Optimization design method for fins of air conditioner evaporator

Abstract

The invention discloses an air conditioner evaporator fin optimization design method which comprises the steps that S1, air conditioner fin multi-dimensional data are collected and preprocessed, and preprocessed air conditioner fin multi-dimensional data are obtained; s2, processing the preprocessed multi-dimensional data of the air conditioner fin into an image, extracting first and second image features, and complementing the first and second image features with the data volume to obtain complemented multi-dimensional data of the air conditioner fin; s3, extracting first, second and third characteristic relations according to the heat efficiency and the gas resistance, and calculating the bending times; s4, multi-dimensional data of the air conditioner fins are screened according to the bending times, the characteristic relation between the notch position and the heat efficiency and the characteristic relation between the notch position and the gas resistance are extracted, and the notch length and the notch depth are calculated; and S5, according to the bending times, the notch positions, the notch lengths and the notch depths obtained through calculation, an air conditioner fin drawing is drawn, and notch machining is conducted. The problems that a traditional air conditioner fin design method excessively depends on artificial experience and data is insufficient can be solved.

Description

Technical Field

[0001] The present invention relates to the technical field of air conditioners, and in particular to an optimization design method for air conditioner evaporator fins. Background Art

[0002] As a key component in the evaporator and condenser, the design of air-conditioning fins directly affects the heat exchange efficiency and energy consumption of the air-conditioning system. However, the traditional air-conditioning fin design method has many disadvantages. It is highly dependent on the designer's experience and intuition, and lacks data support during the design process, resulting in unstable design results. At the same time, the traditional method requires a large number of experimental tests and design iterations, which not only has a long design cycle and high cost, but also has low data utilization and is difficult to fully reflect the changing laws of fin performance. In addition, due to the lack of efficient data analysis and optimization algorithms, traditional methods are difficult to achieve fine optimization of fin performance and can often only stay in a relatively rough design stage. Summary of the invention

[0003] In view of this, the present invention aims to provide an air-conditioning evaporator fin optimization design method to solve the problem that traditional air-conditioning fin design methods are overly dependent on manual experience and insufficient data.

[0004] An air conditioner evaporator fin optimization design method, comprising:

[0005] S1: collecting multi-dimensional data of air conditioner fins and performing preprocessing to obtain preprocessed multi-dimensional data of air conditioner fins;

[0006] S2: converting the preprocessed multi-dimensional data of the air conditioner fins into an image, extracting the first and second image features, and completing the preprocessed multi-dimensional data of the air conditioner fins to obtain completed multi-dimensional data of the air conditioner fins;

[0007] S3: Extract the first, second and third characteristic relationships according to thermal efficiency and gas resistance, and calculate the number of bends;

[0008] S4: Filter the multi-dimensional data of air conditioner fins according to the number of bends, extract the characteristic relationship between the incision position, thermal efficiency and gas resistance, and calculate the incision length and depth;

[0009] S5: Draw the air conditioner fin drawing and perform the cutting process according to the calculated bending times, incision position, incision length and incision depth;

[0010] Furthermore, the step S1 collects multi-dimensional data of air conditioner fins and performs preprocessing to obtain preprocessed multi-dimensional data of air conditioner fins, including:

[0011] S11: Taking the louvered fin as the fin type, collecting multi-dimensional data of the air conditioner fin; each data entry in the multi-dimensional data of the air conditioner fin includes 6 dimensions, namely, the number of bends, the bending angle, the incision position, the incision length, the thermal efficiency and the gas resistance; the number of bends includes 1 time, 2 times and 3 times; the bending angle includes 0°, 30°, 60° and 90°;

[0012] S12: For the multi-dimensional data of the air conditioner fins, the data entries with the same six dimensions are deleted, and then the data entries with missing values ​​are filled by linear interpolation method to obtain the multi-dimensional data of the air conditioner fins after preliminary preprocessing;

[0013] S13: The multi-dimensional data of the air-conditioning fins after preliminary preprocessing is normalized by using the maximum and minimum value method to obtain the multi-dimensional data of the air-conditioning fins after preprocessing.

[0014] Furthermore, the step S2 converts the preprocessed air conditioner fin multi-dimensional data into an image, extracts the first and second image features, and completes the preprocessed air conditioner fin multi-dimensional data to obtain the completed air conditioner fin multi-dimensional data, including:

[0015] S21: converting the preprocessed multi-dimensional data of the air conditioner fin into a 6×n matrix as the first image, wherein n is the total number of data entries in the preprocessed multi-dimensional data of the air conditioner fin; using the number of bends and the bend angle as the horizontal and vertical coordinates, respectively, and using the thermal efficiency and the gas resistance as the pixel points of the horizontal and vertical coordinates, respectively, as the second image; in the second image, each pixel point contains 4 channels, which correspond to the incision position, incision length, thermal efficiency and gas resistance of the horizontal and vertical coordinates, respectively;

[0016] Step S21 converts the data table into an image, laying a data foundation for the subsequent convolutional neural network; and the number of bends and the bend angle are used as the horizontal and vertical coordinates, respectively, and the thermal efficiency and the gas resistance are used as the pixel points of the horizontal and vertical coordinates, respectively. This new data organization method is more helpful for the convolutional neural network to learn the potential relationship between the number of bends, the bend angle, the thermal efficiency and the gas resistance;

[0017] S22: Extract features from the first image and the second image respectively to obtain first image features and second image features. The calculation method is:

[0018] F 1 =MLP(CNN 1*6 (P 1 ))

[0019] F 2 =CNN(ConcatFeature(CNN 6*6 (P2 )))

[0020] Among them, F 1 is the first image feature, MLP() is the fully connected layer operation, CNN 1*6 () is the convolutional neural network operation with a convolution kernel size of 1*6, P 1 is the pre-processed multi-dimensional data of the air conditioner fin, i.e., the first image, F 2 is the second image feature, CNN() is a horizontal convolutional neural network operation, ConcatFeature() is a feature map concatenation operation, CNN 6*6 () is the vertical convolutional neural network operation with a convolution kernel size of 6*6, P 2 is the second image; CNN(ConcatFeature(CNN 6*6 () is the vertical and horizontal convolution operation proposed by the present invention;

[0021] Through the 1×6 convolution kernel, CNN can perform detailed scanning and feature extraction for each row of the first image (i.e., the preprocessed multi-dimensional data matrix of the air conditioner fins);

[0022] For the second image, the present invention proposes a calculation method of vertical and horizontal convolution, using a 6×6 convolution kernel for vertical convolution, and then through feature map splicing and horizontal convolution, it realizes the comprehensive capture of the global and local features of the image;

[0023] S23: According to the extracted first image features and second image features, the preprocessed multi-dimensional data of the air conditioner fins is supplemented to obtain supplemented multi-dimensional data of the air conditioner fins, and the calculation method is:

[0024] F = Concat(F 1 , F 2 )

[0025]

[0026] Among them, F is the fusion feature of the first and second images, Concat() is the splicing operation, is the forward data feature, LSTM() is the long short-term memory network operation, x 1 , x 2 , ..., x i are the first, second, ..., i-th positive data in the preprocessed multi-dimensional data of the air conditioner fins, i is the data entry index, is the backward data feature, x -1 , x -2 , ..., x i+1are the first to last, the second to last, ..., up to the positive i+1th data in the preprocessed multi-dimensional data of the air conditioner fins, x′ i+1 The multi-dimensional data of the i+1th air conditioner fin is completed;

[0027] The above calculation is an iterative calculation. Each round of iterative calculation will generate a completed data. After each completed multi-dimensional data of the air conditioner fin is calculated, it is used as a new pre-processed multi-dimensional data of the air conditioner fin and inserted into x i The right side of , and then use the data at this time as the basis for the next round of calculation; the data entry index i∈[1,n], and follows the iteration round, decreasing by 1;

[0028] After the iterative calculation traverses each pre-processed multi-dimensional data of the air conditioner fin, the iterative calculation ends, and the completed multi-dimensional data of the air conditioner fin is obtained;

[0029] Through iterative calculation, step S23 can gradually fill in the missing values ​​in the data set until the required amount of data is reached. This is not only applicable to the case of fewer data entries, but also can effectively expand the size of the data set while maintaining data consistency and accuracy. At the same time, after each round of iterative calculation, the newly completed data will be inserted into the data set as the basis for the next round of calculation. The data set is dynamically updated during the iteration process, which improves the reliability of the completed data.

[0030] In addition, the strategy of filling in from the right side enables the model to make the most of the real data on the left side in each iteration, ensuring that the completed data is generated as close to the real data as possible, effectively reducing the cumulative error;

[0031] The traditional Bi-LSTM inputs the entire data sequence into the model and is a static data model; the present invention, on the other hand, utilizes two unidirectional LSTMs in different directions, combined with an iterative calculation method. Each round of iteration utilizes the forward and backward LSTMs to process the data respectively, and performs feature fusion and decision-making through MLP, allowing the model to dynamically utilize existing data during the filling process and adjust according to the filling effect, thereby achieving efficient and accurate completion of the pre-processed multi-dimensional data of the air-conditioning fins, and providing data support for the subsequent calculation of the number of bends, incision position, incision length and incision depth.

[0032] Furthermore, the step S3 extracts the first, second and third characteristic relationships according to the thermal efficiency and the gas resistance, and calculates the number of bends, including:

[0033] S31: Extract the characteristic relationship between thermal efficiency and the number of bends to obtain the first characteristic relationship, which is calculated as follows:

[0034] The number of bends and thermal efficiency in the completed multi-dimensional data of the air conditioner fins are extracted to obtain a 2*2n matrix, recorded as M η , where η is the thermal efficiency;

[0035] f 1 =CNN(ConcatFeature(CNN 2*2 (M η )))

[0036] Among them, f 1 is the first feature relationship, CNN 2*2 () is a convolutional neural network operation with a convolution kernel size of 2*2;

[0037] S32: Extract the characteristic relationship between gas resistance and the number of bends to obtain a second characteristic relationship, which is calculated as follows:

[0038] The number of bends and gas resistance in the completed multi-dimensional data of the air conditioner fins are extracted to obtain a 2*2n matrix, recorded as M g , where g is the gas resistance;

[0039] f 2 =CNN(ConcatFeature(CNN 2*2 (M g )))

[0040] Among them, f, is the second characteristic relationship;

[0041] S33: Extract the characteristic relationship between thermal efficiency, gas resistance and number of bends to obtain the third characteristic relationship, which is calculated as follows:

[0042] f 3 =MLP(Concat(CNN 4*4 (M η , M g ), f 1 , f 2 ))

[0043] Among them, f 3 For the third feature relationship, CNN 4*4 () is the convolutional neural network operation with a convolution kernel size of 4*4;

[0044] S34: Calculate the number of bends based on the thermal efficiency and gas resistance, combined with the first characteristic relationship, the second characteristic relationship, and the third characteristic relationship. The calculation method is:

[0045] p flip =softmax(W 4 ·(W 1 f 1 +W2 f 2 +W 2 f 3 )+b)

[0046] Among them, p flip is the probability distribution of bending times, softmax() is the softmax function; W 1 , W 2 , W 3 , W 4 are the first, second, third, and fourth weight matrices respectively, and b is the bias vector;

[0047] W 1 Responsible for mapping the first characteristic relationship (the characteristic relationship between thermal efficiency and the number of bends) to the calculation of the number of bends, W 2 It not only participates in the mapping of the second characteristic relationship (the characteristic relationship between gas resistance and bending times), but also 1 Working together, the first and second characteristic relationships are integrated into the calculation of the number of bends, W 3 It is specially used to deal with the third characteristic relationship (the comprehensive characteristic relationship between thermal efficiency, gas resistance and number of bends), W 4 As the weight matrix of the final decision layer, it is responsible for mapping the feature relationships of the previous layers to the probability distribution of the number of bends;

[0048] Calculate the probability distribution of the number of bends, and take the category with the highest probability in the probability distribution of the number of bends as the number of bends; the categories in the probability distribution of the number of bends include: 1 time, 2 times, and 3 times;

[0049] Step S3 extracts the relationship between thermal efficiency and the number of bends (the first characteristic relationship), the relationship between gas resistance and the number of bends (the second characteristic relationship), and the comprehensive characteristic relationship between thermal efficiency, gas resistance and the number of bends (the third characteristic relationship). This multi-level and multi-dimensional feature extraction method can more comprehensively capture the complex relationships in the data and provide strong support for the accurate calculation of the number of bends.

[0050] In addition, through the softmax function, the S3 step outputs the probability distribution of the number of bends, rather than just a single number of bends, so that designers can evaluate the possibility of different numbers of bends based on the probability distribution, which meets the actual needs of air conditioner evaporator fin design.

[0051] For example, when the gas resistance (g) is 120Pa, for sample data 1, when the number of bends is 1, the thermal efficiency (η) is 0.85, and for sample data 2, when the number of bends is 2, the thermal efficiency (η) is 0.87; this indicates that under this specific data set and conditions, when the number of bends increases from 1 to 2, the thermal efficiency increases from 0.85 to 0.87. Increasing the number of bends can improve the thermal efficiency of the air-conditioning fins under the current environment.

[0052] Furthermore, the step S4 filters the multi-dimensional data of the air conditioner fins according to the number of bends, extracts the characteristic relationship between the incision position, thermal efficiency and gas resistance, and calculates the incision length and incision depth, including:

[0053] S41: using the calculated number of bending times as an index, filtering out the completed multi-dimensional data of the air-conditioning fins, and obtaining the filtered multi-dimensional data of the air-conditioning fins;

[0054] S42: Extract the characteristic relationship between the cutout position and the thermal efficiency and gas resistance from the screened multi-dimensional data of the air conditioner fins to obtain the cutout position characteristics. The calculation method is:

[0055] Feature=Concat(CNN 1*3 (M ηg ), GRU 3 (M ηg ))

[0056] Among them, Feature is the incision location feature, CNN 1*3 For the convolutional neural network operation with a convolution kernel size of 1*3, M ηg For the filtered multi-dimensional data of air conditioner fins, GRU 3 () is a horizontal gated recurrent unit operation of length 3;

[0057] S43: Calculate the length of each incision based on the incision position characteristics. The calculation method is:

[0058] F nonlinear =tanh(W 5 *Feature+b′)

[0059] F fused =α·Feature+(1-α)·F nonlinear

[0060]

[0061] Among them, F nonlinear is the nonlinear feature of the cut position, tanh() is the hyperbolic tangent function, W 5 is the nonlinear feature weight matrix, b′ is the nonlinear feature bias vector, Ffused is the nonlinear fusion feature, α is the nonlinear weight parameter, L is the length sequence of each cut, W poly is the polynomial weight matrix, is the transpose of the nonlinear fusion feature, T is the transpose symbol, b poly is the polynomial bias vector;

[0062] S44: Calculate the incision depth based on the incision length and the bending angle. The calculation method is:

[0063] H = MLP (Concat (angle, L))

[0064] Among them, H is the sequence of the depth of each incision, and angle is the bending angle;

[0065] Step S43 first uses the hyperbolic tangent function (tanh) as the activation function to capture the nonlinear relationship in the incision position feature; then the nonlinear fusion feature F fused The calculation of the original incision position feature Feature and the nonlinear feature F nonlinear Weighted fusion is performed to retain the information of the original features and introduce the advantages of nonlinear features, which helps to optimize the calculation results of the incision length. Finally, a polynomial regression model is used to calculate the incision length sequence L and fit the nonlinear relationship between the incision position characteristics and the incision length, thereby improving the accuracy and reliability of the air-conditioning fin design.

[0066] In particular, after predicting the number of bends, if higher heat exchange efficiency, higher corrosion resistance and higher mechanical strength are required, the air conditioner fins need to be cut on both sides. The calculation method for whether the cuts need to be made on the opposite side of the fins is:

[0067] p′ flip =softmax(MLP(angle,L,H))

[0068] Flip = sgn(p′ flip -0.6)

[0069] Among them, p′ flip is the probability distribution of the need for a cut on the other side of the fin, Flip is the sequence of the need for a cut on the other side of the fin, and sgn is the sign function;

[0070] The incision is made at a position greater than zero in Flip, and the incision position is the same as that of the contralateral incision. The incision length and depth are both 1 / 2 of the original incision position.

[0071] For example, it has been calculated that the fin needs to be bent three times, that is, the number of cuts is 3, and the cut positions are recorded as ①②③ respectively. Combining angle, L, and H, calculate p′ flip=[0.1, 0.8, 0.6], indicating that on the other side of the fin, the positions on the opposite side of the fin corresponding to the three cuts (①, ②, ③) have probabilities of needing to be cut respectively [0.1, 0.8, 0.4], and after subtracting 0.6, we get [-0.5, 0.2, 0]. After calculation by the sign function sgn, we get Flip = [-1, 1, 0], that is, the cut is made on the opposite side of the cut position ②, while the cut positions ① and ③ are not cut on the opposite sides;

[0072] Furthermore, the step S5 draws a drawing of the air conditioner fin and performs the cutting process according to the calculated bending times, cutting positions, cutting lengths and cutting depths, including:

[0073] S51: Draw a drawing of the air conditioner fin according to the calculated bending times, incision position, incision length and incision depth;

[0074] S52: Select a laser cutting machine as the processing equipment and perform cutting processing according to the air conditioner fin drawing.

[0075] Compared with the prior art, the advantages of the present invention are:

[0076] (1) The present invention adopts the idea of ​​deep learning. First, the multi-dimensional data of the air-conditioning fin is reconstructed, and the idea of ​​image processing is introduced to complete the data, which reduces the requirements for the data set and has stronger universality. In addition, based on the completed data, the present invention uses deep learning to mine the relationship between the data and calculate the number of bends, incision position, incision length and incision depth. The whole process is more intelligent and supported by scientific data.

[0077] (2) The present invention proposes a data processing method of vertical and horizontal convolution, which first uses vertical convolution to extract local information into multiple feature maps, and then performs horizontal convolution on the feature maps to analyze the interaction between the feature maps. Compared with traditional convolutional neural networks, the vertical and horizontal convolution proposed in the present invention has a stronger ability to comprehensively extract local information.

[0078] (3) The present invention proposes an iterative calculation data completion method based on a long short-term memory network, which can enrich the data volume while ensuring accuracy; in addition, the strategy of filling in from the end of the entire data enables the model to make maximum use of the real data on the left in each iteration, ensuring that the completed data is generated on the basis of being as close to the real data as possible, effectively reducing the cumulative error, and providing data support for the subsequent calculation of the number of bends, incision position, incision length and incision depth. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] Figure 1 A schematic diagram of a flow chart of an air-conditioning evaporator fin optimization design method provided by the present invention;

[0080] Figure 2 A schematic diagram of the algorithm flow of the vertical and horizontal convolution provided by the present invention;

[0081] Figure 3 A schematic diagram of the algorithm flow of the LSTM iterative calculation provided by the present invention. DETAILED DESCRIPTION

[0082] The present invention is further described below in conjunction with the accompanying drawings, but the present invention is not limited in any way. Any changes or substitutions made based on the teachings of the present invention belong to the protection scope of the present invention.

[0083] Example 1: A method for optimizing the design of air conditioner evaporator fins, such as Figure 1 As shown, the following steps are included:

[0084] S1: collecting multi-dimensional data of air conditioner fins and performing preprocessing to obtain preprocessed multi-dimensional data of air conditioner fins;

[0085] S11: Taking the louvered fin as the fin type, collecting multi-dimensional data of the air conditioner fin; each data entry in the multi-dimensional data of the air conditioner fin includes 6 dimensions, namely, the number of bends, the bending angle, the incision position, the incision length, the thermal efficiency and the gas resistance; the number of bends includes 1 time, 2 times and 3 times; the bending angle includes 0°, 30°, 60° and 90°;

[0086] S12: For the multi-dimensional data of the air conditioner fins, the data entries with the same six dimensions are deleted, and then the data entries with missing values ​​are filled by linear interpolation method to obtain the multi-dimensional data of the air conditioner fins after preliminary preprocessing;

[0087] S13: The multi-dimensional data of the air-conditioning fins after preliminary preprocessing is normalized by using the maximum and minimum value method to obtain the multi-dimensional data of the air-conditioning fins after preprocessing.

[0088] S2: converting the preprocessed multi-dimensional data of the air conditioner fins into an image, extracting the first and second image features, and completing the preprocessed multi-dimensional data of the air conditioner fins to obtain completed multi-dimensional data of the air conditioner fins;

[0089] S21: converting the preprocessed multi-dimensional data of the air conditioner fin into a 6×n matrix as the first image, wherein n is the total number of data entries in the preprocessed multi-dimensional data of the air conditioner fin; using the number of bends and the bend angle as the horizontal and vertical coordinates, respectively, and using the thermal efficiency and the gas resistance as the pixel points of the horizontal and vertical coordinates, respectively, as the second image; in the second image, each pixel point contains 4 channels, which correspond to the incision position, incision length, thermal efficiency and gas resistance of the horizontal and vertical coordinates, respectively;

[0090] Step S21 converts the data table into an image, laying a data foundation for the subsequent convolutional neural network; and the number of bends and the bend angle are used as the horizontal and vertical coordinates, respectively, and the thermal efficiency and the gas resistance are used as the pixel points of the horizontal and vertical coordinates, respectively. This new data organization method is more helpful for the convolutional neural network to learn the potential relationship between the number of bends, the bend angle, the thermal efficiency and the gas resistance;

[0091] S22: Extract features from the first image and the second image respectively to obtain first image features and second image features. The calculation method is:

[0092] F 1 =MLP(CNN 1*6 (P 1 ))

[0093] F 2 =CNN(ConcatFeature(CNN 6*6 (P 2 )))

[0094] Among them, F 1 is the first image feature, MLP() is the fully connected layer operation, CNN 1*6 () is the convolutional neural network operation with a convolution kernel size of 1*6, P 1 is the pre-processed multi-dimensional data of the air conditioner fin, i.e., the first image, F 2 is the second image feature, CNN() is a horizontal convolutional neural network operation, ConcatFeature() is a feature map concatenation operation, CNN 6*6 () is the vertical convolutional neural network operation with a convolution kernel size of 6*6, P 2 is the second image; CNN(ConcatFeature(CNN 6*6 () is the vertical and horizontal convolution operation proposed by the present invention, such as Figure 2 As shown;

[0095] Through the 1×6 convolution kernel, CNN can perform detailed scanning and feature extraction for each row of the first image (i.e., the preprocessed multi-dimensional data matrix of the air conditioner fins);

[0096] For the second image, the present invention proposes a calculation method of vertical and horizontal convolution, using a 6×6 convolution kernel for vertical convolution, and then through feature map splicing and horizontal convolution, it realizes the comprehensive capture of the global and local features of the image;

[0097] S23: According to the extracted first image features and second image features, the preprocessed multi-dimensional data of the air conditioner fins is supplemented to obtain supplemented multi-dimensional data of the air conditioner fins, and the calculation method is:

[0098] F = Concat(F 1 , F 2 )

[0099]

[0100] Among them, F is the fusion feature of the first and second images, Concat() is the splicing operation, is the forward data feature, LSTM() is the long short-term memory network operation, x 1 , x 2 , ..., x i are the first, second, ..., i-th positive data in the preprocessed multi-dimensional data of the air conditioner fins, i is the data entry index, is the backward data feature, x -1 , x -2 , ..., x i+1 are the first to last, the second to last, ..., up to the positive i+1th data in the preprocessed multi-dimensional data of the air conditioner fins, x′ i+1 The multi-dimensional data of the i+1th air conditioner fin is completed;

[0101] The above calculation method is an iterative calculation. Each round of iterative calculation will generate a completed data. After each completed multi-dimensional data of the air conditioner fin is calculated, it is used as a new pre-processed multi-dimensional data of the air conditioner fin and inserted into x i The right side of , and then use the data at this time as the basis for the next round of calculation; the data entry index i∈[1,n], and follows the iteration round, decreasing by 1;

[0102] After the iterative calculation traverses each pre-processed air conditioner fin multi-dimensional data, the iterative calculation ends and the completed air conditioner fin multi-dimensional data is obtained, such as Figure 3 As shown;

[0103] Through iterative calculation, step S23 can gradually fill in the missing values ​​in the data set until the required amount of data is reached. This is not only applicable to the case of fewer data entries, but also can effectively expand the size of the data set while maintaining data consistency and accuracy. At the same time, after each round of iterative calculation, the newly completed data will be inserted into the data set as the basis for the next round of calculation. The data set is dynamically updated during the iteration process, which improves the reliability of the completed data.

[0104] In addition, the strategy of filling in from the right side enables the model to make the most of the real data on the left side in each iteration, ensuring that the completed data is generated as close to the real data as possible, effectively reducing the cumulative error;

[0105] The traditional Bi-LSTM inputs the entire data sequence into the model, which is a static data model; while the present invention uses two unidirectional LSTMs in different directions, combined with iterative calculation, and uses forward and backward LSTMs to process data respectively in each iteration, and performs feature fusion and decision-making through MLP, allowing the model to dynamically use existing data during the filling process and adjust according to the filling effect, thereby achieving efficient and accurate completion of the pre-processed multi-dimensional data of the air conditioner fins, and providing data support for the subsequent calculation of the number of bends, incision position, incision length and incision depth;

[0106] For example, if there are 100 data items now and I want to add them to 200 data items, i=99, then first 1 to x 99 Input LSTM and combine with F to calculate Starting from the last piece of data, convert the last piece of data x 100 , input another LSTM, calculate Then and Through MLP feature fusion, the first completed data x′ is obtained 100 , and insert it into the original data sequence at x 99 The right side, that is, x′ 100 The position is at x 99 The right side of x 100 At this time, the data length changes from 100 to 101, and the first round of calculation ends;

[0107] At the beginning of the second round of calculation, based on the 101 data at the end of the first round, the xth 1 to x 98 Input LSTM and combine with F to calculate Starting from the last data, x the last 3 data 99 , x′ 100 、x 100 , input another LSTM, calculate Then and Through MLP feature fusion, the second completed data x′ is obtained 99 , and insert it into the original data sequence at x 98 The right side, that is, x′ 99 The position is at x 98 The right side of x 99 At this time, the data length changes from 101 to 102, and the second round of calculation ends; the subsequent iterations are analogous.

[0108] S3: Extract the first, second and third characteristic relationships according to thermal efficiency and gas resistance, and calculate the number of bends;

[0109] S3 1: Extract the characteristic relationship between thermal efficiency and bending times to obtain the first characteristic relationship, which is calculated as follows:

[0110] The number of bends and thermal efficiency in the completed multi-dimensional data of the air conditioner fins are extracted to obtain a 2*2n matrix, recorded as M η , where η is the thermal efficiency;

[0111] f 1 =CNN(ConcatFeature(CNN 2*2 (M η )))

[0112] Among them, f 1 is the first feature relationship, CNN 2*2 () is a convolutional neural network operation with a convolution kernel size of 2*2;

[0113] S32: Extract the characteristic relationship between gas resistance and the number of bends to obtain a second characteristic relationship, which is calculated as follows:

[0114] The number of bends and gas resistance in the completed multi-dimensional data of the air conditioner fins are extracted to obtain a 2*2n matrix, recorded as M g , where g is the gas resistance;

[0115] f 2 =CNN(ConcatFeature(CNN 2*2 (M g )))

[0116] Among them, f 2 is the second characteristic relationship;

[0117] S33: Extract the characteristic relationship between thermal efficiency, gas resistance and number of bends to obtain the third characteristic relationship, which is calculated as follows:

[0118] f 3 =MLP(Concat(CNN 4*4 (M η , M g ), f 1 , f 2 ))

[0119] Among them, f 3 For the third feature relationship, CNN 4*4 () is the convolutional neural network operation with a convolution kernel size of 4*4;

[0120] S34: Calculate the number of bends based on the thermal efficiency and gas resistance, combined with the first characteristic relationship, the second characteristic relationship, and the third characteristic relationship. The calculation method is:

[0121] p flip =softmax(W 4 ·(W 1 f 1 +W 2 f 2 +W 2 f 3 )+b)

[0122] Among them, p flip is the probability distribution of bending times, softmax() is the softmax function; W 1 , W 2 , W 3 , W 4 are the first, second, third, and fourth weight matrices respectively, and b is the bias vector;

[0123] W 1 Responsible for mapping the first characteristic relationship (the characteristic relationship between thermal efficiency and the number of bends) to the calculation of the number of bends, W 2 It not only participates in the mapping of the second characteristic relationship (the characteristic relationship between gas resistance and bending times), but also 1 Working together, the first and second characteristic relationships are integrated into the calculation of the number of bends, W 3 It is specially used to deal with the third characteristic relationship (the comprehensive characteristic relationship between thermal efficiency, gas resistance and number of bends), W 4 As the weight matrix of the final decision layer, it is responsible for mapping the feature relationships of the previous layers to the probability distribution of the number of bends;

[0124] Calculate the probability distribution of the number of bends, and take the category with the highest probability in the probability distribution of the number of bends as the number of bends; the categories in the probability distribution of the number of bends include: 1 time, 2 times, and 3 times;

[0125] Step S3 extracts the relationship between thermal efficiency and the number of bends (the first characteristic relationship), the relationship between gas resistance and the number of bends (the second characteristic relationship), and the comprehensive characteristic relationship between thermal efficiency, gas resistance and the number of bends (the third characteristic relationship). This multi-level and multi-dimensional feature extraction method can more comprehensively capture the complex relationships in the data and provide strong support for the accurate calculation of the number of bends.

[0126] In addition, through the softmax function, the S3 step outputs the probability distribution of the number of bends, rather than just a single number of bends, so that designers can evaluate the possibility of different numbers of bends based on the probability distribution, which meets the actual needs of air conditioner evaporator fin design.

[0127] S4: Filter the multi-dimensional data of air conditioner fins according to the number of bends, extract the characteristic relationship between the incision position, thermal efficiency and gas resistance, and calculate the incision length and depth;

[0128] S41: using the calculated number of bending times as an index, filtering out the completed multi-dimensional data of the air-conditioning fins, and obtaining the filtered multi-dimensional data of the air-conditioning fins;

[0129] S42: Extract the characteristic relationship between the cutout position and the thermal efficiency and gas resistance from the screened multi-dimensional data of the air conditioner fins to obtain the cutout position characteristics. The calculation method is:

[0130] Feature=Concat(CNN 1*3 (M ηg ), GRU 3 (M ηg ))

[0131] Among them, Feature is the incision location feature, CNN 1*3 For the convolutional neural network operation with a convolution kernel size of 1*3, M ηg For the filtered multi-dimensional data of air conditioner fins, GRU 3 () is a horizontal gated recurrent unit operation of length 3;

[0132] S43: Calculate the length of each incision based on the incision position characteristics. The calculation method is:

[0133] F nonlinear =tanh(W 5 *Feature+b′)

[0134] F fused =α·Feature+(1-α)·F nonlinear

[0135]

[0136] Among them, F nonlinear is the nonlinear feature of the cut position, tanh() is the hyperbolic tangent function, W 5 is the nonlinear feature weight matrix, b′ is the nonlinear feature bias vector, F fused is the nonlinear fusion feature, α is the nonlinear weight parameter, L is the length sequence of each cut, W poly is the polynomial weight matrix, is the transpose of the nonlinear fusion feature, T is the transpose symbol, b poly is the polynomial bias vector;

[0137] S44: Calculate the incision depth based on the incision length and the bending angle. The calculation method is:

[0138] H = MLP (Concat (angle, L))

[0139] Among them, H is the sequence of the depth of each incision, and angle is the bending angle;

[0140] Step S43 first uses the hyperbolic tangent function (tanh) as the activation function to capture the nonlinear relationship in the incision position feature; then the nonlinear fusion feature F fused The calculation of the original incision position feature Feature and the nonlinear feature F nonlinear Weighted fusion is performed to retain the information of the original features and introduce the advantages of nonlinear features, which helps to optimize the calculation results of the incision length. Finally, a polynomial regression model is used to calculate the incision length sequence L and fit the nonlinear relationship between the incision position characteristics and the incision length, thereby improving the accuracy and reliability of the air-conditioning fin design.

[0141] In particular, after predicting the number of bends, if higher heat exchange efficiency, higher corrosion resistance and higher mechanical strength are required, the air conditioner fins need to be cut on both sides. The calculation method for whether the cuts need to be made on the opposite side of the fins is:

[0142] p′ flip =softmax(MLP(angle,L,H))

[0143] Flip = sgn(p′ flip -0.6)

[0144] Among them, p′ flip is the probability distribution of the need for a cut on the other side of the fin, Flip is the sequence of the need for a cut on the other side of the fin, and sggn is the sign function;

[0145] The incision is made at a position greater than zero in Flip, and the incision position is the same as that of the contralateral incision. The incision length and depth are both 1 / 2 of the original incision position.

[0146] For example, it has been calculated that the fin needs to be bent three times, that is, the number of cuts is 3, and the cut positions are recorded as ①②③ respectively. Combining angle, L, and H, calculate p′ flip =[0.1, 0.8, 0.6], which means that on the other side of the fin, the probabilities of the positions on the opposite side of the fin corresponding to the three incisions (①, ②, ③) needing to be cut are [0.1, 0.8, 0.4] respectively. After subtracting 0.6, we get [-0.5, 0.2, 0]. After calculation by the sign function sgn, we get Flip = [-1, 1, 0], that is, the incision is made on the side opposite to the incision position ②, while the incisions ① and ③ are not made on the sides opposite to the incision positions.

[0147] S5: Draw the air conditioner fin drawing and perform the cutting process according to the calculated bending times, incision position, incision length and incision depth;

[0148] S51: Draw a drawing of the air conditioner fin according to the calculated bending times, incision position, incision length and incision depth;

[0149] S52: Select a laser cutting machine as the processing equipment and perform cutting processing according to the air conditioner fin drawing.

[0150] It should be noted that the serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages and disadvantages of the embodiments. And the terms "including", "comprising" or any other variants thereof in this article are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "including a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0151] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes a number of instructions for a terminal device (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in multiple embodiments of the present invention.

[0152] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. An air conditioner evaporator fin optimization design method, characterized in that: The following steps are involved: S1: collecting multi-dimensional data of air conditioner fins and performing preprocessing to obtain preprocessed multi-dimensional data of air conditioner fins; S2: converting the preprocessed multi-dimensional data of the air conditioner fins into an image matrix, extracting the first and second image features, and completing the preprocessed multi-dimensional data of the air conditioner fins to obtain completed multi-dimensional data of the air conditioner fins; S3: Extract the first, second and third characteristic relationships according to thermal efficiency and gas resistance, and calculate the number of bends; S4: Filter the multi-dimensional data of air conditioner fins according to the number of bends, extract the characteristic relationship between the incision position, thermal efficiency and gas resistance, and calculate the incision length and depth; S5: Draw the air conditioner fin drawing and perform the cutting process according to the calculated bending times, incision position, incision length and incision depth.

2. The air conditioner evaporator fin optimization design method according to claim 1, characterized in that: The S1 step includes: S11: Taking the louver fin as the fin type, collecting the air conditioner fin data; each data entry in the air conditioner fin data contains 6 dimensions, namely, the number of bends, the bending angle, the incision position, the incision length, the thermal efficiency and the gas resistance; the number of bends includes 1 time, 2 times and 3 times; the bending angle includes 0°, 30°, 60° and 90°; S12: For the multi-dimensional data of the air conditioner fins, the data entries with the same six dimensions are deleted, and then the data entries with missing values ​​are filled by linear interpolation method to obtain the multi-dimensional data of the air conditioner fins after preliminary preprocessing; S13: The multi-dimensional data of the air-conditioning fins after preliminary preprocessing is normalized by using the maximum and minimum value method to obtain the multi-dimensional data of the air-conditioning fins after preprocessing.

3. The air conditioner evaporator fin optimization design method according to claim 2, characterized in that: The S2 step includes: First, the pre-processed multi-dimensional data of the air conditioner fins is converted into a 6×n matrix as the first image; the number of bends and the bending angle are used as the horizontal and vertical coordinates, respectively, and the thermal efficiency and gas resistance are used as the pixel points of the horizontal and vertical coordinates, respectively, as the second image; Secondly, feature extraction is performed on the first image and the second image respectively; a convolutional neural network with a convolution kernel size of 1*6 combined with a fully connected layer is used to extract features of the first image, and a vertical and horizontal convolution is used to extract features of the second image to obtain features of the first image and features of the second image; Again, based on the extracted first image features and second image features, iterative calculation is performed in combination with the long short-term memory network. Each round of iteration generates a completed data. When the iterative calculation traverses all the preprocessed multi-dimensional data of the air-conditioning fins, the iterative calculation ends and the completed multi-dimensional data of the air-conditioning fins is obtained.

4. The air conditioner evaporator fin optimization design method according to claim 2, characterized in that: The S2 step includes: S21: converting the preprocessed multi-dimensional data of the air conditioner fin into a 6×n matrix as the first image, wherein n is the total number of data entries in the preprocessed multi-dimensional data of the air conditioner fin; using the number of bends and the bend angle as the horizontal and vertical coordinates, respectively, and using the thermal efficiency and the gas resistance as the pixel points of the horizontal and vertical coordinates, respectively, as the second image; in the second image, each pixel point contains 4 channels, which correspond to the incision position, incision length, thermal efficiency and gas resistance of the horizontal and vertical coordinates, respectively; S22: Extract features from the first image and the second image respectively to obtain first image features and second image features. The calculation method is: F1=MLP(CNN 1*6 (P1)) F2=CNN(ConcatFeature(CNN 6*6 (P2))) Among them, F1 is the first image feature, MLP() is the fully connected layer operation, and CNN 1*6 () is a convolutional neural network operation with a convolution kernel size of 1*6, P1 is the preprocessed multi-dimensional data of the air conditioner fins, F2 is the second image feature, CNN() is a horizontal convolutional neural network operation, ConcatFeature() is a feature map splicing operation, CNN 6*6 () is the vertical convolutional neural network operation with a convolution kernel size of 6*6, and P2 is the second image; S23: According to the extracted first image features and second image features, the preprocessed multi-dimensional data of the air conditioner fins is supplemented to obtain supplemented multi-dimensional data of the air conditioner fins, and the calculation method is: F=Concat(F1,F2) Among them, F is the fusion feature of the first and second images, Concat() is the splicing operation, is the forward data feature, LSTM() is the long short-term memory network operation, x1, x2,…, x i are the first, second, ..., i-th positive data in the preprocessed multi-dimensional data of the air conditioner fins, i is the data entry index, is the backward data feature, x -1 ,x -2 ,…,x i+1 are the first to last, the second to last, ..., up to the positive i+1th data in the preprocessed multi-dimensional data of the air conditioner fins, x' i+1 The multi-dimensional data of the i+1th air conditioner fin is completed; The above calculation method is an iterative calculation. Each round of iterative calculation will generate a completed data. After each completed multi-dimensional data of the air conditioner fin is calculated, it is used as a new pre-processed multi-dimensional data of the air conditioner fin and inserted into x i The right side of , and then use the data at this time as the basis for the next round of calculation; the data entry index i∈[1,n], and follows the iteration round, decreasing by 1; After the iterative calculation traverses each preprocessed multi-dimensional data of the air-conditioning fin, the iterative calculation ends, and the completed multi-dimensional data of the air-conditioning fin is obtained.

5. The air conditioner evaporator fin optimization design method according to claim 4, characterized in that: The S3 step includes: S31: Extract the characteristic relationship between thermal efficiency and the number of bends to obtain the first characteristic relationship, which is calculated as follows: The number of bends and thermal efficiency in the completed multi-dimensional data of the air conditioner fins are extracted to obtain a 2*2n matrix, recorded as M η , where η is the thermal efficiency; f1=CNN(ConcatFeature(CNN 2*2 (M η ))) Among them, f1 is the first feature relationship, CNN 2*2 () is a convolutional neural network operation with a convolution kernel size of 2*2; S32: Extract the characteristic relationship between gas resistance and the number of bends to obtain a second characteristic relationship, which is calculated as follows: The number of bends and gas resistance in the completed multi-dimensional data of the air conditioner fins are extracted to obtain a 2*2n matrix, recorded as M g , where g is the gas resistance; f2=CNN(ConcatFeature(CNN 2*2 (M g ))) Among them, f2 is the second characteristic relationship; S33: Extract the characteristic relationship between thermal efficiency, gas resistance and number of bends to obtain the third characteristic relationship, which is calculated as follows: f3=MLP(Concat(CNN 4*4 (M η ,M g ),f1,f2)) Among them, f3 is the third feature relationship, CNN 4*4 () is the convolutional neural network operation with a convolution kernel size of 4*4; S34: Calculate the number of bends based on the thermal efficiency and gas resistance, combined with the first characteristic relationship, the second characteristic relationship, and the third characteristic relationship. The calculation method is: p flip =softmax(W4·(W1f1+W2f2+W2f3)+b) Among them, p flip is the probability distribution of the number of bends, softmax() is the softmax function; W1, W2, W3, W4 are the first, second, third, and fourth weight matrices respectively, and b is the bias vector; The probability distribution of the number of bends is calculated, and the category with the highest probability in the probability distribution of the number of bends is taken as the number of bends; the categories in the probability distribution of the number of bends include: 1 time, 2 times, and 3 times.

6. The air conditioner evaporator fin optimization design method according to claim 5, characterized in that: The S4 step includes: S41: using the calculated number of bending times as an index, filtering out the completed multi-dimensional data of the air-conditioning fins, and obtaining the filtered multi-dimensional data of the air-conditioning fins; S42: Extract the characteristic relationship between the cutout position and the thermal efficiency and gas resistance from the screened multi-dimensional data of the air conditioner fins to obtain the cutout position characteristics. The calculation method is: Feature=Concat(CNN 1*3 (M ηg ),GRU3(M ηg )) Among them, Feature is the incision location feature, CNN 1*3 For the convolutional neural network operation with a convolution kernel size of 1*3, M ηg The multi-dimensional data of air conditioner fins after screening, GRU3() is a horizontal gated recurrent unit operation with a length of 3; S43: Calculate the length of each incision based on the incision position characteristics. The calculation method is: F nonlinear =tanh(W5*Feature+b') F fused =α·Feature+(1-α)·F nonlinear Among them, F nonlinear is the nonlinear feature of the incision position, tanh() is the hyperbolic tangent function, W5 is the nonlinear feature weight matrix, b' is the nonlinear feature bias vector, F fused is the nonlinear fusion feature, α is the nonlinear weight parameter, L is the length sequence of each cut, W poly is the polynomial weight matrix, is the transpose of the nonlinear fusion feature, T is the transpose symbol, b poly is the polynomial bias vector; S44: Calculate the incision depth based on the incision length and the bending angle. The calculation method is: H = MLP (Concat (angle, L)) Among them, H is the sequence of the depth of each incision, and angle is the bending angle.

7. The air conditioner evaporator fin optimization design method according to claim 6, characterized in that: The S5 step includes: S51: Draw a drawing of the air conditioner fin according to the calculated bending times, incision position, incision length and incision depth; S52: Select a laser cutting machine as the processing equipment and perform cutting processing according to the air conditioner fin drawing.

Images