Data Transmission Method for the Integration of In-Vehicle Battery Management and Vehicle Network Security

Abstract

The present invention discloses a data transmission method for the integration of in-vehicle battery management and vehicle network security, including: a sending end, data acquisition and preliminary integration to form multi-dimensional data; based on a first neural network model and the multi-dimensional data, obtaining corresponding high-dimensional features; encrypting the high-dimensional features and then transmitting them; a receiving end, based on a second neural network model and the high-dimensional features, obtaining a final result instruction; wherein, the neural network model is an algorithm model with neurons as the computing unit, and the neural network model is composed of a first neural network model and a second neural network model. The present invention combines the randomness of neurons with traditional encryption methods, introduces a correction bit and a mechanism for determining randomness parameters based on the current timestamp, and significantly improves the security of the data transmission system.

Description

Technical Field

[0001] The present invention belongs to the field of vehicle network security, and more specifically, relates to a data transmission method for the integration of in-vehicle battery management and vehicle network security. Background Art

[0002] Currently, in-vehicle battery management systems (BMS) and vehicle network security systems mainly rely on traditional encryption technologies and data transmission protocols, and ensure that data is not stolen or tampered with during transmission through hardware encryption chips, digital signatures, and dynamic key updates. At the same time, neural computing models such as deep neural networks, convolutional neural networks (CNNs), and recurrent neural networks (RNNs) have been widely used in fields such as autonomous driving and energy management for real-time data analysis and decision support. However, there are still vulnerabilities in these systems in terms of data protection: once a hacker obtains the encryption key or cracks the encryption algorithm, they may affect the system's determination by tampering with the feature mapping, and traditional protection measures are difficult to provide comprehensive protection against the variability introduced by the randomness of neural network models.

[0003] During the process of autonomous driving, the vehicle obtains real-time images of the surrounding area through image sensors, generates specific control instructions in combination with vehicle status information, and then adjusts the charging and discharging strategies of the battery module to achieve efficient energy management. At the same time, the battery control unit collects key parameters such as temperature, voltage, current, internal resistance, SOC, and SOH, and transmits them to the central control unit and remote monitoring platform in real time through the vehicle network for fault diagnosis and safety interlock protection. In these scenarios, data transmission not only needs to ensure high speed and low latency, but also must ensure that the data is not maliciously tampered with or intercepted during transmission, so as not to affect autonomous driving decisions and battery management strategies. Summary of the Invention

[0004] To solve the deficiencies in the prior art, the purpose of the present invention is to address the above-mentioned defects, and further propose a data transmission method for the integration of in-vehicle battery management and vehicle network security.

[0005] The present invention adopts the following technical solutions.

[0006] The first aspect of the present invention discloses a data transmission method for the integration of in-vehicle battery management and vehicle network security, including:

[0007] At the sending end, data collection and preliminary fusion are performed to form multi-dimensional data;

[0008] Based on the first neural network model and the multi-dimensional data, the corresponding high-dimensional features are obtained;

[0009] The high-dimensional features are encrypted and then transmitted;

[0010] The receiving end obtains the final result instruction based on the second neural network model and the high-dimensional features;

[0011] Among them, the neural network model is an algorithm model with neurons as the computing unit, and the neural network model is composed of a first neural network model and a second neural network model.

[0012] Furthermore, the algorithm model with neurons as the computing unit is CNN, RNN or DNN.

[0013] Furthermore, the result instruction is based on the data generated by the vehicle in real time and is obtained according to the output of the neural network model.

[0014] Furthermore, all neurons in the neural network algorithm are regarded as all nodes of a graph, and the connection relationship of neurons constitutes the directed edges of the graph, then the high-dimensional feature is the cut directed edge.

[0015] Furthermore, the high-dimensional feature is the feature map output by the middle layer in the neural network model.

[0016] Furthermore, in the neural network model, the input layer and the middle layer are set at the sending end, and the output layer is set at the receiving end.

[0017] Furthermore, the high-dimensional feature also includes a check bit; it is used for the receiving end to verify whether the data is correct; among them, the sending end includes an output layer, and the check bit is obtained based on the high-dimensional feature and all the random parameters in the collapsed output layer.

[0018] The second aspect of the present invention discloses a data transmission system integrating in-vehicle battery management and vehicle network security, including: an in-vehicle battery management system and a vehicle network security system;

[0019] When the in-vehicle battery management system and the vehicle network security system exchange data, they act as the sending end and the receiving end for each other;

[0020] The sending end is used for data acquisition and preliminary fusion to form multi-dimensional data; and based on the first neural network model and the multi-dimensional data, obtaining the corresponding high-dimensional features; and encrypting the high-dimensional features and then transmitting them to the receiving end;

[0021] The receiving end is used for obtaining the final result instruction based on the second neural network model and the high-dimensional features;

[0022] Among them, the neural network model is an algorithm model with neurons as the computing unit, and the neural network model is composed of a first neural network model and a second neural network model.

[0023] The third aspect of the present invention discloses a terminal, including a processor and a storage medium; characterized in that:

[0024] The storage medium is used to store instructions;

[0025] The processor is configured to operate according to the instructions to execute the steps of the method described in the first aspect.

[0026] A fourth aspect of the present invention discloses a computer-readable storage medium having a computer program stored thereon, characterized in that when the program is executed by a processor, the steps of the method described in the first aspect are implemented.

[0027] The beneficial effects of the present invention are as follows. Compared with the prior art, the present invention has the following advantages:

[0028] By combining the randomness of neurons with traditional encryption methods and introducing a mechanism for correcting bits and determining randomness parameters based on the current timestamp, the present invention significantly improves the security of the data transmission system. At the sending end, not only the feature map data is transmitted, but also the correction bit information is attached, enabling the receiving end to verify the data integrity and ensuring that even if there is randomness in the feature map, the actual operation result can be accurately restored. In this way, when cracking the encrypted data, hackers not only need to perform brute-force calculations, but also have to simulate the operation process of the entire neural network, greatly increasing the cracking difficulty and enhancing the protection ability of the entire vehicle battery management and vehicle network security system, effectively ensuring the safe and stable operation of autonomous driving and energy management. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 It is a schematic diagram of a neural network model under data transmission integrating vehicle battery management and vehicle network security according to an embodiment of the present invention.

[0030] Figure 2 It is a schematic diagram of a data transmission system integrating vehicle battery management and vehicle network security according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0031] The following further describes the present application with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and should not be used to limit the protection scope of the present application.

[0032] In the present invention, a vehicle battery management system (BMS) at least includes: a battery module, a battery control unit (BCU), a communication interface, and other modules. The battery module constitutes the basic unit of the battery pack, and each battery cell has a certain voltage and capacity. Multiple cells are combined into a module according to a certain structure and then integrated to form the overall battery pack; the battery control unit, as the core processor, is responsible for collecting sensor data, running battery management algorithms, judging the states of each unit, and executing protection and balancing strategies; the communication interface transmits battery data to the vehicle network security system through communication protocols such as CAN, LIN, and Ethernet to achieve data interaction and collaborative control.

[0033] The vehicle network security system at least includes: an in-vehicle gateway module, an in-vehicle control module, etc. The in-vehicle gateway module serves as the data center among various in-vehicle subsystems, realizing the interconnection and interoperability of different networks (such as the powertrain network and the infotainment network), and at the same time providing basic data filtering functions; the in-vehicle control module is used to generate precise control instructions to optimize the overall operation state of the vehicle, and at the same time encrypts the data and ensures the secure cooperation among subsystems. This module can, during the autonomous driving process, identify information such as roads, obstacles, and pedestrians based on real-time image data, and combine the current power and energy state of the vehicle to intelligently adjust the control strategy to achieve coordinated management of battery charging and discharging, energy recovery, and power distribution. In addition, the in-vehicle control module uses hardware or software encryption methods to encrypt the data transmission inside the vehicle and remotely, ensuring that the data is not stolen or tampered with during the transmission process; at the same time, it realizes the identity authentication among various terminal devices.

[0034] The scenarios of data transmission discussed in the present invention can be: in the first scenario, during the autonomous driving process, based on the image sensor, real-time pictures around the vehicle can be obtained, and the in-vehicle control module can analyze these real-time pictures to obtain battery control instructions, so as to control the battery module. The battery control instructions can be, for example: "Adjust the battery output power to 85% and activate the energy regeneration braking". In the second scenario, the battery control unit can acquire key parameters such as temperature, voltage, current, internal resistance, SOC (state of charge), SOH (state of health), etc., and generate data such as battery abnormal data, fault codes, and alarm logs based on the battery control unit.

[0035] The first scenario is essentially to send the result instruction from the vehicle network security system to the BMS, while the second scenario is essentially to transmit the result instruction from the BMS to the vehicle network security system. In the prior art, its steps can all be summarized as steps 1 to 3.

[0036] Step 1: At the sending end, data collection and preliminary fusion are performed to form multi-dimensional data.

[0037] Inside the sending end, real-time status data is collected through various sensors. Usually, clock synchronization technology is required to ensure that the time of each data source is consistent. The edge node is used to perform preliminary filtering and correction on the data to solve the security vulnerabilities caused by data loss or errors.

[0038] Step 2: Multi-dimensional real-time analysis is performed, and the corresponding result instruction is output.

[0039] Step 2 is essentially to select a specific algorithm according to different application scenarios, and then output the result instruction to be sent to the receiving end.

[0040] For example, the first scenario involves the processing of image data. Generally, neural network algorithms such as CNN, RNN, and DNN can be selected, while in the second scenario, algorithms such as decision trees, SVM, and K-NN are often preferred.

[0041] It can be understood that step 1 is actually operations such as data cleaning, normalization, and defect removal. And step 2 is actually based on the trained algorithm model and the multi-dimensional data in step 1 to obtain the corresponding output. These are all common knowledge and will not be elaborated further. In the first scenario, the multi-dimensional data can be the image data obtained by each image sensor, and the corresponding result instruction can be a battery control instruction. In the second scenario, the multi-dimensional data can be key parameters such as temperature, voltage, current, internal resistance, SOC (state of charge), and SOH (state of health), and the corresponding result instructions can be data such as battery anomaly data, fault codes, and alarm logs.

[0042] Step 3: Before sending, the result instruction also needs to be encrypted.

[0043] Since the result instruction is based on the data generated in real time by the vehicle and is obtained according to the output of the neural network model. Therefore, considering the real-time nature of data transmission, the lightweight encryption algorithm (such as the hybrid scheme of ECC and AES) is usually adopted in step 3, which results in the encryption itself being easily brute-forced.

[0044] This paragraph selects the representative neural network algorithm CNN for introduction. CNN mainly includes: the input layer, the middle layer (hidden layer), and the output layer.

[0045] The input layer is used to receive the original data and transfer the data to the network in the form of a vector or tensor. Generally, the input layer does not perform complex operations, but is responsible for data format conversion and normalization to provide a suitable input for the subsequent layers.

[0046] The middle layer is usually the core of CNN, and its main function is to perform feature extraction and data representation conversion. The hidden layer can be further divided into different types of layers, and each layer plays a specific role. These layers can usually be: convolutional layer, pooling layer, fully connected layer, etc.

[0047] The output layer is used to map the features finally extracted by the hidden layer to the output space of the target task. The structure and activation function of the output layer depend on the specific task or specific application scenario.

[0048] It is not difficult to understand that the basic structures of other neural network algorithms (such as: RNN, DNN) are the same, and the only difference is the way of obtaining feature mapping, usually the logic of the middle layer is different.

[0049] Assume that the algorithm selected in step 2 is a neural network algorithm. Then, the traditional method is essentially to take the result instruction (output by the output layer), then encrypt it in step 3, and then transmit it from the sender to the receiver.

[0050] The disadvantage of this approach is that once a hacker masters the encryption rule and cracks the encryption method, they can modify the result instruction.

[0051] Since the result instruction usually has a standardized format, it is easy for a hacker to infer the specific data type corresponding to each value based on the specific values in it (usually these values have a stable fluctuation range).

[0052] For example: If the result instruction is cracked as [0.01; 0.02; 0.95; 0.01; 0.01], then it is not difficult to infer that this result instruction is essentially a prediction result containing 5 cases, among which the probability of the third type of case occurring is 95%.

[0053] It is understandable that there are duplicate characteristics in the data of the result instruction, which means that in the transmission process, in order to prevent data from being tampered with, check bits need to be added.

[0054] The process of adding check bits can be summarized as:

[0055] d2 = check(d,tm)

[0056] Where d is the result instruction, tm should usually be the current timestamp, and d2 is the result instruction containing the check bit (usually naturally including tm).

[0057] However, this approach is actually of no avail, because once a hacker cracks the encryption itself, then cracking the check bit again only linearly increases the cracking difficulty by less than twice.

[0058] Based on this, the present invention proposes a data transmission method for the integration of in-vehicle battery management and vehicle network security. The core idea is to "invisibly" integrate the data to be transmitted by the in-vehicle battery management and vehicle network security system to achieve multi-level security protection for the data to be transmitted, including: at the sender, data collection and preliminary integration to form multi-dimensional data; based on the first neural network model and multi-dimensional data, obtaining corresponding high-dimensional features; encrypting and transmitting the high-dimensional features; at the receiver, based on the second neural network model and high-dimensional features, obtaining the final result instruction; where the neural network model is an algorithm model with neurons as the computing unit, and the neural network model is composed of the first neural network model and the second neural network model.

[0059] It should be understood that the present invention is not very applicable to the second scenario. However, if neural network algorithms such as CNN and RNN are also selected for the second scenario, the present invention can also be applicable to the second scenario.

[0060] In an embodiment of the present invention, the high-dimensional feature is preferably the feature map output by the intermediate layer.

[0061] It should be noted here that the high-dimensional feature can also be selected as a certain output of the intermediate layer. However, the problem with this approach is that, on the one hand, the logic of the intermediate layer is too complex, and each neuron has a large coupling. That is to say, if the intermediate layer is analyzed into a convolutional layer, a pooling layer, a recurrent layer, etc., the internal operation logic is not necessarily serial, it can also be parallel, or a combination of serial and parallel, as Figure 1 shown. Among them, once it is parallel, it is difficult to segment the high-dimensional feature. On the other hand, in a further embodiment (see the content of the check bit below), how to control some random parameters in the intermediate layer, that is, how to cut the intermediate layer, is also an extremely complex problem.

[0062] Figure 1 In [the figure], modules 1 to 12 can respectively refer to a convolutional layer, a pooling layer, etc.

[0063] This paragraph explains "it is difficult to segment the high-dimensional feature". The "segmentation" in this sentence can be analogous to the concept of cut in graph theory. That is: if all neurons in the neural network algorithm are regarded as all nodes of a graph, and the connection relationship of neurons constitutes the directed edges of the graph, then the high-dimensional feature can only be a cut directed edge. Therefore, if there is a complex parallel relationship in the intermediate layer, there may very likely be no cut directed edge.

[0064] It can be understood that the directed edge essentially represents the input and output of the neuron, and the cut directed edge means that its single side constitutes a cut of the graph.

[0065] Specifically, in Figure 1 [the figure], the cut can be c1: {1 -> 2}; it can also be c2: {8 -> 9}, or it can be c3: {3 -> 4; 5 -> 6}, but only 1 -> 2 and 8 -> 9 are cuts formed by a single side, that is, cut directed edges. If 8 -> 9 is selected as the cut directed edge, the first neural network model consists of modules 1 to 8, and the second neural network model consists of modules 9 to 12.

[0066] It can be understood that once the high-dimensional feature is determined, then the first neural network model and the second neural network model after splitting the neural network model will necessarily also be determined.

[0067] Furthermore, the feature map output by the middle layer must be a cut directed edge; correspondingly, the first neural network model is the input layer and the middle layer, which is set at the sending end, and the second neural network model is the output layer, which is set at the receiving end.

[0068] Feature maps (which are vectors) usually have a high degree of randomness, which makes it impossible to clarify the specific meaning of each element in the feature map even if the data is cracked. Most importantly, the activation function included in the output layer is a probability model (not an invertible or non-invertible computational model like matrix operations), which makes hackers feel at a loss when using brute-force cracking methods based on matrix operations. The core of the present invention is essentially to combine the randomness of neurons with traditional encryption methods to form a multi-level security protection structure. Unless hackers completely crack the encryption method and the neural network model, it is impossible to crack the data to be transmitted.

[0069] The expression of the output layer is specifically as follows:

[0070] d = f(W * a + b)

[0071] Where f is the activation function of the output layer, a is the feature map, W is the m*n weight matrix in the output layer, and b is the bias term. It can be understood that m < n, and thus an n-dimensional feature map is implemented on an m-dimensional result instruction.

[0072] The probability model is essentially a double-edged sword. If hackers tamper with the feature map, the receiving end has no way of knowing. Therefore, the data sent by the sending end should not only include the feature map itself, but also include a correction bit. When giving the correction bit, the parameters of randomness should be determined. For example, based on the current timestamp, the parameters of randomness can be collapsed to a specific value.

[0073] Collapse means that a parameter of randomness is determined to be a fixed value. It can be understood that in a neural network model, any randomized parameter must be determined to be a fixed value when finally participating in network calculations to ensure the uniqueness of the output result.

[0074] Assume that the feature map a is represented as the following vector [a1, a2, ..., an]. Then, the original intention of the correction bit can be understood as a parity check bit or error correction coding (to prevent data transmission errors), but in the scenario of the present invention, it can also essentially prevent data from being tampered with.

[0075] It can be understood that whether it is data transmission errors or being tampered with, they are essentially the same, that is, the data transmitted is not the correct and true data.

[0076] To ensure the correctness and authenticity of the transmitted data, the transmitted feature map should actually also include the timestamp tm and the c check bits.

[0077] Here, it should be noted that c can contain multiple elements (for example: c1, c2, c3...), and its positions can be scattered in the feature map. That is to say, the final transmitted feature map may look like this:

[0078] [c1, a1, c2, a2, a3, c3, a4,..., an, tm]

[0079] If c = g(a, tm) is satisfied, where g is the pre-set check rule, which is pre-negotiated and agreed upon by the sender and the receiver. That is to say, if c is not equal to f(a, tm), then it needs to be resent.

[0080] It can be understood that the significance of introducing the timestamp tm is to ensure the randomness of the check bit c.

[0081] In summary, the high-dimensional features also include check bits; which are used by the receiver to verify whether the data is correct; among which, the sender should also include an output layer, and the check bits are obtained based on the high-dimensional features and all the random parameters in the collapsed output layer.

[0082] In this case, when a hacker cracks the password, not only does he have to perform brute-force calculations himself, but also has to simulate the operation process of the neural network model itself, which undoubtedly greatly increases the difficulty of cracking.

[0083] Correspondingly, the present invention also discloses a data transmission system integrating in-vehicle battery management and vehicle network security, as Figure 2 shown, including: an in-vehicle battery management system and a vehicle network security system;

[0084] When the in-vehicle battery management system and the vehicle network security system interact with each other to exchange data, they act as the sender and the receiver respectively; among which, the sender is used for data acquisition and preliminary fusion to form multi-dimensional data; and based on the first neural network model and the multi-dimensional data, to obtain the corresponding high-dimensional features; and after encrypting the high-dimensional features, to transmit them to the receiver; the receiver is used for obtaining the final result instruction based on the second neural network model and the high-dimensional features; the neural network model is an algorithm model with neurons as the computing unit, and the neural network model is composed of the first neural network model and the second neural network model.

[0085] The applicant of the present invention has made a detailed description and illustration of the embodiments of the present invention in conjunction with the accompanying drawings. However, those skilled in the art should understand that the above embodiments are only the preferred implementation schemes of the present invention, and the detailed description is only to help readers better understand the spirit of the present invention, rather than a limitation on the protection scope of the present invention. On the contrary, any improvement or modification made based on the spirit of the present invention should fall within the protection scope of the present invention.

Claims

1. A data transmission method integrating vehicle battery management and vehicle network security, characterized in that: include: At the sending end, data is collected and initially integrated to form multi-dimensional data; Based on the first neural network model and the multi-dimensional data, obtaining corresponding high-dimensional features; Encrypt high-dimensional features before transmitting; The receiving end obtains the final result instruction based on the second neural network model and high-dimensional features; The neural network model is an algorithm model using neurons as calculation units, and the neural network model is composed of a first neural network model and a second neural network model; All neurons in the neural network algorithm are regarded as all nodes of a graph. The connection relationship between neurons constitutes the directed edges of the graph, and the high-dimensional feature is the cut directed edge; The high-dimensional features also include check bits, which are used by the receiving end to verify whether the data is correct. The sending end includes an output layer, and the check bits are obtained based on the high-dimensional features and all random parameters in the collapsed output layer. The random parameters are collapsed based on the current timestamp.

2. A data transmission method integrating vehicle battery management and vehicle network security according to claim 1, characterized in that: The algorithm model that uses neurons as computing units is CNN, RNN or DNN.

3. The data transmission method for integrating vehicle battery management and vehicle network security according to claim 1 is characterized in that: The result instruction is based on data generated by the vehicle in real time and is obtained according to the output of the neural network model.

4. The data transmission method for integrating vehicle battery management and vehicle network security according to claim 1 is characterized in that: High-dimensional features are feature maps output by the intermediate layers in the neural network model.

5. The data transmission method for integrating vehicle battery management and vehicle network security according to claim 4 is characterized in that: In the neural network model, the input layer and the middle layer are set at the sending end, and the output layer is set at the receiving end.

6. A data transmission system integrating vehicle battery management and vehicle network security, applied to any method of claims 1-5, characterized in that: The system includes: an on-board battery management system and a vehicle network security system; When exchanging data between the on-board battery management system and the vehicle network security system, they serve as the sender and receiver to each other; The sending end is used for data collection and preliminary fusion to form multi-dimensional data; and based on the first neural network model and the multi-dimensional data, obtains the corresponding high-dimensional features; and encrypts the high-dimensional features and transmits them to the receiving end; The receiving end is used to obtain the final result instruction based on the second neural network model and the high-dimensional features; Among them, the neural network model is an algorithm model that uses neurons as calculation units, and the neural network model is composed of a first neural network model and a second neural network model.

7. A terminal comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is used to operate according to the instructions to execute the steps of a data transmission method for integrating vehicle battery management and vehicle network security according to any one of claims 1-5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of a data transmission method for integrating vehicle battery management and vehicle network security as described in any one of claims 1 to 5 are implemented.

Images