Production planning management optimization method based on vertical federated learning industrial software docking

Abstract

The present invention discloses a production plan management optimization method based on vertical federated learning industrial software docking. The method first confirms the joint modeling learning task of the vertical federation based on the needs of the enterprise itself, then pre-processes the sample data stream data of the federated learning participating enterprises, then performs sample alignment on the federated learning participating enterprises, and then performs encrypted vertical federated learning with parameter exchange on the federated learning participating enterprises, and finally applies the federated learning prediction results according to the needs of the enterprise to optimize the production plan management of the enterprise; the present invention uses federated learning means to obtain the optimal deployment of the production plan for the next time period, and comprehensively manages enterprise resources based on the industrial Internet industrial software system, improves enterprise productivity and production efficiency, and jointly establishes a more powerful model, thereby effectively improving the lag of current enterprise production plan management.

Description

Technical Field

[0001] The present invention relates to the fields of machine learning and industrial Internet technology, and in particular to a production planning management optimization method based on vertical federated learning industrial software docking. Background Art

[0002] As a product of the deep integration of information technology and manufacturing in the new era, the Industrial Internet can lead the organic integration of the new generation of electronic information technology and advanced manufacturing technology, and promote the informatization revolution of my country's manufacturing enterprises. The Industrial Internet is an important infrastructure for the upgrading of China's manufacturing industry and has a comprehensive, profound and revolutionary impact on future development.

[0003] At present, in manufacturing enterprises with poor information construction, the level of coordinated management of edge modules and local data of production equipment is low, the data acquisition ability is poor, and the production planning management work is mainly based on current production needs. In this case, there is an obvious lag in resource allocation, which seriously affects production efficiency and enterprise productivity, and cannot complete the coordinated production of people, machines and materials. Under the wave of global industrial production informatization, such enterprises urgently need to carry out informatization reform and industrial chain upgrades, and systematically manage and use edge layer production data.

[0004] In manufacturing enterprises with better information construction, resource allocation in industrial production is mainly based on the MES system, PLM system, ERP system and other industrial Internet production management industrial software to assist in setting production plans, resource allocation and plan design. Such enterprises have relatively systematic management of edge device data, but the use of these complex heterogeneous data is still insufficient, and there is a lack of scientific means to use these data for future production plan optimization. In addition, due to data confidentiality and conflicts of interest, the degree of data interoperability between different departments and upstream, midstream and downstream enterprises is low. Enterprises and departments can only allocate resources based on the production information they have, which is lagging and limits the optimization design of production plans.

[0005] Faced with higher requirements from customers and the market for product quality and production capacity, manufacturing companies have begun to realize that they should reform the most basic production management to enhance their competitiveness. Currently, new production planning management methods are emerging one after another, such as the BTO (make to order) method oriented to demand customization, and the JIT (just-in-time) method, a high-quality, low-consumption production method under conditions of multi-variety and small-batch mixed production, which aims to reduce inventory and optimize resource management. Although these methods have effectively improved the resource utilization and industrial efficiency of the company, these production planning methods are essentially planned based on a posteriori knowledge, and the demand-oriented production method also has the lag of production planning.

[0006] In addition to the lag in production plans, there are also difficulties in big data collaboration between upstream and downstream manufacturing companies in the same industry. Most manufacturing companies are still in the initial stage of using their industrial software data, that is, using their own data to optimize their business, and the decision-making level manages it through the company's industrial software system. This situation does not meet the development strategy needs of the company, nor does it meet the relevant requirements of the "Guiding Opinions on Deepening the Development of Industrial Internet with "Internet + Advanced Manufacturing". Therefore, companies have the motivation and desire to use external data to optimize their business; however, for cross-enterprise industrial software docking, privacy issues are the biggest collaboration dilemma, resulting in physical and logical islands of data between companies. Therefore, the present invention proposes a production plan management optimization method based on vertical federated learning industrial software docking to solve the problems existing in the prior art. Summary of the invention

[0007] In response to the above problems, the purpose of the present invention is to propose a production plan management optimization method based on vertical federated learning industrial software docking. This method uses vertical federated learning technology to dock industrial software data, which can make full use of the enterprise's industrial software edge layer data, thereby optimizing production plans and improving production efficiency, solving the lag and collaboration dilemma of production plan management optimization of enterprises in current industrial production.

[0008] In order to achieve the purpose of the present invention, the present invention is implemented by the following technical scheme: a production planning management optimization method based on vertical federated learning industrial software docking, comprising the following steps:

[0009] Step 1: Take upstream, midstream and downstream industrial enterprises or different vertical departments of the same enterprise as the participants of the vertical federated learning of the Industrial Internet. Among the participants of the vertical federated learning of the Industrial Internet, the raw material flow, product flow and information flow in the enterprise production plan are used as learning samples to confirm the learning tasks. After the participants of the vertical federated learning of the Industrial Internet confirm the learning tasks, they retrieve and organize the relevant data in their respective enterprise industrial software systems to complete the enterprise data preparation for the federated learning software docking;

[0010] Step 2: After the enterprise data is prepared, the prepared enterprise data is first cleaned, the overlapping data is deleted, and the missing data is supplemented or strategically filled. Then, the enterprise data is dimensionless by normalization and the data dimension is unified. Then, the enterprise data is randomly sampled to achieve a balanced enterprise data category. Then, the enterprise data is binned by clustering, that is, thresholds are set to classify the sample features of the enterprise data.

[0011] Step 3: Adopt the encrypted sample ID alignment method based on the Freedman protocol, first encrypt the sample feature parameters, then align the encrypted sample feature parameters on the trusted intermediate server, and perform sample alignment between the participants of the federated learning of the sample data stream of the industrial enterprise;

[0012] Step 4: With the cooperation of the trusted intermediate server, all participants in the vertical federated learning of the industrial Internet adopt the vertical federated learning of raw material flow based on Paillier semi-homomorphic encryption. First, the key and data encryption are generated through the intermediate server, and then the enterprise local encryption features are calculated. Then, the global model is updated through the intermediate server. Finally, the enterprise updates the local model according to the global model and outputs the learning results, that is, the predicted value of the future raw material flow characteristics.

[0013] Step 5: All participants in the vertical federated learning of the Industrial Internet apply the learning results output by federated learning according to enterprise needs, design the production plan for the sample data stream in the next time period, optimize production management, optimize production plan management based on prior knowledge, and complete the production plan optimization management based on the vertical federated learning industrial software docking.

[0014] A further improvement is that in the step one, the specific steps for each participant to confirm the learning task are as follows: the upstream and midstream enterprises use different raw material flows as learning samples, and expand the features through the mining capacity, transportation capacity, primary processing capacity and order characteristics of each raw material flow mastered in the upstream enterprise's industrial software system, and the order, effective consumption and inventory characteristics of the same raw material flow mastered in the midstream enterprise's industrial software system, and optimize the allocation of raw material flows according to the learning results; the midstream and downstream enterprises use different products as learning samples to confirm the vertical federated learning tasks between the midstream and downstream enterprises.

[0015] A further improvement is that in step 2, the enterprise data is normalized as follows:

[0016]

[0017] Among them, x is the raw material flow data before normalization, and x' is the raw material flow data after normalization.

[0018] A further improvement is that in the step 2, when the enterprise data is randomly sampled, under-sampling calculations are performed on the majority classes in the enterprise data, and over-sampling calculations are performed on the minority classes in the enterprise data.

[0019] Further improvement lies in: in the step 2, the data binning processing of enterprise data is specifically as follows: the normalized enterprise data is applied with the k-means clustering algorithm, the enterprise data observations are clustered into k categories, and then the binning intervals of each sample feature are divided, and the cluster center is confirmed by the equidistance method, so as to place the enterprise data into the divided binning intervals.

[0020] Further improvements are: The steps of the k-means clustering algorithm are as follows:

[0021] A1. Select the initialized k raw material flow sample data as the initial clustering center a=a1,a2,…a k ;

[0022] A2. For each raw material flow data sample x i , calculate its Euclidean distance to each cluster center a, and put it into the category with the shortest distance;

[0023] A3. For each category a f , recalculate its cluster center based on all raw material flow data samples of this category;

[0024] A4. Repeat steps A2 and A3 until the cluster center of A3 no longer changes, and then complete the k-means clustering.

[0025] A further improvement is that: in step 3, the raw material flow data of enterprise A among the participants is assumed to be X = {x1,…,x k}, the raw material flow data of enterprise B among the participants is Y = {y1,…,y k}, the elements in the input set are all taken from the domain samples of size N, then the sample alignment steps are as follows:

[0026] B1. Enterprise A selects key parameters for a semantically secure homomorphic encryption scheme and publishes its public key and parameters. The plaintext is located in a field containing the representation of N elements of the input domain and uses interpolation to calculate the polynomial The value of , each coefficient k is encrypted by a semantically secure homomorphic encryption scheme, and the result is sent to the ciphertext set {Enc(α0),…,Enc(α k}middle;

[0027] B2. For each y∈Y, enterprise B uses the homomorphic property to calculate the encryption polynomial y, that is, calculate And choose a random value r to calculate the ciphertext Enc(rP(y)+y), and calculate Enc(rP(y)+(y|p y )) to encrypt additional payload data p y , if y is in X, then enterprise A obtains p y, to achieve the effect of encrypted sample alignment;

[0028] Enterprise B randomly arranges the k ciphertexts and sends the result back to Enterprise A;

[0029] B3. Enterprise A decrypts all received key texts and locally outputs all x∈X with corresponding decryption values. Enterprise A and Enterprise B compare and output raw material stream data to complete data alignment based on the encrypted sample ID alignment of the Freedman protocol.

[0030] The beneficial effects of the present invention are as follows: the present invention adopts a federated learning method to realize the data docking of privacy-preserving industrial software, solves the problems of difficulty in docking industrial software between existing industrial Internet enterprises and the existence of physical and logical data islands, realizes data interoperability and win-win cooperation between enterprises, and adopts a vertical federated learning method to perform joint modeling between industrial enterprises, solves the hysteresis of the demand-oriented production planning management method, and the vertical federated learning based on feature and label expansion can guide enterprises to optimize the production plan for the next time period and realize production planning management based on prior knowledge. The encrypted sample ID alignment method based on the Freedman protocol and the vertical federated learning of Paillier semi-homomorphic encryption are adopted to realize two-party joint modeling, which fully utilizes enterprise data while ensuring the privacy of enterprise data, establishes a learning model with better performance than the enterprise's individual modeling, and can better guide enterprises to optimize production planning management. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0032] Figure 1 is a schematic diagram of a method flow in an embodiment of the present invention;

[0033] Figure 2 It is a schematic diagram of the vertical federated learning process in an embodiment of the present invention. DETAILED DESCRIPTION

[0034] The following will be combined with the drawings in 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 creative work are within the scope of protection of the present invention.

[0035] See also Figure 1 , Figure 2 , this embodiment provides a production planning management optimization method based on vertical federated learning industrial software docking, including the following steps:

[0036] Step 1: Take upstream, midstream and downstream industrial enterprises or different vertical departments of the same enterprise as the participants of the vertical federated learning of the Industrial Internet. Among the participants of the vertical federated learning of the Industrial Internet, the raw material flow, product flow and information flow in the enterprise production plan are used as learning samples to confirm the learning tasks. After the participants of the vertical federated learning of the Industrial Internet confirm the learning tasks, they retrieve and organize the relevant data in their respective enterprise industrial software systems to complete the enterprise data preparation for the federated learning software docking;

[0037] This embodiment confirms the raw material flow between upstream and midstream enterprises as the learning task of the federated learning object;

[0038] The specific steps for each participant to confirm the learning task are as follows: Upstream and midstream enterprises use different raw material flows as learning samples, and expand the features through the mining capacity, transportation capacity, primary processing capacity and order characteristics of each raw material flow mastered in the industrial software system of the upstream enterprise, and the order, effective consumption and inventory characteristics of the same raw material flow mastered in the industrial software system of the midstream enterprise, and optimize the allocation of raw material flows according to the learning results. Midstream and downstream enterprises use different products as learning samples to confirm the vertical federated learning tasks between midstream and downstream enterprises;

[0039] Step 2: Due to different industrial software systems or different data collection forms, participating enterprises often have different data formats, which makes the data present in a non-standardized form. Therefore, before establishing a federated learning model, data preprocessing processes such as data cleaning, data normalization and data binning are required to complete data structuring. After the enterprise data is prepared, specifically: first clean the prepared enterprise data, delete the overlapping data, and supplement or strategically fill the missing data, then use normalization to non-dimensionalize the enterprise data and unify the data dimensions, then randomly sample the enterprise data to achieve enterprise data category balance, and then use cluster binning to bin the enterprise data, that is, set thresholds to classify the sample features of the enterprise data;

[0040] The normalization of enterprise data is as follows:

[0041]

[0042] Among them, x is the raw material flow data before normalization, and x' is the raw material flow data after normalization;

[0043] When randomly sampling enterprise data, under-sampling calculations are performed on the majority class in the enterprise data, and over-sampling calculations are performed on the minority class in the enterprise data;

[0044] The specific process of data binning for enterprise data is as follows: apply the k-means clustering algorithm to the normalized enterprise data, cluster the enterprise data observations into k categories, and then divide the binning intervals of each sample feature. Use the equidistance method to confirm the cluster center, and then put the enterprise data into the divided binning intervals. The steps of the k-means clustering algorithm are as follows:

[0045] A1. Select the initialized k raw material flow sample data as the initial clustering center a=a1,a2,…a k ;

[0046] A2. For each raw material flow data sample x i , calculate its Euclidean distance to each cluster center a, and put it into the category with the shortest distance;

[0047] A3. For each category a f , recalculate its cluster center based on all raw material flow data samples of this category;

[0048] A4, repeat steps A2 and A3 until the cluster center of A3 no longer changes, and then complete the k-means clustering;

[0049] Step 3: Adopt the encrypted sample ID alignment method based on the Freedman protocol, first encrypt the sample feature parameters, then align the encrypted sample feature parameters on the trusted intermediate server, and perform sample alignment between the participants of the federated learning of the sample data stream of the industrial enterprise;

[0050] Assume that the raw material flow data of enterprise A among all the participants is X = {x1,…,x k}, the raw material flow data of enterprise B among the participants is Y = {y1,…,y k}, the elements in the input set are all taken from the domain samples of size N, then the sample alignment steps are as follows:

[0051] B1. Enterprise A selects key parameters for a semantically secure homomorphic encryption scheme and publishes its public key and parameters. The plaintext is located in a field containing the representation of N elements of the input domain and uses interpolation to calculate the polynomial The value of , each coefficient k is encrypted through a semantically secure homomorphic encryption scheme, and the result is sent to the ciphertext set {Enc(α0),…,Enc(α k )}middle;

[0052] B2. For each y∈Y, enterprise B uses the homomorphic property to calculate the encryption polynomial y, that is, calculate And choose a random value r to calculate the ciphertext Enc(rP(y)+y), and calculate Enc(rP(y)+(y|p y )) to encrypt additional payload data p y , if y is in X, then enterprise A obtains p y , to achieve the effect of encrypted sample alignment;

[0053] Enterprise B randomly arranges the k ciphertexts and sends the result back to Enterprise A;

[0054] B3. Enterprise A decrypts all received key texts and locally outputs all x∈X with corresponding decrypted values. Enterprise A and Enterprise B compare and output raw material stream data to complete data alignment. Encrypted sample ID alignment based on Freedman protocol

[0055] Step 4: With the cooperation of the trusted intermediate server, all participants in the vertical federated learning of the industrial Internet adopt the vertical federated learning of raw material flow based on Paillier semi-homomorphic encryption. First, the key and data encryption are generated through the intermediate server, and then the enterprise local encryption features are calculated. Then, the global model is updated through the intermediate server. Finally, the enterprise updates the local model according to the global model and outputs the learning results, that is, the predicted value of the future raw material flow characteristics.

[0056] Two participating enterprises A and B use vertical federated learning of raw material flow data based on Paillier semi-homomorphic encryption under the cooperation of a trusted intermediate server. Figure 2 As shown, the specific steps are as follows:

[0057] Enterprise A has raw material flow data characteristics, such as the current cycle purchase volume of a certain raw material flow x A1 , the inventory of the previous cycle x A2 Etc. and labels such as whether the current cycle is short of y A1 etc. Enterprise B also has its raw material flow data characteristics x B With label y B1 (At least one enterprise number is required to have a label y). The two enterprises aim to jointly model through encrypted vertical federated learning, achieve industrial software docking while protecting privacy, and thus optimize production plans. This embodiment classifies this learning task as a vertical logistic regression task;

[0058] After homomorphic encryption encrypts plaintext into ciphertext, calculations can be performed between ciphertexts, and the calculation results are equal to the results of the same calculations on the plaintext after decryption. Paillier semi-homomorphic encryption satisfies the requirement that the addition of ciphertext and the result of the same operation on plaintext are consistent, aiming to complete encryption operations while protecting the privacy of plaintext.

[0059] Since semi-homomorphic encryption does not support exponential encryption calculations, the loss function is approximately taken as:

[0060]

[0061] The gradient is approximately taken as:

[0062]

[0063] C1, the intermediate server C generates a homomorphic encryption key group and distributes the public key to the participating enterprises A and B. The public key is used by the participating enterprises to encrypt the local learning data and obtain the encrypted features [[x A1 ]],[[x A2 ]],[[x A3 ]] and [[x B1 ]],[[x B2 ]],[[x B3 ]]wait;

[0064] C2, Enterprise A will encrypt the features [[x A1 ]],[[x A2 ]],[[x A3 ]] and other data are sent to enterprise B, which will compare the encrypted features of enterprise A with the local encrypted features [[x B1 ]],[[x B2 ]],[[x B3 ]] add the ciphertexts, and according to the Paillier semi-homomorphic encryption property [[u]]+[[v]]=[[u+v]], we get the ciphertext feature set [[x A1 +x A2 +…+x B1 +x B2 +…]], Enterprise B calculates based on the ciphertext feature set And combine local features to calculate the local gradient

[0065] Company B will Sent to enterprise A, which calculates the local gradient based on the local gradient

[0066] Encrypted loss function where [[yω T x]] has been We get the last term [[(ω T x) 2 ]] is calculated as Among them, due to the Paillier semi-homomorphic encryption property n[[u]]=[[nu]], so [[x A xB ]] item can be equivalent to x on the B side B [[x A ]], in summary, in Enterprise A, [[x A ]]After being sent to enterprise B, the operation of the encrypted loss function [[l]] can be completed at enterprise B;

[0067] After completing the local calculation, companies A and B upload the encrypted local gradients and loss function values ​​to the intermediate server;

[0068] C3, the intermediate server uses the private key to decrypt the data uploaded by each participating enterprise, and then and Get the global gradient and update the global model;

[0069] C4, the intermediate server proportionally sends the new and The data are sent to A and B respectively. The enterprises update their local models according to the global model and repeat steps C1, C2, C3 and C4 until the intermediate server determines that the global model converges and sends a termination signal to the participating enterprises.

[0070] Step 5: All participants in the vertical federated learning of the Industrial Internet apply the learning results output by federated learning according to enterprise needs, design the production plan for the sample data stream in the next time period, optimize production management, optimize production plan management based on prior knowledge, and complete the production plan optimization management based on the vertical federated learning industrial software docking.

[0071] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A production planning management optimization method based on vertical federated learning industrial software docking, characterized in that: The following steps are involved: Step 1: Take upstream, midstream and downstream industrial enterprises or different vertical departments of the same enterprise as the participants of the vertical federated learning of the Industrial Internet. Among the participants of the vertical federated learning of the Industrial Internet, the raw material flow, product flow and information flow in the enterprise production plan are used as learning samples to confirm the learning tasks. After the participants of the vertical federated learning of the Industrial Internet confirm the learning tasks, they retrieve and organize the relevant data in their respective enterprise industrial software systems to complete the enterprise data preparation for the federated learning software docking; Step 2: After the enterprise data is prepared, the prepared enterprise data is first cleaned, the overlapping data is deleted, and the missing data is supplemented or strategically filled. Then, the enterprise data is dimensionless by normalization and the data dimension is unified. Then, the enterprise data is randomly sampled to achieve a balanced enterprise data category. Then, the enterprise data is binned by clustering, that is, thresholds are set to classify the sample features of the enterprise data. Step 3: Adopt the encrypted sample ID alignment method based on the Freedman protocol, first encrypt the sample feature parameters, then align the encrypted sample feature parameters on the trusted intermediate server, and perform sample alignment between the participants of the federated learning of the sample data stream of the industrial enterprise; Step 4: With the cooperation of the trusted intermediate server, all participants in the vertical federated learning of the industrial Internet adopt the vertical federated learning of raw material flow based on Paillier semi-homomorphic encryption. First, the key and data encryption are generated through the intermediate server, and then the enterprise local encryption features are calculated. Then, the global model is updated through the intermediate server. Finally, the enterprise updates the local model according to the global model and outputs the learning results, that is, the predicted value of the future raw material flow characteristics. Step 5: All participants in the vertical federated learning of the Industrial Internet apply the learning results output by federated learning according to enterprise needs, design the production plan for the sample data stream in the next time period, optimize production management, optimize production plan management based on prior knowledge, and complete the production plan optimization management based on the vertical federated learning industrial software docking.

2. The production planning management optimization method based on vertical federated learning industrial software docking according to claim 1 is characterized by: In the step one, the specific steps for each participant to confirm the learning task are as follows: the upstream and midstream enterprises use different raw material flows as learning samples, and expand the features through the mining capacity, transportation capacity, primary processing capacity and order characteristics of each raw material flow mastered in the upstream enterprise's industrial software system, and the order, effective consumption and inventory characteristics of the same raw material flow mastered in the midstream enterprise's industrial software system, and optimize the allocation of raw material flows according to the learning results; the midstream and downstream enterprises use different products as learning samples to confirm the vertical federated learning tasks between the midstream and downstream enterprises.

3. The production planning management optimization method based on vertical federated learning industrial software docking according to claim 1 is characterized by: In the step 2, the normalization method of enterprise data is as follows: Among them, x is the raw material flow data before normalization, and x' is the raw material flow data after normalization.

4. The production planning management optimization method based on vertical federated learning industrial software docking according to claim 1 is characterized by: In the step 2, when the enterprise data is randomly sampled, under-sampling calculations are performed on the majority classes in the enterprise data, and over-sampling calculations are performed on the minority classes in the enterprise data.

5. The production planning management optimization method based on vertical federated learning industrial software docking according to claim 1 is characterized by: In the step 2, the enterprise data is binned as follows: the normalized enterprise data is subjected to the k-means clustering algorithm, the enterprise data observations are clustered into k categories, and then the bin intervals of each sample feature are divided, and the cluster center is confirmed by the equidistance method, so that the enterprise data is placed in the divided bin intervals.

6. The production planning management optimization method based on vertical federated learning industrial software docking according to claim 5 is characterized by: The steps of k-means clustering algorithm are as follows: A1. Select the initialized k raw material flow sample data as the initial clustering center a=a1,a2,…a k ; A2. For each raw material flow data sample x i , calculate its Euclidean distance to each cluster center a, and put it into the category with the shortest distance; A3. For each category a f , recalculate its cluster center based on all raw material flow data samples of this category; A4. Repeat steps A2 and A3 until the cluster center of A3 no longer changes, and then complete the k-means clustering.

7. The production planning management optimization method based on vertical federated learning industrial software docking according to claim 1 is characterized by: In step 3, suppose the raw material flow data of enterprise A among the participants is X = {x1,…,x k }, the raw material flow data of enterprise B among the participants is Y = {y1,…,y k }, the elements in the input set are all taken from the domain samples of size N, then the sample alignment steps are as follows: B1. Enterprise A selects key parameters for a semantically secure homomorphic encryption scheme and publishes its public key and parameters. The plaintext is located in a field containing the representation of N elements of the input domain and uses interpolation to calculate the polynomial The value of , each coefficient k is encrypted by a semantically secure homomorphic encryption scheme, and the result is sent to the ciphertext set {Enc(α0),…,Enc(α k )}middle; B2. For each y∈Y, enterprise B uses the homomorphic property to calculate the encryption polynomial y, that is, calculate And choose a random value r to calculate the ciphertext Enc(rP(y)+y), and calculate Enc(rP(y)+(y|p y )) to encrypt additional payload data p y , if y is in X, then enterprise A obtains p y , to achieve the effect of encrypted sample alignment; Enterprise B randomly arranges the k groups of ciphertexts and sends the results back to Enterprise A; B3. Enterprise A decrypts all received key texts and locally outputs all x∈X with corresponding decryption values. Enterprise A and Enterprise B compare and output raw material stream data to complete data alignment based on the encrypted sample ID alignment of the Freedman protocol.

Images