Converter valve temperature prediction method and device based on multi-source data fusion

Abstract

The application discloses a converter valve temperature prediction method and device based on multi-source data fusion, and the method comprises the following steps: obtaining multiple historical operation data of a converter station at n time points, wherein one kind is cooling water outlet valve temperature T o , and a standardized historical operation data matrix D is constructed by using other historical operation data; the operation conditions of the converter station are classified based on the standardized historical operation data matrix D; the operation condition data of different categories are processed by using a BP neural network model, and the relationship between the cooling water outlet valve temperature T o and other historical operation data under different operation condition categories is fitted; real-time converter station operation data are collected, the cooling water outlet valve temperature T o is obtained by using the trained BP neural network, and it is judged whether the predicted value of the cooling water outlet valve temperature T o is within a normal threshold value, and then the operation parameters of the converter station are adjusted correspondingly; the application has the advantages that the converter valve temperature prediction precision is high.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power grid, more particularly to a converter valve temperature prediction method and device based on multi-source data fusion. BACKGROUND

[0002] The ultra-high voltage converter station is an important device for power operation activities, and its safe and stable operation is of great significance to power transmission. However, the operating conditions of the ultra-high voltage converter station and the relationship between various operating parameters are relatively complex, so the research project on the key parameter testing and evaluation technology of the DC control and protection system of the ultra-high voltage converter station based on the digital twin model is carried out to study the key parameters of the DC control and protection system of the ultra-high voltage converter station.

[0003] The converter valve in the ultra-high voltage converter station is the core equipment for AC-DC power conversion, and the current-carrying elements will generate a large amount of heat during normal operation. Part of the heat is discharged to the outdoor by the cooling water circulation system, and the other part is exchanged with the air in the valve hall through natural convection. Once the heat exchange performance of the converter station is abnormal, the cooling water outlet temperature will rise, and in severe cases, it will reach the maximum value of the allowable temperature of the converter valve operation, resulting in the occurrence of forced shutdown events of the DC system. The outlet water temperature of the converter valve is an important indicator reflecting the heat exchange performance of the cooling tower, which is affected by multiple operating parameters of the valve cooling system. Therefore, it is of great significance to evaluate the safety operation margin of the DC system to predict the outlet water temperature of the converter valve by comprehensively considering multiple operating parameters.

[0004] Chinese patent publication No. CN112672594A discloses a converter station valve cooling system inlet valve temperature prediction monitoring method and system, relating to the technical field of power system detection. The method includes obtaining the converter valve DC load rate P, the spray water flow L s , the cooling water flow L f , the cooling tower air intake Q, and the environmental wet bulb temperature T w ; constructing an inlet valve temperature prediction model according to the obtained physical quantities; and monitoring the cooling tower heat exchange performance of the converter valve according to the established inlet valve temperature prediction model. The patent application judges the specific fault reason by spray water flow, cooling water flow, cooling tower air intake, and environmental wet bulb temperature, sends an alarm to remind the operator, and improves the operation reliability of the converter station valve cooling system. First, the patent application predicts the inlet valve temperature, and second, the relationship between the converter station operating data and the cooling water outlet valve temperature is nonlinear and complex, and the data of the converter station under different operating conditions differs greatly. The patent application directly performs extensive fitting, and the prediction result and the actual value will have a large difference, resulting in low prediction accuracy. SUMMARY

[0005] The technical problem to be solved by the present application is that the existing technology converter valve temperature prediction method directly performs extensive fitting, and there is a large difference between the prediction result and the actual value, resulting in low prediction accuracy.

[0006] The present application solves the above technical problems by the following technical means: a converter valve temperature prediction method based on multi-source data fusion, the method comprising:

[0007] Step one: obtain multiple historical operation data of the converter station at n time points, one of which is the cooling water outlet valve temperature T o , and use other historical operation data to construct a standardized historical operation data matrix D;

[0008] Step two: based on the standardized historical operation data matrix D, the operating conditions of the converter station are classified by a fuzzy clustering method;

[0009] Step three: use a BP neural network model to process the operating condition data of different categories, and fit the relationship between the cooling water outlet valve temperature T o and other historical operation data under different operating condition categories;

[0010] Step four: real-time acquisition of the operating data of the converter station, classification of the operating data and input of the trained BP neural network to obtain the cooling water outlet valve temperature T o , and judge whether the predicted value of the cooling water outlet valve temperature T o is within the normal threshold, and then make corresponding adjustments to the operating parameters of the converter station, or input the target value of the cooling water outlet valve temperature into the trained BP neural network, and according to the preset target value of the cooling water outlet valve temperature, the parameter threshold of the operating condition is back calculated.

[0011] The present application classifies the operating conditions of the converter station by a fuzzy clustering method, processes the operating condition data of different categories by a BP neural network model, and fits under different categories, which can achieve a more precise prediction effect.

[0012] Further, the step one comprises:

[0013] The multiple historical operation data of the converter station at n time points include 1 electrical quantity data: DC load P; 2 environmental data: environmental temperature T S , valve hall temperature T P ; 5 water cooling system data: expansion tank water level L P , spray pool water level L S , main waterway conductivity sigma, cooling water inlet valve temperature T i , and cooling water outlet valve temperature T o; the first 7 historical running data are put into a historical running data matrix X of [n x 7] dimension, and the historical running data matrix X is normalized to obtain a standardized historical running data matrix D.

[0014] Further, the step one further comprises:

[0015] S11, collecting the converter station running data at a time interval of 4h, and constructing a converter station historical running data matrix X = [P T S T P L P L S σ T i ] according to the running data at n time points, wherein P = [P1 P2 … P j … P n ] T , P j represents the DC load power at the jth time point, and other running data are the same;

[0016] S12, normalizing the historical running data matrix X to obtain a standardized historical running data matrix D as follows

[0017]

[0018] wherein max(p) is the maximum value of the DC load power at each sampling point, is the unit value of the DC load power at the jth time point after the maximum value normalization, P S is a vector composed of n standardized DC load powers.

[0019] Further, the step two comprises:

[0020] S21, constructing a clustering effectiveness function;

[0021] S22, constructing an objective function, and obtaining the membership matrix U and the clustering center V when the classification number c is from 2 to N, wherein N is the number of converter stations;

[0022] S23, substituting the membership matrix U and the clustering center V corresponding to each c value into the clustering effectiveness function formula respectively, when the clustering effectiveness function value reaches the maximum value, the classification number c * corresponding to this time is the best classification number, the membership matrix U * and the clustering center V * at this time are the membership matrix and the clustering center at the best classification number c * ;

[0023] S24, according to the membership matrix U * , the running states of the N converter stations are divided into c* Class.

[0024] Further, the cluster validity function in the step S21 is

[0025]

[0026] where N is the number of converter stations; c is the number of classes; y = 1, 2, …, c; u xy represents the membership of the xthconverter station operating state belonging to the ythclass, U is a c x N membership matrix composed of u xy , D x is the converter station operating data matrix X x of the xthconverter station operating state, x = 1, 2, …, N; V y is the classification center of the ythclass, is the sum of distances from all converter stations to the converter station center V0and is the possibility partition coefficient.

[0027] Further, the step S22 includes:

[0028] constructing an objective function where m ∈ [1, +∞) is a fuzzy index, d ij = ||x j -v i || represents the Euclidean distance of the sample x j to the cluster center v i of the ithclass, and λ is a Lagrange multiplier.

[0029] The membership of the jthsample data belonging to the ithclass is solved by using the Lagrange multiplier method as

[0030]

[0031] The cluster center v is solved by using the Lagrange multiplier method as

[0032]

[0033] The membership matrix U and the cluster center V of the classification number c from 2 to N are respectively solved by the above formula.

[0034] Further, the cluster center V in the step S22 is:

[0035] V = [V1 V2 … V y … V c ] T

[0036] Vy = [V y1 V y2 … V yt … V y24 ]

[0037] wherein, V is the clustering center, V y is the clustering center of the y-th class, V yt is the feature quantity of the clustering center of the y-th class at the sampling point t.

[0038] Further, the step three comprises:

[0039] The BP neural network model is used to process the operation condition data of different classes, the cooling water outlet valve temperature T o is the output of the BP neural network model, and other operation data is the input, the relationship between the cooling water outlet valve temperature T o and other operation data under different operation condition classes is fitted, and the BP neural network model is trained, the trained model is used as the first BP neural network model, and the other operation data includes the direct current load P, the environment temperature T S , the valve hall temperature T P , the expansion water tank water level L P , the spray pool water level L S , the main waterway conductivity σ, and the cooling water inlet valve temperature T i .

[0040] Alternatively, the cooling water outlet valve temperature T o is the input of the BP neural network model, and the direct current load P, the environment temperature T S , the valve hall temperature T P , the expansion water tank water level L P , the spray pool water level L S , the main waterway conductivity σ, and the cooling water inlet valve temperature T i are the output of the BP neural network model, the corresponding relationship between the cooling water outlet valve temperature T o and the output is obtained by training the massive data, and the trained model is used as the second BP neural network model.

[0041] Further, the step four comprises:

[0042] The operation data of the converter station is collected in real time, the operation data is classified and input into the trained first BP neural network to obtain the cooling water outlet valve temperature T o , it is judged whether the predicted value of the cooling water outlet valve temperature T o is within the normal threshold, and then the operation parameters of the converter station are adjusted accordingly.

[0043] Or, input the cooling water outlet valve temperature T o Target value, according to the preset cooling water outlet valve temperature T o The target value is used to deduce the parameter threshold of the operating condition.

[0044] The application also provides a converter valve temperature prediction device based on multi-source data fusion, which comprises:

[0045] The operation data preprocessing module is used to acquire a plurality of historical operation data of the converter station at n time points, wherein one of the historical operation data is the cooling water outlet valve temperature T o , and the other historical operation data is used to construct a standardized historical operation data matrix D.

[0046] The operation condition classification module is used to classify the operation conditions of the converter station based on the standardized historical operation data matrix D through a fuzzy clustering method.

[0047] The model training module is used to process the operation condition data of different categories by using a BP neural network model, and fit the relationship between the cooling water outlet valve temperature T o and the other historical operation data under different operation condition categories.

[0048] The converter valve temperature prediction module is used to collect the operation data of the converter station in real time, classify the operation data, and input the trained BP neural network to obtain the cooling water outlet valve temperature T o , judge whether the predicted value of the cooling water outlet valve temperature T o is within the normal threshold, and then adjust the operation parameters of the converter station accordingly, or input the cooling water outlet valve temperature target value into the trained BP neural network, and deduce the parameter threshold of the operating condition according to the preset cooling water outlet valve temperature target value.

[0049] Further, the operation data preprocessing module is also used to:

[0050] The plurality of historical operation data of the converter station at n time points comprises one electrical quantity data: DC load P; two environmental data: ambient temperature T S , valve hall temperature T P ; five water cooling system data: expansion tank water level L P , spray pool water level L S , main waterway conductivity sigma, cooling water inlet valve temperature T i , and cooling water outlet valve temperature T o ; the first seven historical operation data are placed in a historical operation data matrix X of [n x 7] dimensions, and the historical operation data matrix X is normalized to obtain a standardized historical operation data matrix D.

[0051] Furthermore, the runtime data preprocessing module is also used for:

[0052] S11. Collect converter station operation data at 4-hour intervals, and construct a historical operation data matrix X = [PT] based on the collected operation data at n time points. S T P L P L S σ T i ], where P = [P1 P2 … P j … P n ] T P j This represents the DC load power at time point j; other operating data are similar.

[0053] S12. Normalize the historical operation data matrix X to obtain the standardized historical operation data matrix D as follows:

[0054]

[0055] Where max(p) is the maximum value of the DC load power at each sampling point at each time moment. P is the per-unit value of the DC load power at j time points after normalization using the maximum value. S It is a vector composed of the power of n standardized DC loads.

[0056] Furthermore, the operating condition classification module is also used for:

[0057] S21. Construct a clustering validity function;

[0058] S22. Construct the objective function and find the membership matrix U and cluster centers V for the number of categories c from 2 to N, where N is the number of converter stations.

[0059] S23. Substitute the membership matrix U and cluster center V corresponding to each c value into the clustering effectiveness function formula. When the clustering effectiveness function reaches its maximum value, the corresponding number of categories c is... * This represents the optimal number of categories, and the corresponding membership matrix U is... * and cluster center V * To achieve the optimal number of categories c * Membership matrix and cluster centers at time;

[0060] S24. Based on the membership matrix U * The operating states of N converter stations are divided into c. * kind.

[0061] Furthermore, the clustering validity function in step S21 is:

[0062]

[0063] where N is the number of converter stations; c is the number of classifications; y = 1, 2, …, c; u xy is the membership of the xthconverter station operating state belonging to the ythclassification, U is a c x N membership matrix composed of u xy , D x is the converter station operating data matrix X x of the xthconverter station operating state, x = 1, 2, …, N; V y is the classification center of the ythclassification, is the sum of distances from all converter stations to the converter station center V0, and is the possibility partition coefficient.

[0064] Further, the step S22 comprises:

[0065] constructing a target function where m ∈ [1, +∞) is a fuzzy index, d ij = ||x j -v i || represents the Euclidean distance of the sample x j to the clustering center v i of the ithclassification, and λ is a Lagrange operator.

[0066] The membership of the jthsample data belonging to the ithclassification is solved by using the Lagrange multiplier method as

[0067]

[0068] The clustering center is solved by using the Lagrange multiplier method as

[0069]

[0070] The membership matrix U and the clustering center V of the classification number c from 2 to N are respectively solved by the above formula.

[0071] Further, the clustering center V in the step S22 is:

[0072] V = [V1 V2 … V y … V c ] T

[0073] V y = [V y1 V y2 … V yt … V y24 ]

[0074] wherein V is a cluster center, V y is a cluster center of the yth class, V yt is a feature quantity of the cluster center of the yth class at a sampling point t.

[0075] Further, the model training module is further configured to:

[0076] process the running condition data of different classes by using a BP neural network model, and T o is an output quantity of the BP neural network model, and other running data is an input quantity, so as to fit the relationship between the cooling water outlet valve temperature T o and the other running data under different running condition classes, and train the BP neural network model, wherein the other running data includes a direct current load P, an environment temperature T S , a valve hall temperature T P , an expansion tank water level L P , a spray pool water level L S , a main waterway conductivity σ, and a cooling water inlet valve temperature T i ;

[0077] Alternatively, T o is an input quantity of the BP neural network model, and the direct current load P, the environment temperature T S , the valve hall temperature T P , the expansion tank water level L P , the spray pool water level L S , the main waterway conductivity σ, and the cooling water inlet valve temperature T i are output quantities of the BP neural network model, a corresponding relationship between the cooling water outlet valve temperature T o and the output quantities is obtained by using a large amount of data to train the BP neural network model, and the trained model is a second BP neural network model.

[0078] Still further, the converter valve temperature prediction module is further configured to:

[0079] collect running data of the converter station in real time, classify the running data, and input the classified running data into the trained first BP neural network to obtain the cooling water outlet valve temperature T o , and determine whether a predicted value of the cooling water outlet valve temperature T o is within a normal threshold, and then adjust the running parameters of the converter station accordingly;

[0080] Alternatively, a target value of the cooling water outlet valve temperature T o is input into the trained second BP neural network, and a parameter threshold of the running condition is obtained by inversely calculating the target value of the cooling water outlet valve temperature T o .

[0081] The present application has the advantages of:

[0082] (1) The present application classifies the operating conditions of the converter station by the fuzzy clustering method, and uses the BP neural network model to process the operating condition data of different categories, and fits under different categories, so as to achieve a more precise prediction effect.

[0083] (2) The present application combines electrical quantities and non-electrical quantities, and comprehensively predicts and evaluates them, which can greatly improve the calculation accuracy. In addition, the valve temperature can be predicted by the operating parameters, and the parameter threshold can be derived by the valve temperature, which has strong functionality.

[0084] (3) The present application realizes the verification of the parameter setting scheme of the converter station based on the FCM clustering and neural network algorithm without additional experiments, so the method is easy to implement, convenient to debug, and has high cost performance.

[0085] (4) The present application can calculate other operating parameter thresholds from the safety boundary constraint of the converter station outlet valve temperature, so as to reasonably configure the operating conditions of the converter station and ensure the safe operation of the system.

[0086] (5) The present application uses the BP neural network model for fitting, which has an advantage in processing nonlinear relationships over traditional polynomial fitting, and the fitting result is more accurate. BRIEF DESCRIPTION OF DRAWINGS

[0087] Figure 1 The flowchart of the converter valve temperature prediction method based on multi-source data fusion provided in Embodiment 1 of the present application. DETAILED DESCRIPTION

[0088] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be described clearly and completely below in combination with the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0089] Embodiment 1

[0090] The converter valve temperature prediction method based on multi-source data fusion comprises:

[0091] S1: Obtain 8 historical operating data of the converter station at n time points by copying the important parameters of the converter station, including one electrical quantity data: DC load P; two environmental data: environmental temperature T S , valve hall temperature T PFive water cooling system data: expansion tank water level L P , spray pool water level L S , main waterway conductivity σ, cooling water inlet valve temperature T i , cooling water outlet valve temperature T o . The first seven historical operation data are put into the historical operation data matrix X of n*7, and the historical operation data matrix X is normalized to obtain the standardized historical operation data matrix D; the specific process is as follows:

[0092] S11, collect the converter station operation data with a time interval of 4h, and construct the converter station historical operation data matrix X according to the operation data of n time points collected

[0093] X=[P T S T P L P L S σ T i ] (1)

[0094] Wherein, P=[P1 P2 … P j … P n ] T (2)

[0095] P j represents the DC load power at the jth time point, and the other operation data is the same;

[0096] S12, the historical operation data matrix X is preprocessed by using the maximum value normalization data standardization method, which is realized by the following formula:

[0097]

[0098] Wherein, different tables represent different converter station operation data, taking the first column DC load power P as an example, max(p) is the maximum value of the DC load power of each sampling point, is the unit value of the DC load power at the jth time point after the maximum value standardization, P S is a vector composed of n standardized DC load powers, and the other columns are the same.

[0099] S2: based on the standardized historical operation data matrix D, the running conditions of the converter station are classified by using the fuzzy clustering method; the specific process is as follows:

[0100] S21, the clustering effectiveness function is constructed by the following formula

[0101]

[0102] Wherein, N is the number of converter stations; c is the number of classification; y=1,2,…,c; uxy The membership of the xth converter station operating state belonging to the yth category is represented as U, which is a c x N order membership matrix composed of u xy x is the converter station operating data matrix of the xth converter station operating state x The maximum normalized unit matrix is adopted, x = 1, 2, …, N; V y is the classification center of the yth category is the sum of distances from all converter stations to the converter station center V0, and is the possibility partition coefficient.

[0103] S22, construct the objective function, and obtain the membership matrix U and the clustering center V of the classification number c from 2 to N, N being the number of converter stations. The specific process is as follows

[0104] The objective function of the FCM clustering algorithm is

[0105]

[0106] c represents the total number of clustering categories, i represents the clustering category index (i = 1, 2, … c); U = [μ ij ] represents a c x n order membership matrix, u ij represents the membership of the jth sample data belonging to the ith category; V = [v i ] represents a c x d order clustering center matrix, v i represents the clustering center of the ith category; m ∈ [1, +∞) is a fuzzy index, and m = 2 is taken; d ij = ||x j -v i || represents the Euclidean distance of the sample x j to the clustering center v i of the ith category.

[0107] The objective function is optimized by using Lagrange multiplier to obtain the following formula

[0108]

[0109] where λ is the Lagrange operator.

[0110] The first order necessary condition for optimization is

[0111]

[0112]

[0113]

[0114] From the above formula (5), we have ​

[0115]

[0116] Substitute formula (7) into formula (4) to get

[0117]

[0118] Thus, there are

[0119]

[0120] Substitute formula (12) into formula (10) to get

[0121]

[0122] When the ith cluster center is x j , d ij =||x j -v i ||=0, otherwise, d ij ≠0, so

[0123]

[0124] Similarly, the Lagrange multiplier method can be applied to solve the cluster center, and formula (9) is simplified to get

[0125]

[0126]

[0127] Using formula (14) and formula (16), the membership matrix U and the cluster center V of the classification number c from 2 to N can be solved respectively.

[0128] After the solution is finished, the cluster center V can be represented as:

[0129] V=[V1 V2 … V y … V c ] T

[0130] V y =[V y1 V y2 … V yt … V y24 ]

[0131] Wherein, V is the cluster center, V y is the classification center of the yth class, and V yt is the feature quantity of the classification center of the yth class at the sampling point t.

[0132] S23, calculate the optimal classification number using the clustering effectiveness function;

[0133] Substitute the membership matrix U and cluster center V corresponding to each c value into the clustering effectiveness function formula. When the clustering effectiveness function value reaches its maximum value, i.e., P'(U) * ;c * = max{P'(U;c)}, where the corresponding number of categories c * This represents the optimal number of categories, and the corresponding membership matrix U is... * and cluster center V * To find the optimal number of categories c * The membership matrix and cluster centers at time, let c = c * U = U * and V = V * ;

[0134] S24. Based on the membership matrix U calculated in step S23, classify the operating states of the N converter stations into c categories. Given the sample matrix and the number of categories, using the FCM clustering algorithm to calculate the membership matrix U and cluster centers V is a current technique; for example, the FCM() function in MATLAB can easily implement this algorithm. By taking values ​​from 2 to n for c, the FCM clustering algorithm is executed iteratively to calculate the membership matrix U and cluster center V corresponding to each c value.

[0135] S3: Use a BP neural network model to process different types of operating condition data and fit the cooling water outlet valve temperature T under different operating condition categories. o The relationship with other historical operational data; the specific process is as follows:

[0136] A BP neural network model was used to process different types of operating condition data, including the cooling water outlet valve temperature T. o Using the output of the BP neural network model and other operating data as input, the cooling water outlet valve temperature T is fitted under different operating conditions. o The relationship between the data and other operating data was analyzed, and the BP neural network model was trained. The trained model was used as the first BP neural network model. The other operating data included DC load P and ambient temperature T. S Valve hall temperature T P Expansion tank water level L P Water level L in the spray tank S Main water circuit conductivity σ, cooling water inlet valve temperature T i These seven quantities;

[0137] Alternatively, the cooling water outlet valve temperature T o The inputs to the BP neural network model are the DC load P and the ambient temperature T. S Valve hall temperature T P Expansion tank water level LP Water level L in the spray tank S Main water circuit conductivity σ, cooling water inlet valve temperature T i These seven quantities are the outputs of the BP neural network model, which is trained using massive amounts of data to obtain the cooling water outlet valve temperature T. o The correspondence between the input and output quantities is established, and the trained model serves as the second BP neural network model. Based on the set maximum and minimum values ​​of the outlet valve temperature, the input outlet valve temperature is increased from the minimum to the maximum value in a step size of 0.01, yielding the corresponding DC load P and ambient temperature T. S Valve hall temperature T P Expansion tank water level L P Water level L in the spray tank S Main water circuit conductivity σ, cooling water inlet valve temperature T i Find the maximum and minimum values ​​of each output quantity, and use them as the parameter thresholds for the new operating condition. Based on this, the operating parameters for the new operating condition can be set reasonably.

[0138] S4: Real-time acquisition of converter station operating data, classification of operating data, and input into a trained BP neural network to obtain the cooling water outlet valve temperature T. o Determine the cooling water outlet valve temperature T o Whether the predicted value is within the normal threshold, and then adjust the converter station operating parameters accordingly; or, input the target value of the cooling water outlet valve temperature into the trained BP neural network, and deduce the parameter thresholds of the operating condition based on the preset target value of the cooling water outlet valve temperature. Corresponding to step S3, step S4 has two execution processes, as follows:

[0139] S41, based on the real-time acquisition of converter station operation data from S1, classify the operation data and determine the DC load P and ambient temperature T corresponding to the target operation state. S Valve hall temperature T P Expansion tank water level L P Water level L in the spray tank S Main water circuit conductivity σ, cooling water inlet valve temperature T i Which cluster category do these seven runtime parameters belong to?

[0140] S42, input the above seven target operating parameters into the trained first BP neural network to obtain the cooling water outlet valve temperature T. o The predicted value.

[0141] S43, determine the cooling water outlet valve temperature T o The system checks whether the predicted value is within the normal threshold; if it exceeds the normal threshold, an early warning is issued, and the target operating parameters are adjusted accordingly. This is mainly based on the DC load P and the ambient temperature T. S Valve hall temperature TP , expansion tank water level L P , spray pool water level L S , main waterway conductivity σ, cooling water inlet valve temperature T i After predicting the cooling water outlet valve temperature by seven electrical and non-electrical quantities, the cooling water outlet valve temperature threshold set by the converter station is compared, and when it exceeds the threshold, the system triggers an alarm. At this time, the DC load P, ambient temperature T S , valve hall temperature T P , expansion tank water level L P , spray pool water level L S , main waterway conductivity σ, cooling water inlet valve temperature T i Seven quantities are reset until they are stable within the safety threshold.

[0142] S44, the cooling water outlet valve temperature target value can also be input accordingly, the operating condition parameter threshold is inversely calculated according to the preset cooling water outlet valve temperature, and the operating parameters of the new operating condition are reasonably set, specifically, the cooling water outlet valve temperature T o target value is input into the trained second BP neural network, and the operating condition parameter threshold is inversely calculated according to the preset cooling water outlet valve temperature T o target value, and each operating parameter is reasonably set within the parameter threshold of the new operating condition.

[0143] The following is a specific experimental process of the application in the cooling valve temperature prediction of the converter station. In this embodiment, there are 734 operating conditions of the converter station. First, according to the FCM clustering result, the 734 operating conditions are divided into two categories, and the neural network model is fitted respectively. The derived model is used for subsequent use. Ten are randomly selected from the 734 operating conditions, and the DC load P, ambient temperature T S , valve hall temperature T P , expansion tank water level L P , spray pool water level L S , main waterway conductivity σ, cooling water inlet valve temperature T i After standardizing the seven operating parameters, they are classified (category 1, category 2) first, and then input into the trained neural network model. The actual value and the predicted value of the cooling water outlet valve temperature T i are compared.

[0144] Table 1 is the actual value of the seven operating parameters; Table 2 is the result of the actual value of the seven operating parameters after standardization; and Table 3 is the actual value and the predicted value of the cooling water outlet valve temperature T i .

[0145] Table 1

[0146] Serial number P(MW) T S (°C) T P (°C) L P (%)]] L S (%)]] σ(μs / cm) T i (°C) 1 1998 32 30 51 94 0.11 34.7 2 1998 21 26 51 92 0.11 34.6 3 2000 24 26 51 93 0.12 34.8 4 1999 29 27 51 92 0.11 34.7 5 2248 25 27 52 93 0.12 34.8 6 1446 14 24 46 94 0.11 34.7 7 299 16 24 39 92 0.08 30.2 8 300 15 24 40 92 0.08 30.3 9 1296 17 23 44 93 0.1 33.9 10 1649 6 18 44 93 0.12 35.1

[0147] Table 2

[0148]

[0149]

[0150] Table 3

[0151]

[0152] As can be seen from Table 3, the predicted value and the actual value are close, and the standard deviation of the predicted value is 0.0201, which indicates that the application has the ability to accurately predict the cooling water outlet valve temperature T i under a certain operating state. When the predicted value is greater than the safety threshold, the system will give a warning, so that the method can be used for parameter configuration by the staff of the converter station.

[0153] Through the above technical scheme, the application considers the nonlinear relationship between the operating parameters of the converter station, and proposes a lean prediction method for the temperature of the converter valve and a calculation method for the safety threshold of the protection parameter based on multi-source data fusion. According to the historical operating data, the FCM clustering method is used to divide the operating state of the converter station into several categories, and the relationship between one output and seven inputs is fitted under different categories. When the converter station is about to reach a new operating state, the seven target operating state quantities are inputted to obtain the predicted value of the cooling water outlet valve temperature T o , and it is detected whether it is within the safety operating threshold: if it is within the safety threshold, the converter station is allowed to operate in the operating state, otherwise the target operating parameter needs to be reconfigured; or the target value of the cooling water outlet valve temperature can be inputted, and the parameter threshold of the operating condition is deduced according to the preset cooling water outlet valve temperature, and the operating parameters of the new condition are reasonably set. By using this method, the safe, stable and economic operation of the converter station can be guaranteed to the greatest extent.

[0154] Embodiment 2

[0155] Based on embodiment 1, the application embodiment 2 further provides a temperature prediction device for the converter valve based on multi-source data fusion, which comprises:

[0156] An operating data preprocessing module is configured to acquire a plurality of historical operating data of the converter station at n time points, one of which is the cooling water outlet valve temperature T o , and to construct a standardized historical operating data matrix D by using other historical operating data;

[0157] An operating condition classification module is configured to classify the operating condition of the converter station based on the standardized historical operating data matrix D by using a fuzzy clustering method;

[0158] The model training module is configured to process the operation condition data of different categories by using the BP neural network model, and fit the cooling water outlet valve temperature T under different operation condition categories o relationship with other historical operation data;

[0159] The converter valve temperature prediction module is configured to collect real-time converter station operation data, classify the operation data, and input the trained BP neural network to obtain the cooling water outlet valve temperature T o determine whether the predicted value of the cooling water outlet valve temperature T o is within the normal threshold, and then adjust the operation parameters of the converter station accordingly, or input the target value of the cooling water outlet valve temperature into the trained BP neural network, and inversely deduce the parameter threshold of the operation condition according to the preset target value of the cooling water outlet valve temperature.

[0160] Specifically, the operation data preprocessing module is further configured to:

[0161] The plurality of historical operation data of the converter station at n time points includes 1 electrical quantity data: DC load P; 2 environmental data: ambient temperature T S , valve hall temperature T P ; 5 water cooling system data: expansion tank water level L P , spray pool water level L S , main waterway conductivity σ, cooling water inlet valve temperature T i , and cooling water outlet valve temperature T o ; the first 7 historical operation data are placed in a historical operation data matrix X of [n x 7] dimensions, and the historical operation data matrix X is normalized to obtain a standardized historical operation data matrix D.

[0162] More specifically, the operation data preprocessing module is further configured to:

[0163] S11, collect the converter station operation data at intervals of 4h, and construct a converter station historical operation data matrix X = [P T S T P L P L S σ T i ] according to the collected operation data at n time points, wherein P = [P1 P2 … P j … P n ] T , P j j represents the DC load power at the jth time point, and the other operation data are the same;

[0164] S12, normalize the historical operation data matrix X to obtain a standardized historical operation data matrix D as follows

[0165]

[0166] Where max(p) is the maximum value of the DC load power at each sampling point at each time moment. P is the per-unit value of the DC load power at j time points after normalization using the maximum value. S It is a vector composed of the power of n standardized DC loads.

[0167] Specifically, the operating condition classification module is also used for:

[0168] S21. Construct a clustering validity function;

[0169] S22. Construct the objective function and find the membership matrix U and cluster center V for the number of categories c from 2 to N, where N is the number of converter stations;

[0170] S23. Substitute the membership matrix U and cluster center V corresponding to each c value into the clustering effectiveness function formula. When the clustering effectiveness function reaches its maximum value, the corresponding number of categories c is... * This represents the optimal number of categories, and the corresponding membership matrix U is... * and cluster center V * To find the optimal number of categories c * Membership matrix and cluster centers at time;

[0171] S24. Based on the membership matrix U * The operating states of N converter stations are divided into c. * kind.

[0172] More specifically, the clustering validity function in step S21 is:

[0173]

[0174] Where N is the number of converter stations; c is the number of categories; y = 1, 2, ..., c; u xy Let U represent the membership degree of the x-th converter station's operating state belonging to the y-th class, where U is a c×N membership matrix. xy Composition, D x X is the converter station operation data matrix X under the operating status of the xth converter station. x The per-unit matrix is ​​normalized using the maximum value, x = 1, 2, ..., N; V y Let y be the classification center of class y. The sum of the distances from all converter stations to the converter station center V0 and It is the probability division coefficient.

[0175] More specifically, step S22 includes:

[0176] Constructing objective function Wherein, m∈[1, +∞) is fuzzy index, d ij =||x j -v i || represents the Euclidean distance of sample x j to the clustering center v i of the i-th class, and λ is Lagrange operator.

[0177] The membership degree of the j-th sample data belonging to the i-th class is solved by using Lagrange multiplier method as follows:

[0178]

[0179] The clustering center V is solved by using Lagrange multiplier method as follows:

[0180]

[0181] The membership degree matrix U and the clustering center V of the classification number c from 2 to N are solved by the above formula respectively.

[0182] Further, the clustering center V in the step S22 is:

[0183] V=[V1 V2 … V y … V c ] T

[0184] V y =[V y1 V y2 … V yt … V y24 ]

[0185] Wherein, V is the clustering center, V y is the classification center of the y-th class, and V yt is the feature quantity of the classification center of the y-th class at the sampling point t.

[0186] Specifically, the model training module is further configured to:

[0187] Process the running condition data of different categories by using the BP neural network model, take the cooling water outlet valve temperature T o as the output quantity of the BP neural network model, take other running data as the input quantity, fit the relationship between the cooling water outlet valve temperature T o and other running data under different running condition categories, and train the BP neural network model, wherein the trained model is taken as the first BP neural network model, and the other running data includes direct current load P, environment temperature T S , valve hall temperature T P , and expansion water tank water level LP , spray pool water level L S , main waterway conductivity σ, cooling water inlet valve temperature T i These seven quantities;

[0188] Or, taking the cooling water outlet valve temperature T o As the input quantity of the BP neural network model, the direct current load P, the environment temperature T S , the valve hall temperature T P , the expansion tank water level L P , the spray pool water level L S , the main waterway conductivity σ, the cooling water inlet valve temperature T i These seven quantities are output quantities of the BP neural network model, and the corresponding relationship between the cooling water outlet valve temperature T o And the output quantity is obtained by training a large amount of data, and the trained model is used as a second BP neural network model.

[0189] More specifically, the converter valve temperature prediction module is further configured to:

[0190] Real-time acquisition of the converter station operation data, classification of the operation data and input of the trained first BP neural network to obtain the cooling water outlet valve temperature T o Judgment of whether the predicted value of the cooling water outlet valve temperature T o Is within the normal threshold, and further adjustment of the operation parameters of the converter station;

[0191] Or, inputting the target value of the cooling water outlet valve temperature T o To the trained second BP neural network, and inversely deducing the parameter threshold of the operation condition according to the preset target value of the cooling water outlet valve temperature T o .

[0192] The above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement to some technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for converter valve temperature prediction based on multi-source data fusion, characterized in that, The method comprises: Step one: obtaining historical operation data of the converter station n at a plurality of time points, wherein one of the historical operation data is cooling water outlet valve temperature , and the other historical operation data is used to construct a standardized historical operation data matrix D ; the converter station n ; the plurality of historical operation data at the plurality of time points comprises 1 electrical quantity data: DC load ; 2 environmental data: ambient temperature , valve hall temperature ; 5 water cooling system data: expansion tank water level , spray pool water level , main waterway conductivity , cooling water inlet valve temperature , cooling water outlet valve temperature ; the first 7 historical operation data are placed in a historical operation data matrix X of dimensions, and the historical operation data matrix X is normalized to obtain a standardized historical operation data matrix D ; Step two: based on the standardized historical operation data matrix D classifying the operating conditions of the converter station by a fuzzy clustering method; step two comprises: S21, constructing a clustering validity function; S22, constructing the target function, and obtaining the number of classifications from 2 to N membership matrix U and clustering center V, N is the number of converter stations; S23, each Substituting the membership matrix U and cluster centers V corresponding to the values ​​into the clustering effectiveness function formula, the number of clusters corresponding to the maximum value of the clustering effectiveness function is determined. This represents the optimal number of categories, and the corresponding membership matrix at this point... and cluster center To achieve the optimal number of categories Membership matrix and cluster centers at time; S24, according to the membership matrix The N converter station operating states are divided into classes; Step three: use BP neural network model to process different categories of operating condition data, and fit the cooling water outlet valve temperature under different operating condition categories relationship with other historical operation data; step three includes: The BP neural network model is used for processing different types of operation condition data, and the cooling water outlet valve temperature is taken as an output quantity of the BP neural network model, and other operation data are taken as input quantities, so as to fit the cooling water outlet valve temperature under different operation condition categories , and the relationship between the cooling water outlet valve temperature and the other operation data is trained, and the trained model is taken as a first BP neural network model, wherein the other operation data include direct current load , environment temperature , valve hall temperature , expansion water tank water level , spray pool water level , main waterway electric conductivity , cooling water inlet valve temperature , and seven quantities. Or, the cooling water outlet valve temperature As the input quantity of the BP neural network model, the direct current load , Ambient temperature , Valve hall temperature , Expansion tank water level , Spray pool water level , Main waterway conductivity , Cooling water inlet valve temperature The seven quantities are the output quantities of the BP neural network model, and the cooling water outlet valve temperature Corresponding relationship between the output quantities is obtained by training a large amount of data, and the trained model is used as the second BP neural network model; Step four: collecting real-time operation data of the converter station, classifying the operation data and inputting the classified operation data into the trained first BP neural network to obtain the cooling water outlet valve temperature , judging whether the predicted value of the cooling water outlet valve temperature is within a normal threshold, and further adjusting the operation parameters of the converter station or inputting the target value of the cooling water outlet valve temperature into the trained second BP neural network to back-calculate the parameter threshold of the operation condition according to the preset target value of the cooling water outlet valve temperature .

2. The multi-source data fusion based converter valve temperature prediction method of claim 1, wherein, The step one further comprises: S11, collect the converter station operation data with 4h as the time interval, and construct the converter station historical operation data matrix according to the operation data of the collected time points n , wherein, , represents the DC load power of the i-th time point, and the other operation data are the same; j ​​ S12, normalize the historical operation data matrix X to obtain a standardized historical operation data matrix D As follows wherein, is the maximum value of the DC load power at each sampling point, is the normalized unit value of the DC load power at j time points, is n is a vector composed of j normalized DC load powers.

3. The multi-source data fusion based converter valve temperature prediction method of claim 1, wherein, The clustering validity function in step S21 is wherein, is the number of converter stations; is the number of categories; y = 1, 2, …, c; represents the membership of the operating state of the jth converter station belonging to the yth category, U is a c x n membership matrix composed of , N is the membership of the operating state of the jth converter station belonging to the yth category, U is a c x n membership matrix composed of , is the membership of the operating state of the jth converter station belonging to the yth category, U is a c x n membership matrix composed of , is the normalized matrix of the operating data of the jth converter station in the operating state ; is the classification center of the yth category, is the sum of the distances from all converter stations to the converter station center and ; is the possibility partition coefficient.

4. The multi-source data fusion based converter valve temperature prediction method of claim 3, wherein, Step S22 comprises: Constructing the objective function wherein, is the fuzzy exponent, denotes the sample to the cluster center of the class , and λ is the Lagrange multiplier; The Lagrange multiplier method is used to solve the membership of the first sample data belonging to the first class Solving the clustering center by using the Lagrange multiplier method The number of categories is solved by the above formula respectively From 2 to N The membership matrix U and clustering center V.

5. The multi-source data fusion based converter valve temperature prediction method of claim 4, wherein, The clustering center V in the step S22 is: wherein V is a cluster center, is a cluster center of the y-th class, is a feature quantity of the cluster center of the y-th class at the sampling point t.

6. Converter valve temperature prediction device based on multi-source data fusion, for implementing the method according to any one of claims 1 to 5, characterized in that, The device comprises: Run the data preprocessing module to obtain data from the converter station. n Multiple historical operating data points at various time points, one of which is the cooling water outlet valve temperature. Construct a standardized historical operation data matrix using other historical operation data. D ; The operating condition classification module is used to classify operating conditions based on a standardized historical operating data matrix. D The operating conditions of converter stations are classified using fuzzy clustering. The model training module is configured to process the running condition data of different categories by using a BP neural network model to fit the cooling water outlet valve temperature under different running condition categories relationship with other historical running data; The converter valve temperature prediction module is used for collecting converter station operation data in real time, classifying the operation data and inputting the classified operation data into the trained BP neural network to obtain the cooling water outlet valve temperature , judging whether the predicted value of the cooling water outlet valve temperature is within a normal threshold value, and further adjusting the operation parameters of the converter station accordingly, or inputting a cooling water outlet valve temperature target value into the trained BP neural network to inversely deduce the parameter threshold value of the operation condition according to the preset cooling water outlet valve temperature target value.

Images