A high performance diesel generator set and intelligent control method thereof

Abstract

The present invention relates to the field of generator set control technology, and discloses a high-performance diesel generator set and an intelligent control method thereof, comprising the following steps: step S201, obtaining the working information and fault record information of y power generation units through an information collector; step S202, constructing a health prediction model in a controller, and inputting the health matrix A into the health prediction model. The present invention comprehensively considers the service life of each diesel generator, the fault conditions that have occurred, and the electrical signal information, and evaluates the health of each diesel generator based on these data. Compared with the traditional evaluation method, the scoring standard is more unified, and a large amount of manpower does not need to be wasted. In addition, after obtaining the health score, the power generation of each diesel generator in the next day is allocated to avoid the phenomenon that the high-complexity work of the generator with low health leads to a significant increase in its failure probability.

Description

Technical Field

[0001] The present invention relates to the technical field of generator set control, and more specifically, to a high-performance diesel generator set and an intelligent control method thereof. Background Art

[0002] Diesel generators undertake important power generation tasks in some remote areas (mainly remote areas far away from power grid coverage, such as mines, field work sites, remote communities, islands, drilling platforms and scientific research bases, etc.). Multiple diesel generators form a power grid to supply power to these areas.

[0003] However, as the working time of diesel generators increases, the probability of failure may greatly increase. If one of the diesel generators suddenly fails during operation, it may cause a short-term power shortage. Therefore, it is necessary to evaluate the health of the diesel generator. The common method is to use manual evaluation, but this method is not only not accurate enough, but also very labor-intensive. After the evaluation results are obtained, it is difficult to allocate the power generation of all diesel generators in the power grid so that some generators with lower health can work at low power. Therefore, a high-performance diesel generator set and its intelligent control method are proposed. Summary of the invention

[0004] The present invention provides a high-performance diesel generator set and an intelligent control method thereof, which solve the technical problems in the above-mentioned background technology.

[0005] The present invention provides a high-performance diesel generator set, comprising:

[0006] A power generation unit for generating electricity using diesel;

[0007] An information collector collects the working information and fault record information of y power generation units;

[0008] y is a custom parameter;

[0009] The controller is used to control the power generation of the power generation unit in the next day according to the information collected by the information collector.

[0010] The present invention also provides an intelligent control method for a high-performance diesel generator set, comprising the following steps:

[0011] Step S201, obtaining working information and fault record information of y power generation units through an information collector;

[0012] The working information of the power generation unit includes: the maximum load information, maximum voltage information, safety circuit information and environmental information of each day in the past year, and the usage time information of the power generation unit;

[0013] The fault record information includes: the fault type information and fault severity information of y power generation units in the past year;

[0014] Step S202, constructing a health prediction model in the controller, and constructing a health matrix A according to the working information and fault record information of multiple power generation units, inputting the health matrix A into the health prediction model, and outputting the health score of each power generation unit;

[0015] The health score ranges from 0 to 100;

[0016] Step S203, adjusting the power generation of each power generation unit in the next day according to the health score of each power generation unit.

[0017] Furthermore, the health matrix A is as follows:

[0018]

[0019] in, Indicates the working information and fault record information of the first power generation unit on the first day; Indicates the working information and fault record information of the first power generation unit on the mth day; Indicates the working information and fault record information of the nth power generation unit on the first day; Indicates the working information and fault record information of the nth power generation unit on the mth day;

[0020] m is 365, and n is a custom parameter.

[0021] Furthermore, the method for obtaining the working information of the power generation unit includes:

[0022] The load of the power generation unit is monitored in real time through the load monitoring equipment, and the maximum load of each day is selected as the maximum load information of each day in the past year;

[0023] The voltage of the power generation unit is monitored in real time by a voltage testing device, and the maximum voltage of each day is selected as the maximum voltage information;

[0024] The current and voltage information of the safety circuit of the power generation unit is monitored in real time every day by current monitoring equipment and voltage testing equipment, and the maximum current and maximum voltage are selected as the safety circuit information;

[0025] The temperature and humidity information of the corresponding area of ​​the power generation unit is monitored in real time through the temperature and humidity testing equipment, and the maximum and minimum temperature and humidity are selected as environmental information;

[0026] Use time information of the power generation unit: the use time information of the power generation unit is determined according to the production information of the power generation unit.

[0027] Furthermore, the health prediction model is constructed based on the long short-term memory network.

[0028] Furthermore, the health prediction model includes m columns of hidden layers and a classifier;

[0029] The working information and fault record information of the power generation unit in the jth column of the health matrix A are input into the jth hidden layer, and the updated features are output;

[0030] Update the feature input classifier and output the health score of each power generation unit;

[0031] 1≤j≤m.

[0032] Furthermore, the calculation formula of the jth hidden layer of the health prediction model includes:

[0033] The forget gate F of the jth hidden layer of the health prediction model j The calculation formula is as follows:

[0034] Fj=σ(xjw F1 +Hj- 1 w F2 +b F )

[0035] In the formula, w F1 、w F2 and b F Represents the forget gate F of the jth hidden layer j The corresponding first, second and third weight parameters; Hj- 1 represents the hidden layer output at the previous moment, σ represents the sigmoid function;

[0036] The input gate I of the jth hidden layer of the health prediction model j The calculation formula is as follows:

[0037] Ij=σ(xjw I1 +Hj- 1 w I2 +b I );

[0038] In the formula, w I1 、w I2 and b I Represents the input gate I of the jth hidden layer j corresponding first, second and third weight parameters;

[0039] The intermediate state C% of the jth hidden layer of the health prediction model j The calculation formula is as follows:

[0040]

[0041] In the formula, w C1 、w C2 and b C Represents the intermediate state of the jth hidden layer The corresponding first, second and third weight parameters; x j Represents the input of the jth hidden layer;

[0042] The output state C of the jth hidden layer of the health prediction model j The calculation formula is as follows:

[0043]

[0044] In the formula, C j-1 represents the output state of the j-1th hidden unit in the first hidden layer; e represents the dot product operation;

[0045] The output gate O of the jth hidden layer of the health prediction model j The calculation formula is as follows:

[0046] Oj=σ(xjw 01 +Hj- 1 w 02 +b 0 )

[0047] In the formula, w 01 、w 02 and b 0 represents the output gate O of the jth hidden layer j corresponding first, second and third weight parameters;

[0048] The output H of the jth hidden layer of the health prediction model j The calculation formula is as follows:

[0049] H j =0 j etanh(C j )

[0050] In the formula, the output gate O j and tanh(C j ) multiply point by point to get the output H j ;

[0051] Further, the working information and fault record information of u power generation units in time period v are obtained, and a training matrix B is constructed;

[0052] Set a sliding window in the training matrix B, the length of the sliding window is m, the width is n, and the step length is 1;

[0053] The sliding window slides continuously horizontally and vertically in the training matrix B to obtain several training samples as sample data for training the health prediction model;

[0054] The health score of each power generation unit corresponding to the training sample is obtained as a sample label for training the health prediction model.

[0055] The objective function is used to improve the accuracy of the health prediction model, and its calculation formula is as follows:

[0056]

[0057] Where L represents the loss value of the health prediction model; N represents the total number of all power generation units; y i represents the actual health score of the ith power generation unit; The health prediction model outputs the health score of the i-th power generation unit.

[0058] Furthermore, the steps of adjusting the power generation and power generation efficiency of the power generation unit are as follows:

[0059] Step S301, dividing the power grid system according to where each power generation unit is located;

[0060] Step S302, obtaining the preset power generation in each power grid system;

[0061] Step S303, presetting the first and second health scores;

[0062] Step S304, the power generation of the power generation unit whose health score is lower than the first health score in the next day shall not exceed the first power generation threshold; the power generation of the power generation unit whose health score is lower than the second health score in the next day shall not exceed the second power generation threshold; the remaining preset power generation is scored by the power generation unit whose health score is higher than the second health score.

[0063] Further with respect to step S302, the working information and fault record information of the power generation unit are obtained through the information collector, and the scheduling optimization of the power generation unit is realized. The specific process is as follows:

[0064] Step 1, information collection;

[0065] Working information includes: maximum load information, maximum voltage information, safety circuit information, environmental information, and usage time of the power generation unit for each day in the past year;

[0066] Fault record information includes: fault type information and fault severity information of the power generation unit in the past year;

[0067] X data matrix, including the working information and fault record information of all power generation units;

[0068] x ij The working information and fault record information of the i-th power generation unit on the j-th day; y is the load forecast value for the next day;

[0069] Step 2, load prediction;

[0070] The Prophet model is used to predict future loads and provide a reference for scheduling optimization. It consists of trend, seasonality and holiday effects. The prediction formula of the model is as follows:

[0071]

[0072] in, is the model prediction value; g(t) is the trend function, which indicates the long-term trend; s(t) is the seasonal component, which indicates the periodic change; h(t) is the holiday effect, which indicates the influence of special dates; ∈t is the error term;

[0073] Define the loss function for parameter updating, which measures the difference between the model prediction and the true value;

[0074]

[0075] in, is the smooth L1 loss function, y i is the true load value, is the predicted load value, i represents the i-th power generation unit;

[0076] Calculate the gradient of the loss function with respect to the parameter θ for parameter update;

[0077]

[0078] in, is the gradient of the loss function, Used to indicate the direction of the prediction error for each sample; is the partial derivative of the predicted value with respect to the parameter;

[0079] Use the quasi-Newton method to update the parameters to minimize the loss function;

[0080]

[0081] Among them, θ ( k+ 1) is the parameter value for the k+1th iteration, θ ( k ) is the parameter value of the kth iteration, H is the quasi-Newton matrix, which can be solved iteratively through the update formula of the quasi-Newton method, H -1 is the inverse of the quasi-Newton matrix, is the gradient of the loss function at the kth iteration;

[0082] Step 3: Transfer learning optimization;

[0083] Using the transfer learning method, the knowledge of the pre-trained model is transferred to the power generation unit scheduling optimization task, which is used to improve the intelligence and adaptability of the scheduling strategy;

[0084] In domain adaptation, the Wasserstein Visual Distance (WVD) is used to select appropriate source domain features to reduce the feature distribution difference between the source domain and the target domain. WVD measures the distance between the two distributions and helps select aligned feature subsets to improve the effect of transfer learning.

[0085]

[0086] Among them, X s and X t It is the feature set of the source domain and the target domain; and The i-th sample in the source domain and the target domain; φ(g) is the mapping function, usually a neural network, and P×Pp is L P norm; m and n represent the number of samples in the source domain and target domain datasets respectively;

[0087] Based on the selected source domain feature subset, feature extraction is performed on the source domain and target domain data to provide input data for model training; a consistency loss function is defined to measure the consistency between the prediction of the auxiliary label classifier and the true label of the target domain;

[0088]

[0089] in, is the consistency loss function, is the true label of the target domain, The prediction value of the auxiliary label classifier;

[0090] By minimizing the consistency loss function and feature distribution differences, domain-adaptive migration scheduling optimization is achieved;

[0091]

[0092] Among them, f is the feature extractor, g is the auxiliary label classifier, and X s and X t are the feature sets of the source domain and the target domain respectively, and λ is a hyperparameter that weighs the losses of the two parts;

[0093] The beneficial effects of the present invention are as follows: the service life of each diesel generator, the fault conditions that have occurred and the electrical signal information are comprehensively considered, and the health of each diesel generator is evaluated based on these data. Compared with the traditional evaluation method, the scoring standard is more unified and does not require a lot of manpower. In addition, after obtaining the health score, the power generation of each diesel generator in the next day is allocated to avoid the phenomenon that the high-complex operation of generators with lower health results in a significantly increased probability of failure.

[0094] The health prediction model based on deep learning can comprehensively consider various influencing factors and accurately evaluate the health status of each power generation unit through the LSTM model. Compared with traditional evaluation methods, this method has more unified standards, does not require a lot of manpower, and improves the objectivity and consistency of the scoring. The Prophet model can accurately predict the load demand for the next day. The model can capture the trend and seasonal factors of load changes and adjust the power generation strategy in advance. The Prophet model uses the quasi-Newton method to update parameters to ensure the rapid convergence and efficient calculation of the prediction model; through load prediction, the working state of the generator set can be adjusted in advance when the environment and load change, avoiding system failures and reduced efficiency caused by emergencies; using the domain adaptation framework of transfer learning, the weights of different sensor data are dynamically adjusted according to the reliability and importance of sensor data to better adapt to different working environments and fault conditions. The WVD technology is used to select appropriate source domain features and introduce a consistency loss function to ensure the efficient application of the transfer learning model in the target domain. Intelligent scheduling optimization not only considers the health score, but also future load forecasts and environmental changes. Avoid excessive use of low-health power generation units, thereby reducing failure rates and maintenance costs. When the health score is low, the workload of the power generation unit is automatically reduced to extend the service life of the equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0095] Figure 1 is an overall flow chart of Embodiment 2 of the present invention;

[0096] Figure 2 is a flow chart of adjusting the power generation and power generation efficiency of a power generation unit according to the present invention;

[0097] Figure 3 It is the flow chart of the intelligent control algorithm of the present invention; DETAILED DESCRIPTION

[0098] The subject matter described herein will now be discussed with reference to example embodiments. It should be understood that the discussion of these embodiments is only to enable those skilled in the art to better understand and implement the subject matter described herein, and the functions and arrangements of the elements discussed may be changed without departing from the scope of protection of the contents of this specification. Each example may omit, replace or add various processes or components as needed. In addition, the features described relative to some examples may also be combined in other examples.

[0099] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in one or more embodiments of the present invention shall have the usual meanings understood by persons with ordinary skills in the field to which the present invention belongs. The words "first", "second" and similar terms used in one or more embodiments of the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. Words such as "include" or "comprise" and the like mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. Words such as "connect" or "connected" and the like are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the object being described changes, the relative positional relationship may also change accordingly.

[0100] Embodiment 1 of the present invention is a high-performance diesel generator set, comprising:

[0101] A power generation unit for generating electricity using diesel;

[0102] An information collector collects the working information and fault record information of y power generation units;

[0103] y is a custom parameter, which is determined by the actual number of power generation units;

[0104] A power generation unit represents a diesel generator;

[0105] The controller is used to control the power generation of the power generation unit in the next day according to the information collected by the information collector.

[0106] The second embodiment of the present invention is that Figure 1-Figure 2 As shown, an intelligent control method for a high-performance diesel generator set is proposed, comprising the following steps:

[0107] Step S201, obtaining working information and fault record information of y power generation units through an information collector;

[0108] The working information of the power generation unit includes: the maximum load information, maximum voltage information, safety circuit information and environmental information of each day in the past year, and the usage time information of the power generation unit;

[0109] The fault record information includes: the fault type information and fault severity information of y power generation units in the past year;

[0110] Step S202, constructing a health prediction model in the controller, and constructing a health matrix A according to the working information and fault record information of multiple power generation units, inputting the health matrix A into the health prediction model, and outputting the health score of each power generation unit;

[0111] The health score ranges from 0 to 100;

[0112] Step S203, adjusting the power generation of each power generation unit in the next day according to the health score of each power generation unit.

[0113] In one embodiment of the present invention, the health matrix A is as follows:

[0114]

[0115] in, Indicates the working information and fault record information of the first power generation unit on the first day; Indicates the working information and fault record information of the first power generation unit on the mth day; Indicates the working information and fault record information of the nth power generation unit on the first day; Indicates the working information and fault record information of the nth power generation unit on the mth day;

[0116] m is 365, and n is a custom parameter.

[0117] In one embodiment of the present invention, a method for acquiring working information of a power generation unit includes:

[0118] The load of the power generation unit is monitored in real time through the load monitoring equipment, and the maximum load of each day is selected as the maximum load information of each day in the past year;

[0119] The voltage of the power generation unit is monitored in real time by a voltage testing device, and the maximum voltage of each day is selected as the maximum voltage information;

[0120] The current and voltage information of the safety circuit of the power generation unit is monitored in real time every day by current monitoring equipment and voltage testing equipment, and the maximum current and maximum voltage are selected as the safety circuit information;

[0121] The temperature and humidity information of the corresponding area of ​​the power generation unit is monitored in real time through the temperature and humidity testing equipment, and the maximum and minimum temperature and humidity are selected as environmental information;

[0122] Use time information of the power generation unit: the use time information of the power generation unit is determined according to the production information of the power generation unit.

[0123] In one embodiment of the present invention, the health prediction model is constructed based on a long short-term memory network.

[0124] In one embodiment of the present invention, the health prediction model includes m columns of hidden layers and a classifier;

[0125] The working information and fault record information of the power generation unit in the jth column of the health matrix A are input into the jth hidden layer, and the updated features are output;

[0126] Update the feature input classifier and output the health score of each power generation unit;

[0127] 1≤j≤m.

[0128] In one embodiment of the present invention, the calculation formula of the jth hidden layer of the health prediction model includes:

[0129] The forget gate F of the jth hidden layer of the health prediction model j The calculation formula is as follows:

[0130] Fj=σ(xjw F1 +Hj- 1 w F2 +b F )

[0131] In the formula, w F1 、w F2 and b F Represents the forget gate F of the jth hidden layer j The corresponding first, second and third weight parameters; Hj- 1 represents the hidden layer output at the previous moment, σ represents the sigmoid function;

[0132] The input gate I of the jth hidden layer of the health prediction model j The calculation formula is as follows:

[0133] Ij=σ(xjw I1 +Hj- 1 w I2 +b I );

[0134] In the formula, w I1 、w I2 and b IRepresents the input gate I of the jth hidden layer j corresponding first, second and third weight parameters;

[0135] The intermediate state of the jth hidden layer of the health prediction model The calculation formula is as follows:

[0136]

[0137] In the formula, w C1 、w C2 and b C Represents the intermediate state of the jth hidden layer The corresponding first, second and third weight parameters; x j Represents the input of the jth hidden layer;

[0138] The output state C of the jth hidden layer of the health prediction model j The calculation formula is as follows:

[0139]

[0140] In the formula, C j-1 represents the output state of the j-1th hidden unit in the first hidden layer; e represents the dot product operation;

[0141] The output gate O of the jth hidden layer of the health prediction model j The calculation formula is as follows:

[0142] Oj=σ(xjw 01 +Hj- 1 w 02 +b 0 )

[0143] In the formula, w 01 、w 02 and b 0 represents the output gate O of the jth hidden layer j corresponding first, second and third weight parameters;

[0144] The output H of the jth hidden layer of the health prediction model j The calculation formula is as follows:

[0145] H j =0 j etanh(C j )

[0146] In the formula, the output gate O j and tanh(C j ) multiply point by point to get the output H j ;

[0147] In one embodiment of the present invention, the working information and fault record information of u power generation units in a time period v are obtained, and a training matrix B is constructed;

[0148] Set a sliding window in the training matrix B, the length of the sliding window is m, the width is n, and the step length is 1;

[0149] The sliding window slides continuously horizontally and vertically in the training matrix B to obtain several training samples as sample data for training the health prediction model;

[0150] The health score of each power generation unit corresponding to the training sample is obtained as a sample label for training the health prediction model.

[0151] The objective function is used to improve the accuracy of the health prediction model, and its calculation formula is as follows:

[0152]

[0153] Where L represents the loss value of the health prediction model; N represents the total number of all power generation units; y i represents the actual health score of the ith power generation unit; The health prediction model outputs the health score of the i-th power generation unit.

[0154] In one embodiment of the present invention, the steps of adjusting the power generation and power generation efficiency of the power generation unit are as follows:

[0155] Step S301, dividing the power grid system according to where each power generation unit is located;

[0156] Step S302, obtaining the preset power generation in each power grid system;

[0157] Step S303, presetting the first and second health scores;

[0158] Step S304, the power generation of the power generation unit whose health score is lower than the first health score in the next day shall not exceed the first power generation threshold; the power generation of the power generation unit whose health score is lower than the second health score in the next day shall not exceed the second power generation threshold; the remaining preset power generation is scored by the power generation unit whose health score is higher than the second health score.

[0159] In an implementation case of the present invention, the working information and fault record information of the power generation unit are obtained by the information collector, so as to realize the scheduling optimization of the power generation unit. The specific process is as follows:

[0160] Step 1, information collection;

[0161] Working information includes: maximum load information, maximum voltage information, safety circuit information, environmental information, and usage time of the power generation unit for each day in the past year;

[0162] Fault record information includes: fault type information and fault severity information of the power generation unit in the past year;

[0163] X data matrix, including the working information and fault record information of all power generation units;

[0164] x ij The working information and fault record information of the i-th power generation unit on the j-th day; y is the load forecast value for the next day;

[0165] Step 2, load prediction;

[0166] The Prophet model is used to predict future loads and provide a reference for scheduling optimization. It consists of trend, seasonality and holiday effects. The prediction formula of the model is as follows:

[0167]

[0168] in, is the model prediction value; g(t) is the trend function, which indicates the long-term trend; s(t) is the seasonal component, which indicates the periodic change; h(t) is the holiday effect, which indicates the influence of special dates; εt is the error term;

[0169] Define the loss function for parameter updating, which measures the difference between the model prediction and the true value;

[0170]

[0171] in, is the smooth L1 loss function, y i is the true load value, is the predicted load value, i represents the i-th power generation unit;

[0172] Calculate the gradient of the loss function with respect to the parameter θ for parameter update;

[0173]

[0174] in, is the gradient of the loss function, Used to indicate the direction of the prediction error for each sample; is the partial derivative of the predicted value with respect to the parameter;

[0175] Use the quasi-Newton method to update the parameters to minimize the loss function;

[0176]

[0177] Among them, θ ( k+ 1) is the parameter value for the k+1th iteration, θ ( k) is the parameter value of the kth iteration, H is the quasi-Newton matrix, which can be solved iteratively through the update formula of the quasi-Newton method, H -1 is the inverse of the quasi-Newton matrix, is the gradient of the loss function at the kth iteration;

[0178] Step 3: Transfer learning optimization;

[0179] Using the transfer learning method, the knowledge of the pre-trained model is transferred to the power generation unit scheduling optimization task, which is used to improve the intelligence and adaptability of the scheduling strategy;

[0180] In domain adaptation, the Wasserstein Visual Distance (WVD) is used to select appropriate source domain features to reduce the feature distribution difference between the source domain and the target domain. WVD measures the distance between the two distributions and helps select aligned feature subsets to improve the effect of transfer learning.

[0181]

[0182] Among them, X s and X t It is the feature set of the source domain and the target domain; and The i-th sample in the source domain and the target domain; φ(g) is the mapping function, usually a neural network, and P×Pp is L P norm; m and n represent the number of samples in the source domain and target domain datasets respectively;

[0183] Based on the selected source domain feature subset, feature extraction is performed on the source domain and target domain data to provide input data for model training; a consistency loss function is defined to measure the consistency between the prediction of the auxiliary label classifier and the true label of the target domain;

[0184]

[0185] in, is the consistency loss function, is the true label of the target domain, The prediction value of the auxiliary label classifier;

[0186] By minimizing the consistency loss function and feature distribution differences, domain-adaptive migration scheduling optimization is achieved;

[0187] Among them, f is the feature extractor, g is the auxiliary label classifier, and X s and X t are the feature sets of the source domain and the target domain respectively, and λ is a hyperparameter that weighs the losses of the two parts;

[0188] Numerical case analysis:

[0189] There are three generating units (DGU1, DGU2, DGU3), and the working information and fault record information of each generating unit are as follows:

[0190] A=[[800,230,5,30,85,24],[750,225,2,32,80,24],...]#DGU1

[0191] [[850,240,3,28,75,22],[830,235,1,30,70,22],...]#DGU2

[0192] [[820,235,4,31,78,20],[800,230,2,33,73,20],...]#DGU3

[0193] The input data passes through the LSTM model and outputs the health scores: DGU1:85, DGU2:90, DGU3:80;

[0194] Prophet load forecast: Forecasted load for the next day: DGU1:780, DGU2:840, DGU3:810;

[0195] Transfer learning scheduling optimization: select features through WVD, use dynamic weight adjustment mechanism, and update weights: DGU1: 0.3, DGU2: 0.4, DGU3: 0.3;

[0196] According to the health score and load forecast, optimize the dispatch and adjust the power generation distribution:

[0197] DGU1: 780*0.3=234;

[0198] DGU2: 840*0.4=336;

[0199] DGU3: 810*0.3=243;

[0200] The above describes an embodiment of the present embodiment, but the present embodiment is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present embodiment, ordinary technicians in this field can also make many forms, all of which are within the protection of the present embodiment.

Claims

1. An intelligent control method for a high-performance diesel generator set, characterized in that: The generator set comprises: A power generation unit for generating electricity using diesel; An information collector, which is used to collect the working information and fault record information of y power generation units; y is a custom parameter; A controller, which is used to control the power generation of the power generation unit in the next day according to the information collected by the information collector; The control method comprises the following steps: Step S201, obtaining working information and fault record information of y power generation units through an information collector; The working information of the power generation unit includes: the maximum load information, maximum voltage information, safety circuit information and environmental information of each day in the past year, and the usage time information of the power generation unit; The fault record information includes: the fault type information and fault severity information of y power generation units in the past year; Step S202, constructing a health prediction model in the controller, and constructing a health matrix A according to the working information and fault record information of multiple power generation units, inputting the health matrix A into the health prediction model, and outputting the health score of each power generation unit; The health score ranges from 0 to 100; Step S203, adjusting the power generation of each power generation unit in the next day according to the health score of each power generation unit; The health matrix A is as follows: in, Indicates the working information and fault record information of the first power generation unit on the first day; Indicates the working information and fault record information of the first power generation unit on the mth day; Indicates the working information and fault record information of the nth power generation unit on the first day; Indicates the working information and fault record information of the nth power generation unit on the mth day; m is 365, n is a custom parameter; The method for obtaining the working information of the power generation unit includes: The load of the power generation unit is monitored in real time through the load monitoring equipment, and the maximum load of each day is selected as the maximum load information of each day in the past year; The voltage of the power generation unit is monitored in real time by a voltage testing device, and the maximum voltage of each day is selected as the maximum voltage information; The current and voltage information of the safety circuit of the power generation unit is monitored in real time every day by current monitoring equipment and voltage testing equipment, and the maximum current and maximum voltage are selected as the safety circuit information; The temperature and humidity information of the corresponding area of ​​the power generation unit is monitored in real time through the temperature and humidity testing equipment, and the maximum and minimum temperature and humidity are selected as environmental information; Use time information of the power generation unit: the use time information of the power generation unit is determined according to the production information of the power generation unit.

2. The intelligent control method of a high-performance diesel generator set according to claim 1 is characterized in that: The health prediction model is built based on long short-term memory network.

3. The intelligent control method of a high-performance diesel generator set according to claim 1 is characterized in that: The health prediction model includes m columns of hidden layers and a classifier; The working information and fault record information of the power generation unit in the jth column of the health matrix A are input into the jth hidden layer, and the updated features are output; Update the feature input classifier and output the health score of each power generation unit; 1≤j≤m.

4. The intelligent control method of a high-performance diesel generator set according to claim 3 is characterized in that: The calculation formula of the jth hidden layer of the health prediction model includes: The forget gate F of the jth hidden layer of the health prediction model j The calculation formula is as follows: F j =σ(x j w F1 +H j-1 w F2 +b F ) In the formula, w F1 、w F2 and b F Represents the forget gate F of the jth hidden layer j The corresponding first, second and third weight parameters; H j-1 represents the hidden layer output at the previous moment, σ represents the sigmoid function; The input gate I of the jth hidden layer of the health prediction model j The calculation formula is as follows: I j =σ(x j w I1 +H j-1 w I2 +b I ); In the formula, w I1 、w I2 and b I Represents the input gate I of the jth hidden layer j corresponding first, second and third weight parameters; The intermediate state of the jth hidden layer of the health prediction model The calculation formula is as follows: In the formula, w C1 、w C2 and b C Represents the intermediate state of the jth hidden layer The corresponding first, second and third weight parameters; x j Represents the input of the jth hidden layer; The output state C of the jth hidden layer of the health prediction model j The calculation formula is as follows: In the formula, C j-1 represents the output state of the j-1th hidden unit in the first hidden layer; ⊙ represents the dot product operation; The output gate O of the jth hidden layer of the health prediction model j The calculation formula is as follows: O j =σ(x j w 01 +H j-1 w 02 +b0) In the formula, w 01 、w 02 and b0 represents the output gate O of the jth hidden layer j corresponding first, second and third weight parameters; The output H of the jth hidden layer of the health prediction model j The calculation formula is as follows: H j =0 j ☉tanh(C j ) In the formula, the output gate O j and tanh(C j ) multiply point by point to get the output H j .

5. The intelligent control method of a high-performance diesel generator set according to claim 1, characterized in that: Obtain the working information and fault record information of u power generation units in time period v, and construct a training matrix B; Set a sliding window in the training matrix B, the length of the sliding window is m, the width is n, and the step length is 1; The sliding window slides continuously horizontally and vertically in the training matrix B to obtain several training samples as sample data for training the health prediction model; The health score of each power generation unit corresponding to the training sample is obtained as a sample label for training the health prediction model.

6. The intelligent control method of a high-performance diesel generator set according to claim 1, characterized in that: The objective function is used to improve the accuracy of the health prediction model, and its calculation formula is as follows: Where L represents the loss value of the health prediction model; N represents the total number of all power generation units; y i represents the actual health score of the ith power generation unit; The health prediction model outputs the health score of the i-th power generation unit.

7. The intelligent control method of a high-performance diesel generator set according to claim 1, characterized in that: The steps to adjust the power generation and power generation efficiency of the power generation unit are as follows: Step S301, dividing the power grid system according to where each power generation unit is located; Step S302, obtaining the preset power generation in each power grid system; Step S303, presetting the first and second health scores; Step S304, the power generation of the power generation unit whose health score is lower than the first health score in the next day shall not exceed the first power generation threshold; the power generation of the power generation unit whose health score is lower than the second health score in the next day shall not exceed the second power generation threshold; the remaining preset power generation is scored by the power generation unit whose health score is higher than the second health score.

8. The intelligent control method of a high-performance diesel generator set according to claim 1, characterized in that: The information collector is used to obtain the working information and fault record information of the power generation unit, and the scheduling optimization of the power generation unit is realized. The specific process is as follows: Step 1, information collection; Working information includes: maximum load information, maximum voltage information, safety circuit information, environmental information, and usage time of the power generation unit for each day in the past year; Fault record information includes: fault type information and fault severity information of the power generation unit in the past year; X data matrix, including the working information and fault record information of all power generation units; x ij The working information and fault record information of the i-th power generation unit on the j-th day; y is the load forecast value for the next day; Step 2, load prediction; The Prophet model is used to predict future loads and provide a reference for scheduling optimization. It consists of trend, seasonality and holiday effects. The prediction formula of the model is as follows: in, is the model prediction value; g(t) is the trend function, which indicates the long-term trend; s(t) is the seasonal component, which indicates the periodic change; h(t) is the holiday effect, which indicates the impact of special dates; t is the error term; Define the loss function for parameter updating, which measures the difference between the model prediction and the true value; Where L is the smooth L1 loss function, y i is the true load value, is the predicted load value, i represents the i-th power generation unit; Calculate the gradient of the loss function with respect to the parameter θ for parameter update; in, is the gradient of the loss function, Used to indicate the direction of the prediction error for each sample; is the partial derivative of the predicted value with respect to the parameter; Use the quasi-Newton method to update the parameters to minimize the loss function; Among them, θ (k+1) is the parameter value for the k+1th iteration, θ (k) is the parameter value of the kth iteration, H is the quasi-Newton matrix, which can be solved iteratively through the update formula of the quasi-Newton method, H -1 is the inverse of the quasi-Newton matrix, is the gradient of the loss function at the kth iteration; Step 3: Transfer learning optimization; Using the transfer learning method, the knowledge of the pre-trained model is transferred to the power generation unit scheduling optimization task, which is used to improve the intelligence and adaptability of the scheduling strategy; In domain adaptation, the Wasserstein Visual Distance (WVD) is used to select appropriate source domain features to reduce the feature distribution difference between the source domain and the target domain. WVD measures the distance between the two distributions and helps select aligned feature subsets to improve the effect of transfer learning. Among them, X s and X t It is the feature set of the source domain and the target domain; and The i-th sample in the source domain and the target domain; φ(·) is the mapping function, usually a neural network, ||·|| p YesL P norm; m and n represent the number of samples in the source domain and target domain datasets respectively; Based on the selected source domain feature subset, feature extraction is performed on the source domain and target domain data to provide input data for model training; a consistency loss function is defined to measure the consistency between the prediction of the auxiliary label classifier and the true label of the target domain; Among them, L consistency is the consistency loss function, is the true label of the target domain, The prediction value of the auxiliary label classifier; By minimizing the consistency loss function and feature distribution differences, domain-adaptive migration scheduling optimization is achieved; Among them, f is the feature extractor, g is the auxiliary label classifier, and X s and X t are the feature sets of the source domain and the target domain respectively, and λ is a hyperparameter that weighs the losses of the two parts.

Images