Automatic detection and early warning method for abnormity of transformer substation

Abstract

The invention provides a substation abnormity automatic detection and early warning method, and relates to the field of safety early warning, and the method comprises the following steps: obtaining a target feature vector corresponding to a target transformer in a target time window; inputting the target feature vector into a target AI model to obtain a predicted anomaly occurrence probability corresponding to the target feature vector output by the target AI model within the target time length; wherein the target AI model is obtained according to the following steps: obtaining a first data set D; obtaining a second data set H =; and training the initial AI model according to D and H to obtain a target AI model. When the initial AI model is trained, the negative sample is enhanced while the effectiveness of the negative sample is guaranteed, so that the target AI model is more accurate when obtaining the prediction anomaly occurrence probability.

Description

Technical Field

[0001] The present application relates to the field of safety warning, and in particular to a method for automatic detection and warning of abnormalities in a substation. Background Art

[0002] The substation contains several transformers. The transformer is an electric device that uses electromagnetic principles to convert AC power into different voltage, current and other parameters. Since the transformer oil in the transformer is a flammable liquid, the transformer may spontaneously combust, which in turn leads to fire accidents in the substation. In the prior art, when providing fire warnings for substations, AI models are often used to predict the probability of fires in the future. When selecting negative samples for model training, transformers that have caught fire need to be used. However, the number of transformers that have actually caught fire is relatively small, so there are fewer negative samples. Therefore, the accuracy of the existing AI models is relatively low. Summary of the invention

[0003] In view of this, the present application provides a method for automatic detection and early warning of substation abnormalities, which at least partially solves the technical problems existing in the prior art. The technical solution adopted by the present application is:

[0004] According to a first aspect of the present application, a method for automatic detection and early warning of abnormality in a substation is provided, comprising:

[0005] S100, obtaining a target feature vector corresponding to a target transformer within a target time window;

[0006] S200, inputting the target feature vector into a target AI model to obtain a predicted abnormality occurrence probability corresponding to the target feature vector output by the target AI model within a target time length;

[0007] The target AI model is obtained according to the following steps:

[0008] S210, obtaining a first data set D = (D1, D2, ..., Dδ, ..., Dζ); δ = 1, 2, ..., ζ; wherein ζ is the number of first-category transformers; the first-category transformer is a transformer whose current state is a fire; Dδ is the first sub-data set corresponding to the δth first-category transformer; Dδ = (Dδ1, Dδ2, ..., Dδp, ..., Dδq); p = 1, 2, ..., q; q is the number of first time windows of the dth first-category transformer, and q is the number of days corresponding to the target time length; Dδp is the feature vector corresponding to the δth first-category transformer in the pth first time window; the end time of the first time window corresponding to Dδp is the time point corresponding to the time when the δth first-category transformer is determined to have a fire before the current state is determined to have a fire, and the time from the zero point of the day when the fire occurred is (p-1)*24 hours; the first time window corresponding to Dδp is the same length as the target time window; the probability of occurrence of a marked abnormality corresponding to each feature vector in D is 1;

[0009] S220, obtaining a second data set H = (H1, H2, ..., Hσ, ..., Hr); σ = 1, 2, ..., r; wherein r is the number of second-category transformers; the second-category transformers are transformers whose current state is that no fire has occurred; Hσ is the feature vector corresponding to the σ-th second-category target transformer in the second time window; the second time window has the same length as the target time window; the probability of occurrence of a marked abnormality corresponding to each feature vector in H is 0;

[0010] S230, training the initial AI model according to D and H to obtain the target AI model.

[0011] This application has at least the following beneficial effects:

[0012] In the present application, when training the target AI model, since the number of transformers where fires have occurred is small, in order to improve the accuracy of the target AI model when obtaining negative samples, it is necessary to perform sample enhancement on the negative samples. For the same transformer where fires have occurred, multiple first time windows are determined, each first time window corresponds to a feature vector, the length of the first time window is equal to the length of the first target time window, and the end time of each first time window is different, but the end time of each first time window is within the target time length. Specifically, the end time of the first time window is before zero o'clock on the day of the fire, and the time point corresponding to the time point from zero o'clock on the day of the fire is (p-1)*24 hours (p is the pth first time window) is the end time of the pth first time window. Therefore, it can be ensured that the time when the transformer fire occurs is within the target time length after the end of each first time window, which ensures the effectiveness of the negative samples and enhances the negative samples at the same time, so that the target AI model is more accurate when obtaining the predicted probability of abnormal occurrence.

[0013] When obtaining positive samples, since most transformers are normal and no fire has occurred, sample enhancement is not required, so several second-class transformers (transformers that have not had a fire) are selected, and the feature vector corresponding to each second-class transformer in the second time window is obtained, and the probability of abnormal occurrence of the mark corresponding to each feature vector is set to 0. Here, the second time window has the same length as the first target time window, and the end time of the second time window is any time point before the current time and the length from the current time is less than the target time length. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0015] Figure 1 A flowchart of an embodiment of a substation abnormality early warning method provided by the present application;

[0016] Figure 2 Flowchart of a method for acquiring the target AI model of the substation abnormality warning method provided in this application. DETAILED DESCRIPTION

[0017] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of this application.

[0018] like Figure 1 FIG. 1 is a flow chart of a method for abnormal early warning of a substation provided according to an embodiment of the present application. Figure 1 and Figure 2 As shown, the following steps are included:

[0019] S100, obtaining a target feature vector corresponding to a target transformer within a target time window.

[0020] S200, input the target feature vector into the target AI model to obtain the predicted abnormality occurrence probability corresponding to the target feature vector output by the target AI model within the target time length.

[0021] The target AI model is obtained according to the following steps:

[0022] S210, obtain a first data set D = (D1, D2, ..., Dδ, ..., Dζ); δ = 1, 2, ..., ζ; wherein ζ is the number of first-category transformers; the first-category transformer is a transformer whose current state is that a fire has occurred; Dδ is the first sub-data set corresponding to the δth first-category transformer; Dδ = (Dδ1, Dδ2, ..., Dδp, ..., Dδq); p = 1, 2, ..., q; q is the number of first time windows of the dth first-category transformer, and q is the number of days corresponding to the target time length; Dδp is the feature vector corresponding to the δth first-category transformer in the pth first time window; the end time of the first time window corresponding to Dδp is the time point corresponding to the time when the δth first-category transformer is determined to have a fire before the current state is determined to have a fire, and the time length from the zero point of the day when the fire occurred is (p-1)*24 hours; the first time window corresponding to Dδp is the same length as the target time window; the probability of occurrence of a marked abnormality corresponding to each feature vector in D is 1.

[0023] S220, obtain a second data set H = (H1, H2, ..., Hσ, ..., Hr); σ = 1, 2, ..., r; wherein r is the number of second-category transformers; the second-category transformers are transformers whose current state is that no fire has occurred; Hσ is the feature vector corresponding to the σ-th second-category target transformer in the second time window; the second time window has the same length as the target time window; the probability of occurrence of a marked abnormality corresponding to each feature vector in H is 0. Here, the end time of the second time window is any time point before the current time and the length from the current time is less than the target time length.

[0024] S230, training the initial AI model according to D and H to obtain the target AI model.

[0025] Since the number of transformers that have caught fire is relatively small, in order to improve the accuracy of the target AI model when obtaining negative samples, this embodiment needs to perform sample enhancement on the negative samples. For the same transformer that has caught fire, multiple first time windows are determined, each first time window corresponds to a feature vector, the length of the first time window is equal to the length of the first target time window, and the end time of each first time window is different, but the end time of each first time window is within the target time length. Specifically, the end time of the pth first time window is before midnight on the day of the fire, and the time point corresponding to the time from midnight on the day of the fire is (p-1)*24 hours (p is the pth first time window). Therefore, it can be ensured that the time when the transformer fire occurs is within the target time length after the end of each first time window, thereby ensuring the effectiveness of the negative samples and enhancing the negative samples, thereby improving the accuracy of the target AI model.

[0026] When obtaining positive samples, since most transformers are normal and no fire has occurred, sample enhancement is not required, so several second-class transformers (transformers that have not had a fire) are selected, and the feature vector corresponding to each second-class transformer in the second time window is obtained, and the probability of abnormal occurrence of the mark corresponding to each feature vector is set to 0. Here, the second time window has the same length as the first target time window, and the end time of the second time window is any time point before the current time and the length from the current time is less than the target time length.

[0027] In some optional implementations of some embodiments, the target AI model may also be obtained according to the following steps:

[0028] S240, obtain the first data set D = (D1, D2, ..., Dδ, ..., Dζ); δ = 1, 2, ..., ζ; wherein ζ is the number of first-category transformers; the first-category transformer is the transformer whose current state is that a fire has occurred; Dδ is the first sub-data set corresponding to the δth first-category transformer; Dδ = (Dδ1, Dδ2, ..., Dδp, ..., Dδq); p = 1, 2, ..., q; q is the number of time windows of the dth first-category transformer, and q is the number of days corresponding to the target time length; Dδp is the number of time windows of the δth first-category transformer in the pth first time window The feature vector corresponding to the port; the end time of the first time window corresponding to Dδp is the time point corresponding to (p-1)*24 hours before the current state of the δth first-type transformer is determined to be on fire, and the time from the zero point of the day when the fire occurs is the same as the length of the target time window; the probability of occurrence of the tag anomaly corresponding to Dδp is (tag1=0, tag2=0, …, tagp=1, …, tagq=0); where tagp is the probability of a fire occurring on the pth day after the end time of the first time window corresponding to Dδp;

[0029] S250, obtaining a second data set H = (H1, H2, ..., Hσ, ..., Hr); σ = 1, 2, ..., r; wherein r is the number of second-category transformers; the second-category transformers are transformers whose current state is that no fire has occurred; Hσ is the feature vector corresponding to the σ-th second-category target transformer in the second time window; the second time window has the same length as the target time window; the probability of occurrence of a marked abnormality corresponding to each feature vector in H is 0;

[0030] S260, train the initial AI model according to D and H to obtain the target AI model.

[0031] Here, the labeling method for the feature vector is different from that in the previous embodiment. The labels are marked according to the probability of a fire occurring every day after the end time of the first time window corresponding to Dδp. As the label of the feature vector, compared with the previous embodiment, the model is trained more accurately using the probability of a fire occurring every day within the target time length.

[0032] In an exemplary embodiment of the present application, step S100 includes:

[0033] S110, obtaining a target area identifier of an area where a target substation is located.

[0034] S120: If the target area identifier indicates that the area where the target substation is located is an industrial power consumption area, proceed to step S130.

[0035] It should be noted that the target area identifier is used to identify the area type of the area where the target substation is located, and can identify the area where the target substation is located as a residential power consumption area or an industrial power consumption area.

[0036] S130, obtain the maximum ambient temperature identification set A=(A1, A2, ..., Aa, ..., Ab) corresponding to the target transformer TRA1 in the target substation in the first target time window; a=1, 2, ..., b; wherein the target substation is located in the industrial power consumption area; b is the number of first time periods arranged continuously in the first target time window; Aa is the maximum ambient temperature identification corresponding to TRA1 in the ath first time period; the end time of the first target time window is the current time.

[0037] In some embodiments, the length of the first target time window is b*ΔT1, ΔT1 is the length of each first time period, and the length of the first target window is less than the continuous working time length of the transformer. Specifically, step S130 includes:

[0038] S131, obtaining a maximum ambient temperature set AZ=(AZ1, AZ2, ..., AZa, ..., AZb) corresponding to TRA1 in the first target time window; wherein AZa is the maximum ambient temperature corresponding to TRA1 in the ath first time period;

[0039] S132, determining A according to AZ; wherein Aa is the temperature identifier corresponding to AZa;

[0040] Aa meets the following conditions:

[0041] AZa<TEMP1, then Aa=0;

[0042] TEMP1≤AZa<TEMP2, then Aa=1;

[0043] TEMP2≤AZa<TEMP3, then Aa=2;

[0044] TEMP3≤AZa<TEMP4, then Aa=3;

[0045] TEMP4≤AZa, then Aa=4;

[0046] TEMP1<TEMP2<TEMP3<TEMP4;

[0047] TEMP2-TEMP1>TEMP3-TEMP2>TEMP4-TEMP3;

[0048] Among them, TEMP1, TEMP2, TEMP3, and TEMP4 are preset temperature thresholds.

[0049] In this embodiment, the maximum ambient temperatures corresponding to all first time periods in the first target time window are binned through the above-mentioned multiple temperature intervals. This is because at different temperatures, the equivalent numerical changes in temperature have different effects on the target transformer. For example, when the temperature is below 90°C, it is the normal operating temperature interval of the transformer, so the temperature change in this temperature interval has basically no effect on the target transformer, but after the temperature reaches 105°C, the increase in temperature will have a greater impact on the transformer. When the temperature rises to 120°C, the life of the transformer is only 2 years. Therefore, in this embodiment, multiple temperature ranges of AZa are determined, and the corresponding final values ​​are different, and the length of the temperature interval will gradually decrease (i.e., TEMP2-TEMP1>TEMP3-TEMP2>TEMP4-TEMP3), so that the temperature mark can more accurately reflect the impact of its corresponding maximum ambient temperature on the target transformer.

[0050] S140, obtaining a workpiece information set B=(B1, B2, ..., Ba, ..., Bb) corresponding to TRA1 in the first target time window; wherein Ba is the workpiece information corresponding to TRA1 in the ath first time period.

[0051] In some embodiments, Ba = (Ba1, Ba2, ..., Bae, ..., Baf); e = 1, 2, ..., f; wherein, Bae is the sub-workpiece information corresponding to the workpiece of the e-th target type in TRA1 in the a-th first time period; f is the number of target types; Bae = (NGae, ZGae, EDae); NGae is the working years of the target workpiece in the e-th target type in TRA1 in the a-th first time period; the target workpiece in the e-th target type is the workpiece with the highest probability of scrapping among the workpieces of the e-th target type; ZGae is the maximum working years of the target workpiece in the e-th target type in TRA1 in the a-th first time period; EDae is the rated ambient temperature corresponding to the target workpiece in the e-th target type in TRA1.

[0052] The scrap probability BF of the above target workpiece meets the following conditions: BF=YEAR1 / YEARmax; wherein YEAR1 is the working life of the target workpiece; and YEARmax is the maximum working life of the target workpiece.

[0053] It should be noted that since the maximum service life of the transformer is different at different working temperatures, overheating of the transformer will lead to a reduction in the maximum service life, so the ZGae in this application is obtained by calculation rather than directly obtained according to the workpiece manual. The specific acquisition method is as follows: obtain the average operating temperature of TRA1 in the ath first time period, traverse the preset mapping table based on the average operating temperature of TRA1 in the ath first time period, and determine ZGae. The calculated ZGae is used as part of the target vector, taking into account the impact of the actual operating temperature on the maximum service life of the workpiece. Compared with the method of directly obtaining it according to the workpiece manual, it can better reflect the actual maximum service life of each workpiece. Furthermore, the target AI model is more accurate when obtaining the predicted probability of abnormal occurrence V.

[0054] In this embodiment, when determining the factors in the workpiece information set, the workpiece type corresponding to the relevant workpiece that may cause spontaneous combustion of the transformer in TRA1 is first taken as the target type; then the workpiece with the highest probability of scrapping is selected from all the workpieces of each target type as the target workpiece of the target type, and finally the working years, maximum working years and rated temperature identification of the target workpiece corresponding to each target type in each first time period are used as the workpiece information set corresponding to TRA1 in the first target time window, which characterizes the working years status and optimal working temperature of the target workpiece in TRA1 that may cause spontaneous combustion within a relatively short period of time at the current time, and then determines the aging degree of the target workpiece and the difference between the actual working temperature and the optimal working temperature, and determines whether it will induce spontaneous combustion of the transformer, so that the target AI model is more accurate when obtaining the predicted abnormality occurrence probability V.

[0055] In addition, the maximum service life of the above-mentioned target workpiece is a maximum service life determined based on the average working temperature of the current first time period, rather than being directly determined according to the instructions of the workpiece. The influence of the actual working temperature of the workpiece on the maximum service life of the workpiece is taken into consideration, and the maximum service life of the target workpiece determined in this way is more accurate.

[0056] S150, based on A and B, determine the target feature vector M = (A1, B1, A2, B2, ..., Aa, Ba, ..., Ab, Bb) corresponding to TRA1.

[0057] S160, input M into the target AI model to obtain the predicted abnormality occurrence probability V corresponding to TRA1 output by the target AI model within the future target time length; V meets the following conditions: V = x1*A1+y1*B1+x2*A2+y2*B2+…+xa*Aa+ya*Ba+…+xb*Ab+yb*Bb; wherein x1, x2,…, xa,…, xb, y1, y2,…, ya,…, yb are the parameters obtained by the target AI model after training.

[0058] In some optional implementations of some embodiments, after step S160, the process further includes: if V is greater than a preset probability threshold, outputting prompt information.

[0059] In an exemplary embodiment of the present application, before step S150, the method further includes:

[0060] S151, obtain the estimated temperature identification set C = (C1, C2, ..., Cc, ..., Cd) corresponding to TRA1 in the first target time window; c = 1, 2, ..., d; wherein Cc is the estimated maximum ambient temperature identification of the cth day after the current time corresponding to TRA1; d is the estimated number of days after the current time;

[0061] At this time, the step S150 includes:

[0062] Based on A, B and C, determine the target feature vector M = (A1, B1, A2, B2, ..., Aa, Ba, ..., Ab, Bb, C1, C2, ..., Cc, ..., Cd) corresponding to TRA1.

[0063] Furthermore, V meets the following conditions:

[0064] V=x1*A1+y1*B1+x2*A2+y2*B2+…+xa*Aa+ya*Ba+…+xb*Ab+yb*Bb+z1*C1+z2*C2+…+zc*Cc+…+zd*Cd; wherein z1, z2,…, zc,…, zd are the parameters of the target AI model obtained after training.

[0065] In this embodiment, the estimated maximum ambient temperature identifier of each day within d days after the current time corresponding to TRA1 is obtained, and this is used as a part of the target feature vector, which can reflect the heat dissipation of the transformer of the target transformer under the influence of the maximum ambient temperature within d days after the current time. This is based on the maximum ambient temperature change before the current time and in a period of time short to the current time and the workpiece information change of the target transformer, combined with the estimated maximum ambient temperature of each day within d days after the current time. The time span is longer, including before and after the current time, so that the finally determined V is more accurate.

[0066] The method for obtaining the estimated temperature identification set C corresponding to TRA1 in the first target time window is as follows:

[0067] S1511, obtaining the estimated ambient temperature set CZ=(CZ1, CZ2, ..., CZc, ..., CZd) corresponding to TRA1 in the first target time window; wherein CZc is the estimated maximum ambient temperature of the cth day after the current time corresponding to TRA1;

[0068] S1512, determining C according to CZ; wherein Cc is the temperature identifier corresponding to CZc;

[0069] Cc meets the following conditions:

[0070] CZc<TEMP1, then Cc=0;

[0071] TEMP1≤CZc<TEMP2, then Cc=1;

[0072] TEMP2≤CZc<TEMP3, then Cc=2;

[0073] TEMP3≤CZc<TEMP4, then Cc=3;

[0074] TEMP4≤CZc, then Cc=4;

[0075] TEMP1<TEMP2<TEMP3<TEMP4;

[0076] TEMP2-TEMP1>TEMP3-TEMP2>TEMP4-TEMP3;

[0077] Among them, TEMP1, TEMP2, TEMP3, and TEMP4 are preset temperature thresholds.

[0078] In this embodiment, the estimated maximum ambient temperature corresponding to each day within d days after the current time is divided into boxes through the above-mentioned multiple temperature intervals. This is because at different temperatures, the equivalent numerical change of temperature has different effects on the target transformer. For example, when the temperature is below 90°C, it is the normal working temperature range of the transformer. Therefore, the change of temperature in this temperature range has basically no effect on the target transformer, but after the temperature reaches 105°C, the increase in temperature will have a greater impact on the transformer. When the temperature rises to 120°C, the life of the transformer is only 2 years. Therefore, in this embodiment, multiple temperature ranges of CZc are determined, and the corresponding final values ​​are different, and the length of the temperature interval will gradually decrease (i.e., TEMP2-TEMP1>TEMP3-TEMP2>TEMP4-TEMP3), so that the temperature mark can more accurately reflect the impact of its corresponding estimated maximum ambient temperature on the target transformer.

[0079] In an exemplary embodiment of the present application, before step S150, the method further includes:

[0080] S152, obtain the maintenance information set X=(X1, X2, ..., Xg, ..., Xh) corresponding to TRA1 in the second target time window; g=1, 2, ..., h; wherein h is the number of second time periods arranged continuously in the second target time window; Xg is the maintenance information corresponding to TRA1 in the g-th second time period; Xg=(XJg, XGg); XJg is the number of inspections performed by TRA1 in the g-th second time period; XGg is the workpiece replacement information of the workpiece replaced by TRA1 in the g-th second time period; XGg=(XGg1, XGg2, ..., XGgi, ..., XGgj); i=1, 2, ..., j; XGgi is the workpiece replacement information of the i-th workpiece replaced by TRA1 in the g-th second time period; j is the total number of workpieces replaced by TRA1; XGgi=(XGTgi, XGQgi); wherein, XGTgi is the workpiece replacement time of the i-th workpiece replaced by TRA1 in the g-th second time period; XGQgi is the replacement reason identifier of the i-th workpiece replaced by TRA1 in the g-th second time period; the length of the second target time window is h*ΔT2, ΔT2 is the length of each second time period; (h*ΔT2)>(b*ΔT1), and ΔT2>ΔT1.

[0081] At this time, the step S150 includes:

[0082] According to A, B, C and X, determine the target feature vector M = (A1, B1, A2, B2,…, Aa, Ba,…, Ab, Bb, C1, C2,…, Cc,…, Cd, X1, X2,…, Xg,…, Xh) corresponding to TRA1.

[0083] Furthermore, V satisfies the following condition V=x1*A1+y1*B1+x2*A2+y2*B2+…+xa*Aa+ya*Ba+…+xb*Ab+yb*Bb+z1*C1+z2*C2+…+zc*Cc+…+zd*Cd+k1*X1+k2*X2+…+kg*Xg+…+kh*Xh; among which, k1, k2,…, kg,…, kh are the parameters obtained by the target AI model after training.

[0084] In this embodiment, the maintenance information set X corresponding to TRA1 in the second target time window is obtained, and X includes the maintenance information corresponding to each second time period. Specifically, the maintenance information includes the number of inspections of TRA1 in each time period, and the workpiece replacement information of the workpiece replaced by TRA1 in each time period. Here, the workpiece replacement information includes the replacement time and replacement reason identifier of each replaced workpiece, wherein the replacement reason identifier identifies different replacement reasons, and the replacement reasons include wear, aging, etc. In summary, X reflects the transformer maintenance situation and workpiece replacement details corresponding to each second time period of TRA1 in the second target time window. If, in a second time period of the second target time window, TRA1 is inspected or a key workpiece of TRA1 (the above-mentioned target workpiece, i.e., the relevant workpiece that causes the transformer to spontaneously combust) is replaced and the replacement time is close to the current time, the probability of TRA1 spontaneous combustion will be reduced, otherwise, the probability of TRA1 spontaneous combustion will be increased, so that the target AI model is more accurate when obtaining the predicted probability of abnormal occurrence V.

[0085] It should be noted that the frequency of maintenance and workpiece replacement is low, so in order to obtain as much maintenance information and workpiece replacement information as possible, the length of the second target time window in this embodiment is much greater than the length of the first target time window; and in order to include as much valid information as possible in each second time period, the length of the second time period is also much greater than the length of the first time period.

[0086] In the present application, the parameter combination method in M ​​is not limited to the above-mentioned implementation mode. As an example: M = (A1, B1, A2, B2, …, Aa, Ba, …, Ab, Bb, C1, C2, …, Cc, …, Cd, X1, X2, …, Xg, …, Xh) or M = (A1, B1, A2, B2, …, Aa, Ba, …, Ab, Bb, X1, X2, …, Xg, …, Xh).

[0087] In an exemplary embodiment of the present application, the step S120 includes: if the target area identifier identifies that the area where the target substation is located is a residential power consumption area, then proceeding to step S170:

[0088] S170, obtain the maximum ambient temperature identification set E=(E1, E2, ..., Ea, ..., Eb) corresponding to the target transformer TRA2 in the target substation within the first target time window; a=1, 2, ..., b; wherein b is the number of first time periods arranged continuously in the first target time window; Ea is the maximum ambient temperature identification corresponding to TRA2 in the ath first time period; the end time of the first target time window is the current time; the length of the first target time window is b*ΔT1, and ΔT1 is the length of each first time period.

[0089] In some embodiments, step S170 includes:

[0090] S171, obtaining the maximum ambient temperature set EZ corresponding to TRA2 in the first target time window = (EZ1, EZ2, ..., EZa, ..., EZb); wherein EZa is the maximum ambient temperature corresponding to TRA2 in the ath first time period;

[0091] S172, determining E according to EZ; wherein Ea is the temperature identifier corresponding to EZa;

[0092] Ea meets the following conditions:

[0093] EZa<TEMP1, then Ea=0;

[0094] TEMP1≤EZa<TEMP2, then Ea=1;

[0095] TEMP2≤EZa<TEMP3, then Ea=2;

[0096] TEMP3≤EZa<TEMP4, then Ea=3;

[0097] TEMP4≤EZa, then Ea=4;

[0098] TEMP1<TEMP2<TEMP3<TEMP4;

[0099] TEMP2-TEMP1>TEMP3-TEMP2>TEMP4-TEMP3;

[0100] Among them, TEMP1, TEMP2, TEMP3, and TEMP4 are preset temperature thresholds.

[0101] In this embodiment, the maximum ambient temperatures corresponding to all first time periods in the first target time window are binned through the above-mentioned multiple temperature intervals. This is because at different temperatures, the equivalent numerical changes in temperature have different effects on the target transformer. For example, when the temperature is below 90°C, it is the normal operating temperature interval of the transformer. Therefore, the temperature change in this temperature interval has basically no effect on the target transformer, but after the temperature reaches 105°C, the increase in temperature will have a greater impact on the transformer. When the temperature rises to 120°C, the life of the transformer is only 2 years. Therefore, in this embodiment, multiple temperature ranges of EZa are determined, and the corresponding final values ​​are different, and the length of the temperature interval will gradually decrease (i.e., TEMP2-TEMP1>TEMP3-TEMP2>TEMP4-TEMP3), so that the temperature mark can more accurately reflect the impact of its corresponding maximum ambient temperature on the target transformer.

[0102] S180, obtain the transformer working information set F = (F1, F2, ..., Fa, ..., Fb) corresponding to TRA2 in the first target time window; Fa is the transformer working information corresponding to TRA2 in the a-th first time period; Fa = (PFa, TFa, QFa); wherein PFa is the maximum power consumption of TRA2 in the a-th first time period; TFa is the maximum operating temperature of TRA2 in the a-th first time period; QFa is the weather type identifier corresponding to TRA2 in the a-th first time period.

[0103] S190, obtaining basic transformer information J = (Pmax, Pmin, L, S) of TRA2; wherein Pmax is the maximum output power of TRA2 within the first target time window; Pmin is the minimum output power of TRA2 within the first target time window; L is the number of households connected to TRA2; and S is the service life of TRA2.

[0104] S1000, determine the target feature vector N = (E1, F1, E2, F2, ..., Ea, Fa, ..., Eb, Fb, Pmax, Pmin, L, S) corresponding to TRA2 based on E, F and J.

[0105] S1100, input N into the target AI model to obtain the predicted abnormality occurrence probability U corresponding to TRA2 output by the target AI model within the future target time length; wherein, U meets the following conditions: U = β1*E1+s1*F1+β2*E2+s2*F2+…+βa*Ea+sa*Fa+…+βb*Eb+sb*Fb+m*Pmaβ+n*Pmin+v*L+w*S; wherein, β1, β2,…, βa,…, βb, s1, s2,…, sa,…, sb, m, n, v, w are the parameters obtained by the target AI model after training.

[0106] Here, L meets the following conditions:

[0107] LZ<NUM1, then L=0;

[0108] NUM1≤LZ<NUM2, then L=1;

[0109] NUM2≤LZ<NUM3, then L=2;

[0110] NUM3≤LZ<NUM4, then L=3;

[0111] NUM4≤LZ, then L=4;

[0112] NUM1<NUM2<NUM3<NUM4;

[0113] Among them, LZ is the number of households connected to TRA2; NUM1, NUM2, NUM3, and NUM4 are preset household number thresholds.

[0114] Here, in order to make the parameters in the target feature vector similar in magnitude, the above L is subjected to data binning.

[0115] In some optional implementations of some embodiments, after step S1100, the process further includes: if U is greater than a preset probability threshold, outputting prompt information.

[0116] In this embodiment, for the target transformer TRA2 in the target substation located in the residential power consumption area, the maximum ambient temperature identification set, transformer working information set and transformer basic information corresponding to the first target time window are obtained, and the maximum ambient temperature identification set, transformer working information set and transformer basic information corresponding to the first target time window are used as the target feature vector of TRA2, and the number and end time of the first time period of the time window are controlled at the same time, so that the parameters in the target feature vector can reflect the maximum ambient temperature change of the target transformer in a period of time shorter than the current time, weather change, output power change of the target transformer when working, workload change, service life, maximum working temperature change and other information. Here, since TRA2 is located in the residential power consumption area, and the characteristics of the substation in the residential power consumption area are that the power use has high and low peaks, and the transformer has rest time, so the output power and maximum working temperature of TRA2 vary greatly, and the output power of TRA2 during the peak power consumption period (daytime) is greater than the output power of TRA2 during the low power consumption period (nighttime), and the maximum working temperature is the same. According to the maximum ambient temperature, weather type identification, maximum / minimum output power and maximum operating temperature corresponding to each first time period in the first target time window, the heat dissipation of TRA2 can be determined as part of the target feature vector of TRA2, so that the final determined U is more accurate. In addition, basic information such as the number of households connected to TRA2, the maximum power consumption and service life are also important factors that cause TRA2 to spontaneously combust. The more households connected or the greater the maximum power consumption, the greater the current operating load of TRA2, and the greater the possibility of spontaneous combustion. The longer the service life, the more serious the aging of the internal workpieces of TRA2, and the greater the possibility of spontaneous combustion. Therefore, this application uses the above-mentioned maximum ambient temperature identification set, transformer working information set and transformer basic information as the target feature vector of TRA2, so that the target AI model is more accurate when obtaining the predicted abnormality probability U.

[0117] In some optional implementations of some embodiments, before step S1000, the method further includes:

[0118] S1010, obtaining the estimated temperature set G=(G1, G2, ..., Gc, ..., Gd) corresponding to TRA2 in the first target time window; c=1, 2, ..., d; wherein Gc is the estimated maximum ambient temperature identifier of the cth day after the current time corresponding to TRA2; d is the estimated number of days after the current time;

[0119] At this time, step S1000 includes:

[0120] According to E, F, G and J, determine the target feature vector N = (E1, F1, E2, F2,…, Ea, Fa,…, Eb, Fb, G1, G2,…, Gc,…, Gd, Pmax, Pmin, L, S) corresponding to TRA2.

[0121] U meets the following conditions: U = β1*E1+s1*F1+β2*E2+s2*F2+…+βa*Ea+sa*Fa+…+βb*Eb+sb*Fb+t1*G1+t2*G2+…+tc*Gc+…+td*Gd+m*Pmaβ+n*Pmin+v*L+w*S; among which, t1, t2,…, tc,…, td are the parameters obtained by the target AI model after training.

[0122] In this embodiment, the estimated maximum ambient temperature identifier of each day within d days after the current time corresponding to TRA2 is obtained, and this is used as a part of the target feature vector, which can reflect the transformer heat dissipation condition of the target transformer under the influence of the maximum ambient temperature within d days after the current time. This is based on the maximum ambient temperature change before the current time and in a period of time short to the current time and the workpiece information change of the target transformer, combined with the estimated maximum ambient temperature of each day within d days after the current time, and the time span is longer, including before the current time and after the current time, so that the finally determined U is more accurate.

[0123] The method for obtaining the estimated temperature identification set G corresponding to TRA2 in the first target time window is as follows:

[0124] S1011, obtaining an estimated ambient temperature set GZ=(GZ1, GZ2, ..., GZc, ..., GZd) corresponding to TRA2 in the first target time window; wherein GZc is the estimated maximum ambient temperature of TRA2 corresponding to the current time and the cth day after that;

[0125] S1012, determining G according to GZ; wherein Gc is the temperature identifier corresponding to GZc;

[0126] Gc meets the following criteria:

[0127] GZc<TEMP1, then Gc=0;

[0128] TEMP1≤GZc<TEMP2, then Gc=1;

[0129] TEMP2≤GZc<TEMP3, then Gc=2;

[0130] TEMP3≤GZc<TEMP4, then Gc=3;

[0131] TEMP4≤GZc, then Gc=4;

[0132] TEMP1<TEMP2<TEMP3<TEMP4;

[0133] TEMP2-TEMP1>TEMP3-TEMP2>TEMP4-TEMP3;

[0134] Among them, TEMP1, TEMP2, TEMP3, and TEMP4 are preset temperature thresholds.

[0135] In this embodiment, the estimated maximum ambient temperature corresponding to each day within d days after the current time is divided into boxes through the above-mentioned multiple temperature intervals. This is because at different temperatures, the equivalent numerical change of temperature has different effects on the target transformer. For example, when the temperature is below 90°C, it is the normal working temperature range of the transformer. Therefore, the change of temperature in this temperature range has basically no effect on the target transformer, but after the temperature reaches 105°C, the increase in temperature will have a greater impact on the transformer. When the temperature rises to 120°C, the life of the transformer is only 2 years. Therefore, in this embodiment, multiple temperature ranges of GZc are determined, and the corresponding final values ​​are different, and the length of the temperature interval will gradually decrease (i.e., TEMP2-TEMP1>TEMP3-TEMP2>TEMP4-TEMP3), so that the temperature mark can more accurately reflect the impact of its corresponding estimated maximum ambient temperature on TRA2.

[0136] In some optional implementations of some embodiments, before step S1000, the method further includes:

[0137] S1020, obtaining a maintenance information set Y=(Y1, Y2, ..., Yg, ..., Yh) corresponding to TRA2 in the second target time window; g=1, 2, ..., h; wherein h is the number of second time periods arranged consecutively in the second target time window; Yg is maintenance information corresponding to TRA2 in the g-th second time period Yg=(YJg, YGg); YJg is the number of inspections performed by TRA2 in the g-th second time period; YGg is workpiece replacement information of workpieces replaced by TRA2 in the g-th second time period; YGg=(YGg1, YGg2, ..., YGgθ, ..., YGgε); θ=1, 2, ..., ε; Among them, YGgθ is the workpiece replacement information of the θth workpiece replaced by TRA2 in the gth second time period; ε is the total number of workpieces replaced included in TRA2; YGgθ=(YGTgθ, YGQgθ); wherein, YGTgθ is the workpiece replacement time of the θth workpiece replaced by TRA2 in the gth second time period; YGQgθ is the replacement reason identifier of the θth workpiece replaced by TRA2 in the gth second time period; the length of the second target time window is h*ΔT2, ΔT2 is the length of each second time period; (h*ΔT2)>(b*ΔT1), and ΔT2>ΔT1.

[0138] At this time, step S1000 includes:

[0139] According to E, F, G, J and Y, determine the target feature vector N = (E1, F1, E2, F2,…, Ea, Fa,…, Eb, Fb, G1, G2,…, Gc,…, Gd, Y1, Y2,…, Yg,…, Yh, Pmax, Pmin, L, S) corresponding to TRA2.

[0140] U meets the following conditions: U = β1*E1+s1*F1+β2*E2+s2*F2+…+βa*Ea+sa*Fa+…+βb*Eb+sb*Fb+π1*Y1+π2*Y2+…+πg*Yg+…+πh*Yh+m*Pmaβ+n*Pmin+v*L+w*S; among which, π1, π2,…, πg,…, πh are the parameters obtained by the target AI model after training.

[0141] In this embodiment, the maintenance information set Y corresponding to TRA2 in the second target time window is obtained, and Y includes the maintenance information corresponding to each second time period. Specifically, the maintenance information includes the number of inspections of TRA2 in each time period, and the workpiece replacement information of the workpiece replaced by TRA2 in each time period. Here, the workpiece replacement information includes the replacement time and replacement reason identifier of each replaced workpiece, wherein the replacement reason identifier identifies different replacement reasons, and the replacement reasons include wear, aging, etc. In summary, Y reflects the transformer maintenance situation and workpiece replacement details corresponding to each second time period of TRA2 in the second target time window. If, in a second time period of the second target time window, TRA2 is inspected or a key workpiece of TRA2 (the above-mentioned target workpiece, i.e., the relevant workpiece that causes the transformer to spontaneously combust) is replaced and the replacement time is close to the current time, the probability of TRA2 spontaneous combustion will be reduced, otherwise, the probability of TRA2 spontaneous combustion will be increased, so that the target AI model is more accurate when obtaining the predicted abnormality occurrence probability U.

[0142] It should be noted that the frequency of maintenance and workpiece replacement is low, so in order to obtain as much maintenance information and workpiece replacement information as possible, the length of the second target time window in this embodiment is much greater than the length of the first target time window; and in order to include as much valid information as possible in each second time period, the length of the second time period is also much greater than the length of the first time period.

[0143] In the present application, the parameter combination method in N is not limited to the above-mentioned implementation mode. As an example: N = (E1, F1, E2, F2, ..., Ea, Fa, ..., Eb, Fb, G1, G2, ..., Gc, ..., Gd, Y1, Y2, ..., Yg, ..., Yh, Pmax, Pmin, L, S) or N = (E1, F1, E2, F2, ..., Ea, Fa, ..., Eb, Fb, Y1, Y2, ..., Yg, ..., Yh, Pmax, Pmin, L, S).

[0144] In addition, the above-mentioned figures are only schematic illustrations of the processes included in the method according to the exemplary embodiments of the present application, and are not intended to be limiting. It is easy to understand that the processes shown in the above-mentioned figures do not indicate or limit the time sequence of these processes. In addition, it is also easy to understand that these processes can be performed synchronously or asynchronously, for example, in multiple modules.

[0145] It should be noted that, although several modules or units of the device for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. On the contrary, the features and functions of one module or unit described above can be further divided into multiple modules or units to be embodied.

[0146] The above are only specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. A method for automatic detection and early warning of substation abnormalities, characterized in that: The method comprises: S100, obtaining a target feature vector corresponding to a target transformer in a substation within a target time window; S200, input the target feature vector into the target AI model to obtain the predicted abnormality occurrence probability corresponding to the target feature vector output by the target AI model within the target time length.

2. A method for automatic detection and early warning of substation abnormality according to claim 1, characterized in that: The target AI model is obtained according to the following steps: S210, obtaining a first data set D = (D1, D2, ..., Dδ, ..., Dζ); δ = 1, 2, ..., ζ; wherein ζ is the number of first-category transformers; the first-category transformer is a transformer whose current state is a fire; Dδ is the first sub-data set corresponding to the δth first-category transformer; Dδ = (Dδ1, Dδ2, ..., Dδp, ..., Dδq); p = 1, 2, ..., q; q is the number of first time windows of the dth first-category transformer, and q is the number of days corresponding to the target time length; Dδp is the feature vector corresponding to the δth first-category transformer in the pth first time window; the end time of the first time window corresponding to Dδp is the time point corresponding to the time when the δth first-category transformer is determined to have a fire before the current state is determined to have a fire, and the time from the zero point of the day when the fire occurred is (p-1)*24 hours; the first time window corresponding to Dδp is the same length as the target time window; the probability of occurrence of a marked abnormality corresponding to each feature vector in D is 1; S220, obtaining a second data set H = (H1, H2, ..., Hσ, ..., Hr); σ = 1, 2, ..., r; wherein r is the number of second-category transformers; the second-category transformers are transformers whose current state is that no fire has occurred; Hσ is the feature vector corresponding to the σ-th second-category target transformer in the second time window; the second time window has the same length as the target time window; the probability of occurrence of a marked abnormality corresponding to each feature vector in H is 0; S230, training the initial AI model according to D and H to obtain the target AI model.

3. A method for automatic detection and early warning of substation abnormality according to claim 2, characterized in that: The end time of the second time window is any time point before the current time and whose length from the current time is less than the target time length.

4. A method for automatic detection and early warning of abnormality in a substation according to claim 1, characterized in that: The target AI model is obtained according to the following steps: S240, obtain the first data set D = (D1, D2, ..., Dδ, ..., Dζ); δ = 1, 2, ..., ζ; wherein ζ is the number of first-category transformers; the first-category transformer is the transformer whose current state is that a fire has occurred; Dδ is the first sub-data set corresponding to the δth first-category transformer; Dδ = (Dδ1, Dδ2, ..., Dδp, ..., Dδq); p = 1, 2, ..., q; q is the number of time windows of the dth first-category transformer, and q is the number of days corresponding to the target time length; Dδp is the number of time windows of the δth first-category transformer in the pth first time window The feature vector corresponding to the port; the end time of the first time window corresponding to Dδp is the time point corresponding to (p-1)*24 hours before the current state of the δth first-type transformer is determined to be on fire, and the time from the zero point of the day when the fire occurs is the same as the length of the target time window; the probability of occurrence of the tag anomaly corresponding to Dδp is (tag1=0, tag2=0, …, tagp=1, …, tagq=0); where tagp is the probability of a fire occurring on the pth day after the end time of the first time window corresponding to Dδp; S250, obtaining a second data set H = (H1, H2, ..., Hσ, ..., Hr); σ = 1, 2, ..., r; wherein r is the number of second-category transformers; the second-category transformers are transformers whose current state is that no fire has occurred; Hσ is the feature vector corresponding to the σ-th second-category target transformer in the second time window; the second time window has the same length as the target time window; the probability of occurrence of a marked abnormality corresponding to each feature vector in H is 0; S260, train the initial AI model according to D and H to obtain the target AI model.

5. The method for automatic detection and early warning of abnormality in a substation according to claim 1, characterized in that: The step S100 includes: S110, obtaining a target area identifier of an area where a target substation is located; S120, if the target area identifier indicates that the area where the target substation is located is an industrial power consumption area, proceed to step S130; S130, obtaining a maximum ambient temperature identifier set A=(A1, A2, ..., Aa, ..., Ab) corresponding to a target transformer TRA1 in a target substation in a first target time window; a=1, 2, ..., b; wherein the target substation is located in an industrial power consumption area; b is the number of first time periods arranged continuously in the first target time window; Aa is the maximum ambient temperature identifier corresponding to TRA1 in the ath first time period; the end time of the first target time window is the current time; S140, obtaining a workpiece information set B corresponding to TRA1 in the first target time window = (B1, B2, ..., Ba, ..., Bb); wherein Ba is the workpiece information corresponding to TRA1 in the ath first time period; S150, according to A and B, determine the target feature vector M corresponding to TRA1 = (A1, B1, A2, B2, ..., Aa, Ba, ..., Ab, Bb); S160, input M into the target AI model to obtain the predicted abnormality occurrence probability V corresponding to TRA1 output by the target AI model within the future target time length; V meets the following conditions: V = x1*A1+y1*B1+x2*A2+y2*B2+…+xa*Aa+ya*Ba+…+xb*Ab+yb*Bb; wherein x1, x2,…, xa,…, xb, y1, y2,…, ya,…, yb are the parameters obtained by the target AI model after training.

6. A method for automatic detection and early warning of abnormality in a substation according to claim 5, characterized in that: The length of the first target time window is b*ΔT1, where ΔT1 is the length of each first time period.

7. A method for automatic detection and early warning of abnormality in a substation according to claim 5, characterized in that: The steps S130, including: S131, obtaining a maximum ambient temperature set AZ=(AZ1, AZ2, ..., AZa, ..., AZb) corresponding to TRA1 in the first target time window; wherein AZa is the maximum ambient temperature corresponding to TRA1 in the ath first time period; S132, determining A according to AZ; wherein Aa is the temperature identifier corresponding to AZa; Aa meets the following conditions: AZa<TEMP1, then Aa=0; TEMP1≤AZa<TEMP2, then Aa=1; TEMP2≤AZa<TEMP3, then Aa=2; TEMP3≤AZa<TEMP4, then Aa=3; TEMP4≤AZa, then Aa=4; TEMP1<TEMP2<TEMP3<TEMP4; TEMP2-TEMP1>TEMP3-TEMP2>TEMP4-TEMP3; Among them, TEMP1, TEMP2, TEMP3, and TEMP4 are preset temperature thresholds.

8. A method for automatic detection and early warning of abnormality in a substation according to claim 5, characterized in that: The step S120 includes: if the target area identifier indicates that the area where the target substation is located is a residential electricity consumption area, then proceeding to step S170: S170, obtaining a maximum ambient temperature identifier set E=(E1, E2, ..., Ea, ..., Eb) corresponding to the target transformer TRA2 in the target substation within the first target time window; a=1, 2, ..., b; wherein b is the number of first time periods arranged continuously within the first target time window; Ea is the maximum ambient temperature identifier corresponding to TRA2 in the ath first time period; the end time of the first target time window is the current time; the length of the first target time window is b*ΔT1, where ΔT1 is the length of each first time period; S180, obtaining a transformer operating information set F=(F1, F2, ..., Fa, ..., Fb) corresponding to TRA2 in the first target time window; Fa is the transformer operating information corresponding to TRA2 in the a-th first time period; Fa=(PFa, TFa, QFa); wherein PFa is the maximum power consumption of TRA2 in the a-th first time period; TFa is the maximum operating temperature of TRA2 in the a-th first time period; QFa is the weather type identifier corresponding to TRA2 in the a-th first time period; S190, obtaining basic transformer information J=(Pmax, Pmin, L, S) of TRA2; wherein Pmax is the maximum output power of TRA2 within the first target time window; Pmin is the minimum output power of TRA2 within the first target time window; L is the number of households connected to TRA2; S is the service life of TRA2; S1000, determine the target feature vector N corresponding to TRA2 according to E, F and J = (E1, F1, E2, F2, ..., Ea, Fa, ..., Eb, Fb, Pmax, Pmin, L, S); S1100, input N into the target AI model to obtain the predicted abnormality occurrence probability U corresponding to TRA2 output by the target AI model within the future target time length; wherein, U meets the following conditions: U = β1*E1+s1*F1+β2*E2+s2*F2+…+βa*Ea+sa*Fa+…+βb*Eb+sb*Fb+m*Pmaβ+n*Pmin+v*L+w*S; wherein, β1, β2,…, βa,…, βb, s1, s2,…, sa,…, sb, m, n, v, w are the parameters obtained by the target AI model after training.

9. A method for automatic detection and early warning of abnormality in a substation according to claim 8, characterized in that: The step S170 includes: S171, obtaining the maximum ambient temperature set EZ corresponding to TRA2 in the first target time window = (EZ1, EZ2, ..., EZa, ..., EZb); wherein EZa is the maximum ambient temperature corresponding to TRA2 in the ath first time period; S172, determining E according to EZ; wherein Ea is the temperature identifier corresponding to EZa; Ea meets the following conditions: EZa<TEMP1, then Ea=0; TEMP1≤EZa<TEMP2, then Ea=1; TEMP2≤EZa<TEMP3, then Ea=2; TEMP3≤EZa<TEMP4, then Ea=3; TEMP4≤EZa, then Ea=4; TEMP1<TEMP2<TEMP3<TEMP4; TEMP2-TEMP1>TEMP3-TEMP2>TEMP4-TEMP3; Among them, TEMP1, TEMP2, TEMP3, and TEMP4 are preset temperature thresholds.

Images