Noise control method and system for a diesel engine set based on support vector machines

Abstract

The present application relates to the field of noise control, and relates to a diesel engine unit noise control method based on a support vector machine, comprising: modal decomposition and frequency domain screening of a unit noise data set to obtain a noise frequency spectrum feature group set; sequentially performing multi-modal feature extraction and feature dimension filtering operations on a unit operation data set to obtain a standard state feature group set; using the standard state feature group set and the noise frequency spectrum feature group set to perform random particle training on a preset regression vector machine model to obtain a unit noise model; calculating a real-time noise frequency spectrum feature group corresponding to real-time unit operation data of a target diesel engine unit according to the unit noise model, performing sound source matching on the real-time noise frequency spectrum feature group to obtain a real-time noise sound source result, and performing noise control on the target diesel engine unit according to the real-time noise sound source result. The present application also proposes a diesel engine unit noise control system based on a support vector machine. The present application can improve the efficiency of noise control.

Description

Technical Field

[0001] This invention relates to the field of noise control technology, and in particular to a noise control method and system for diesel generator sets based on support vector machines. Background Technology

[0002] A diesel engine is an internal combustion engine that uses diesel fuel to convert the heat energy released from fuel combustion into electrical or mechanical energy. It is widely used in agricultural machinery, construction machinery, automobile industry, marine machinery, power industry and other fields. However, diesel engine units generate a lot of noise when in use, so noise control of diesel engine units is necessary.

[0003] Existing noise control methods for diesel engines are mostly based on simple physical sound insulation. These methods use fixed sound insulation materials or equipment to physically isolate and reduce noise from the diesel engine unit. They also utilize sound insulation materials around the diesel engine unit to absorb noise and use vibration isolation materials such as springs or dampers to reduce vibration noise. However, in practical applications, the noise generated by diesel engine units often comes from multiple sources, such as mechanical vibration, exhaust noise, or fan noise. Moreover, the noise generated in each operating state is different. Noise control methods based on simple physical sound insulation cannot achieve targeted noise control for each noise source, which may lead to low efficiency in noise control. Summary of the Invention

[0004] This invention provides a noise control method and system for diesel generator sets based on support vector machines, the main purpose of which is to solve the problem of low efficiency in noise control.

[0005] To achieve the above objectives, the present invention provides a noise control method for diesel generator sets based on support vector machines, comprising:

[0006] The previously acquired historical diesel generator set dataset was sequentially split and cleaned to obtain the generator set noise dataset and the generator set operation dataset.

[0007] Modal decomposition is performed on the unit noise dataset to obtain a noise mode feature set. Frequency domain filtering is then performed on the noise mode feature set to obtain a noise spectrum feature set. Specifically, the frequency domain filtering of the noise mode feature set to obtain the noise spectrum feature set includes: selecting noise mode feature sets one by one from the noise mode feature set as target noise mode feature sets, and calculating the mode spectrum coefficients corresponding to the target noise mode feature sets using the following modal spectrum coefficient algorithm:

[0008]

[0009] Among them, w j(a, b) refers to the coefficient value of the j-th modal spectral coefficient in the modal spectral coefficient group under the scale parameter a and displacement parameter b, where a is the scale parameter, b is the displacement parameter, j is the index, and ∞ represents infinity. j (t) refers to the feature value of the j-th target noise mode feature in the target noise mode feature group at time number t, where t is the time number, z() is the conjugate function symbol, e is the Euler number, i is the imaginary symbol, ω0 is the preset frequency parameter, and d is the integral symbol; spectral features are extracted from the modal spectral coefficient group to obtain the primary spectral feature group; cross-relation filtering is performed on the primary spectral feature group to obtain the noise spectral feature group, and all the noise spectral feature groups are collected into a noise spectral feature group set;

[0010] Multimodal feature extraction and feature dimension filtering operations are performed sequentially on the unit operation dataset to obtain a standard state feature set;

[0011] The pre-defined regression vector machine model is trained using random particles using the standard state feature set and the noise spectrum feature set to obtain the unit noise model.

[0012] The real-time noise spectrum feature group corresponding to the real-time unit operation data of the target diesel unit is calculated based on the unit noise model. The real-time noise spectrum feature group is then matched with the source to obtain the real-time noise source result. The target diesel unit is then subjected to noise control based on the real-time noise source result.

[0013] Optionally, the step of sequentially performing data splitting and data cleaning operations on the pre-acquired historical diesel generator set dataset to obtain a generator set noise dataset and a generator set operation dataset includes:

[0014] The historical diesel generator set dataset is hash-mapped to obtain a diesel hash signature set;

[0015] The intersection-union similarity of the diesel hash signature set is calculated to obtain the data similarity results;

[0016] Based on the data similarity results, the historical diesel generator set dataset is deduplicated to obtain a deduplicated diesel generator set dataset.

[0017] Data attribute groups are extracted from the de-dust diesel engine group dataset, and missing data matching is performed on the de-dust diesel engine group dataset according to the data attribute groups to obtain the standard diesel engine group dataset.

[0018] The standard diesel generator set dataset is split into a generator set noise dataset and a generator set operation dataset based on the data attribute group.

[0019] Optionally, the step of performing intersection-union similarity calculation on the diesel hash signature set to obtain data similarity results includes:

[0020] One by one, select the diesel hash signatures in the diesel hash signature set as the target diesel hash signature, and one by one select the diesel hash signatures in the diesel hash signature set other than the target diesel hash signature as the matching diesel hash signature;

[0021] The intersection of the target diesel hash signature and the matching diesel hash signature is calculated to obtain the target signature intersection value.

[0022] The union of the target diesel hash signature and the matching diesel hash signature is calculated to obtain the union value of the target signature;

[0023] The ratio of the intersection value of the target signatures to the union value of the target signatures is used as the target similarity, and all target similarities are aggregated into a data similarity result.

[0024] Optionally, the step of performing mode decomposition on the unit noise dataset to obtain a noise mode feature set includes:

[0025] Each unit noise data point in the unit noise dataset is selected as the target unit noise data, and the maximum noise line and the minimum noise line are extracted from the target unit noise data respectively.

[0026] The average noise line is calculated based on the target unit noise data, the maximum noise line, and the minimum noise line.

[0027] The target unit noise data is filtered for detail using the average noise line to obtain primary detail noise data.

[0028] The primary detailed noise data is iteratively filtered for extreme value features to obtain noise mode feature groups, and all noise mode feature groups are aggregated into a noise mode feature set.

[0029] Optionally, the iterative extreme value feature filtering of the primary detail noise data to obtain a noise mode feature set includes:

[0030] Determine whether the primary detail noise data satisfies the preset extreme value characteristic conditions;

[0031] If not, the primary detail noise data is used as the target unit noise data, and the steps of extracting the maximum noise line and the minimum noise line from the target unit noise data are returned.

[0032] If so, the primary detail noise data is added as a noise modal feature to a preset primary feature group, and it is determined whether the number of features in the primary feature group is equal to a preset number threshold.

[0033] If not, then use the primary detail noise to perform detail filtering on the target unit noise data to obtain the target filtered unit noise, and return to the step of extracting the maximum noise line and the minimum noise line from the target unit noise data respectively.

[0034] If so, the primary feature set is taken as the noise mode feature set.

[0035] Optionally, the step of performing cross-relation filtering on the primary spectral feature group to obtain the noise spectral feature group includes:

[0036] Each primary spectral feature in the primary spectral feature group is paired up to obtain a spectral feature pair group;

[0037] Each spectral feature pair in the aforementioned spectral feature pair group is selected as the target spectral feature pair, and the feature information value of the target spectral feature pair is calculated using the following feature information algorithm:

[0038]

[0039] Where M(X,Y) refers to the feature information value of the target spectral feature pair composed of spectral feature X and spectral feature Y, X and Y are the spectral features of the target spectral feature pair, x is any possible value of the spectral feature X, y is any possible value of the spectral feature Y, p(x,y) is the joint probability that the spectral feature X is x and the spectral feature Y is y, log() is the logarithmic function, p(x) is the marginal probability corresponding to the spectral feature X being x, and p(y) is the marginal probability corresponding to the spectral feature Y being y;

[0040] All feature information values ​​are arranged in descending order to form a feature information value sequence. The feature information value sequence is then filtered by a position threshold to obtain a feature information value group.

[0041] The target spectral feature pairs corresponding to the feature information value groups are aggregated into standard spectral feature pairs, and all spectral features in the standard spectral feature pairs are aggregated into noise spectral feature groups.

[0042] Optionally, the step of sequentially performing multimodal feature extraction and feature dimension filtering operations on the unit operation dataset to obtain a standard state feature set includes:

[0043] The speed feature set, load feature set, temperature feature set, and pressure feature set are extracted from the unit operation dataset, respectively.

[0044] The speed feature set, the load feature set, the temperature feature set, and the pressure feature set are combined into a primary state feature set;

[0045] Perform global feature normalization on the primary state feature set to obtain the secondary state feature set.

[0046] Principal component screening is performed on the secondary state feature set to obtain the standard state feature set.

[0047] Optionally, the step of using the standard state feature set and the noise spectrum feature set to perform random particle training on a preset regression vector machine model to obtain a unit noise model includes:

[0048] An initial set of particle parameters is randomly generated, and the initial particle parameters in the initial set are selected one by one to update the parameters of the preset regression vector machine model, so as to obtain the updated vector machine model.

[0049] The regression spectral feature set corresponding to the standard state feature set is calculated using the updated vector machine model.

[0050] The performance index of the updated vector machine model is calculated based on the regression spectral feature set and the noise spectral feature set.

[0051] The initial particle parameters in the initial particle parameter set are iteratively updated according to the performance index to obtain the standard particle parameters.

[0052] The parameters of the regression vector machine model are updated using the standard particle parameters to obtain the unit noise model.

[0053] Optionally, the step of performing sound source matching on the real-time noise spectrum feature group to obtain real-time noise source results includes:

[0054] Each real-time noise spectrum feature in the real-time noise spectrum feature group is selected as a target real-time spectrum feature, and the target real-time spectrum feature is regularized to obtain a regularized real-time spectrum feature.

[0055] Edge decision-making is performed on the regularized real-time spectral features to obtain real-time decision features;

[0056] The real-time decision features are normalized to obtain the real-time sound source code;

[0057] The real-time spectral features of the target are labeled using the real-time sound source coding to obtain real-time noise sources, and all real-time noise sources are aggregated into a real-time noise source result.

[0058] To address the aforementioned problems, the present invention also provides a noise control system for a diesel generator set based on a support vector machine, the system comprising:

[0059] The data splitting module is used to sequentially split and clean the pre-acquired historical diesel generator set dataset to obtain the generator set noise dataset and the generator set operation dataset.

[0060] The frequency domain filtering module is used to perform mode decomposition on the unit noise dataset to obtain a noise mode feature set, and to perform frequency domain filtering on the noise mode feature set to obtain a noise spectrum feature set. The step of performing frequency domain filtering on the noise mode feature set to obtain the noise spectrum feature set includes: selecting noise mode feature sets one by one from the noise mode feature set as target noise mode feature sets, and calculating the mode spectrum coefficient set corresponding to the target noise mode feature set using the following mode spectrum coefficient algorithm:

[0061]

[0062] Among them, w j (a, b) refers to the coefficient value of the j-th modal spectral coefficient in the modal spectral coefficient group under the scale parameter a and displacement parameter b, where a is the scale parameter, b is the displacement parameter, j is the index, and ∞ represents infinity. j (t) refers to the feature value of the j-th target noise mode feature in the target noise mode feature group at time number t, where t is the time number, z() is the conjugate function symbol, e is the Euler number, i is the imaginary symbol, ω0 is the preset frequency parameter, and d is the integral symbol; spectral features are extracted from the modal spectral coefficient group to obtain the primary spectral feature group; cross-relation filtering is performed on the primary spectral feature group to obtain the noise spectral feature group, and all the noise spectral feature groups are collected into a noise spectral feature group set;

[0063] The feature filtering module is used to sequentially perform multimodal feature extraction and feature dimension filtering operations on the unit operation dataset to obtain a standard state feature set.

[0064] The particle training module is used to perform random particle training on the preset regression vector machine model using the standard state feature set and the noise spectrum feature set to obtain the unit noise model.

[0065] The noise control module is used to calculate the real-time noise spectrum feature group corresponding to the real-time unit operation data of the target diesel unit according to the unit noise model, perform sound source matching on the real-time noise spectrum feature group to obtain the real-time noise source result, and perform noise control on the target diesel unit according to the real-time noise source result.

[0066] This invention improves data accuracy by sequentially splitting and cleaning a pre-acquired historical diesel generator set dataset to obtain a generator set noise dataset and a generator set operation dataset. It also distinguishes between the input and output of the training set, facilitating subsequent model training. Modal decomposition of the generator set noise dataset decomposes the diesel generator set noise signal, facilitating noise source identification. Frequency domain filtering of the noise modal feature set identifies highly correlated noise features, improving the accuracy of subsequent noise source analysis and model training. Furthermore, sequential multimodal feature extraction and feature dimension filtering of the generator set operation dataset extracts and standardizes features from different modes of diesel generator set operation, improving the accuracy of subsequent model training.

[0067] By using the standard state feature set and the noise spectrum feature set to train a preset regression vector machine model with random particles, global parameter search can be achieved, avoiding the trap of getting trapped in local optima, thus improving the efficiency and robustness of the unit noise model. By calculating the real-time noise spectrum feature set corresponding to the real-time unit operation data of the target diesel unit based on the unit noise model, the real-time noise distribution of the diesel unit can be predicted. By performing sound source matching on the real-time noise spectrum feature set, the real-time noise source results can be obtained, and the noise magnitude and distribution of each noise component of the diesel unit can be determined. By performing noise control on the target diesel unit based on the real-time noise source results, targeted noise reduction of the diesel unit can be achieved, thereby improving the efficiency of diesel unit noise control. Therefore, the diesel unit noise control method and system based on support vector machines proposed in this invention can solve the problem of low efficiency in noise control. Attached Figure Description

[0068] Figure 1 This is a flowchart illustrating a noise control method for a diesel generator set based on a support vector machine, according to an embodiment of the present invention.

[0069] Figure 2 This is a flowchart illustrating the calculation of data similarity results according to an embodiment of the present invention.

[0070] Figure 3 This is a schematic diagram of the process for extracting noise modal feature sets according to an embodiment of the present invention;

[0071] Figure 4 A functional block diagram of a noise control system for a diesel generator set based on a support vector machine, provided in an embodiment of the present invention;

[0072] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0073] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0074] This application provides a noise control method for diesel generator sets based on support vector machines. The execution entity of the noise control method for diesel generator sets based on support vector machines includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application embodiment: a server, a terminal, etc. In other words, the noise control method for diesel generator sets based on support vector machines can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0075] Reference Figure 1 The diagram shown is a flowchart illustrating a noise control method for a diesel engine unit based on a support vector machine according to an embodiment of the present invention. In this embodiment, the noise control method for the diesel engine unit based on a support vector machine includes:

[0076] S1. Perform data splitting and data cleaning operations on the pre-acquired historical diesel generator set dataset in sequence to obtain the generator set noise dataset and the generator set operation dataset.

[0077] In this embodiment of the invention, each historical diesel engine set data in the historical diesel engine set dataset corresponds to the operating status monitoring data and corresponding noise monitoring data of the diesel engine set under a certain working mode. Each historical diesel engine set data in the historical diesel engine set dataset corresponds to each unit noise data in the unit noise dataset and each unit operation data in the unit operation dataset. The unit noise data corresponds to data such as generator load, fuel consumption, generator voltage and current, engine speed, coolant temperature, oil temperature, and cylinder pressure of the diesel engine set under a certain working mode. The unit noise data also corresponds to the sound signal collected at a fixed position directly in front of the diesel engine set under a certain working mode.

[0078] In this embodiment of the invention, the step of sequentially performing data splitting and data cleaning operations on the pre-acquired historical diesel generator set dataset to obtain a generator set noise dataset and a generator set operation dataset includes:

[0079] The historical diesel generator set dataset is hash-mapped to obtain a diesel hash signature set;

[0080] The intersection-union similarity of the diesel hash signature set is calculated to obtain the data similarity results;

[0081] Based on the data similarity results, the historical diesel generator set dataset is deduplicated to obtain a deduplicated diesel generator set dataset.

[0082] Data attribute groups are extracted from the de-dust diesel engine group dataset, and missing data matching is performed on the de-dust diesel engine group dataset according to the data attribute groups to obtain the standard diesel engine group dataset.

[0083] The standard diesel generator set dataset is split into a generator set noise dataset and a generator set operation dataset based on the data attribute group.

[0084] In detail, the step of hash mapping the historical diesel generator set dataset to obtain a diesel hash signature set refers to using a hash mapping algorithm to calculate the diesel hash signature of each historical diesel generator set data in the historical diesel generator set dataset, and then aggregating all the diesel hash signatures into a diesel hash signature set. The hash mapping algorithm can be MD5, SHA-256, or BLAKE3, etc.

[0085] Specifically, refer to Figure 2 As shown, the step of performing intersection-union similarity calculation on the diesel hash signature set to obtain data similarity results includes:

[0086] S21. Select diesel hash signatures from the diesel hash signature set one by one as target diesel hash signatures, and select diesel hash signatures from the diesel hash signature set other than the target diesel hash signatures one by one as matching diesel hash signatures.

[0087] S22. Calculate the intersection of the target diesel hash signature and the matching diesel hash signature to obtain the target signature intersection value;

[0088] S23. Perform a union calculation on the target diesel hash signature and the matching diesel hash signature to obtain the target signature union value;

[0089] S24. The ratio of the intersection value of the target signatures to the union value of the target signatures is taken as the target similarity, and all target similarities are aggregated into a data similarity result.

[0090] Specifically, the target signature intersection value refers to the number of intersection items in the target diesel hash signature and the matching diesel hash signature, and the target signature union value refers to the number of union items in the target diesel hash signature and the matching diesel hash signature.

[0091] Specifically, the step of deduplicating the historical diesel generator set dataset based on the data similarity results to obtain a deduplicated diesel generator set dataset refers to selecting diesel hash signatures with a target similarity greater than a preset similarity threshold in the data similarity results as duplicate hash signatures, and filtering out the diesel generator set data corresponding to the duplicate hash signatures as duplicate diesel generator set data from the historical diesel generator set dataset to obtain a deduplicated diesel generator set dataset.

[0092] In detail, the data attribute group is an attribute group composed of the attributes of various data in the de-weighted diesel engine group dataset. The attributes include, for example, noise, temperature, pressure, current, and voltage. Specifically, the data attribute group can be extracted from the de-weighted diesel engine group dataset using a unit matching method. The missing data matching of the de-weighted diesel engine group dataset based on the data attribute group means sequentially matching the de-weighted diesel engine group data in the de-weighted diesel engine group dataset with the data attribute group. If the matching fails, the corresponding de-weighted diesel engine group data is deleted as missing data to obtain the standard diesel engine group dataset.

[0093] Specifically, splitting the standard diesel generator set dataset into a generator set noise dataset and a generator set operation dataset according to the data attribute group means extracting the generator set noise dataset from the standard diesel generator set dataset using the data attribute group and compiling the remaining data into a generator set operation dataset.

[0094] In this embodiment of the invention, by sequentially splitting and cleaning the pre-acquired historical diesel generator set dataset, a generator set noise dataset and a generator set operation dataset are obtained. This can improve the accuracy of the data and distinguish the input and output of the training set, which facilitates the subsequent model training steps.

[0095] S2. Perform mode decomposition on the noise dataset of the unit to obtain a noise mode feature set, and perform frequency domain filtering on the noise mode feature set to obtain a noise spectrum feature set.

[0096] In this embodiment of the invention, each noise mode feature group in the noise mode feature group set corresponds to a noise feature of a unit noise data in the unit noise dataset. Each noise mode feature in the noise mode feature group corresponds to a noise feature of a mode, namely, a noise feature of a specific waveform, frequency distribution, and amplitude distribution.

[0097] In this embodiment of the invention, reference is made to Figure 3 As shown, the modal decomposition of the unit noise dataset to obtain a noise modal feature set includes:

[0098] S31. Select the unit noise data in the unit noise dataset one by one as the target unit noise data, and extract the maximum noise line and the minimum noise line from the target unit noise data respectively.

[0099] S32. Calculate the average noise line based on the target unit noise data, the maximum noise line, and the minimum noise line;

[0100] S33. Use the average noise line to perform detail filtering on the target unit noise data to obtain primary detail noise data;

[0101] S34. Iteratively filter the extreme value features of the primary detail noise data to obtain noise mode feature groups, and gather all the noise mode feature groups into a noise mode feature set.

[0102] Specifically, the maximum noise line refers to the curve or broken line segment formed by connecting all the maximum values ​​of the signals in the target unit noise data, and the minimum noise line refers to the curve or broken line segment formed by connecting all the minimum values ​​of the signals in the target unit noise data. The step of using the average noise line to perform detail filtering on the target unit noise data to obtain primary detail noise data means subtracting the average noise line from the target unit noise data to obtain primary detail noise data.

[0103] Specifically, the iterative extreme value feature filtering of the primary detail noise data to obtain a noise mode feature set includes:

[0104] Determine whether the primary detail noise data satisfies the preset extreme value characteristic conditions;

[0105] If not, the primary detail noise data is used as the target unit noise data, and the steps of extracting the maximum noise line and the minimum noise line from the target unit noise data are returned.

[0106] If so, the primary detail noise data is added as a noise modal feature to a preset primary feature group, and it is determined whether the number of features in the primary feature group is equal to a preset number threshold.

[0107] If not, then use the primary detail noise to perform detail filtering on the target unit noise data to obtain the target filtered unit noise, and return to the step of extracting the maximum noise line and the minimum noise line from the target unit noise data respectively.

[0108] If so, the primary feature set is taken as the noise mode feature set.

[0109] Specifically, the extreme value characteristic conditions include that the mean of any fixed point on the primary detail noise data is zero and the number of extreme points of the primary detail noise data at any fixed point is less than and equal to the number of times the extreme points are eliminated.

[0110] Specifically, the step of performing frequency domain filtering on the noise modal feature set to obtain the noise spectral feature set includes:

[0111] Each noise mode feature group in the noise mode feature set is selected as a target noise mode feature group, and the modal spectrum coefficient group corresponding to the target noise mode feature group is calculated using the following modal spectrum coefficient algorithm:

[0112]

[0113] Among them, w j (a, b) refers to the coefficient value of the j-th modal spectral coefficient in the modal spectral coefficient group under the scale parameter a and displacement parameter b, where a is the scale parameter, b is the displacement parameter, j is the index, and ∞ represents infinity. j (t) refers to the feature value of the j-th target noise mode feature in the target noise mode feature group at time number t, where t is the time number, z() is the conjugate function symbol, e is the Euler number, i is the imaginary number symbol, ω0 is the preset frequency parameter, and d is the integral symbol.

[0114] Spectral features are extracted from the modal spectral coefficient set to obtain a primary spectral feature set;

[0115] The primary spectral feature groups are cross-referenced to obtain noise spectral feature groups, and all the noise spectral feature groups are aggregated into a noise spectral feature group set.

[0116] In detail, the modal spectrum coefficient algorithm can analyze the frequency characteristics of noise modal features at different scales and displacements, thereby improving the recognizability of features and preserving detailed features.

[0117] Specifically, a primary spectral feature set can be obtained by calculating the mean and variance characteristics of each modal spectral coefficient in the modal spectral coefficient set or by using energy statistics to extract spectral features from the modal spectral coefficient set.

[0118] Specifically, the step of performing cross-relation filtering on the primary spectral feature group to obtain the noise spectral feature group includes:

[0119] Each primary spectral feature in the primary spectral feature group is paired up to obtain a spectral feature pair group;

[0120] Each spectral feature pair in the aforementioned spectral feature pair group is selected as the target spectral feature pair, and the feature information value of the target spectral feature pair is calculated using the following feature information algorithm:

[0121]

[0122] Where M(X,Y) refers to the feature information value of the target spectral feature pair composed of spectral feature X and spectral feature Y, X and Y are the spectral features of the target spectral feature pair, x is any possible value of the spectral feature X, y is any possible value of the spectral feature Y, p(x,y) is the joint probability that the spectral feature X is x and the spectral feature Y is y, log() is the logarithmic function, p(x) is the marginal probability corresponding to the spectral feature X being x, and p(y) is the marginal probability corresponding to the spectral feature Y being y;

[0123] All feature information values ​​are arranged in descending order to form a feature information value sequence. The feature information value sequence is then filtered by a position threshold to obtain a feature information value group.

[0124] The target spectral feature pairs corresponding to the feature information value groups are aggregated into standard spectral feature pairs, and all spectral features in the standard spectral feature pairs are aggregated into noise spectral feature groups.

[0125] In detail, by using a feature information algorithm to calculate the feature information value of the target spectral feature pair, the correlation between the two spectral features can be determined based on the probability distribution of various possible values ​​of the features, thereby achieving feature selection.

[0126] Specifically, the step of performing position threshold filtering on the feature information value sequence to obtain a feature information value group refers to selecting a preset number of feature information values ​​with the highest positions in the feature information value sequence and assembling them into a feature information value group.

[0127] In this embodiment of the invention, by performing modal decomposition on the noise dataset of the generator set, the noise signal of the diesel generator set can be decomposed, thereby facilitating the subsequent identification of the noise source. By performing frequency domain filtering on the noise modal feature set, highly correlated noise features can be selected, thereby improving the accuracy of subsequent noise source analysis and model training.

[0128] S3. Perform multimodal feature extraction and feature dimension filtering operations on the unit operation dataset in sequence to obtain a standard state feature set.

[0129] In this embodiment of the invention, each standard state feature group in the standard state feature group set corresponds to a unit operation data in the unit operation dataset. Each standard state feature in the standard state feature group represents an operating state feature of the diesel engine unit during operation, such as engine speed feature, cooling water temperature feature, lubricating oil pressure feature, and exhaust temperature feature.

[0130] In this embodiment of the invention, the step of sequentially performing multimodal feature extraction and feature dimension filtering operations on the unit operation dataset to obtain a standard state feature set includes:

[0131] The speed feature set, load feature set, temperature feature set, and pressure feature set are extracted from the unit operation dataset, respectively.

[0132] The speed feature set, the load feature set, the temperature feature set, and the pressure feature set are combined into a primary state feature set;

[0133] Perform global feature normalization on the primary state feature set to obtain the secondary state feature set.

[0134] Principal component screening is performed on the secondary state feature set to obtain the standard state feature set.

[0135] Specifically, each speed feature in the speed feature set, each load feature in the load feature set, each temperature feature in the temperature feature set, and each pressure feature in the pressure feature set corresponds to each unit operation data in the unit operation dataset. The speed feature refers to the characteristic of the engine speed of the diesel engine unit, the load feature refers to the load of the engine, and the characteristics of the generator voltage, current, etc., the temperature feature refers to the characteristics of the engine cooling water temperature, engine oil temperature, and exhaust temperature, etc., and the pressure feature refers to the characteristics of cylinder pressure, lubricating oil pressure, combustion pressure, etc.

[0136] In detail, a global pooling layer can be used to perform global feature normalization on the primary state feature set, that is, to normalize and unify the feature length and dimension of each primary state feature in the primary state feature set. Principal component analysis can be used to perform principal component screening on the secondary state feature set to obtain a standard state feature set.

[0137] In this embodiment of the invention, by sequentially performing multimodal feature extraction and feature dimension filtering operations on the unit operation dataset, features of different modes during diesel unit operation can be extracted and standardized, thereby improving the accuracy of subsequent model training.

[0138] S4. Using the standard state feature set and the noise spectrum feature set, perform random particle training on the preset regression vector machine model to obtain the unit noise model.

[0139] In this embodiment of the invention, the regression vector machine model refers to the Support Vector Machine for Regression (SVR), a machine learning model for regression problems. The regression vector machine model finds a regression function to predict a continuous target variable. The regression vector machine model finds the optimal regression function by minimizing the prediction error and keeping the error limit in mind.

[0140] In this embodiment of the invention, the step of using the standard state feature set and the noise spectrum feature set to perform random particle training on a preset regression vector machine model to obtain a unit noise model includes:

[0141] An initial set of particle parameters is randomly generated, and the initial particle parameters in the initial set are selected one by one to update the parameters of the preset regression vector machine model, so as to obtain the updated vector machine model.

[0142] The regression spectral feature set corresponding to the standard state feature set is calculated using the updated vector machine model.

[0143] The performance index of the updated vector machine model is calculated based on the regression spectral feature set and the noise spectral feature set.

[0144] The initial particle parameters in the initial particle parameter set are iteratively updated according to the performance index to obtain the standard particle parameters.

[0145] The parameters of the regression vector machine model are updated using the standard particle parameters to obtain the unit noise model.

[0146] In detail, each initial particle parameter in the initial particle parameter set corresponds to a set of model parameters of the regression vector machine model. The performance index of the updated vector machine model can be calculated using the mean square error algorithm based on the regression spectral feature set and the noise spectral feature set.

[0147] Specifically, the step of iteratively updating each initial particle parameter in the initial particle parameter set according to the performance index to obtain standard particle parameters refers to updating the position and velocity of each initial particle parameter in the initial particle parameter set and the position of neighboring particles according to the performance index. The standard particle parameters refer to the initial particle parameters with the globally optimal position. The Particle Swarm Optimization (PSO) algorithm can be used to update the position and velocity of each initial particle parameter in the initial particle parameter set and the position of neighboring particles according to the performance index.

[0148] In this embodiment of the invention, by using the standard state feature set and the noise spectrum feature set to perform random particle training on the preset regression vector machine model, global parameter search can be achieved, avoiding the trap of getting into local optima, and improving the efficiency and robustness of the unit noise model.

[0149] S5. Calculate the real-time noise spectrum feature group corresponding to the real-time unit operation data of the target diesel unit according to the unit noise model, perform sound source matching on the real-time noise spectrum feature group to obtain the real-time noise source result, and perform noise control on the target diesel unit according to the real-time noise source result.

[0150] In this embodiment of the invention, the real-time unit operation data refers to the real-time unit operation data of the target diesel engine, including engine speed characteristics, cooling water temperature characteristics, lubricating oil pressure characteristics, and exhaust temperature characteristics, etc. The real-time noise spectrum characteristic group is the noise spectrum characteristic group of the target diesel engine predicted based on the real-time unit operation data.

[0151] In this embodiment of the invention, the step of performing sound source matching on the real-time noise spectrum feature group to obtain real-time noise source results includes:

[0152] Each real-time noise spectrum feature in the real-time noise spectrum feature group is selected as a target real-time spectrum feature, and the target real-time spectrum feature is regularized to obtain a regularized real-time spectrum feature.

[0153] Edge decision-making is performed on the regularized real-time spectral features to obtain real-time decision features;

[0154] The real-time decision features are normalized to obtain the real-time sound source code;

[0155] The real-time spectral features of the target are labeled using the real-time sound source coding to obtain real-time noise sources, and all real-time noise sources are aggregated into a real-time noise source result.

[0156] Specifically, the support vectors of the support vector machine model trained with a large number of labeled spectral features of noise sources can be used to make edge decisions on the regular real-time spectral features to obtain real-time decision features. The real-time sound source coding is the normalized coding result of the noise sound source, and the real-time sound source coding is a decimal between 0 and 1.

[0157] In detail, noise control of the target diesel engine unit based on the real-time noise source results refers to targeted noise control based on the source location and magnitude of each real-time noise source in the real-time noise source results. For example, for the exhaust noise of the diesel engine unit, a fixed sound insulation material or sound insulation structure is set according to the noise magnitude.

[0158] In this embodiment of the invention, by calculating the real-time noise spectrum feature group corresponding to the real-time operation data of the target diesel generator set according to the generator set noise model, the distribution of real-time noise of the diesel generator set can be predicted. By performing sound source matching on the real-time noise spectrum feature group, the real-time noise source results can be obtained, and the noise magnitude and distribution of each noise component of the diesel generator set can be determined. By performing noise control on the target diesel generator set according to the real-time noise source results, targeted noise reduction of the diesel generator set can be achieved, thereby improving the efficiency of diesel generator set noise control.

[0159] This invention improves data accuracy by sequentially splitting and cleaning a pre-acquired historical diesel generator set dataset to obtain a generator set noise dataset and a generator set operation dataset. It also distinguishes between the input and output of the training set, facilitating subsequent model training. Modal decomposition of the generator set noise dataset decomposes the diesel generator set noise signal, facilitating noise source identification. Frequency domain filtering of the noise modal feature set identifies highly correlated noise features, improving the accuracy of subsequent noise source analysis and model training. Furthermore, sequential multimodal feature extraction and feature dimension filtering of the generator set operation dataset extracts and standardizes features from different modes of diesel generator set operation, improving the accuracy of subsequent model training.

[0160] By using the standard state feature set and the noise spectrum feature set to train a preset regression vector machine model with random particles, global parameter search can be achieved, avoiding the trap of getting stuck in local optima. This improves the efficiency and robustness of the unit noise model. By calculating the real-time noise spectrum feature set corresponding to the real-time unit operation data of the target diesel unit based on the unit noise model, the real-time noise distribution of the diesel unit can be predicted. By performing sound source matching on the real-time noise spectrum feature set, the real-time noise source results can be obtained, and the noise magnitude and distribution of each noise component of the diesel unit can be determined. By performing noise control on the target diesel unit based on the real-time noise source results, targeted noise reduction of the diesel unit can be achieved, thereby improving the efficiency of diesel unit noise control. Therefore, the diesel unit noise control method based on support vector machines proposed in this invention can solve the problem of low efficiency in noise control.

[0161] like Figure 4 The diagram shown is a functional block diagram of a noise control system for a diesel generator set based on a support vector machine, provided in an embodiment of the present invention.

[0162] The noise control system 100 for a diesel generator set based on support vector machines described in this invention can be installed in an electronic device. Depending on the functions implemented, the noise control system 100 for the diesel generator set based on support vector machines may include a data splitting module 101, a frequency domain filtering module 102, a feature filtering module 103, a particle training module 104, and a noise control module 105. The module described in this invention can also be called a unit, which refers to a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, and which are stored in the memory of the electronic device.

[0163] In this embodiment, the functions of each module / unit are as follows:

[0164] The data splitting module 101 is used to sequentially split and clean the pre-acquired historical diesel generator set dataset to obtain the generator set noise dataset and the generator set operation dataset.

[0165] The frequency domain filtering module 102 is used to perform modal decomposition on the unit noise dataset to obtain a noise modal feature set, and to perform frequency domain filtering on the noise modal feature set to obtain a noise spectrum feature set. The step of performing frequency domain filtering on the noise modal feature set to obtain the noise spectrum feature set includes: selecting noise modal feature sets one by one from the noise modal feature set as target noise modal feature sets, and calculating the modal spectrum coefficient set corresponding to the target noise modal feature set using the following modal spectrum coefficient algorithm:

[0166]

[0167] Among them, w j (a, b) refers to the coefficient value of the j-th modal spectral coefficient in the modal spectral coefficient group under the scale parameter a and displacement parameter b, where a is the scale parameter, b is the displacement parameter, j is the index, and ∞ represents infinity. j (t) refers to the feature value of the j-th target noise mode feature in the target noise mode feature group at time number t, where t is the time number, z() is the conjugate function symbol, e is the Euler number, i is the imaginary symbol, ω0 is the preset frequency parameter, and d is the integral symbol; spectral features are extracted from the modal spectral coefficient group to obtain the primary spectral feature group; cross-relation filtering is performed on the primary spectral feature group to obtain the noise spectral feature group, and all the noise spectral feature groups are collected into a noise spectral feature group set;

[0168] The feature filtering module 103 is used to sequentially perform multimodal feature extraction and feature dimension filtering operations on the unit operation dataset to obtain a standard state feature set.

[0169] The particle training module 104 is used to perform random particle training on a preset regression vector machine model using the standard state feature set and the noise spectrum feature set to obtain a unit noise model.

[0170] The noise control module 105 is used to calculate the real-time noise spectrum feature group corresponding to the real-time unit operation data of the target diesel unit according to the unit noise model, perform sound source matching on the real-time noise spectrum feature group to obtain the real-time noise source result, and perform noise control on the target diesel unit according to the real-time noise source result.

[0171] In detail, the modules in the noise control system 100 for diesel generator sets based on support vector machines described in this embodiment of the invention employ the same methods as described above. Figures 1 to 3 The noise control method for diesel generator sets based on support vector machines described in the article uses the same technical means and can produce the same technical effect, so it will not be repeated here.

[0172] In the several embodiments provided by this invention, it should be understood that the disclosed devices, systems, and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0173] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0174] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0175] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0176] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) refers to the theories, methods, technologies, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0177] Furthermore, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or systems described in the system embodiments may also be implemented by a single unit or system through software or hardware. Terms such as "first," "second," etc., are used to indicate names and do not indicate any specific order.

[0178] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A noise control method for a diesel generator set based on support vector machines, characterized in that, The method includes: The previously acquired historical diesel generator set dataset was sequentially split and cleaned to obtain the generator set noise dataset and the generator set operation dataset. Modal decomposition is performed on the unit noise dataset to obtain a noise mode feature set. Frequency domain filtering is then performed on the noise mode feature set to obtain a noise spectrum feature set. Specifically, the frequency domain filtering of the noise mode feature set to obtain the noise spectrum feature set includes: selecting noise mode feature sets one by one from the noise mode feature set as target noise mode feature sets, and calculating the mode spectrum coefficients corresponding to the target noise mode feature sets using the following modal spectrum coefficient algorithm: ， in, This refers to the coefficient value of the j-th modal spectral coefficient in the modal spectral coefficient group under the scale parameter a and the displacement parameter b, where a is the scale parameter, b is the displacement parameter, and j is the index. The symbol for infinity. This refers to the feature value of the j-th target noise mode feature in the target noise mode feature group at time number t, where t is the time number. It is the symbol for the conjugate function, e is the Euler number, and i is the symbol for the imaginary number. Here, d is a preset frequency parameter, and d is the integral sign; spectral features are extracted from the modal spectral coefficient group to obtain a primary spectral feature group; cross-relation filtering is performed on the primary spectral feature group to obtain a noise spectral feature group, and all the noise spectral feature groups are aggregated into a noise spectral feature group set; The unit operation dataset is sequentially subjected to multimodal feature extraction and feature dimension filtering operations to obtain a standard state feature set, including: The speed feature set, load feature set, temperature feature set, and pressure feature set are extracted from the unit operation dataset, respectively. The speed feature set, the load feature set, the temperature feature set, and the pressure feature set are combined into a primary state feature set; Perform global feature normalization on the primary state feature set to obtain the secondary state feature set. Principal component screening is performed on the secondary state feature set to obtain the standard state feature set. The pre-defined regression vector machine model is trained with random particles using the standard state feature set and the noise spectrum feature set to obtain the unit noise model. The real-time noise spectrum feature group corresponding to the real-time unit operation data of the target diesel unit is calculated based on the unit noise model. The real-time noise spectrum feature group is then matched with the source to obtain the real-time noise source result. The target diesel unit is then subjected to noise control based on the real-time noise source result.

2. The noise control method for diesel generator sets based on support vector machines as described in claim 1, characterized in that, The process involves sequentially splitting and cleaning the pre-acquired historical diesel generator set dataset to obtain a generator set noise dataset and a generator set operation dataset, including: The historical diesel generator set dataset is hash-mapped to obtain a diesel hash signature set; The intersection-union similarity of the diesel hash signature set is calculated to obtain the data similarity results; Based on the data similarity results, the historical diesel generator set dataset is deduplicated to obtain a deduplicated diesel generator set dataset. Data attribute groups are extracted from the de-dust diesel engine group dataset, and missing data matching is performed on the de-dust diesel engine group dataset according to the data attribute groups to obtain the standard diesel engine group dataset. The standard diesel generator set dataset is split into a generator set noise dataset and a generator set operation dataset based on the data attribute group.

3. The noise control method for diesel generator sets based on support vector machines as described in claim 2, characterized in that, The step of performing intersection-union similarity calculation on the diesel hash signature set to obtain data similarity results includes: One by one, select the diesel hash signatures in the diesel hash signature set as the target diesel hash signature, and one by one select the diesel hash signatures in the diesel hash signature set other than the target diesel hash signature as the matching diesel hash signature; The intersection of the target diesel hash signature and the matching diesel hash signature is calculated to obtain the target signature intersection value. The union of the target diesel hash signature and the matching diesel hash signature is calculated to obtain the union value of the target signature; The ratio of the intersection value of the target signatures to the union value of the target signatures is used as the target similarity, and all target similarities are aggregated into a data similarity result.

4. The noise control method for diesel generator sets based on support vector machines as described in claim 1, characterized in that, The modal decomposition of the unit noise dataset to obtain a noise modal feature set includes: Each unit noise data point in the unit noise dataset is selected as the target unit noise data, and the maximum noise line and the minimum noise line are extracted from the target unit noise data respectively. The average noise line is calculated based on the target unit noise data, the maximum noise line, and the minimum noise line. The target unit noise data is filtered for detail using the average noise line to obtain primary detail noise data. The primary detailed noise data is iteratively filtered for extreme value features to obtain noise mode feature groups, and all noise mode feature groups are aggregated into a noise mode feature set.

5. The noise control method for diesel generator sets based on support vector machines as described in claim 4, characterized in that, The iterative extreme value feature filtering of the primary detail noise data to obtain a noise mode feature group includes: Determine whether the primary detail noise data satisfies the preset extreme value characteristic conditions; If not, the primary detail noise data is used as the target unit noise data, and the steps of extracting the maximum noise line and the minimum noise line from the target unit noise data are returned. If so, the primary detail noise data is added as a noise modal feature to a preset primary feature group, and it is determined whether the number of features in the primary feature group is equal to a preset number threshold. If not, then use the primary detail noise to perform detail filtering on the target unit noise data to obtain the target filtered unit noise, and return to the step of extracting the maximum noise line and the minimum noise line from the target unit noise data respectively. If so, the primary feature set is taken as the noise mode feature set.

6. The noise control method for diesel generator sets based on support vector machines as described in claim 1, characterized in that, The step of performing cross-relation filtering on the primary spectral feature group to obtain the noise spectral feature group includes: Each primary spectral feature in the primary spectral feature group is paired up to obtain a spectral feature pair group; Each spectral feature pair in the aforementioned spectral feature pair group is selected as the target spectral feature pair, and the feature information value of the target spectral feature pair is calculated using the following feature information algorithm: ， Where M(X,Y) refers to the feature information value of the target spectral feature pair composed of spectral feature X and spectral feature Y, X and Y are the spectral features of the target spectral feature pair, x is any possible value of spectral feature X, y is any possible value of spectral feature Y, and p(x,y) is the joint probability that spectral feature X is x and spectral feature Y is y. Let p(x) be a logarithmic function, where p(x) is the marginal probability corresponding to the spectral feature X taking the value x, and p(y) is the marginal probability corresponding to the spectral feature Y taking the value y. All feature information values ​​are arranged in descending order to form a feature information value sequence. The feature information value sequence is then filtered by a position threshold to obtain a feature information value group. The target spectral feature pairs corresponding to the feature information value groups are aggregated into standard spectral feature pairs, and all spectral features in the standard spectral feature pairs are aggregated into noise spectral feature groups.

7. The noise control method for diesel generator sets based on support vector machines as described in claim 1, characterized in that, The step of using the standard state feature set and the noise spectrum feature set to perform random particle training on a preset regression vector machine model to obtain a unit noise model includes: An initial set of particle parameters is randomly generated, and the initial particle parameters in the initial set are selected one by one to update the parameters of the preset regression vector machine model, so as to obtain the updated vector machine model. The regression spectral feature set corresponding to the standard state feature set is calculated using the updated vector machine model. The performance index of the updated vector machine model is calculated based on the regression spectral feature set and the noise spectral feature set. The initial particle parameters in the initial particle parameter set are iteratively updated according to the performance index to obtain the standard particle parameters. The parameters of the regression vector machine model are updated using the standard particle parameters to obtain the unit noise model.

8. The noise control method for diesel generator sets based on support vector machines as described in claim 1, characterized in that, The step of performing sound source matching on the real-time noise spectrum feature group to obtain real-time noise source results includes: Each real-time noise spectrum feature in the real-time noise spectrum feature group is selected as a target real-time spectrum feature, and the target real-time spectrum feature is regularized to obtain a regularized real-time spectrum feature. Edge decision-making is performed on the regularized real-time spectral features to obtain real-time decision features; The real-time decision features are normalized to obtain the real-time sound source code; The real-time spectral features of the target are labeled using the real-time sound source coding to obtain real-time noise sources, and all real-time noise sources are aggregated into a real-time noise source result.

9. A noise control system for a diesel generator set based on support vector machines, characterized in that, The system includes: The data splitting module is used to sequentially split and clean the pre-acquired historical diesel generator set dataset to obtain the generator set noise dataset and the generator set operation dataset. The frequency domain filtering module is used to perform mode decomposition on the unit noise dataset to obtain a noise mode feature set, and to perform frequency domain filtering on the noise mode feature set to obtain a noise spectrum feature set. The step of performing frequency domain filtering on the noise mode feature set to obtain the noise spectrum feature set includes: selecting noise mode feature sets one by one from the noise mode feature set as target noise mode feature sets, and calculating the mode spectrum coefficient set corresponding to the target noise mode feature set using the following mode spectrum coefficient algorithm: , in, This refers to the coefficient value of the j-th modal spectral coefficient in the modal spectral coefficient group under the scale parameter a and the displacement parameter b, where a is the scale parameter, b is the displacement parameter, and j is the index. The symbol for infinity. This refers to the feature value of the j-th target noise mode feature in the target noise mode feature group at time number t, where t is the time number. It is the symbol for the conjugate function, e is the Euler number, and i is the symbol for the imaginary number. Here, d is a preset frequency parameter, and d is the integral sign; spectral features are extracted from the modal spectral coefficient group to obtain a primary spectral feature group; cross-relation filtering is performed on the primary spectral feature group to obtain a noise spectral feature group, and all the noise spectral feature groups are aggregated into a noise spectral feature group set; The feature filtering module is used to sequentially perform multimodal feature extraction and feature dimension filtering operations on the unit operation dataset to obtain a standard state feature set; including: The speed feature set, load feature set, temperature feature set, and pressure feature set are extracted from the unit operation dataset, respectively. The speed feature set, the load feature set, the temperature feature set, and the pressure feature set are combined into a primary state feature set; Perform global feature normalization on the primary state feature set to obtain the secondary state feature set. Principal component screening is performed on the secondary state feature set to obtain the standard state feature set. The particle training module is used to perform random particle training on the preset regression vector machine model using the standard state feature set and the noise spectrum feature set to obtain the unit noise model. The noise control module is used to calculate the real-time noise spectrum feature group corresponding to the real-time unit operation data of the target diesel unit according to the unit noise model, perform sound source matching on the real-time noise spectrum feature group to obtain the real-time noise source result, and perform noise control on the target diesel unit according to the real-time noise source result.

Images