Electric vehicles (EVs) are highly favored by governments worldwide and can significantly reduce the air pollution generated by fuel-driven vehicles (account for 17% of global CO2 emissions). Research indicates that using EVs instead of traditional fuel-powered vehicles can reduce CO2 emissions by 70% [1]. With popularity of energy-saving and environmentally friendly EVs, they become an essential component of intelligent transportation systems (ITS). Due to the rapid development of ITS, vehicle-to-grid (V2G) is emerging as a promising service in ITS [2–4]. V2G provides mobile and distributed power for ITS and smart grid systems and reduces their dependence on nonrenewable energy. Furthermore, in V2G networks, as distributed energy storage elements, EVs can purchase power at valley hours and send power back to the smart grid at peak hours to achieve ‘peak shaving and valley filling’ to stabilize the power supply of ITS [5–7]. EVs can also generate profits by buying at a low price and selling at a high price [8]. This bidirectional power transmission produces many records, which can be analyzed by the control center to provide valuable services, such as charging/discharging scheduling, dynamic pricing, and optimal power dispatching [9–11].

However, these records can also cause a series of security and privacy issues, for example, the identities and locations of EVs and the amounts of charging and discharging [12,13]. These security and privacy issues are significant obstacles to the development of V2G in ITS. Especially when power is injected into the grid, information about power injection is highly sensitive. For example, network attackers may maliciously intercept relevant information, and power injection data may be tampered with to damage the ITS and smart grid services. In addition, each link from EVs to the control center is likely to be threatened by attacks. Measures should be taken to ensure the availability, integrity, confidentiality, and immutability of information in V2G networks.

Related studies have been conducted on the privacy and security issues of V2G networks. For example, from the perspective of identity authentication, Saxena et al. [14] presented a mutual authentication scheme for protecting the privacy of EV information by employing bilinear pairing technology. Still, its bilinear pairing implementation is costly. Abdallah et al. [15] constructed a secure authentication and privacy-preserving V2G connection scheme that leverages symmetric and public-private keys to authenticate identity. Tao et al. [16] investigated a lightweight protocol and developed capacity-based secure access authentication for the IoT that can efficiently satisfy security and privacy preservation requirements. Elliptic curve cryptography (ECC)-based schemes were also used in V2G networks. For example, Liang et al. [17] proposed a group authentication protocol by employing elliptic curve Diffie-Hellman and bilinear pairing, which effectively realizes security authentication in a V2G network. Fan et al. [18] developed a three-factor user authentication scheme, which significantly improves the robustness of the network by integrating one-way hash functions, bitwise exclusive OR (XOR) operations and ECC. Work [19] studied an anonymous key distribution scheme based on ECC, but its high computational overhead is unadaptable for the resource-constrained IoT environment.

The traditional centralized mechanism relies on a trusted third party to manage every energy transaction. Each transaction is vulnerable to a series of security threats in this scenario, such as single-point failure, denial of service attacks, and privacy leakage. Therefore, some schemes, such as [20–22], introduced blockchain technology for energy transactions because blockchains have the properties of decentralization, anonymity and immutability to provide an effective solution for V2G networks. Liu et al. [23] proposed a cross-domain identity authentication scheme based on a blockchain, which utilizes the encryption algorithm SM9 to guarantee the security and privacy required by V2G networks. Similarly, Kang et al. [24] addressed security and privacy issues in peer-to-peer (P2P) energy transactions by employing a consortium blockchain. Garg et al. [25] studied a combination mechanism of an ECC encryption algorithm and blockchain technology, which provides secure and anonymous energy transactions in V2G.

Although a secure identity authentication protocol can provide privacy preservation for users, fine-grained power consumption data are also sensitive due to their correlation with users’ activities when EVs inject power into the grid. From the perspective of power consumption data, Tonyali et al. [26] developed a meter data confusion scheme for protecting consumer privacy by concealing meter data. However, it does not involve specified V2G networks. Mahmoud et al. [27] presented a power injection scheme for smart grid system that utilizes homomorphic encryption to aggregate power injection bids from the storage unit at the local gateway. Unfortunately, this scheme cannot ensure the privacy of power injection data. Accordingly, Zhang et al. [28] constructed a privacy-aware sensing data aggregation scheme for protecting the power injection information, but it involves many expensive bilinear pairing operations. Next, Zhang et al. [29] proposed a 5G-based communication and power injection scheme for privacy protection in V2G networks, which adopts the novel aggregation technology named ‘hash-then-homomorphic’ to further aggregate blinded bids in various time slots. The above schemes focus on the bidding prices of users but fail to consider the amounts of power that users can inject into the grid. Although they protect users’ power injection and consumption data, security remains the major challenge of V2G networks.

In contrast to the previously established solutions, from the perspective of security and privacy preservations of identity and private injection data, this paper proposes a blockchain-enabled secure and privacy-preserving data aggregation scheme for fog-based ITS by combining blockchain technology and fog computing. The significant contributions are summarized as follows:

A fog computing and blockchain co-aware three-tier aggregation framework for power injection data is constructed, which provides firm support for the V2G network to realize secure, reliable and efficient power injection.

Finally, extensive simulation results show that the proposed scheme has lower computational costs than the previously established schemes while providing security and privacy protections. It is more suitable for ITS with latency-sensitive applications and limited resources.

The remainder of this paper is structured as follows. Section 2 describes the fog computing and blockchain co-aware three-tier aggregation network model for ITS. Section 3 presents a blockchain-enabled secure and privacy-preserving data aggregation scheme. Then, a simple security analysis and detailed performance evaluation of the experimental results are conducted in Section 4. Finally, Section 5 presents the conclusions of this study.

This section constructs a fog computing and blockchain co-aware three-tier aggregation network model for ITS, consisting of five types of entities: EVs, charging stations (CSs), fog nodes, a data center and a trusted authority (TA). Consider a city as an example. The city is divided into m subareas, and each subarea usually has w EVs. For simplicity, this paper uses symbol EVij(0≤i≤w,0≤j≤m) to represent the i^th EV in the j^th region. All m∗w EVs constitute the data sensing layer. Meanwhile, each subarea employs a fog node which is responsible for collecting and aggregating power injection data from the sensing layer. The symbol Fogj(0≤j≤m) represents the fog node deployed in subarea j, and all m fog nodes constitute the data aggregation layer, which is located at the edge of the ITS. Fig. 1 illustrates this framework.

Data Sensing Layer: The data sensing layer is located on the user side and includes EVs and CSs. EVs mainly refer to automobiles, motorcycles, ships, aircraft and other vehicles powered by batteries. CSs can provide charging or discharging services for vehicle batteries equipped with smart meters to record the corresponding data. In this layer, EVs can buy energy from the grid at a low price and inject the remaining power into the grid at a higher price to earn profits.

Data Aggregation Layer: The data aggregation layer is located on the edge side of the ITS. It is mainly composed of many fog nodes in all subareas. This layer selects an aggregation node in each time slot according to the remaining energy, and the remaining nodes are regarded as ordinary fog nodes. When the CS uploads the power injection data of an EV, the corresponding ordinary fog node is responsible for aggregating these data, generating a block, adding it to the blockchain through the consensus mechanism, and subsequently transmitting it to an aggregation node. The aggregation node is responsible for aggregating the received data from ordinary fog nodes (namely, for executing secondary aggregation of the power injection data) in this layer and encapsulating the data into a block with other relevant information. Then, the newly generated block is appended to the blockchain through a consensus mechanism and uploaded to the application supporting layer, where it awaits the decryption and analysis operations on the data center.

Application Supporting Layer: The application supporting layer refers to the data and control center, which contains a cloud server and is mainly responsible for decrypting and analyzing the data uploaded from the lower layer.

TA: TA is mainly used to generate and allocate public parameters and keys for entities. At the same time, it generates a Bloom filter by collecting the pseudonyms of EVs and fog nodes and sends the Bloom filter to the corresponding entities.

The network model mainly considers the following three types of network threats: (1) threats on fog nodes and the data center: fog nodes and the cloud server are considered honest but curious. Namely, they strictly abide by the service protocol, but at the same time, they will attempt to learn all detailed information from received data. In addition, fog nodes and the cloud server are easy to capture; (2) threats on the communication link: there is a potential hazard that an attacker may obtain private user data through eavesdropping on communication links; (3) threats of active attacks: an attacker may damage the authenticity and integrity of transmission data by launching active attacks (such as tampering, forgery, or replay).

This section presents a blockchain-enabled secure and privacy-preserving data aggregation scheme for fog-based ITS. This scheme includes five parts: system initialization, power injection request, EV-chain generation, fog-chain generation and application support.

The focus of this section is to analyze the security and privacy of the developed scheme and evaluate its performance while guaranteeing the security and privacy of collected data.

As described in Section 2, there are three main types of network threats: honest-but-curious processing nodes, link eavesdropping and active attacks. To resist first two types of threats, Paillier homomorphic encryption is utilized to encrypt the power injection data to counteract eavesdropping attacks and prevent information from leaking to honest-but-curious fog and cloud nodes. Although the cloud server can derive the aggregation result of each subarea, it cannot recover the power injection data of each EV. Consequently, the proposed scheme can guarantee confidentiality for EV power injection data. Intending to resist active attacks from attackers, the signature mechanism with a timestamp in the proposed scheme guarantees the integrity and validity of private data. Furthermore, the Bloom filter can ensure the identity anonymity and authenticity of valid nodes. As a result, the developed scheme guarantees the integrity and validity of private data and provides identity protection for EVs and fog nodes.

To analyze the performance of the proposed scheme, this paper compares the computational costs of this scheme with two previously established schemes: privacy-aware data aggregation (PADA) [28] and efficient privacy-preserving communication and power injection (ePPCP) [29]. To facilitate the explanation, with the same definitions in reference [30], it represents TE1, TE2, TM and TP as the exponential operation in zn2∗, the exponential operations, multiplication operations and bilinear pairings G, respectively. In our simulation scenario, it assumes that there is one control center in the application supporting layer. The number of fog nodes is 50, and the number of EVs in each subarea ranges from 0 to 1000. The parameters are randomly generated within their ranges. The experiment is performed 1000 times to evaluate the average value of the simulation results. The code is implemented using the pairing-based cryptography (PBC) library, and all the simulations are performed on a laptop with an Intel Core i5-7200U 2.5 GHz CPU and 8.00 GB RAM. Table 1 lists their time costs in the execution process.

In the PADA scheme, the generation of a power request packet <Power−req−UC> requires a computational cost of 2TM; in the privacy aggregation stage, the required computational cost is (2w+1)TP+(w+2)TM+TE2+2TE1. After receiving the packet <Power−req−UC>, the gateway needs 2TP+TM to verify the packet and generate a new packet. Meanwhile, a computational cost of wTP+(w+2)TM+TE1 is required during the privacy aggregation phase. Next, each storage unit incurs a computational cost of 2Tp+4TM+TE2+(k+2)TE1, where k represents the number of time slots. Since this scheme focuses only on the power supply of a specified time slot, it sets k = 1 for comparison. Therefore, the computational costs at the utility company (UC), gateway (GW) and power storage unit (PSU) are (2w+1)TP+(w+4)TM+TE2+2TE1, (w+2)TP+(w+3)TM+TE1 and 2wTP+4wTM+wTE2+3wtTE1, respectively.

In the ePPCP scheme, to generate power requests, a computational cost of TE1 is incurred. In the privacy aggregation phase, the computational cost is (3w+6)TE1. After receiving the data packet <Power−req−UC>, the gateway incurs a computational cost of 3TE1 to verify the packet. In the privacy aggregation stage, the gateway incurs the computational cost of (5w+2)TE1. To successfully bid, each PSU incurs a computational cost of (k+8)TE1. Then, the computational costs at UC, GW, and PSU are (3w+7)TE1, (5w+5)TE1 and (k+8)TE1, respectively.

To generate the power request <Power−req−fog>, the cloud server incurs a computational cost of TE1; while in the privacy aggregation stage, the required computational cost is TM+TE1. Here the cloud server is equivalent to the UC of the above solutions, and the scheme transfers this part to be implemented in the cloud server. After receiving the data packet <Power−req−fog>, the fog node needs the cost of 2wTE1 to verify the packet and generate a new signature. During the aggregation phase, the fog node incurs the cost of 2wTM+wTE1. Here the fog node is equivalent to the gateway in the above solutions. Subsequently, the data packet is received at the EV. A cost of wTE1 is required to verify the packet, and the EV incurs a cost of 2wTM+wTE1 to protect the privacy of EV. Here the EV is equivalent to the PSU in above schemes. The computational costs at the cloud server, fog node, and EV are wTM+(w+1)TE1, 2wTM+3wTE1 and w(2TM+2TE1), respectively. The cost comparisons are compared in Table 2. It can be observed that the developed scheme outperforms the other two benchmark schemes.

Through the above analysis, the developed solution has advantages over computational costs, but these advantages are not unconditional. For example, to achieve secure and reliable power injection, this paper adopts blockchain technology which usually associates with specific memory and bandwidth consumptions, but this trade-off is acceptable and well worth for the security and privacy preservation improvements. To show the above results more intuitively, the following figures are shown.

As shown in Fig. 4, the computational cost at the UC is directly proportional to the number of EVs. Compared with PADA and ePPCP, the computational cost at the UC of the proposed scheme is lower, and as the number of EVs increases, this advantage is enhanced. This is mainly because PADA uses expensive bilinear pairing calculations to generate packets, and ePPCP uses many exponential operations. Compared to PADA and ePPCP, the proposed scheme effectively avoids these operations, thereby reducing computational costs.

As shown in Fig. 5, the computational cost at the fog (GW) is also proportional to the number of EVs. Meanwhile, compared with those of schemes PADA and ePPCP, the computational cost of the proposed scheme is lower, and this advantage is strengthened with the increase in the number of EVs. This is mainly because PADA uses pairing calculations to verify the data packets generated by the UC in the privacy aggregation phase, the PADA scheme also uses pairing operations and many multiplication operations. At the same time, ePPCP uses many exponential operations to verify data packets and aggregate private data. Compared to PADA and ePPCP, the proposed scheme effectively avoids pairing operations and simplifies the operations as much as possible; thus, it can reduce the computational costs significantly.

Results similar to those in the above two figures are presented in Fig. 6, which shows the computational cost of the proposed scheme is lower than those of the other two schemes at the EV. PADA utilizes many bilinear pairing operations to protect private data and verify the integrity and authenticity of packets broadcasted by the gateway. Meanwhile, in ePPCP, many exponential operations are used, which consume a substantial amount of time.

As depicted in Fig. 7, the total computational cost is significantly reduced in our proposed scheme while providing security and privacy protections. The proposed scheme is more suitable for ITS with latency-sensitive applications and is also adapted to deploying devices with limited resources.

Existing privacy protection schemes tend to focus on the security of identity, while ignoring that fine-grained power consumption data are also sensitive. And many privacy protection schemes of power consumption data require high computational costs. To achieve the efficient aggregation of EV injection data with security and privacy guarantees, this paper proposes a blockchain-enabled secure and privacy-preserving data aggregation scheme for fog-based ITS. The proposed scheme constructs a secure and efficient aggregation framework by combining fog computing and blockchain technology. Then, it uses the Bloom filter and lightweight signature mechanism to build a secure and anonymous registration and authentication mechanism for preventing forgery attacks from malicious nodes. Furthermore, Paillier homomorphic encryption is implemented to encrypt the power injection data to ensure the confidentiality of the data. Finally, the performance evaluation shows that the proposed scheme has a lower computational cost with security and privacy guarantees. It is assumed that all EVs are benign and do not consider the impact of malicious EVs uploading fake private data on system performance. In the future, this paper will combine blockchain and smart contract technology to defend against malicious poisoning attacks.