An intelligent screening method for compressor piping schemes for refrigeration equipment

Abstract

The present invention discloses an intelligent screening method for refrigeration equipment compressor piping schemes, comprising the following steps: 1) given a compressor compartment model and data on the start and end points of each pipeline, a plurality of different piping layout schemes are obtained; 2) feature data of each scheme is extracted and formed into a characteristic sample; 3) the characteristic samples are normalized; 4) the training samples are input into a clustering algorithm to calculate multiple centers; 5) neural networks of different sample classes are trained using the centers and samples, and these networks are combined; 6) the trained results are used to screen test schemes and output results that meet the requirements. The present invention can solve the problem of screening refrigeration equipment compressor piping that requires qualitative and empirical analysis and has large feature differences, thereby providing more reasonable piping schemes for compressor piping and improving the efficiency of piping design.

Description

Technical Field

[0001] The present invention relates to an intelligent screening method for compressor piping schemes of refrigeration equipment, which is used to implement the screening of compressor piping layout schemes. Background Art

[0002] Refrigeration equipment compressor piping operates under the excitation of a specific compressor frequency. Without a specific vibration reduction configuration design, failures such as leakage and pipe breakage are prone to occur. While some current piping planning methods address the layout of refrigeration equipment compressor piping, the compressor vibration reduction configuration requirements are qualitative and empirical, and the solutions generated by current planning methods must be screened to meet these requirements. Furthermore, compressor piping configurations are complex and diverse, and the characteristics of different excellent solutions vary significantly. Current methods struggle to accommodate multiple types of solutions, easily leading to missed or incorrect selections. Summary of the Invention

[0003] The purpose of the present invention is to solve the problem of screening compressor piping solutions and provide an intelligent screening method for refrigeration equipment compressor piping solutions, so as to quickly, efficiently and accurately screen out piping solutions that meet the configuration requirements, thereby providing a more reasonable piping solution for compressor piping and improving the efficiency of piping design work.

[0004] In order to solve the above technical problems, the following technical solutions are adopted

[0005] The present invention provides a method for screening piping layout schemes for refrigeration equipment compressors, which comprises the following steps:

[0006] Step 1: Obtain the compressor cabin model and the data of the starting and ending points of each pipeline, and randomly generate J different pipeline layout schemes B = [b1, b2, ..., b j ,…,b J ], where b j represents the layout of the jth pipeline, and P j,1 Indicates the starting point of the jth pipeline, P j,i represents the i-th inflection point of the j-th pipeline, n j Indicates the end point number of the jth pipeline; P j,i =[x i ,y i ,z i ] T , x i ,y i ,z i Point P of the jth pipeline j,i The three-dimensional coordinates in the Cartesian coordinate system; J is the total number of layout options;

[0007] Step 2: Extract the layout plan b of the jth pipeline j The characteristic sequence t j , thus obtaining the feature set T = [t1, t2,…, t j ,…,t J ] T After normalization, the normalized feature set S is obtained, where s j represents the jth normalized feature sequence;

[0008] Step 3: Divide the normalized feature set S into training samples and test samples And the training samples Given a set of scores in, represents the j1th feature sequence in the training sample, and express The pth feature in express The corresponding score, Represents the j2th feature sample in the test sample, j1=1,2,…,J1, j2=1,2,…,J2, J1 represents the training sample The number of feature sequences in J2 represents the test sample The number of feature samples in , and J1+J2=J;

[0009] Step 4: Set the total number of categories to N and train the samples Clustering is performed to divide the training samples into Divided into N category sets [A1, A2, ..., A n ,…,A N ] and its corresponding cluster centers [c1,c2,…,c n ,…,c N ], where A n represents the nth category set, and Indicates A n The lth characteristic sequence in, l=1,2,…,L n , L n Indicates A n The total number of characteristic sequences in c n Represents the nth category set A n The cluster center of c n =[μ n,1 ,μ n,2 ,…,μ n,p ,…,μ n,P ],μ n,p Indicates c nThe pth feature of , n = 1, 2, ..., N;

[0010] Step 5: From N cluster centers [c1,c2,…,c n ,…,c N ] and select K qualified cluster centers as K RBF neural networks [R1, R2, ..., R k ,…,R K ], and then K RBF neural networks [R1, R2, …, R k ,…,R K ] are combined into a neural network model R; thus using the N category set [A1, A2, ..., A n ,…,A N ] Train the neural network model R to obtain a trained neural network model R′; wherein, R k represents the kth RBF neural network;

[0011] Step 6: Place the test sample Input the trained neural network model into R for processing to filter out excellent feature sample sets Then the qth excellent feature sample The corresponding pipeline scheme is Thus, the set of pipelines that meet the vibration reduction configuration requirements of the compressor is obtained as

[0012] The intelligent screening method for refrigeration equipment compressor piping schemes according to the present invention is also characterized in that the neural network model R in step 5 is trained according to the following steps:

[0013] Step 5.1: According to A n The lth feature sequence in The score y n,l judge Whether it meets the requirements of the vibration reduction configuration, thus obtaining A n The number of characteristic sequences that meet the requirements of vibration reduction configuration and A n The ratio of the total number of feature sequences in is taken as the nth category set A n The pass rate η n ;

[0014] Step 5.2: Set the minimum pass rate as η pass , find all qualified rates greater than the minimum qualified rate η pass The category set of [A1′, A2′, …, A k ′,…,A′ K ], k=1,2,…,K, the remaining unqualified category sets form the K+1th category set A′K+1 Among them, A k ′ represents the kth qualified category set; then the input set of the kth neural network is defined as in, is the mth feature sequence of the kth neural network input set, M is The total number of characteristic sequences; The corresponding score set is recorded as in, is the mth feature sequence of the kth neural network score;

[0015] Step 5.3: The cluster centers corresponding to the qualified category set [c1′,c2′,…,c′ k ,…,c′ K ] are respectively used as the center of each sub-neural network to construct each RBF neural network [R1, R2, ..., R k ,…,R K ], where c′ k Represents the kth qualified category set A k ′ corresponds to the cluster center;

[0016] Step 5.3.1: Use formula (1) to get the kth RBF neural network R k The activation function f of the pth feature k :

[0017]

[0018] In formula (1), express The mth feature sequence in The pth feature of k,p Represents the qualified class set A k The center of ′ k The pth feature of k,p Represents the kth RBF neural network R k The p-th variance of

[0019] Step 5.3.2: Use formula (2) to get the kth RBF neural network R k The output function

[0020]

[0021] In formula (2), w k,p Represents the kth RBF neural network R k The pth weight, w k,0 Represents the kth RBF neural network Rk The offset value of is the feature sequence The estimated score value;

[0022] Step 5.3.3: Use formula (3) to get the kth RBF neural network R k The error loss function e k :

[0023]

[0024] Step 5.4: Take the kth cluster center c′ k The pth feature μ′ in k,p The maximum distance between the pth feature and the rest of the cluster centers is used as the kth RBF neural network R k The p-th variance σ k,p ;

[0025] Step 5.5: Use formula (4) to obtain the kth RBF neural network R k The weight set W k :

[0026]

[0027] In formula (4), Φ k is the kth RBF neural network R k The generalized kernel vector of is the kernel vector Φ k The mth kernel function of

[0028]

[0029] Step 5.6: Input the normalized feature set S into the neural network model R for prediction and calculate the j-th feature sequence s j The Mahalanobis distance to the cluster center corresponding to each RBF neural network, and the RBF neural network R corresponding to the minimum distance is selected min As the jth feature sequence s j The prediction network R min Output the jth feature sequence s j The estimated score value; then the corresponding estimated score value is output by the prediction network corresponding to each feature sequence in the normalized feature set S in the neural network model R; min∈[1,K].

[0030] In step 6, excellent feature samples are screened out according to the following steps:

[0031] Step 6.1: Calculate the test sample The j2th feature sample in There are K cluster centers [c1′,c2′,…,c′ k ,…,c′ K ] and find the cluster center c′ corresponding to the minimum distance min ;

[0032] Step 6.2: Take the j2th feature sample Input the min cluster center c′ min The corresponding RBF neural network R min Score prediction is performed in

[0033] Step 6.3: Set the passing score to γ, if It represents the j2th feature sample Satisfy the configuration requirements and serve as an excellent feature sample, otherwise, it represents the j2th feature sample Failure to meet configuration requirements;

[0034] Step 6.4: Filter out all excellent feature samples according to the process of steps 6.1 to 6.3.

[0035] The electronic device of the present invention includes a memory and a processor, and is characterized in that the memory is used to store a program that supports the processor to execute the intelligent screening method, and the processor is configured to execute the program stored in the memory.

[0036] The present invention provides a computer-readable storage medium, wherein a computer program is stored on the computer-readable storage medium, and the computer program executes the steps of the intelligent screening method when the computer program is executed by a processor.

[0037] Compared with the prior art, the present invention has the following advantages:

[0038] 1. The present invention uses a combined RBF neural network to solve the problem that a single neural network is difficult to screen out excellent pipeline solutions with large feature differences, thereby avoiding the problem of missing excellent solutions.

[0039] 2. The present invention uses a clustering algorithm to pre-calculate the center and variance of the neural network, which reduces the difficulty of subsequent neural network training, improves training efficiency, and saves computing resources.

[0040] 3. The present invention uses a specific class set as the training sample of the corresponding neural network, which reduces the dimension of information, eliminates the interference of non-similar information, and improves the training speed of the neural network, thereby improving the efficiency of pipeline design work.

[0041] 4. The present invention selects a suitable subnetwork for prediction based on the distance from the sample to the center, thereby improving the accuracy of the prediction and reducing the misselection of unqualified solutions. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 is a schematic diagram of the compressor compartment and pipeline connection points of the present invention;

[0043] Figure 2 Schematic diagram of the pipeline inflection point of the present invention;

[0044] Figure 3 It is a schematic diagram of the sub-network training process of the present invention;

[0045] Figure 4 It is a schematic diagram of the neural network prediction process of the present invention;

[0046] Figure 5 Flowchart of the method of the present invention. DETAILED DESCRIPTION

[0047] The following is a detailed description of an embodiment of the present invention. This embodiment is implemented based on the technical solution of the present invention, but the protection scope of the present invention is not limited to the following embodiment.

[0048] In this embodiment, considering that the piping of the refrigeration equipment compressor has a vibration reduction configuration requirement, this requirement is qualitative and empirical, and it is difficult to obtain directly by the path planning algorithm. It is necessary to screen the schemes, and there are also large differences between the excellent schemes, which increases the difficulty of screening. Therefore, a method for screening the piping layout schemes of the refrigeration equipment compressor is proposed, such as Figure 5 As shown, specifically including:

[0049] Step 1: Obtain the compressor cabin model and the data of the starting and ending points of each pipeline, and randomly generate J different pipeline layout schemes B = [b1, b2, ..., b j ,…,b J ], where b j represents the layout of the jth pipeline, and P j,1 Indicates the starting point of the jth pipeline, P j,i represents the i-th inflection point of the j-th pipeline, n j Indicates the end point number of the jth pipeline; P j,i =[x i ,y i ,z i ] T , x i ,y i ,z i Point P of the jth pipeline j,i The three-dimensional coordinates in the Cartesian coordinate system; J is the total number of layout options;

[0050] In this embodiment, the established Cartesian coordinate system and the typical compressor cabin model are as follows: Figure 1 As shown in the figure, 1 and 1' represent the starting and ending points of the first pipeline, and 2 and 2' represent the starting and ending points of the second pipeline. Figure 2 shown.

[0051] Step 2: Extract the layout plan b of the jth pipeline j The characteristic sequence t j , thus obtaining the feature set T = [t1, t2,…, t j ,…,t J ] T After normalization, the normalized feature set S is obtained; where s j Represents the j-th normalized feature sequence.

[0052] Step 3: Divide the normalized feature set S into training samples and test samples And the training samples Given a set of scores in, represents the j1th feature sequence in the training sample, and express The pth feature in express The corresponding score, Represents the j2th feature sample in the test sample, j1=1,2,…,J1, j2=1,2,…,J2, J1 represents the training sample The number of feature sequences in J2 represents the test sample The number of feature samples in , and J1+J2=J;

[0053] Step 4: Set the total number of categories to N and train the samples Clustering is performed to divide the training samples into Divided into N category sets [A1, A2, ..., A n ,…,A N ] and its corresponding cluster centers [c1,c2,…,c n ,…,c N ], where A n represents the nth category set, and Indicates A n The lth characteristic sequence in, l=1,2,…,L n , L n Indicates A nThe total number of characteristic sequences in c n Represents the nth category set A n The cluster center of c n =[μ n,1 ,μ n,2 ,…,μ n,p ,…,μ n,P ],μ n,p Indicates c n The pth feature of , n = 1, 2, ..., N;

[0054] In this example, the K-Means clustering algorithm is used to classify the samples and obtain the center of each class. Other clustering algorithms are also applicable here.

[0055] Step 5: From N cluster centers [c1,c2,…,c n ,…,c N ] and select K qualified cluster centers as K RBF neural networks [R1, R2, ..., R k ,…,R K ], and then K RBF neural networks [R1, R2, …, R k ,…,R K ] are combined into a neural network model R; thus using the N category set [A1, A2, ..., A n ,…,A N ] Train the neural network model R to obtain a trained neural network model R′; wherein, R k represents the kth RBF neural network;

[0056] Step 5.1: According to A n The lth feature sequence in The score y n,l judge Whether it meets the requirements of the vibration reduction configuration, thus obtaining A n The number of characteristic sequences that meet the requirements of vibration reduction configuration and A n The ratio of the total number of feature sequences in is taken as the nth category set A n The pass rate η n ;

[0057] Step 5.2: Set the minimum pass rate as η pass , find all qualified rates greater than the minimum qualified rate η pass The category set of [A1′, A2′, …, A k ′,…,A′ K ], k=1,2,…,K, the remaining unqualified category sets form the K+1th category set A′ K+1 Among them, Ak ′ represents the kth qualified category set; then the input set of the kth neural network is defined as in, is the mth feature sequence of the kth neural network input set, M is The total number of characteristic sequences; The corresponding score set is recorded as in, is the mth feature sequence of the kth neural network score;

[0058] In this example, the pass rate η is set pass =0.5, that is, as long as more than half of the pipelines in the set are qualified pipelines, the set is considered to be a qualified set.

[0059] Step 5.3: The cluster centers corresponding to the qualified category set [c1′,c2′,…,c′ k ,…,c′ K ] are respectively used as the center of each sub-neural network to construct each RBF neural network [R1, R2, ..., R k ,…,R K ], where c k ′ represents the kth qualified category set A k ′ corresponds to the cluster center;

[0060] Step 5.3.1: Use formula (1) to get the kth RBF neural network R k The activation function f of the pth feature k :

[0061]

[0062] In formula (1), express The mth feature sequence in The pth feature of k,p Represents the qualified class set A k The center c of ′ k ′’s p-th feature; σ k,p Represents the kth RBF neural network R k The p-th variance of

[0063] Step 5.3.2: Use formula (2) to get the kth RBF neural network R k The output function

[0064]

[0065] In formula (2), wk,p Represents the kth RBF neural network R k The pth weight, w k,0 Represents the kth RBF neural network R k The offset value of is the feature sequence The estimated score value;

[0066] Step 5.3.3: Use formula (3) to get the kth RBF neural network R k The error loss function e k :

[0067]

[0068] Most of the pipelines that meet the requirements are in the qualified class set. Constructing the neural network with the qualified class center as the center can ensure that more qualified pipelines of different types are included.

[0069] Step 5.4: Take the kth cluster center c k The pth feature μ′ in ′ k,p The maximum distance between the pth feature and the rest of the cluster centers is used as the kth RBF neural network R k The p-th variance σ k,p ;

[0070] Step 5.5: Use formula (4) to obtain the kth RBF neural network R k The weight set W k :

[0071]

[0072] In formula (4), Φ k is the kth RBF neural network R k The generalized kernel vector of is the kernel vector Φ k The mth kernel function of

[0073]

[0074] by The kth sub-network is trained for the sample to ensure that the kth neural network includes more excellent data in the kth qualified class, while excluding all solutions that do not meet the requirements, and reducing the training data set to improve training efficiency. The flowchart of the sub-network training is as follows Figure 3 shown.

[0075] Step 5.6: Input the normalized feature set S into the neural network model R for prediction and calculate the j-th feature sequence s jThe Mahalanobis distance to the cluster center corresponding to each RBF neural network, and the RBF neural network R corresponding to the minimum distance is selected min As the jth feature sequence s j The prediction network R min Output the jth feature sequence s j The estimated score value; then the corresponding estimated score value is output by the prediction network corresponding to each feature sequence in the normalized feature set S in the neural network model R; min∈[1,K].

[0076] The final training result is equivalent to a combination of K neural network models, rather than a neural network with K hidden layers.

[0077] Step 6: Place the test sample Input the trained neural network model into R for processing to filter out excellent feature sample sets Then the qth excellent feature sample The corresponding pipeline scheme is Thus, the set of pipelines that meet the vibration reduction configuration requirements of the compressor is obtained as

[0078] Step 6.1: Calculate the test sample The j2th feature sample in There are K cluster centers [c1′,c2′,…,c′ k ,…,c′ K ] and find the cluster center c′ corresponding to the minimum distance min ;

[0079] Step 6.2: Take the j2th feature sample Input the min cluster center c′ min The corresponding RBF neural network R min Score prediction is performed in

[0080] Step 6.3: Set the passing score to γ, if It represents the j2th feature sample Satisfy the configuration requirements and serve as an excellent feature sample, otherwise, it represents the j2th feature sample Failure to meet configuration requirements;

[0081] Step 6.4: Filter out all excellent feature samples according to the process of steps 6.1 to 6.3.

[0082] The selection of neural networks requires first determining the selected sub-network based on the minimum distance between the sample and the qualified class center, and then using the sub-network for valuation. The flow chart is as follows: Figure 4shown.

[0083] In this embodiment, an electronic device includes a memory and a processor. The memory is used to store a program that supports the processor to execute the above-mentioned intelligent screening method. The processor is configured to execute the program stored in the memory.

[0084] In this embodiment, a computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above-mentioned intelligent screening method are executed.

Claims

1. A method for screening piping layout schemes for refrigeration equipment compressors, characterized in that: The following steps are involved: Step 1: Obtain the compressor cabin model and the data of the starting and ending points of each pipeline, and randomly generate J different pipeline layout schemes B = [b1, b2, ..., b j ,…,b J ], where b j represents the layout of the jth pipeline, and P j,1 represents the starting point of the jth pipeline, P j,i represents the i-th inflection point of the j-th pipeline, n j Indicates the end point number of the jth pipeline; P j,i =[x i ,y i ,z i ] T , x i ,y i ,z i Point P of the jth pipeline j,i The three-dimensional coordinates in the Cartesian coordinate system; J is the total number of layout options; Step 2: Extract the layout plan b of the jth pipeline j The characteristic sequence t j , thus obtaining the feature set T = [t1, t2,…, t j ,…,t J ] T After normalization, the normalized feature set S is obtained, where s j represents the jth normalized feature sequence; Step 3: Divide the normalized feature set S into training samples and test samples And the training samples Given a set of scores in, represents the j1th feature sequence in the training sample, and λ j1,k express The pth feature in express The corresponding score, Represents the j2th feature sample in the test sample, j1=1,2,…,J1, j2=1,2,…,J2, J1 represents the training sample The number of feature sequences in J2 represents the test sample The number of feature samples in , and J1+J2=J; Step 4: Set the total number of categories to N and train the samples Clustering is performed to divide the training samples into Divided into N category sets [A1, A2, ..., A n ,…,A N ] and its corresponding cluster centers [c1,c2,…,c n ,…,c N ], where A n represents the nth category set, and Indicates A n The lth characteristic sequence in, l=1,2,…,L n , L n Indicates A n The total number of characteristic sequences in c n Represents the nth category set A n The cluster center of c n =[μ n,1 ,μ n,2 ,…,μ n,p ,…,μ n,P ],μ n,p Indicates c n The pth feature of , n = 1, 2, ..., N; Step 5: From N cluster centers [c1,c2,…,c n ,…,c N ] and select K qualified cluster centers as K RBF neural networks [R1, R2, ..., R k ,…,R K ], and then K RBF neural networks [R1, R2, …, R k ,…,R K ] are combined into a neural network model R; thus using the N category set [A1, A2, ..., A n ,…,A N ] Train the neural network model R to obtain a trained neural network model R′; wherein, R k represents the kth RBF neural network; Step 6: Place the test sample Input the trained neural network model into R for processing to filter out excellent feature sample sets Then the qth excellent feature sample The corresponding pipeline scheme is Thus, the set of pipelines that meet the vibration reduction configuration requirements of the compressor is obtained as 2. The intelligent screening method for refrigeration equipment compressor piping scheme according to claim 1 is characterized in that: The neural network model R in step 5 is trained according to the following steps: Step 5.1: According to A n The lth feature sequence in The score y n,l judge Whether it meets the requirements of the vibration reduction configuration, thus obtaining A n The number of characteristic sequences that meet the requirements of vibration reduction configuration and A n The ratio of the total number of feature sequences in is taken as the nth category set A n The pass rate η n ; Step 5.2: Set the minimum pass rate as η pass , find all qualified rates greater than the minimum qualified rate η pass The category set of [A1′, A2′, …, A k ′,…,A′ K ], k=1,2,…,K, the remaining unqualified category sets form the K+1th category set A′ K+1 ; Among them, A k ′ represents the kth qualified category set; then the input set of the kth neural network is defined as in, is the mth feature sequence of the kth neural network input set, M is The total number of characteristic sequences; The corresponding score set is recorded as in, is the mth feature sequence of the kth neural network score; Step 5.3: The cluster centers corresponding to the qualified category set [c1′,c2′,…,c′ k ,…,c′ K ] are respectively used as the center of each sub-neural network to construct each RBF neural network [R1, R2, ..., R k ,…,R K ], where c′ k Represents the kth qualified category set A k ′ corresponds to the cluster center; Step 5.3.1: Use formula (1) to get the kth RBF neural network R k The activation function f of the pth feature k : In formula (1), express The mth feature sequence in The pth feature of k,p Represents the qualified class set A k The center of ′ c′ k The pth feature of k,p Represents the kth RBF neural network R k The p-th variance of Step 5.3.2: Use formula (2) to get the kth RBF neural network R k The output function In formula (2), w k,p Represents the kth RBF neural network R k The pth weight, w k,0 Represents the kth RBF neural network R k The offset value of is the feature sequence The estimated score value; Step 5.3.3: Use formula (3) to get the kth RBF neural network R k The error loss function e k : Step 5.4: Take the kth cluster center c′ k The pth feature μ′ in k,p The maximum distance between the pth feature and the rest of the cluster centers is used as the kth RBF neural network R k The p-th variance σ k,p ; Step 5.5: Use formula (4) to obtain the kth RBF neural network R k The weight set W k : In formula (4), Φ k is the kth RBF neural network R k The generalized kernel vector of is the kernel vector Φ k The mth kernel function of Step 5.6: Input the normalized feature set S into the neural network model R for prediction and calculate the j-th feature sequence s j The Mahalanobis distance to the cluster center corresponding to each RBF neural network, and the RBF neural network R corresponding to the minimum distance is selected min As the jth feature sequence s j The prediction network R min Output the jth feature sequence s j The estimated score value; then the corresponding estimated score value is output by the prediction network corresponding to each feature sequence in the normalized feature set S in the neural network model R; min∈[1,K].

3. The intelligent screening method for refrigeration equipment compressor piping scheme according to claim 1 is characterized in that: In step 6, excellent feature samples are screened out according to the following steps: Step 6.1: Calculate the test sample The j2th feature sample in There are K cluster centers [c1′,c2′,…,c′ k ,…,c′ K ] and find the cluster center c′ corresponding to the minimum distance min ; Step 6.2: Take the j2th feature sample Input the min cluster center c′ min The corresponding RBF neural network R min Score prediction is performed in Step 6.3: Set the passing score to γ, if It represents the j2th feature sample Satisfy the configuration requirements and serve as an excellent feature sample, otherwise, it represents the j2th feature sample Failure to meet configuration requirements; Step 6.4: Filter out all excellent feature samples according to the process of steps 6.1 to 6.

3.

4. An electronic device comprising a memory and a processor, characterized in that: The memory is used to store a program that supports the processor to execute the screening method according to claim 1, 2 or 3, and the processor is configured to execute the program stored in the memory.

5. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the screening method according to claim 1, 2 or 3 are performed.

Images