A container silent power station and noise control method thereof

Abstract

The invention discloses a container silent power station and a noise control method thereof, comprising a container body, a sound-absorbing floor fixedly installed at the bottom end of the inner wall of the container body, a monitoring cabinet fixedly installed at one side of the top of the sound-absorbing floor, a first muffler fixedly installed at one side of the monitoring cabinet, an air-conditioning body fixedly installed at the top of the inner wall of the container body, and a second muffler fixedly installed at one side of the air-conditioning body. The invention arranges a sound-absorbing floor and a mute strip, wherein the sound-absorbing floor absorbs the noise generated by the power station equipment in the container from the bottom of the container, the mute strip absorbs the noise generated by the power station equipment in the container from the top of the container, and a decibel monitor monitors the decibel in the container in real time to avoid excessive noise in the container power station; and by arranging a plurality of mufflers, if the decibel value monitored by the decibel monitor is too large, the muffler installed beside the running equipment is started to weaken and reduce the sound.

Description

Technical Field

[0001] The present invention relates to the technical field of container power stations, and in particular to a container silent power station and a noise control method thereof. Background Art

[0002] With the popularity of electric vehicles, charging infrastructure has become an important part of promoting the development of electric vehicles. Among the many charging methods, container power stations are gradually emerging as an innovative charging solution. The improvement of their layout has important strategic significance. First of all, container power stations have a wider range of applicability. Traditional power stations require stable power supply from the power grid, but in some remote areas or places with inconvenient transportation, the problem of insufficient power supply will become a problem in the construction of charging stations. Container power stations adopt a modular design and can be built at any location. The layout can be adjusted at any time according to demand, so it can better meet the charging needs of different regions and different time periods. Secondly, container power stations help to improve charging efficiency. Traditional charging methods require a long waiting time for charging, while power stations can achieve fast battery replacement, greatly shortening the charging time. The improvement of the layout of container power stations means that more power stations will enter the market. Electric car owners can complete battery replacement more conveniently and quickly during travel, which improves charging efficiency and enhances the user experience of electric vehicles.

[0003] However, traditional container power plants have the following disadvantages:

[0004] In traditional container power plants, multiple devices operate in a closed environment. Since the container is not a soundproof structure, this leads to excessive noise in the container power plant, causing immeasurable damage to the user's body. Summary of the invention

[0005] The purpose of the present invention is to provide a container silent power station and a noise control method thereof, so as to solve the problem proposed in the above background technology that multiple devices in a traditional container power station operate in a closed environment. Since the container is a non-soundproof structure, this leads to excessive noise in the container power station, causing immeasurable damage to the user's body.

[0006] To achieve the above-mentioned objectives, the present invention provides the following technical solutions: a container silent power station, comprising a container body, a sound-absorbing floor is fixedly installed at the bottom end of the inner wall of the container body, a monitoring cabinet is fixedly installed on one side of the top of the sound-absorbing floor, a first muffler is fixedly installed on one side of the monitoring cabinet, an air-conditioning body is fixedly installed on the top of the inner wall of the container body, a second muffler is fixedly installed on one side of the air-conditioning body, a battery cabinet located on one side of the monitoring cabinet is fixedly installed on the sound-absorbing floor, a third muffler is fixedly installed on one side of the battery cabinet, a transformer is fixedly installed on the other side of the top of the sound-absorbing floor, a fourth muffler is fixedly installed on one side of the transformer, a current transformer located on one side of the transformer is fixedly installed on the sound-absorbing floor, a fifth muffler is fixedly installed on one side of the current transformer, and a silent strip is fixedly installed on the top of the container body.

[0007] As a preferred technical solution of the present invention, the sound-absorbing floor includes a cellulose board, an organic synthetic board, a metal board, a composite board, a particle board, a porous board and a substrate, the bottom end of the cellulose board is fixedly connected to the top end of the organic synthetic board, the bottom end of the organic synthetic board is fixedly connected to the top end of the metal plate, the bottom end of the metal plate is fixedly connected to the top end of the composite board, the bottom end of the composite board is fixedly connected to the top end of the particle board, the bottom end of the particle board is fixedly connected to the top end of the porous board, and the bottom end of the porous plate is fixedly connected to the top end of the substrate.

[0008] As a preferred technical solution of the present invention, the cellulose board is made of gypsum board material, the organic synthetic board is made of polyurethane fiber sound-absorbing board material, the metal plate is made of metal foam material, the composite board is made of wood-plastic sound-absorbing board material, the particle board is made of perlite sound-absorbing board material, the porous board is made of open-pore rubber board material, and the substrate is made of solid wood board material.

[0009] As a preferred technical solution of the present invention, several of the noise-canceling strips include a first noise-canceling pad, a second noise-canceling pad, a third noise-canceling pad and a fourth noise-canceling pad, the bottom end of the first noise-canceling pad is fixedly connected to the top end of the second noise-canceling pad, the bottom end of the second noise-canceling pad is fixedly connected to the top end of the third noise-canceling pad, and the bottom end of the third noise-canceling pad is fixedly connected to the top end of the fourth noise-canceling pad.

[0010] As a preferred technical solution of the present invention, the first sound-absorbing pad is made of mineral wool board material, the second sound-absorbing pad is made of sound-insulating foam material, the third sound-absorbing pad is made of sound-insulating cotton material, and the fourth sound-absorbing pad is made of sound-insulating felt material. The first sound-absorbing pad achieves sound absorption effect through the high-density fiber structure and pore shape of the mineral wool board, the second sound-absorbing pad achieves sound absorption effect through the porous structure and surface shape of the foam, the third sound-absorbing pad achieves sound absorption effect through the high density of the fiber and the size and shape of the pores, and the fourth sound-absorbing pad absorbs sound through the high density of the fiber and the friction in the air.

[0011] As a preferred technical solution of the present invention, the four corners at the top of the container body and the four corners at the bottom of the container body are fixedly installed with lifting blocks, and the middle parts of the eight lifting blocks are provided with lifting holes. When transporting the container body, the hook on the lifting equipment is hooked in the lifting hole, and the container body is lifted and transported by the lifting blocks.

[0012] As a preferred technical solution of the present invention, sealing doors are hinged at both ends of one side of the container body, direction shells are fixedly installed at both ends of one side of the two sealing doors, a pull rod is rotatably connected between every two direction shells facing the two pull rods, an assembly block is fixedly installed in the middle of the two pull rods, a handle is fixedly installed on one side of the two assembly blocks, and limit blocks are fixedly installed at both ends of the two pull rods. The user rotates the pull rod along the direction shell, and the pull rod drives the assembly block to move synchronously to adjust the direction of the handle, and the handle drives the sealing door to unfold along the container body.

[0013] As a preferred technical solution of the present invention, a plurality of drainage grooves are provided on both sides of the bottom of the container body, a plurality of heat dissipation grooves are provided on both sides of the container body, a decibel monitor is fixedly installed on the surface of the container body, and the decibel monitor monitors the decibel in the container body in real time. The opening of the drainage grooves facilitates the flow of water at the bottom of the container body.

[0014] The present invention provides a method for controlling noise of a container silent power station, comprising the following steps:

[0015] Step 1: Decibel monitoring: The decibel monitor monitors the decibel level in the container in real time. If the decibel level is not high, the sound-absorbing floor and the mute strip continue to absorb the decibel level in the container.

[0016] Step 2: Determine the running equipment: If the monitored decibel is too high, determine the equipment running in the container body, specifically whether the monitoring cabinet is running, whether the air conditioner body is running, whether the battery cabinet is running, whether the transformer is running, and whether the converter is running;

[0017] Step 3, start the silencer: start the silencer on the running equipment to eliminate and absorb the noise generated by the running equipment. If the monitoring cabinet is running, start the first silencer; if the air conditioner body is running, start the second silencer; if the battery cabinet is running, start the third silencer; if the transformer is running, start the fourth silencer; if the converter is running, start the fifth silencer.

[0018] Furthermore, in steps 1 to 3, the silencing intensity of the silencer is controlled by an intelligent control algorithm, and the specific steps are as follows;

[0019] Step 1: Data collection and preprocessing;

[0020] Sound sensor: collects sound data inside and around the container power station, including equipment operation sound and environmental noise; uses spectral subtraction to remove background noise to improve the quality of sound data;

[0021] S(w,t)=X(w,t)-λ·N(w)

[0022] Where S(w,t) represents the amplitude of the noisy sound signal at time t and frequency w; N(w) represents the estimated noise amplitude at frequency w; λ represents an over-reduction factor used to adjust the degree of noise reduction;

[0023] Step 2: feature processing;

[0024] Convert the processed sound signal into Mel spectrum, which is an input format that can be processed by convolutional neural network (CNN);

[0025]

[0026] Among them, MFCC(t) is the Mel frequency cepstral coefficient at time t, S k (t) is the power spectrum at time t and frequency band k, K represents the number of filter banks, is the sum of all frequency bands k from 1 to K, log(S k (t)) is the power spectrum S k (t) takes the logarithm, is the cosine function, where i is the index of the cepstral coefficient; the CNN is set up as follows:

[0027] Convolutional Layer:

[0028] The first convolution layer: 64 3x3 convolution kernels, stride 1, using ReLU activation function;

[0029] The second convolution layer: 128 3x3 convolution kernels, stride 1, using ReLU activation function;

[0030] The third convolution layer: 256 3x3 convolution kernels, stride 1, using ReLU activation function;

[0031] Pooling layer:

[0032] Each convolutional layer is followed by a 2x2 max pooling layer to reduce the size of the feature map;

[0033] For CNN, the operation of each layer can be expressed as:

[0034] Z l+1 =f(W l *X l +b l )

[0035] Among them, X l The input of layer l, W l The weight matrix (convolution kernel) of the lth layer, b l The bias of the lth layer, * convolution operation, f activation function, Z l+1 The output of the l+1th layer, on the feature map of the lth layer input, performs the maximum pooling operation on each pooling window to obtain the output of the l+1th layer;

[0036]

[0037] Among them, P l+1 The pooled output of layer l+1, Perform the maximum value operation on k and l, Z l [k,l] The value of the l-th layer input at position [k,l], k and l are the two-dimensional position indexes within the pooling window, indicating the row and column positions within the local area of ​​the feature map;

[0038] Step 3: Data distribution optimization;

[0039] The domain adaptation network DAN is used for domain adaptation. This type of network optimizes the generalization ability of the model by minimizing the distribution difference between the source domain and the target domain.

[0040]

[0041] Among them, MMD 2 (D S ,D T ) represents the maximum mean difference between the data distribution of the source domain and the data distribution of the target domain; D S Data distribution in the source domain, D T The data distribution of the target domain, n is the number of source domain samples, m is the number of target domain samples, For all samples in the source domain To sum, For all samples in the target domain Sum, φ(·) kernel function, maps the input sample to a high-dimensional feature space, The i-th sample in the source domain, The tth sample in the target domain, ||·|| 2 The square norm of a vector, used to measure the Euclidean distance between vectors;

[0042] Step 4: Transformer processing;

[0043] After CNN feature extraction, the features are input into the Transformer model; this can fully utilize the advantages of Transformer in processing long sequence data and improve the recognition accuracy of sound events;

[0044]

[0045] Among them, Attention(Q,K,V) represents the output of the attention mechanism, Q, K, V represent the query, key, and value matrices, which represent the query vector, key vector, and value vector of the current time step respectively, d k is the dimension of the key vector, used to scale the dot product to prevent excessive internal layer activation values, softmax, normalization function, to ensure that the sum of the output weights is 1, QK T The query matrix Q and the transposed matrix K of the key matrix K T The dot product of , which represents the similarity between the query vector and the key vector, Apply the Softmax function to the scaled dot product result to get the attention weight, Apply the attention weights to the value matrix V to get the final output of the attention mechanism;

[0046] In order to pay attention to the frequency changes and temporal dynamic characteristics of the sound signal at the same time, the multi-head attention structure of the Transformer model is introduced;

[0047] MultiHead(Q,K,V)=Concat(head 1 ,...,head h )W O

[0048] head i =Attention(QW i Q ,KW i K ,VW i V )

[0049] Among them, the output of MultiHead (Q, K, V) multi-head attention mechanism, head i The i-th attention head, the number of head attention heads, W i Q , W i K , W i V The linear transformation matrix of the query, key, and value of the i-th head, W O The linear transformation matrix of all head outputs, the Concat operation, is used to merge the outputs of each head, and the Attention (QW i Q ,KW i K ,VW i V ) is the attention mechanism output of the i-th attention head;

[0050] The trained model is integrated into the central control system of the container power station to process sound data in real time, and the noise control strategy and the working status of the silencer are adjusted according to the output of the model.

[0051] Compared with the prior art, the present invention has the following beneficial effects:

[0052] 1. By setting up sound-absorbing floors and silent strips, the sound-absorbing floors absorb the noise generated by the power station equipment in the container from the bottom of the container, and the silent strips absorb the noise generated by the power station equipment in the container from the top of the container. The decibel monitor monitors the decibel in the container in real time to avoid excessive noise in the container power station;

[0053] 2. By setting up multiple silencers, if the decibel value detected by the decibel monitor is too large, the silencer installed next to the running equipment will be activated to weaken and reduce the sound.

[0054] 3. By using intelligent control algorithms, noise levels can be monitored and precisely adjusted in real time to suit different working environments and needs. For example, when abnormal noise is detected, the system can automatically adjust the output of the silencer to reduce unnecessary background noise, thereby effectively protecting the hearing of workers and improving the comfort of the working environment. Intelligent control systems are able to optimize the operating status of equipment based on real-time data and avoid excessive use of equipment such as silencers, thereby saving energy and reducing operating costs. By reducing noise pollution in the working environment, intelligent control systems help comply with strict health and safety regulations and reduce the risk of occupational diseases caused by excessive noise. This is particularly important for ensuring the health of employees and avoiding legal liabilities arising from violations of regulations. Reducing the noise impact of power stations on the surrounding environment helps to enhance the corporate image and improve relations with the community. This is especially important for power stations located in or close to residential areas, helping to reduce resident complaints and improve corporate social responsibility. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 is a cross-sectional view of the present invention;

[0056] Figure 2 It is a three-dimensional diagram of the container body of the present invention;

[0057] Figure 3 It is a side view of the container body of the present invention;

[0058] Figure 4 It is a structural schematic diagram of the sound-absorbing floor of the present invention;

[0059] Figure 5 It is a schematic diagram of the structure of the mute strip of the present invention;

[0060] Figure 6 is a flow chart of the present invention;

[0061] Figure 7 This is a muffler control flow chart of the present invention.

[0062] In the figure: 1. container body; 2. drainage trough; 3. heat dissipation trough; 4. sealing door; 5. pull rod; 6. direction shell; 7. limit block; 8. assembly block; 9. handle; 10. lifting hole; 11. lifting block; 12. mute strip; 121. first mute pad; 122. second mute pad; 123. third mute pad; 124. fourth mute pad; 13. monitoring cabinet; 14. first muffler; 15. air conditioner body; 16. second muffler; 17. battery cabinet; 18. third muffler; 19. sound-absorbing floor; 191. cellulose board; 192. organic synthesis board; 193. metal plate; 194. composite board; 195. particle board; 196. porous plate; 197. substrate; 20. transformer; 21. fourth muffler; 22. converter; 23. fifth muffler; 24. decibel monitor. DETAILED DESCRIPTION

[0063] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0064] See also Figure 1-6 The present invention provides a container silent power station, comprising a container body 1, a sound-absorbing floor 19 is fixedly installed at the bottom end of the inner wall of the container body 1, a monitoring cabinet 13 is fixedly installed on one side of the top of the sound-absorbing floor 19, a first muffler 14 is fixedly installed on one side of the monitoring cabinet 13, an air-conditioning body 15 is fixedly installed on the top of the inner wall of the container body 1, a second muffler 16 is fixedly installed on one side of the air-conditioning body 15, a battery cabinet 17 located on one side of the monitoring cabinet 13 is fixedly installed on the sound-absorbing floor 19, a third muffler 18 is fixedly installed on one side of the battery cabinet 17, a transformer 20 is fixedly installed on the other side of the top of the sound-absorbing floor 19, a fourth muffler 21 is fixedly installed on one side of the transformer 20, a current transformer 22 located on one side of the transformer 20 is fixedly installed on the sound-absorbing floor 19, a fifth muffler 23 is fixedly installed on one side of the current transformer 22, and a silent strip 12 is fixedly installed on the top of the container body 1.

[0065] The sound-absorbing floor 19 includes a cellulose board 191, an organic synthetic board 192, a metal board 193, a composite board 194, a particle board 195, a porous board 196 and a substrate 197. The bottom end of the cellulose board 191 is fixedly connected to the top end of the organic synthetic board 192, the bottom end of the organic synthetic board 192 is fixedly connected to the top end of the metal plate 193, the bottom end of the metal plate 193 is fixedly connected to the top end of the composite board 194, the bottom end of the composite board 194 is fixedly connected to the top end of the particle board 195, the bottom end of the particle board 195 is fixedly connected to the top end of the porous board 196, and the bottom end of the porous board 196 is fixedly connected to the top end of the substrate 197.

[0066] The cellulose board 191 is made of gypsum board material, the organic synthetic board 192 is made of polyurethane fiber sound-absorbing board material, the metal board 193 is made of metal foam material, the composite board 194 is made of wood-plastic sound-absorbing board material, the particle board 195 is made of perlite sound-absorbing board material, the porous board 196 is made of open-pore rubber board material, and the base plate 197 is made of solid wood board material.

[0067] Several silent strips 12 include a first silent pad 121, a second silent pad 122, a third silent pad 123 and a fourth silent pad 124. The bottom end of the first silent pad 121 is fixedly connected to the top end of the second silent pad 122, the bottom end of the second silent pad 122 is fixedly connected to the top end of the third silent pad 123, and the bottom end of the third silent pad 123 is fixedly connected to the top end of the fourth silent pad 124.

[0068] The first sound-absorbing pad 121 is made of mineral wool board material, the second sound-absorbing pad 122 is made of sound-insulating foam material, the third sound-absorbing pad 123 is made of sound-insulating cotton material, and the fourth sound-absorbing pad 124 is made of sound-insulating felt material. The first sound-absorbing pad 121 achieves sound absorption effect through the high-density fiber structure and pore shape of the mineral wool board, the second sound-absorbing pad 122 achieves sound absorption effect through the porous structure and surface shape of the foam, the third sound-absorbing pad 123 achieves sound absorption effect through the high density of fibers and the size and shape of pores, and the fourth sound-absorbing pad 124 absorbs sound through the high density of fibers and friction in the air.

[0069] Lifting blocks 11 are fixedly installed on the four corners at the top and the four corners at the bottom of the container body 1, and lifting holes 10 are opened in the middle of the eight lifting blocks 11. When transporting the container body 1, the hook on the lifting equipment is hooked in the lifting hole 10, and the container body 1 is lifted and transported through the lifting blocks 11.

[0070] Sealed doors 4 are hinged at both ends of one side of the container body 1, and direction shells 6 are fixedly installed at both ends of one side of the two sealed doors 4. A pull rod 5 is rotatably connected between every two opposite direction shells 6, and an assembly block 8 is fixedly installed in the middle of the two pull rods 5. A handle 9 is fixedly installed on one side of the two assembly blocks 8. Limit blocks 7 are fixedly installed at both ends of the two pull rods 5. The user rotates the pull rod 5 along the direction shell 6, and the pull rod 5 drives the assembly block 8 to move synchronously to adjust the direction of the handle 9, and the handle 9 drives the sealed door 4 to unfold along the container body 1.

[0071] A plurality of drainage grooves 2 are provided on both sides of the bottom of the container body 1, a plurality of heat dissipation grooves 3 are provided on both sides of the container body 1, a decibel monitor 24 is fixedly installed on the surface of the container body 1, and the decibel monitor 24 monitors the decibel in the container body 1 in real time. The opening of the drainage grooves 2 facilitates the flow of water at the bottom of the container body 1.

[0072] The present invention provides a method for controlling noise of a container silent power station, comprising the following steps:

[0073] Step 1, decibel monitoring: the decibel monitor 24 monitors the decibel level in the container body 1 in real time. If the decibel level is not high, the sound-absorbing floor 19 and the mute strip 12 continue to absorb the decibel level in the container body 1.

[0074] Step 2, determine the running equipment: if the monitored decibel is too high, determine the equipment running in the container body 1, specifically determine whether the monitoring cabinet 13 is running, whether the air conditioner body 15 is running, whether the battery cabinet 17 is running, whether the transformer 20 is running, and whether the converter 22 is running;

[0075] Step three, start the silencer: start the silencer on the running equipment to eliminate and absorb the noise generated by the running equipment. If the monitoring cabinet 13 is running, start the first silencer 14; if the air conditioner body 15 is running, start the second silencer 16; if the battery cabinet 17 is running, start the third silencer 18; if the transformer 20 is running, start the fourth silencer 21; if the converter 22 is running, start the fifth silencer 23.

[0076] In the present invention, the decibel monitor 24 monitors the decibel in the container body 1 in real time. If the decibel is not high, the sound-absorbing floor 19 and the silent strip 12 continue to absorb the decibel in the container body 1; if the monitored decibel is too high, the equipment running in the container body 1 is determined, specifically, whether the monitoring cabinet 13 is running, whether the air-conditioning body 15 is running, whether the battery cabinet 17 is running, whether the transformer 20 is running, and whether the converter 22 is running; the mufflers on the running equipment are started to eliminate and absorb the noise generated by the running equipment. If the monitoring cabinet 13 is running, the first muffler 14 is started; if the air-conditioning body 15 is running, the second muffler 16 is started; if the battery cabinet 17 is running, the third muffler 18 is started; if the transformer 20 is running, the fourth muffler 21 is started; if the converter 22 is running, the fifth muffler 23 is started.

[0077] Furthermore, in steps 1 to 3, the silencing intensity of the silencer is controlled by an intelligent control algorithm, and the specific steps are as follows;

[0078] Step 1: Data collection and preprocessing;

[0079] Sound sensor: collects sound data inside and around the container power station, including equipment operation sound and environmental noise; uses spectral subtraction to remove background noise to improve the quality of sound data;

[0080] S(w,t)=X(w,t)-λ·N(w)

[0081] Where S(w,t) represents the amplitude of the noisy sound signal at time t and frequency w; N(w) represents the estimated noise amplitude at frequency w; λ represents an over-reduction factor used to adjust the degree of noise reduction;

[0082] Step 2: feature processing;

[0083] Convert the processed sound signal into Mel spectrum, which is an input format that can be processed by convolutional neural network (CNN);

[0084]

[0085] Among them, MFCC(t) is the Mel frequency cepstral coefficient at time t, S k (t) is the power spectrum at time t and frequency band k, K represents the number of filter banks, is the sum of all frequency bands k from 1 to K, log(S k (t)) is the power spectrum S k (t) takes the logarithm, is the cosine function, where i is the index of the cepstral coefficient; the CNN is set up as follows:

[0086] Convolutional Layer:

[0087] The first convolution layer: 64 3x3 convolution kernels, stride 1, using ReLU activation function;

[0088] The second convolution layer: 128 3x3 convolution kernels, stride 1, using ReLU activation function;

[0089] The third convolution layer: 256 3x3 convolution kernels, stride 1, using ReLU activation function;

[0090] Pooling layer:

[0091] Each convolutional layer is followed by a 2x2 max pooling layer to reduce the size of the feature map;

[0092] For CNN, the operation of each layer can be expressed as:

[0093] Z l+1 =f(W l *X l +b l )

[0094] Among them, X l The input of layer l, W l The weight matrix (convolution kernel) of the lth layer, b l The bias of the lth layer, * convolution operation, f activation function, Z l+1 The output of the l+1th layer, on the feature map of the lth layer input, performs the maximum pooling operation on each pooling window to obtain the output of the l+1th layer;

[0095]

[0096] Among them, P l+1 The pooled output of layer l+1, Perform the maximum value operation on k and l, Z l[k,l] The value of the l-th layer input at position [k,l], k and l are the two-dimensional position indexes within the pooling window, indicating the row and column positions within the local area of ​​the feature map;

[0097] Step 3: Data distribution optimization;

[0098] The domain adaptation network DAN is used for domain adaptation. This type of network optimizes the generalization ability of the model by minimizing the distribution difference between the source domain and the target domain.

[0099]

[0100] Among them, MMD 2 (D S ,D T ) represents the maximum mean difference between the data distribution of the source domain and the data distribution of the target domain; D S Data distribution in the source domain, D T The data distribution of the target domain, n is the number of source domain samples, m is the number of target domain samples, For all samples in the source domain To sum, For all samples in the target domain Sum, φ(·) kernel function, maps the input sample to a high-dimensional feature space, The i-th sample in the source domain, The tth sample in the target domain, ||·|| 2 The square norm of a vector, used to measure the Euclidean distance between vectors;

[0101] Step 4: Transformer processing;

[0102] After CNN feature extraction, the features are input into the Transformer model; this can fully utilize the advantages of Transformer in processing long sequence data and improve the recognition accuracy of sound events;

[0103]

[0104] Among them, Attention(Q,K,V) represents the output of the attention mechanism, Q, K, V represent the query, key, and value matrices, which represent the query vector, key vector, and value vector of the current time step respectively, d k is the dimension of the key vector, used to scale the dot product to prevent excessive internal layer activation values, softmax, normalization function, to ensure that the sum of the output weights is 1, QK T The query matrix Q and the transposed matrix K of the key matrix K T The dot product of , which represents the similarity between the query vector and the key vector, Apply the Softmax function to the scaled dot product result to get the attention weight, Apply the attention weights to the value matrix V to get the final output of the attention mechanism;

[0105] In order to pay attention to the frequency changes and temporal dynamic characteristics of the sound signal at the same time, the multi-head attention structure of the Transformer model is introduced;

[0106] MultiHead(Q,K,V)=Concat(head 1 ,...,head h )W O

[0107] head i =Attention(QW i Q ,KW i K ,VW i V )

[0108] Among them, the output of MultiHead (Q, K, V) multi-head attention mechanism, head i The i-th attention head, the number of head attention heads, W i Q , W i K , W i V The linear transformation matrix of the query, key, and value of the i-th head, W O The linear transformation matrix of all head outputs, the Concat operation, is used to merge the outputs of each head, and the Attention (QW i Q ,KW i K ,VW i V ) is the attention mechanism output of the i-th attention head;

[0109] The trained model is integrated into the central control system of the container power station to process sound data in real time, and the noise control strategy and the working status of the silencer are adjusted according to the output of the model.

[0110] Numerical case analysis;

[0111] Five sound sensors are deployed in the container power station. Each sensor collects 1,000 data points in one minute. Each data point represents the original waveform amplitude of the sound. The sound data of sensor 1 at t seconds is: X 1(t) = 0.5sin(2π·440·t) + 0.02sin(2π·1000·t) (contains noise);

[0112] The noise is mainly high-frequency components: N(w) = 0.02, the frequency is 120HZ, and the spectral subtraction method is used for preprocessing, the over-subtraction factor λ = 1.1, and the denoised signal S(w,t) is calculated;

[0113] A Mel filter bank with 40 filter groups was used to obtain the Mel spectrum at each time point;

[0114] CNN convolution and pooling operations: Layer 1 input X 1 , output Z 2 , the second layer input Z 2 , output Z 3 , the third layer input Z 3 , output Z 4 , each layer is followed by a max pooling layer;

[0115] Transformer processing: Input: Feature Z from CNN 4 ,Multi-head attention processing, uses 4 heads, output: the category of the sound event, "normal operation", "abnormal noise";

[0116] Noise reduction control:

[0117] At time point t=1s, the model outputs "abnormal noise" and the silencer strength increases by 20%.

[0118] At time t = 2s, the model output is "normal operation" and the muffler strength is reduced by 10%;

[0119] Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments, or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A container silent power station, comprising a container body (1), characterized in that: A sound-absorbing floor (19) is fixedly mounted on the bottom end of the inner wall of the container body (1), a monitoring cabinet (13) is fixedly mounted on one side of the top of the sound-absorbing floor (19), a first muffler (14) is fixedly mounted on one side of the monitoring cabinet (13), an air-conditioning body (15) is fixedly mounted on the top end of the inner wall of the container body (1), a second muffler (16) is fixedly mounted on one side of the air-conditioning body (15), and a battery cabinet (14) located on one side of the monitoring cabinet (13) is fixedly mounted on the sound-absorbing floor (19). 7), a third muffler (18) is fixedly mounted on one side of the battery cabinet (17), a transformer (20) is fixedly mounted on the other side of the top of the sound-absorbing floor (19), a fourth muffler (21) is fixedly mounted on one side of the transformer (20), a current transformer (22) located on one side of the transformer (20) is fixedly mounted on the sound-absorbing floor (19), a fifth muffler (23) is fixedly mounted on one side of the current transformer (22), and a mute strip (12) is fixedly mounted on the top of the container body (1); The sound-absorbing floor (19) comprises a cellulose board (191), an organic synthesis board (192), a metal board (193), a composite board (194), a particle board (195), a porous board (196) and a base board (197), wherein the bottom end of the cellulose board (191) is fixedly connected to the top end of the organic synthesis board (192), the bottom end of the organic synthesis board (192) is fixedly connected to the top end of the metal board (193), the bottom end of the metal board (193) is fixedly connected to the top end of the composite board (194), the bottom end of the composite board (194) is fixedly connected to the top end of the particle board (195), the bottom end of the particle board (195) is fixedly connected to the top end of the porous board (196), and the bottom end of the porous board (196) is fixedly connected to the top end of the base board (197); The cellulose board (191) is made of a gypsum board material, the organic synthetic board (192) is made of a polyurethane fiber sound-absorbing board material, the metal board (193) is made of a metal foam material, the composite board (194) is made of a wood-plastic sound-absorbing board material, the particle board (195) is made of a perlite sound-absorbing board material, the porous board (196) is made of an open-pore rubber board material, and the base plate (197) is made of a solid wood board material.

2. A container silent power station according to claim 1, characterized in that: Several of the silent strips (12) include a first silent pad (121), a second silent pad (122), a third silent pad (123) and a fourth silent pad (124); the bottom end of the first silent pad (121) is fixedly connected to the top end of the second silent pad (122); the bottom end of the second silent pad (122) is fixedly connected to the top end of the third silent pad (123); and the bottom end of the third silent pad (123) is fixedly connected to the top end of the fourth silent pad (124).

3. A container silent power station according to claim 2, characterized in that: The first soundproofing pad (121) is made of mineral wool board material, the second soundproofing pad (122) is made of sound insulation foam material, the third soundproofing pad (123) is made of sound insulation cotton material, and the fourth soundproofing pad (124) is made of sound insulation felt material.

4. A container silent power station according to claim 1, characterized in that: The four corners at the top of the container body (1) and the four corners at the bottom of the container body (1) are fixedly mounted with lifting blocks (11), and the middle parts of the eight lifting blocks (11) are each provided with a lifting hole (10).

5. The container silent power station according to claim 1, characterized in that: Both ends of one side of the container body (1) are hinged with sealed doors (4), both ends of one side of the two sealed doors (4) are fixedly installed with direction shells (6), between each two opposite direction shells (6) there is a pull rod (5) rotatably connected, the middle of the two pull rods (5) are fixedly installed with assembly blocks (8), one side of the two assembly blocks (8) are fixedly installed with handles (9), and both ends of the two pull rods (5) are fixedly installed with limit blocks (7).

6. The container silent power station according to claim 1, characterized in that: A plurality of drainage grooves (2) are provided on both sides of the bottom of the container body (1), a plurality of heat dissipation grooves (3) are provided on both sides of the container body (1), and a decibel monitor (24) is fixedly installed on the surface of the container body (1).

7. A method for controlling noise in a container silent power station according to any one of claims 1 to 6, characterized in that: The following steps are involved: Step 1, decibel monitoring: the decibel monitor (24) monitors the decibel level in the container body (1) in real time. If the decibel level is not high, the sound-absorbing floor (19) and the mute strip (12) continue to absorb the decibel level in the container body (1); Step 2, determining the running equipment: if the monitored decibel is too high, determine the running equipment in the container body (1), specifically, determine whether the monitoring cabinet (13) is running, whether the air conditioner body (15) is running, whether the battery cabinet (17) is running, whether the transformer (20) is running, and whether the converter (22) is running; Step 3, start the silencer: start the silencer on the running equipment to eliminate and absorb the noise generated by the running equipment. If the monitoring cabinet (13) is running, start the first silencer (14); if the air conditioner body (15) is running, start the second silencer (16); if the battery cabinet (17) is running, start the third silencer (18); if the transformer (20) is running, start the fourth silencer (21); if the converter (22) is running, start the fifth silencer (23); Step three also includes controlling the silencing intensity of the silencer through a control algorithm, and the specific steps are as follows; Step 1: Data collection and preprocessing; Sound sensor: collects sound data inside and around the container power station, including equipment operation sound and environmental noise; uses spectral subtraction to remove background noise to improve the quality of sound data; S(w,t)=X(w,t)-λ·N(w) Where S(w,t) represents the amplitude of the noisy sound signal at time t and frequency w; N(w) represents the estimated noise amplitude at frequency w; λ represents an over-reduction factor used to adjust the degree of noise reduction; Step 2: feature processing; Convert the processed sound signal into Mel spectrum, which is an input format that can be processed by convolutional neural network (CNN); Among them, MFCC(t) is the Mel frequency cepstral coefficient at time t, S k (t) is the power spectrum at time t and frequency band k, K represents the number of filter banks, is the sum of all frequency bands k from 1 to K, log(S k (t)) is the power spectrum S k (t) takes the logarithm, is the cosine function, where i is the index of the cepstral coefficient; the CNN is set up as follows: Convolutional Layer: The first convolution layer: 64 3x3 convolution kernels, stride 1, using ReLU activation function; The second convolution layer: 128 3x3 convolution kernels, stride 1, using ReLU activation function; The third convolution layer: 256 3x3 convolution kernels, stride 1, using ReLU activation function; Pooling layer: Each convolutional layer is followed by a 2x2 max pooling layer to reduce the size of the feature map; For CNN, the operation of each layer can be expressed as: Z l+1 =f(W l *X l +b l ) Among them, X l The input of layer l, W l The weight matrix (convolution kernel) of the lth layer, b l The bias of the lth layer, * convolution operation, f activation function, Z l+1 The output of the l+1th layer, on the feature map of the lth layer input, performs the maximum pooling operation on each pooling window to obtain the output of the l+1th layer; Among them, P l+1 The pooled output of layer l+1, Perform the maximum value operation on k and l, Z l [k,l] The value of the l-th layer input at position [k,l], k and l are the two-dimensional position indexes within the pooling window, indicating the row and column positions within the local area of ​​the feature map; Step 3: Data distribution optimization; The domain adaptation network DAN is used for domain adaptation. This type of network optimizes the generalization ability of the model by minimizing the distribution difference between the source domain and the target domain. Among them, MMD 2 (D S ,D T ) represents the maximum mean difference between the data distribution of the source domain and the data distribution of the target domain; D S Data distribution in the source domain, D T The data distribution of the target domain, n is the number of source domain samples, m is the number of target domain samples, For all samples in the source domain To sum, For all samples in the target domain Sum, φ(·) kernel function, maps the input sample to a high-dimensional feature space, The i-th sample in the source domain, The tth sample in the target domain, ||·|| 2 The square norm of a vector, used to measure the Euclidean distance between vectors; Step 4: Transformer processing; After CNN feature extraction, the features are input into the Transformer model; this can fully utilize the advantages of Transformer in processing long sequence data and improve the recognition accuracy of sound events; Among them, Attention(Q,K,V) represents the output of the attention mechanism, Q, K, V represent the query, key, and value matrices, which represent the query vector, key vector, and value vector of the current time step respectively, d k is the dimension of the key vector, used to scale the dot product to prevent excessive internal layer activation values, softmax, normalization function, to ensure that the sum of the output weights is 1, QK T The query matrix Q and the transposed matrix K of the key matrix K T The dot product of , which represents the similarity between the query vector and the key vector, Apply the Softmax function to the scaled dot product result to get the attention weight, Apply the attention weights to the value matrix V to get the final output of the attention mechanism; In order to pay attention to the frequency changes and temporal dynamic characteristics of the sound signal at the same time, the multi-head attention structure of the Transformer model is introduced; MultiHead(Q,K,V)=Concat(head1,...,head h )W O head i =Attention(QW i Q ,KW i K ,VW i V ) Among them, the output of MultiHead (Q, K, V) multi-head attention mechanism, head i The i-th attention head, the number of head attention heads, W i Q , W i K , W i V The linear transformation matrix of the query, key, and value of the i-th head, W O The linear transformation matrix of all head outputs, the Concat operation, is used to merge the outputs of each head, and the Attention (QW i Q ,KW i K ,VW i V ) is the attention mechanism output of the i-th attention head; The trained model is integrated into the central control system of the container power station to process sound data in real time, and the noise control strategy and the working status of the silencer are adjusted according to the output of the model.

Images