquired by the data acquisition terminal and acquisition channel of the transformer substation system.

[0042] The transformer can collect many parameters reflecting the state of the transformer in the running state, such as the voltage Uh flowing into the high-voltage side of the transformer, the current Ih flowing into the high-voltage side of the transformer, the voltage Ul flowing out of the low-voltage side of the transformer, the current Il flowing out of the low-voltage side of the transformer, the oil temperature Yw of the transformer, the vibration condition Zf of the transformer, the core temperature Tw of the transformer, the grounding current Ij of the transformer, and the partial discharge current.

[0043] In addition to data collection, the data collection module 100 also needs to process the data. The data collection module 100 forms a sample data set D0 by obtaining historical running data of the transformer. The historical running data is data from the previous day, or data from a week ago or a month ago. At the same time, the data collection module 100 also needs to obtain data of the current day or recent data, such as data of a certain day within a week or a month as an evaluation sample Ps. The process of data collection and processing by the data collection module 100 includes the following steps.

[0044] In the system, the sample data S in the sample data set D0 is a sequence data, S=(s1, s2, s3, s4, … Sn), where n is the length of the sequence data. The above sequence can contain one time start point or multiple time start points. For example, s1 is data collected at T0, s2 is data collected at T0+Tj, s3 is data collected at T0+2Tj, and so on, until Sn is data collected at T0+(n-1)Tj. Where Tj is the interval time of collecting data.

[0045] Step 1, determine the extraction date and extract the data of 24 hours of the day, including all the parameters of the transformer and the time label of the parameter.

[0046] Step 2, determine the time span and interval time Tj of the sample data S. For example, the time span of the sample data S is 24 hours, Tj=1 hour, that is, a data point is collected every hour within 24 hours, and the sequence of the sample data S contains 24 data points.

[0047] The time span can be set to different periods, such as 8-12 hours, 12-18 hours, etc., and Tj can also be half an hour, 15 minutes, two hours, etc.

[0048] More preferably, two time spans are selected, i.e. the sample data S is composed of two series of data spliced into one series. The two time spans are the stable time span Ts and the fluctuation time span Tf. From the daily routine, it can be known that after the early morning, the night time is relatively stable because there are fewer users and the load is relatively stable. Meanwhile, in the morning and afternoon, there are more users and the devices are turned on and off frequently due to the production needs, so the load is relatively fluctuant. Therefore, 0-6 o'clock can be selected as the relatively stable detection time span, and 12 o'clock-18 o'clock can be selected as the relatively fluctuant detection time span. For example, in order to reduce the data amount, the stable time span Ts can be 2 o'clock-3 o'clock, the fluctuation time span Tf can be 14 o'clock-15 o'clock, and Tj=10 minutes, so that the final sample data S contains 12 data. The operation amount is reduced and the operation precision is improved by shortening the time span and the interval time.

[0049] In the embodiment, the stable time span Ts and the fluctuation time span Tf can be different, e.g. Tf>Ts or Tf=2Ts, Tf=3Ts in order to better reflect the state of the transformer.

[0050] Step 3, after the time span and the interval time are determined, the data of the transformer parameters can be extracted according to the determined time points in the data of the extraction date. For example, the data extracted by s1 is (t, Uh, Ih, Ul, Il, Yw, Zf, Fw), wherein t is the time information, which can also be the number of data, e.g. t=1 in s1.

[0051] Further, some transformers are in all-weather operation state, and there are fluctuations all day long, and the relatively stable time is short. In this case, the stable time span Ts and the fluctuation time span Tf can be determined by the following method.

[0052] Step 1, the time length of one day is divided into m sections, each section has a time length of 5-15 minutes, and each section is sequentially numbered from zero.

[0053] Step 2, j data b1, b2, b3……bj in the a section are collected, wherein a=1, 2, 3……m. a is the number of the section, and b1, b2, b3……bj are the data in the section.

[0054] Since this step is used to detect that the transformer is in a relatively stable operation state, the data collected can be singularized, without collecting data of all parameters of the transformer like sample data S. Since the change of current can well reflect the fluctuation of load, the current is collected. Generally, the transformer in operation is mostly a three-phase transformer, in order to calculate simply, the data of one phase is selected, and the A phase is taken as an example in this embodiment. Taking j=4 and the length of each time period as 12 minutes as an example, b1 is the current of the A phase at the 3rd minute, b2 is the current of the A phase at the 6th minute, b3 is the current of the A phase at the 9th minute, and b4 is the current of the A phase at the 12th minute.

[0055] Step 3, calculation:

[0056]

[0057] and The sample data is composed of Wherein a is the label of the time period, and b is the variance of b1, b2, b3 and b4. is the average of b1, b2, b3 and b4.

[0058] Step 4, converting the sample data into sample data and composing a sample data set, and then using a clustering algorithm to cluster the data set to obtain different clusters. The clustering algorithm can adopt K-means clustering, hierarchical clustering, k-means clustering algorithm, etc. Each cluster represents a collection in a different fluctuation state or different average load period.

[0059] Step 5, extracting sample data of the cluster representing different fluctuation states. According to the sample data, the number a of sample data B is obtained. In the sample data of the cluster, whether there is a number that is sequentially increased or sequentially decreased is found. For example, a sample is extracted, the number of which is 3, and then whether the sample contains a sample with the number 2 or 4 is found in the sample, and the number of samples with the numbers 2, 3 and 4 is approximately the same, or the difference between the numbers is within 10, then the time periods with the numbers 2, 3 and 4 can be selected to form a stable time period Ts. Taking this embodiment as an example, the length of each time period is 12 minutes, and the length of the stable time period Ts is 36 minutes.

[0060] If not, another cluster is selected and step 5 is repeated until a time period meeting the requirements is found to form a stable time period.

[0061] Step 6, converting the sample data into sample data (b) and composing a sample data set, and then using a clustering algorithm to cluster the data set to obtain different clusters. Each cluster represents a collection in a different fluctuation state period.

[0062] Step7、extracting sample data of cluster class with large fluctuation. According to sample data, obtain the number a of sample data B. Extract a sample with number a, obtain the average value of sample with number a-1, the average value of sample with number a+1 and the average value of sample with number a, if at least two of the three average values are different, then select the three to form the fluctuation period Tf. If the three are the same, then select another sample in the cluster class to extract. At the same time, select the label in the cluster class, if it appears in step Step5, then do not select the period of the label. Then repeat step step7 and reselect a number to make the above judgment, until a suitable group of time periods is selected to form the fluctuation period.

[0063] In the above manner, more accurate relative stable and relative fluctuation data can be obtained. In the above manner, three time periods of labels are not necessarily selected to form the detection period, four or five time periods can also be selected, and the above is only an example of the method of selecting three time periods to form the detection period. After determining the selected time period, the long time period formed by the selected time periods is intercepted on the 24-hour time line according to the number of the time period as the stable time period and the fluctuation time period.

[0064] After obtaining the stable time period and the fluctuation time period of the detection period, data points of the two time periods are collected and form sequence data, and then the two sequence data are combined into one sequence or form a matrix parameter. For example, the sequence data of the stable time period is (p11, p12, p13, p14, p15), and the sequence data of the fluctuation time period is (p21, p22, p23, p24, p25), which can form a long sequence data (p11, p12, p13, p14, p15, p21, p22, p23, p24, p25). At this time, the sample data S=(p11, p12, p13, p14, p15, p21, p22, p23, p24, p25).

[0065] The sequence data can reflect the performance of the transformer in the relative stable state and the relative fluctuation state, and can accurately reflect the overall operation state of the transformer, while reducing a large amount of data and improving the operation efficiency.

[0066] In this embodiment, three time periods are selected to form the detection period, the length of the time period in this embodiment is 12 minutes, and three is 36 minutes. One stable detection period and one fluctuation detection period are 72 minutes. The model can be trained by general minute data, and the operation time length can be greatly reduced by less than two hours of data compared with 24 hours of data, and the efficiency is improved. At the same time, 10 sample data selected in 24 hours may cause the accuracy of the operation result to be reduced due to too long interval time. After determining the time period, 10 data points are selected to form sample data.

[0067] Taking the data s1 in the embodiment as an example, s1 contains a plurality of characteristic data t, Uh, Ih, Ul, Il, Yw, Zf, Fw, wherein t is time information, Uh is the voltage flowing into the high-voltage side of the transformer, Ih is the current flowing into the high-voltage side of the transformer, Ul is the voltage flowing out of the low-voltage side of the transformer, Il is the current flowing out of the low-voltage side of the transformer, Yw is the oil temperature of the transformer, Zf is the vibration parameter of the transformer, and Fw is the core temperature of the transformer. The above parameters can effectively reflect the operation characteristics and faults of the transformer. In addition, other data such as core grounding current, partial discharge data, gas concentration, etc. can also be selected. That is, the data contained in s1 is not only the above data, but also other data. The above data is only example data. At the same time, in order to reduce the operation complexity and improve the operation efficiency, a number of characteristic data less than the example in the embodiment can also be selected, such as only setting three characteristic data of high-voltage current, oil temperature and vibration. System evaluation can also be performed.

[0068] The sample data set D0 containing a large amount of sample data S is finally obtained through the data acquisition module 100. In the process of obtaining the sample data S, the time span and the time interval are determined, that is, the time label of the first parameter a of S is determined, such as s1 is the data at 15:00 on March 3, s2 is the data at 18:00 on March 3, and so on. Therefore, when collecting the evaluation sample Ps, the data is collected at the same time point. First, the extraction date of the evaluation sample Ps is determined. The extraction date of Ps is the data of the current day or the data of the previous day or the data of a certain day within one week before the current day or the data of a certain day within one month before the current day. After determining the extraction date, the data is extracted according to the time point of each data of the sample data S to form Ps.

[0069] In the process of collecting data, the data of abnormal operation, fault and maintenance state of the transformer is removed, and the data of the collection day is the data of the normal operation state of the transformer. The sample data set D0 composed of the sample data S belongs to the original data set. The sample data set D0 is applied to the model construction module 200 to construct a corresponding transformer normal operation model, wherein the model construction module 200 applies a recurrent neural network algorithm to construct the transformer normal operation model.

[0070] In constructing a transformer normal operation model based on a recurrent neural network (RNN) model, a long short-term memory network (LSTM) can be used as a variant of RNN. LSTM performs well in processing time series data and can better capture long-term dependencies. In the model construction module, three gating units of LSTM, including an input gate, a forget gate, and an output gate, and a cell state for storing information are set. In constructing the model, sample data S = (t, Uh, Ih, Ul, Il, Yw, Zf, Fw) is converted into (Uh, Ih, Ul, Il, Yw, Zf, Fw) and used as an input vector x t of time step t. Then, the LSTM model will use these input vectors to update its internal state and generate the corresponding output.

[0071] First, the weight matrix and bias vector are initialized. Then, forward propagation is performed. For each time step t, the forget gate output f t , the input gate output i t , the cell state update , the cell state C t , and the output gate output o t are calculated according to the current input x t and the hidden state h t-1 of the previous time step. Then, the hidden state is calculated according to the cell state of the current time step and the output of the output gate, and the hidden state h t of the current time step is calculated. Then, backpropagation is performed to calculate the gradient of the loss function with respect to the model parameters and update the model parameters. By stacking multiple LSTM units to construct a deep recurrent neural network, the complex features and patterns in time series data can be better captured.

[0072] During training, cross-validation, parameter tuning, and other techniques can be used to optimize model performance and ensure that the model can accurately represent the normal operating state of the transformer. After the model is constructed, the prediction analysis module can use the model to predict data and evaluate the state of the transformer.

[0073] In this embodiment, the parameters of each data point include Uh, Ih, Ul, Il, Yw, Zf, and Fw. After constructing the transformer normal operation model, the mutual correlation between the above parameters can be reflected. During the training of the model, different evaluation features are selected to train sub-models with specific prediction functions. For example, if the evaluation feature is oil temperature Yw, the sub-model is an oil temperature prediction model, which mainly trains the correlation characteristics of oil temperature Yw with Uh, Ih, Ul, Il, Zf, and Fw. The data predicted by the model is the prediction sample Py, and the actually collected data is the evaluation sample Ps. By comparing the similarity of the prediction sample and the evaluation sample, a result is obtained to reflect the state of the transformer. The specific steps include the following.

[0074] Firstly, an evaluation feature needs to be determined. Hereinafter, taking the current flowing out of the low-voltage side of the transformer as an evaluation feature for example to illustrate.

[0075] After the evaluation feature is determined, the evaluation sample needs to be processed. After the data is extracted, the evaluation sample Ps is obtained according to the detection period determined when the model is constructed and the data collection point determined. Taking the detection period composed of the above-mentioned stable period and fluctuation period as an example, the sample data S=(p11, p12, p13, p14, p15, p21, p22, p23, p24, p25). At this time, the evaluation sample Ps=(p11', p12', p13', p14', p15', p21', p22', p23', p24', p25'). Taking p11' as an example, the original sample p11=(t, Uh, Ih, Ul, Il, Yw, Zf, Fw), and in the evaluation sample data p11'=(t, Uh, Ih, Ul, Yw, Zf, Fw). The extracted low-voltage side current constitutes an evaluation sequence Xp=(s11, s12, s13, s14, s15, s21, s22, s23, s24, s25), wherein s11=(t11, Ilp1), s12=(t12, Ilp2) …… Ilp1 is the transformer low-voltage side current collected at the time point t11, and Ilp2 is the transformer low-voltage side current collected at the time point t12. The current value here is the true value. Pp is an evaluation sequence composed of the true collected data, which is used to compare with the predicted sequence Xy composed of the predicted sample Py.

[0076] The evaluation sample Ps is input into the transformer normal operation model sub-model for predicting the low-voltage side current of the transformer to predict the low-voltage side current of the transformer. The predicted sample Py is obtained. In the predicted sample Py, the time point and the low-voltage side current value are extracted to constitute a predicted sequence Xy=(y11, y12, y13, y14, y15, y21, y22, y23, y24, s25), wherein y11=(t11, Ily1), y12=(t12, I1y2) …… Ily1 is the transformer low-voltage side current at the time point t11 predicted by the model, and I1y2 is the transformer low-voltage side current collected at the time point t12 predicted by the model.

[0077] After the model prediction and sequence data extraction, the prediction sequence Xy=(y11, y12, y13, y14, y15, y21, y22, y23, y24, y25) and the evaluation sequence Xp=(s11, s12, s13, s14, s15, s21, s22, s23, s24, s25) are obtained. At this time, the transformer can be evaluated by comparing the similarity of the two sequences. Dynamic time warping (DTW) can be used to compare the two sequences. At the same time, since Xy and Xp are equal-length sequences, the similarity can be compared by calculating the Euclidean distance. The following is an example of calculating the Euclidean distance.

[0078] To facilitate the calculation, the subscripts are converted, Xy=(y1, y2, y3, y4, y5, y6, y7, y8, y9, y10) and Xp=(s1, s2, s3, s4, s5, s6, s7, s8, s9, s10). The calculation is as follows:

[0079]

[0080] Dsy is the Euclidean distance of the two sequences. The similarity is:

[0081]

[0082] where Dmax is the maximum distance, which can be the range of the difference between the elements of the two sequences, i.e., the maximum difference. Then, the similarity is converted to a percentage as the evaluation score of the transformer.

[0083] The closer the evaluation score is to 100%, the better the performance of the transformer. Due to the limitations of sample data and prediction algorithms, the evaluation score may not reach 100%, so it may be in a normal state at 95% or in a normal state at 98%. This value can be obtained through multiple cross-validation during the training process of the model to get a relatively true value. For example, G% represents the system running normally. If the evaluation score obtained after prediction is 10 percentage points less than G%, it means that the transformer may be in an abnormal state or some performance indicators are declining, at which time the device can be selected for maintenance, or other algorithms can be used to determine faults and abnormalities.

[0084] Further, in order to more accurately reflect the state of the transformer, multiple evaluation features can be set. For example, three evaluation features are set, the first one is the current I1 flowing out of the low voltage side of the transformer, the second one is the transformer oil temperature Yw, and the third one is the transformer vibration Zf. Through the above prediction method, three evaluation scores E1%, E2%, and E2% can be obtained. When evaluating the transformer, the average of the three can be taken as the evaluation score of the state of the transformer. When the transformer is fault warned, the lowest evaluation score is selected as the evaluation score of the transformer. At this time, the abnormality of the transformer can be found in time, and maintenance can be carried out in time to avoid accidents.

[0085] In summary, the above is only a preferred embodiment of the present application, and is not intended to limit the scope of the present application. Based on the above description, relevant personnel can make various changes and modifications without deviating from the technical idea of the present application. The technical scope of the present application is not limited to the content of the specification. Any equivalent changes and modifications of shape, structure, features and spirit within the scope of the present application shall be included in the scope of the claims of the present application.

Claims

1. A transformer comprehensive monitoring and evaluation system based on big data analysis, characterized in that: It comprises a data acquisition module (100), a model construction module (200) and a prediction analysis module (300); The model construction module (200) applies a recurrent neural network algorithm to construct a transformer normal operation model based on the normal operation data obtained by the data acquisition module (100); After determining the evaluation features, the prediction analysis module (300) applies the data collected by the data acquisition module (100) to form an evaluation sample; through the evaluation sample, an evaluation sequence is extracted, and a prediction sequence is obtained by predicting through the transformer normal operation model; Then, by comparing the similarity of the evaluation sequence and the prediction sequence, and converting the similarity into a percentage as the evaluation score of the transformer; During the data acquisition process: first, determine the collection points of the sequence data by the determined interval time and the starting time point of the time span; then, according to the determined collection points, collect data to form sample data within the day; The time span includes a stable period and a fluctuation period, The method for obtaining the stable period is: Step 1, divide the time of a day into m segments and sequentially number them; Step 2, collect j data b1, b2, b3…bj in the a-th segment, where a=1, 2, 3…m; b1, b2, b3…bj are data collected at equal time intervals within the segment; Step 3, Calculate the average and variance b: constituent sample data ​ Step 4, converting the sample data into and forming a sample data set, and then using a clustering algorithm to cluster the data set to obtain clusters reflecting different fluctuation states and average loads; Step 5, for sample data in one cluster, obtain the number a of sample data B according to the original sample data; then, in the sample data of the cluster, check whether there is a number that increases one by one or decreases one by one; If there is, then the time period with consecutive numbers is the stable period; if not, select another cluster and repeat step 5; The method for obtaining the fluctuation period is: Step 6, converting the sample data into (b) and composing a sample data set, and then using a clustering algorithm to cluster the data set to obtain clusters reflecting different fluctuation states; Step 7, extract sample data of a cluster that reacts more greatly, extract a sample with number a, and obtain the average value of a group of time periods centered on a; if the average values of at least two time periods are different, select the group of time periods to form the fluctuation period.

2. The transformer comprehensive monitoring and evaluation system based on big data analysis according to claim 1, characterized in that: The time span includes at least one detection period.

3. The transformer comprehensive monitoring and evaluation system based on big data analysis according to claim 1, characterized in that: In step Step 1: each segment is 5-15 minutes long.

4. The transformer comprehensive monitoring and evaluation system based on big data analysis according to claim 1, characterized in that: b1, b2, b3…bj are the numerical values of an electrical parameter at different time points.

5. The transformer comprehensive monitoring and evaluation system based on big data analysis according to claim 1, characterized in that: At least three evaluation features are set, and the average of the evaluation scores corresponding to all evaluation features is taken as the evaluation score of the transformer performance.

6. The transformer comprehensive monitoring and evaluation system based on big data analysis according to claim 1, characterized in that: If the number a in the selected group of time periods belongs to the cluster in Step 5, repeat step 7 and extract a new number for judgment.

Images