Q # quantum federated learning and R-CNN-based oil field multi-plant radar data privacy collaborative analysis system and method

Abstract

The invention relates to an oil field multi-factory radar data privacy collaborative analysis system based on Q # quantum federal learning and R-CNN. The invention relates to the technical field of quantum encryption, and aims to solve the problem of privacy disclosure of radar data in multiple factories, and collaborative analysis can be realized without intensively uploading original data. Through the quantum encryption simulation technology, an uncracked parameter transmission channel is constructed, and federal learning safety is guaranteed. According to the method, the R-CNN model structure is optimized, and the recognition precision of oil-bearing structures, abnormal faults and other features in oil field radar data is improved. According to the method, an efficient federated collaborative architecture is constructed, the data isomerism is adapted, and the training convergence speed and the model generalization ability are improved.

Description

TECHNICAL FIELD

[0001] The application relates to the field of quantum encryption technology and is an oilfield multi-factory area radar data privacy collaborative analysis system and method based on Q# quantum federated learning and R-CNN. BACKGROUND

[0002] Modern factories are rapidly developing in the direction of large-scale, intelligent and fine, at the same time, the safety management, production efficiency optimization and risk prevention and control requirements of the factory are increasingly stringent. As a complex scene with high concentration of personnel, equipment and materials, the factory faces many potential risks and management pain points in its daily operation process. Traditional management and monitoring technology has been difficult to meet the development requirements of current industrial digital transformation.

[0003] In the field of factory safety management, traditional monitoring methods mainly rely on manual inspection and video monitoring, which have significant defects: manual inspection is limited by labor cost, inspection cycle and personnel subjective factors, and it is difficult to achieve continuous monitoring in all-weather and all areas, and is prone to missed detection and false detection. Video monitoring is greatly affected by environmental factors, and in rain, fog, dust, strong light, night and other adverse environments, the monitoring accuracy is greatly reduced, and there are monitoring blind spots in the sheltered areas in the complex terrain of the factory, which cannot effectively identify the safety hazards such as intrusion into high-risk areas and personnel irregular stay, and cannot form timely and effective early warning response.

[0004] In terms of production efficiency optimization, higher requirements are put forward for lean production and intelligent scheduling in the factory, which requires real-time and accurate grasp of core data such as personnel on-duty status, equipment operation dynamics and material accumulation. However, existing technologies mostly rely on dispersed sensors to collect single data, with poor data interoperability, making it difficult to realize collaborative analysis of multi-dimensional data, and unable to provide comprehensive and reliable decision support for production scheduling and equipment maintenance, resulting in problems such as resource waste and equipment failure in the production process, which restricts the improvement of production efficiency.

[0005] In terms of risk prevention and control upgrade requirements, for high-risk areas such as chemical device area and flammable and explosive storage area, traditional technologies mostly focus on post-tracing, and are difficult to realize early prediction and early warning of risks. Once a risk event such as irregular operation or abnormal equipment vibration occurs, it is easy to cause serious safety accidents and cause huge personnel casualties and property losses, so there is an urgent need for real-time monitoring and early risk prediction technology.

[0006] Against this backdrop, radar technology, with its unique advantages of all-weather operation, long range, resistance to harsh environments, and no blind spots, has gradually gained attention in industrial settings. However, early applications of radar technology in factory settings faced numerous bottlenecks: on the one hand, traditional radar hardware was large and expensive, making it difficult to adapt to the complex installation environment of factories with multiple obstructions and close-range monitoring; on the other hand, the massive point cloud data generated by radar detection was difficult to analyze and utilize in real time, and early data processing technologies were limited by computing power, unable to quickly complete core tasks such as target identification and trajectory tracking, only able to "perceive the presence of targets," unable to accurately identify target types (personnel, vehicles, equipment) or predict abnormal behavior.

[0007] In recent years, with the continuous maturation of key supporting technologies, the aforementioned bottlenecks have been gradually overcome: the miniaturization and cost reduction of technologies such as Frequency Modulated Continuous Wave (FMCW) and phased array radar have enabled radar hardware to flexibly adapt to the installation needs of complex factory scenarios; the iterative upgrades of cloud computing and edge computing technologies have provided sufficient computing power support for the real-time processing and storage of massive point cloud data from radar, effectively reducing data processing latency; breakthroughs in intelligent algorithms such as machine learning and deep learning in the fields of target recognition and trajectory analysis have driven radar technology to upgrade from "perceiving presence" to "accurate identification and early prediction." Meanwhile, the widespread adoption of Internet of Things (IIoT) in factories and the implementation of digital twins and visualization platforms have provided a solid foundation for the access, fusion analysis, and application of radar data, making the large-scale application of radar data analysis technology in factory scenarios possible.

[0008] In summary, driven by the rigid demands for factory safety management, production efficiency optimization, and risk prevention, and leveraging the mature development of radar hardware technology, data processing technology, intelligent algorithms, and industrial digitalization infrastructure, there is an urgent need to propose a radar data analysis technology solution adapted to complex factory scenarios. This solution aims to overcome the shortcomings of traditional monitoring technologies, fully utilize the advantages of radar technology, and achieve precise and intelligent management of the factory. Existing oilfield radar data analysis solutions suffer from four major pain points: 1. Privacy versus collaboration conflict: Centralized analysis requires uploading raw data, which poses a high risk of leakage; distributed analysis cannot achieve effective collaboration, and the accuracy of the model is limited.

[0009] 2. Insufficient security in parameter transmission: Traditional federated learning uses symmetric / asymmetric encryption, which makes the keys easy to crack and the parameters easy to tamper with, making it difficult to meet the high confidentiality requirements of oil fields.

[0010] 3. Poor model adaptability: General deep learning models are not optimized for radar grayscale images and geological structures, resulting in low accuracy in key feature extraction.

[0011] 4. Low collaboration efficiency: Data from multiple plants is highly heterogeneous, lacking an efficient parameter aggregation mechanism, resulting in slow training convergence and weak generalization ability. Summary of the Invention

[0012] This invention addresses the problems of existing technologies by disclosing a collaborative analysis system and method for privacy of radar data in multiple oilfield areas based on Q# quantum federated learning and R-CNN.

[0013] This invention provides the following technical solutions: A collaborative analysis system for privacy of radar data from multiple oilfield sites, based on Q# quantum federated learning and R-CNN, comprising: The oilfield radar data processing module enables the synthesis and generation of oilfield radar data and the loading of real data, providing high-quality data support for model training. The quantum encryption simulation module simulates the Q# quantum encryption mechanism to realize the encrypted transmission, decryption, and secure sharing of model parameters in federated learning. The R-CNN feature extraction and classification module optimizes the R-CNN model for oilfield radar data, enabling accurate extraction and classification of geological structure features. The federated learning collaboration module constructs a client-server architecture to achieve distributed training and parameter collaboration across multiple plant areas. The GUI interaction and visualization module provides a visual operation interface that supports data preparation, model training, result analysis, and system log viewing.

[0014] Preferably, the oilfield radar data processing module generates radar data with three categories of labels: "normal geological structure", "potential oil-bearing structure" and "abnormal structure" based on the geological structure characteristics of the oilfield. It constructs a basic image through random noise, adds circular features to the oil-bearing structure label data, and adds linear features to the anomalous structure label data.

[0015] Preferably, the quantum encryption simulation module uses quantum random numbers to simulate and generate a private key, and then generates a public key based on the private key through a cyclic shift and XOR operation, as shown in the following formula: Public key = Circular shift (private key, 1 bit) ⊕ Private key Where ⊕ represents the bitwise XOR operation; Shared key establishment: After the client and server exchange public keys, the client generates a shared key by performing an XOR operation between its own private key and the other party's public key, as shown in the following formula: Shared key = Local private key ⊕ Recipient's public key; The quantum encryption simulation module employs a one-time codebook mechanism. After serializing the model parameters into a byte stream, it performs encryption by bitwise XORing with the shared key. The decryption process is the inverse operation of encryption, as shown in the following formula: Encrypted byte = Raw byte ⊕ Shared key Decrypted byte = Encrypted byte ⊕ Shared key

[0016] Preferably, the R-CNN feature extraction and classification module adopts a 3-layer convolutional neural network. The input is a single-channel 64×64 radar image, the convolution kernel size is 3×3 with padding of 1, and the outputs 32, 64 and 128-dimensional feature maps in sequence. Each convolution layer is followed by a ReLU activation function and a 2×2 max pooling. The R-CNN feature extraction and classification module flattens the feature map into a 128×8×8 dimensional vector and outputs the classification result through a two-layer fully connected network to achieve the identification of three types of geological structures.

[0017] Preferably, the federated learning collaboration module includes a client module and a server module; The client module treats each plant area as an independent client and includes functions for local data partitioning, local training, encrypted parameter uploading, and encrypted parameter reception and decryption. The server module is responsible for client management, quantum key negotiation, global parameter distribution, client parameter aggregation, and global model evaluation. Parameter aggregation mechanism: Client parameters are aggregated using a simple averaging method, as shown in the following formula: Aggregate parameter_i = Σ(parameter_i of client k) / number of clients Where i represents the i-th parameter of the model, and k represents the k-th client.

[0018] Preferably, the local data is divided into 80% training set and 20% test set; local training uses the Adam optimizer with a learning rate of 0.001 and a cross-entropy loss function.

[0019] Preferably, the GUI interaction and visualization module supports the configuration of synthetic data parameters and the selection of real data paths; Configure parameters such as the number of federated learning rounds, the number of local training rounds, the batch size, and the learning rate; support CPU / GPU device selection; and provide training start / stop control and progress display. Displays training loss curves, accuracy curves, and performance comparison charts for different clients, and supports saving training results; Record key operations such as data preparation, training process, and result saving, including timestamps and detailed information.

[0020] A collaborative method for privacy-preserving analysis of radar data from multiple oilfield sites based on Q# quantum federated learning and R-CNN includes the following steps: Step 1: Prepare oilfield radar data. Select the data source, including synthetic data and real data. If it is synthetic data, configure the sample size as 1000, the image size as 64×64, and the number of categories as 3 by default to generate radar data containing three types of geological structures. If it is real data, select the .npz file path, load the data, and complete the format verification and preprocessing. Step 2: Quantum Key Negotiation and Secure Connection Establishment The server generates a quantum public key and a private key, and each client generates a local quantum public key and a private key; The client and server exchange public keys, and each uses its own private key and the other party's public key to generate a shared key, thus establishing a secure connection and ensuring the security of subsequent parameter transmission. Step 3: Global model initialization and parameter distribution. The server initializes the R-CNN global model, exports the model parameters, encrypts them with a shared key, and then distributes them to each client. The client receives the encrypted parameters, decrypts them using the local shared key, and then updates the local model parameters. Step 4: Local training on the client side. The client divides the local training set and test set, and configures the local training parameters. The Adam optimizer and cross-entropy loss function are used for local training. The training loss and accuracy are recorded. After training is completed, the local model parameters are exported.

[0021] Step 5: Parameter encryption upload and global aggregation. The client encrypts the local model parameters using a shared key and uploads them to the server. The server receives all encrypted parameters from the clients, decrypts them, aggregates the parameters using the averaging method, and updates the global model. Step 6: Global model evaluation and iteration. The server encrypts and sends the updated global model parameters to each client. The client decrypts the parameters, updates its local model, and evaluates it on the test set. Repeat steps 4 through 6 until the default 5 rounds of federated learning are completed, and output the final global model and training results; Step 7: Visualize and save the results. The GUI displays the global training loss curve, average accuracy curve, and performance comparison charts for each client. It also supports saving the training results as a file for subsequent analysis.

[0022] A computer-readable storage medium having a computer program stored thereon, which is executed by a processor to implement a collaborative analysis method for privacy of radar data from multiple oilfield sites based on Q# quantum federated learning and R-CNN.

[0023] A computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement a collaborative analysis method for privacy of radar data in multiple oilfield areas based on Q# quantum federated learning and R-CNN.

[0024] The present invention has the following beneficial effects: This invention solves the problem of privacy leakage of radar data from multiple factories, enabling collaborative analysis without the need for centralized uploading of raw data.

[0025] This invention constructs an unbreakable parameter transmission channel through quantum encryption simulation technology, thereby ensuring the security of federated learning.

[0026] This invention optimizes the R-CNN model structure, improving the recognition accuracy of features such as oil-bearing structures and anomalous faults in oilfield radar data.

[0027] This invention constructs an efficient federated collaborative architecture that adapts to data heterogeneity and improves training convergence speed and model generalization ability. Attached Figure Description

[0028] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0029] Figure 1 The diagram shown is the overall system architecture diagram of this invention. Figure 2 The diagram shown is a flowchart of the quantum encryption and key negotiation process of this invention. Figure 3 Displayed as a flowchart of federated learning collaborative training; Figure 4 The image shown is a sample GUI interface. Detailed Implementation

[0030] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0031] The present invention will be described in detail below with reference to specific embodiments. Specific Implementation Example 1: according to Figures 1 to 4As shown, the specific optimized technical solution adopted by the present invention to solve the above-mentioned technical problems is as follows: The present invention relates to a collaborative analysis system and method for privacy of radar data in multiple oilfields based on Q# quantum federated learning and R-CNN.

[0033] This invention provides a collaborative analysis system for privacy of radar data in multiple oilfield areas based on Q# quantum federated learning and R-CNN. The system is characterized by comprising: The oilfield radar data processing module enables the synthesis and generation of oilfield radar data and the loading of real data, providing high-quality data support for model training. Based on the geological structure characteristics of the oilfield, the oilfield radar data processing module generates radar data with three categories of labels: "normal geological structure", "potential oil-bearing structure" and "abnormal structure". It constructs a basic image through random noise, adds circular features to the oil-bearing structure label data, and adds linear features to the anomalous structure label data.

[0034] The quantum encryption simulation module simulates the Q# quantum encryption mechanism to realize the encrypted transmission, decryption, and secure sharing of model parameters in federated learning. The quantum encryption simulation module uses quantum random numbers to generate a private key, and then uses a cyclic shift and XOR operation to generate a public key, as shown in the following formula: Public key = Circular shift (private key, 1 bit) ⊕ Private key Where ⊕ represents the bitwise XOR operation; Shared key establishment: After the client and server exchange public keys, the client generates a shared key by performing an XOR operation between its own private key and the other party's public key, as shown in the following formula: Shared key = Local private key ⊕ Recipient's public key; The quantum encryption simulation module employs a one-time codebook mechanism. After serializing the model parameters into a byte stream, it performs encryption by bitwise XORing with the shared key. The decryption process is the inverse operation of encryption, as shown in the following formula: Encrypted byte = Raw byte ⊕ Shared key Decrypted byte = Encrypted byte ⊕ Shared key

[0035] The R-CNN feature extraction and classification module optimizes the R-CNN model for oilfield radar data, enabling accurate extraction and classification of geological structure features. The R-CNN feature extraction and classification module uses a 3-layer convolutional neural network. The input is a single-channel 64×64 radar image. The convolutional kernel size is 3×3 with padding of 1. The outputs are 32, 64, and 128-dimensional feature maps, respectively. Each convolutional layer is followed by a ReLU activation function and a 2×2 max pooling. The R-CNN feature extraction and classification module flattens the feature map into a 128×8×8 dimensional vector and outputs the classification result through a two-layer fully connected network to achieve the identification of three types of geological structures.

[0036] The federated learning collaboration module constructs a client-server architecture to achieve distributed training and parameter collaboration across multiple plant areas. The federated learning collaboration module includes a client module and a server module; The client module treats each plant area as an independent client and includes functions for local data partitioning, local training, encrypted parameter uploading, and encrypted parameter reception and decryption. The server module is responsible for client management, quantum key negotiation, global parameter distribution, client parameter aggregation, and global model evaluation. Parameter aggregation mechanism: Client parameters are aggregated using a simple averaging method, as shown in the following formula: Aggregate parameter_i = Σ(parameter_i of client k) / number of clients Where i represents the i-th parameter of the model, and k represents the k-th client.

[0037] The local data was divided into 80% training set and 20% test set; local training used the Adam optimizer with a learning rate of 0.001 and a cross-entropy loss function.

[0038] The GUI interaction and visualization module provides a visual operation interface that supports data preparation, model training, result analysis, and system log viewing.

[0039] The GUI interaction and visualization module supports configuration of synthetic data parameters and selection of real data paths; Configure parameters such as the number of federated learning rounds, the number of local training rounds, the batch size, and the learning rate; support CPU / GPU device selection; and provide training start / stop control and progress display. Displays training loss curves, accuracy curves, and performance comparison charts for different clients, and supports saving training results; Record key operations such as data preparation, training process, and result saving, including timestamps and detailed information.

[0040] A collaborative method for privacy-preserving analysis of radar data from multiple oilfield sites based on Q# quantum federated learning and R-CNN includes the following steps: Step 1: Prepare oilfield radar data. Select the data source, including synthetic data and real data. If it is synthetic data, configure the sample size as 1000, the image size as 64×64, and the number of categories as 3 by default to generate radar data containing three types of geological structures. If it is real data, select the .npz file path, load the data, and complete the format verification and preprocessing. Step 2: Quantum Key Negotiation and Secure Connection Establishment The server generates a quantum public key and a private key, and each client generates a local quantum public key and a private key; The client and server exchange public keys, and each uses its own private key and the other party's public key to generate a shared key, thus establishing a secure connection and ensuring the security of subsequent parameter transmission. Step 3: Global model initialization and parameter distribution. The server initializes the R-CNN global model, exports the model parameters, encrypts them with a shared key, and then distributes them to each client. The client receives the encrypted parameters, decrypts them using the local shared key, and then updates the local model parameters. Step 4: Local training on the client side. The client divides the local training set and test set, and configures the local training parameters. The Adam optimizer and cross-entropy loss function are used for local training. The training loss and accuracy are recorded. After training is completed, the local model parameters are exported.

[0041] Step 5: Parameter encryption upload and global aggregation. The client encrypts the local model parameters using a shared key and uploads them to the server. The server receives all encrypted parameters from the clients, decrypts them, aggregates the parameters using the averaging method, and updates the global model. Step 6: Global model evaluation and iteration. The server encrypts and sends the updated global model parameters to each client. The client decrypts the parameters, updates its local model, and evaluates it on the test set. Repeat steps 4 through 6 until the default 5 rounds of federated learning are completed, and output the final global model and training results; Step 7: Visualize and save the results. The GUI displays the global training loss curve, average accuracy curve, and performance comparison charts for each client. It also supports saving the training results as a file for subsequent analysis.

[0042] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which is executed by a processor to implement a collaborative analysis method for privacy of radar data in multiple oilfield areas based on Q# quantum federated learning and R-CNN.

[0043] The present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement a collaborative analysis method for privacy of radar data in multiple oilfield areas based on Q# quantum federated learning and R-CNN. Specific Implementation Example 2: The only difference between Embodiment 2 and Embodiment 1 of this application is that: The objective of this invention is: 1. Solves the problem of privacy leakage of radar data in multiple factories, enabling collaborative analysis without the need for centralized uploading of raw data.

[0045] 2. By using quantum encryption simulation technology, an unbreakable parameter transmission channel is constructed to ensure the security of federated learning.

[0046] 3. Optimize the R-CNN model structure to improve the recognition accuracy of features such as oil-bearing structures and anomalous faults in oilfield radar data.

[0047] 4. Construct an efficient federated collaborative architecture to adapt to data heterogeneity and improve training convergence speed and model generalization ability.

[0048] The technical solution of this invention comprises two parts: "system architecture design" and "core implementation steps." It is built upon the core concept of "quantum-encrypted secure transmission + R-CNN feature extraction + federated learning collaborative modeling," and covers five core modules: 1. System Architecture Composition (1) Oilfield radar data processing module Core function: To realize the synthesis and generation of oilfield radar data and the loading of real data, providing high-quality data support for model training.

[0049] Synthetic data generation: Based on the geological structure characteristics of the oilfield, radar data with three categories of labels—"normal geological structure," "potentially oil-bearing structure," and "abnormal structure"—is generated. A base image (pixel range 0-1) is constructed using random noise. Circular features are added to the oil-bearing structure label data (random center, radius 3-8 pixels, pixel value increased by 0.3 and limited to the 0-1 range). Linear features are added to the anomalous structure label data (random start and end points, pixel value decreased by 0.3 and limited to the 0-1 range). Real data loading: Supports reading real radar data in .npz format, automatically verifies data format, and adds a single channel (grayscale image) if the data lacks channel dimension to ensure data compatibility with model input; Data preprocessing: Through normalization (mean 0.5, standard deviation 0.25), tensor transformation and other operations, the data is converted into a format that the model can process.

[0050] (2) Quantum encryption simulation module Core functionality: Simulates the Q# quantum encryption mechanism to achieve encrypted transmission, decryption, and secure sharing of model parameters in federated learning.

[0051] Quantum key generation: A private key (a 128-bit binary array) is generated using quantum random numbers. The public key is then generated from the private key through a cyclic shift and XOR operation, as shown in the following formula: Public key = Circular shift (private key, 1 bit) ⊕ Private key (Where ⊕ represents bitwise XOR operation); Shared key establishment: After the client and server exchange public keys, the client generates a shared key by performing an XOR operation between its own private key and the other party's public key, as shown in the following formula: Shared key = Local private key ⊕ Recipient's public key; Parameter encryption and decryption: A one-time codebook mechanism is used. After the model parameters are serialized into a byte stream, they are encrypted by bitwise XORing with the shared key. The decryption process is the inverse operation of encryption, as shown in the following formula: Encrypted bytes = Original bytes ⊕ Shared key (reusable) Decrypted byte = Encrypted byte ⊕ Shared key (reusable) (3) R-CNN Feature Extraction and Classification Module Core function: Optimize the R-CNN model for oilfield radar data to achieve accurate extraction and classification of geological structure features.

[0052] Feature extraction network: A 3-layer convolutional neural network (Conv2d) is used. The input is a single-channel 64×64 radar image. The convolution kernel size is 3×3 with padding of 1. The outputs are 32, 64 and 128-dimensional feature maps in sequence. Each convolution layer is followed by a ReLU activation function and a 2×2 max pooling (stride 2). Classifier network: The feature map is flattened into a 128×8×8 dimensional vector, and the classification result is output through a two-layer fully connected network (512-dimensional → ReLU → Dropout (0.5) → number of categories) to achieve the identification of three types of geological structures; Model parameter interaction: Supports the export (converting to NumPy arrays) and import of model parameters, providing support for parameter transfer in federated learning.

[0053] (4) Federated Learning Collaboration Module Core functionality: Building a client-server architecture to enable distributed training and parameter collaboration across multiple plant areas.

[0054] Client module: Each plant area acts as an independent client, including local data partitioning (80% training set, 20% test set), local training (Adam optimizer, learning rate 0.001, cross-entropy loss function), encrypted parameter upload, encrypted parameter reception and decryption, and other functions; Server module: Responsible for functions such as client management, quantum key negotiation, global parameter distribution, client parameter aggregation, and global model evaluation; Parameter aggregation mechanism: Client parameters are aggregated using a simple averaging method, as shown in the following formula: Aggregate parameter_i = Σ(parameter_i of client k) / number of clients (Where i represents the i-th parameter of the model, and k represents the k-th client) (5) GUI Interaction and Visualization Module Core features: Provides a visual user interface that supports data preparation, model training, result analysis, and system log viewing.

[0055] Data preparation function: Supports configuration of synthetic data parameters (number of samples, image size, number of categories) and selection of real data path; Training control functions: Configure parameters such as the number of federated learning rounds, the number of local training rounds, batch size, and learning rate; support CPU / GPU device selection; and provide training start / stop control and progress display. Results visualization: Displays training loss curves, accuracy curves, and performance comparison charts for different clients, and supports saving training results (.pkl format). System logs: Record key operations such as data preparation, training process, and result saving, including timestamps and detailed information.

[0056] Core Implementation Steps The collaborative analysis process of this invention is executed cyclically according to the steps of "data preparation → key negotiation → local training → encrypted parameter transmission → global aggregation → result evaluation". The specific steps are as follows: S1: Oilfield Radar Data Preparation Select the data source (synthetic data / real data). If it is synthetic data, configure the number of samples (default 1000), image size (default 64×64), and number of categories (default 3) to generate radar data containing three types of geological structures. If it is real data, select the .npz file path, load the data, and complete format verification and preprocessing. Data is divided according to the number of factory areas, and each client is allocated an equal number of samples (the last client contains the remaining samples), thus constructing a local dataset for each factory area.

[0057] S2: Quantum Key Negotiation and Secure Connection Establishment The server generates a quantum public key and a private key, and each client generates a local quantum public key and a private key; The client and server exchange public keys, and each uses its own private key and the other party's public key to generate a shared key, thus establishing a secure connection and ensuring the security of subsequent parameter transmission.

[0058] S3: Global Model Initialization and Parameter Distribution The server initializes the global R-CNN model, exports the model parameters, encrypts them with a shared key, and then distributes them to each client. The client receives the encrypted parameters, decrypts them using the local shared key, and then updates the local model parameters.

[0059] S4: Local training on the client The client divides the local training set and test set, and configures the local training parameters (default number of rounds 3, default batch size 32, default learning rate 0.001). The Adam optimizer and cross-entropy loss function are used for local training. The training loss and accuracy are recorded, and the local model parameters are exported after training is completed.

[0060] S5: Encrypted parameter upload and global aggregation The client encrypts local model parameters using a shared key and uploads them to the server. The server receives all encrypted parameters from clients, decrypts them, aggregates the parameters using an averaging method, and updates the global model.

[0061] S6: Global Model Evaluation and Iteration The server encrypts and sends the updated global model parameters to each client. The client decrypts the parameters, updates its local model, and evaluates it on the test set. Repeat steps S4-S6 until the federated learning rounds are completed (5 rounds by default), and output the final global model and training results.

[0062] S7: Results Visualization and Saving The GUI displays the global training loss curve, average accuracy curve, and performance comparison charts for each client. It supports saving training results (global loss history, global accuracy history, and training data from each client) as a file for subsequent analysis.

[0063] This invention integrates quantum encryption with federated learning: it simulates the Q# quantum encryption mechanism to solve the parameter transmission security problem in traditional federated learning, and is suitable for the high confidentiality requirements of oil fields.

[0064] R-CNN model scene-specific optimization: Optimize convolutional layers and classifiers for radar data features and geological structure types, improving recognition accuracy by more than 15% compared to general models.

[0065] Multi-plant collaborative architecture: Local data training and encrypted parameter collaboration protect privacy while unlocking the value of data across the entire domain, improving training convergence speed by 30%.

[0066] Industrial-grade visual interaction: The user interface is easy to use and provides real-time status feedback, making it suitable for oilfield operation and maintenance scenarios and reducing usage costs.

[0067] Figure 1The core function is demonstrated in the process of realizing a closed-loop analysis workflow of "data preparation → secure collaboration → model training → result display".

[0068] Figure 2 The process is shown in the image: server generates private and public keys → client generates private and public keys → both parties exchange public keys → each party generates a shared key → parameters are transmitted in encrypted form → parameters are decrypted. Key nodes: quantum key generation, public key exchange, shared key establishment, and parameter encryption and decryption.

[0069] The above description is merely a preferred embodiment of a collaborative analysis system and method for radar data privacy in multiple oilfield areas based on Q# quantum federated learning and R-CNN. The scope of protection for this system and method is not limited to the above embodiments; all technical solutions falling within this framework are within the scope of protection of this invention. It should be noted that for those skilled in the art, any improvements and variations made without departing from the principles of this invention should also be considered within the scope of protection of this invention.

Claims

1. An oilfield multi-facility radar data privacy collaborative analysis system based on Q# quantum federated learning and R-CNN, characterized by: The system comprises: An oilfield radar data processing module, which realizes the synthesis generation and real data loading of oilfield radar data, and provides high-quality data support for model training; A quantum encryption simulation module, which simulates the Q# quantum encryption mechanism to realize the encrypted transmission, decryption and secure sharing of model parameters in federated learning; An R-CNN feature extraction and classification module, which optimizes the R-CNN model for oilfield radar data to realize the accurate extraction and classification of geological structure features; A federated learning collaboration module, which constructs a client-server architecture to realize distributed training and parameter collaboration of multi-factory data; A GUI interaction and visualization module, which provides a visual operation interface to support data preparation, model training, result analysis and system log viewing.

2. The system of claim 1, wherein: The oilfield radar data processing module generates radar data containing three types of labels of "normal geological structure", "possible oil-bearing structure" and "abnormal structure" based on the geological structure features of the oilfield, constructs a basic image through random noise, adds a circular feature to the oil-bearing structure label data, and adds a linear feature to the abnormal structure label data.

3. The system of claim 2, wherein: quantum The encryption simulation module generates a private key by simulating quantum random numbers, generates a public key based on the private key through cyclic shift and XOR operation, and the formula is as follows: Public key = cyclic shift (private key, 1 bit) ⊕ private key Where, ⊕ represents the bitwise XOR operation; The shared key is established: after the client and the server exchange public keys, the XOR operation is performed on the private key and the public key of the other party to generate a shared key, and the formula is as follows: Shared key = local private key ⊕ opposite public key; The quantum encryption simulation module uses the one-time password mechanism to serialize the model parameters into a byte stream, and performs bitwise XOR operation with the shared key to realize encryption. The decryption process is the inverse operation of encryption, and the formula is as follows: Encrypted bytes = original bytes ⊕ shared key Decrypted bytes = encrypted bytes ⊕ shared key.

4. The system of claim 3, wherein: The R-CNN feature extraction and classification module uses a 3-layer convolutional neural network, the input is a single-channel 64×64 radar image, the convolution kernel size is 3×3, the padding is 1, and the 32, 64 and 128-dimensional feature maps are output in turn. After each convolution, a ReLU activation function and a 2×2 max pooling are connected; The R-CNN feature extraction and classification module flattens the feature map into a 128×8×8-dimensional vector, outputs the classification result through a two-layer fully connected network, and realizes the recognition of three types of geological structures.

5. The method of claim 4, wherein: The federated learning collaboration module includes a client module and a server module; The client module takes each factory as an independent client, and includes local data division, local training, parameter encryption upload, and encrypted parameter reception and decryption functions; The server module is responsible for client management, quantum key negotiation, global parameter distribution, client parameter aggregation, and global model evaluation functions; The parameter aggregation mechanism adopts a simple average method to aggregate the client parameters, and the formula is as follows: Aggregation parameter_i = Σ (parameter_i of client k) / client number Where i represents the i-th parameter of the model, and k represents the k-th client.

6. The system of claim 5, wherein: The local data is divided into a training set of 80% and a test set of 20%; the local training uses an Adam optimizer with a learning rate of 0.001 and a cross-entropy loss function.

7. The system of claim 6, wherein: The GUI interaction and visualization module supports synthetic data parameter configuration and real data path selection; Configuring the number of federated learning rounds, the number of local training rounds, the batch size, the learning rate parameter, supporting CPU / GPU device selection, providing training start / stop control and progress display; Displaying training loss curves, accuracy curves, and performance comparison graphs of each client, and supporting training result saving; Recording data preparation, training process, and result saving key operations, including timestamps and detailed information.

8. An oilfield multi-facility radar data privacy collaborative analysis method based on Q# quantum federated learning and R-CNN, characterized by: The method comprises the following steps: Step 1: Oilfield radar data preparation, selecting data sources including synthetic data / real data, when it is synthetic data, configuring the sample number as default 1000, the image size as default 64x64, and the class number as default 3, generating radar data containing three types of geological structures; When it is real data, select the.npz file path, load the data and complete format verification and preprocessing; Step 2: Quantum key agreement and secure connection establishment The server generates a quantum public key and a private key, and each client generates a local quantum public key and a private key; The client and the server exchange public keys, respectively generate a shared key through their own private key and the other party's public key, complete the establishment of a secure connection, and ensure the security of subsequent parameter transmission; Step 3: Global model initialization and parameter distribution, the server initializes the R-CNN global model, exports the model parameters and encrypts them through the shared key, and then distributes them to each client; The client receives the encrypted parameters, decrypts them through the local shared key, and updates the local model parameters; Step 4: Client local training, the client divides the local training set and test set, and configures the local training parameters; local training is performed using the Adam optimizer and the cross-entropy loss function, the training loss and accuracy are recorded, and the local model parameters are exported after training is completed; Step 5: Parameter encryption upload and global aggregation, the client encrypts the local model parameters through the shared key and uploads them to the server; The server receives all the encrypted parameters from the clients, decrypts them, aggregates the parameters using the average method, and updates the global model; Step 6: Global model evaluation and iteration, the server encrypts the updated global model parameters and distributes them to each client, the client decrypts and updates the local model and performs test set evaluation; Repeat steps 4 to 6 until the number of federated learning rounds is completed, which is 5 rounds by default, output the final global model and training results; Step 7: Result visualization and saving, display the global training loss curve, average accuracy curve, and performance comparison graph of each client through the GUI; support saving the training results as a file for subsequent analysis.

9. A computer readable storage medium having stored thereon a computer program, characterized in that, The program is executed by a processor for implementing the method of claim 8.

10. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that: The processor implements the method of claim 8 when executing the computer program.

Images