Robust Compression Distributed Economic Scheduling Algorithm and System Based on Chaotic Encryption
By constructing a directed network and introducing a robust compressed distributed economic scheduling algorithm with chaotic encryption technology, the problems of high communication overhead and insufficient security in directed non-convex networks are solved, realizing efficient and secure power grid scheduling, which is suitable for modern smart grids.
Patent Information
- Application Number
- CN202611286955.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-24
- Publication Date
- 2026-11-10
Smart Images

Figure CN123073979A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of smart grid technology and provides a robust compressed distributed economic dispatch algorithm and system based on chaotic encryption. Background Technology
[0002] In the early stages of smart grid development, the system structure was relatively simple, and economic dispatch problems could be efficiently solved mainly by relying on traditional centralized dispatch methods.
[0003] With the rapid development of modern power grids and the integration of massive distributed energy resources, the system is increasingly exhibiting a significant trend towards decentralization. Traditional centralized algorithms are gradually becoming inadequate in handling such large and complex global scheduling tasks, failing to meet the growing practical needs of modern power grids. Therefore, researchers have begun to turn their attention to the emerging field of distributed architecture and have conducted in-depth explorations in economic dispatch.
[0004] In traditional centralized dispatch architectures, the underlying power generation equipment is responsible for collecting data and transmitting this information to a central controller for unified calculation. However, when the number of underlying devices is large, the centralized data upload creates a significant communication bottleneck and places stringent demands on the central controller's computing power. More critically, if the central controller fails, the entire dispatch system will be completely paralyzed.
[0005] In contrast, distributed scheduling architectures offer significant advantages. Under this architecture, all generating units within the power grid are equal, interacting and cooperating with neighboring nodes solely through the underlying communication network using local information. Without intervention from a central control unit, each node can collaboratively obtain the optimal solution that satisfies global constraints. This mechanism not only effectively avoids the risk of single points of failure, improving the system's resilience and robustness, but also significantly reduces the computational load of centralized systems. Given the shortcomings of centralized architectures and the advantages of distributed architectures, distributed strategies have been increasingly widely applied in modern power grid resource optimization.
[0006] However, as smart grid applications continue to expand, the network environment and constraints faced by actual dispatching are becoming increasingly complex. Early distributed economic dispatching strategies were mostly limited to ideal undirected communication networks, and their core framework heavily relied on strict convexity assumptions.
[0007] In real physical power grids, on the one hand, due to the physical limitations of communication links, information exchange between nodes often exhibits an asymmetric directed topology; on the other hand, due to the widespread presence of nonlinear devices, the system typically displays highly nonconvex characteristics. This makes existing algorithms based on undirected communication and strict convexity assumptions difficult to apply to complex power grid environments, becoming a major bottleneck restricting the in-depth application of this technology.
[0008] Furthermore, although distributed architecture avoids the overhead of centralized transmission, the communication overhead and privacy and security risks caused by frequent interactions between nodes cannot be ignored as the number of nodes continues to increase.
[0009] At the algorithmic level, existing technologies attempt to improve efficiency and reduce overhead by introducing momentum acceleration mechanisms or communication compression frameworks, but often fail to combine adaptive momentum techniques with efficient communication compression to address directed non-convex scenarios. At the security level, existing defenses such as homomorphic encryption or state decomposition are vulnerable to complex attacks and inevitably generate significant additional communication overhead. In contrast, while chaotic encryption can balance security and estimation accuracy, its integrated application in complex power grid scheduling is still immature. Finally, many current advanced scheduling algorithms remain at the theoretical level, with a severe lack of high-speed simulations and system verification based on parallel computing hardware such as field-programmable gate arrays (FPGAs).
[0010] In summary, the existing technology has the following problems: 1. Existing distributed optimization algorithms that combine adaptive step size technology usually assume that the communication network is undirected, thus relying heavily on symmetric double random matrices for information exchange, which is difficult to directly apply to unbalanced directed networks with limited links and clear directionality in actual power systems. 2. Existing traditional distributed economic scheduling algorithms are easily constrained by high communication bandwidth overhead during frequent iterative interactions, and their convergence speed is often limited under large-scale networks, making it difficult to deploy them in actual large-scale smart grids or microgrid clusters. 3. Existing distributed algorithms lack underlying security protection mechanisms when nodes engage in high-frequency information exchange. Core operating parameters are easily exposed in the communication link, posing a very high risk of data privacy leakage. Summary of the Invention
[0011] In view of this, the purpose of this invention is to provide a robust compressed distributed economic scheduling algorithm based on chaotic encryption, which aims to solve the three problems existing in the prior art.
[0012] To achieve the above objectives, the present invention provides the following technical solution: A robust compressed distributed economic scheduling algorithm based on chaotic encryption is proposed for distributed economic scheduling in directed networks, comprising: A directed network is constructed based on the network topology, and row random matrices and column random matrices are constructed based on this unbalanced directed network for communication between generators; A four-dimensional hyperchaotic Lorentz system is set up and combined with an XOR encryption mechanism to construct a lightweight chaotic encryption strategy. This strategy encrypts the compressed data before the generator node communicates, and the receiving end decrypts and restores the ciphertext based on the synchronization key after acquiring the data, thereby completing the subsequent system state update. Initialize the variables for each generator, including local estimation variables, gradient tracking variables, communication compression auxiliary variables, dynamic scaling variables, first-order momentum variables, and second-order momentum variables, and initialize the hybrid key. In each iteration, each generator node adopts an adaptive momentum strategy, using the first-order momentum term to track historical gradient information and adaptively adjusting the iteration step size based on the second-order momentum term, and then completing the update of the local estimated variables. Retrieve encrypted information pushed by neighboring nodes, decrypt and restore the encrypted information using the synchronization key, and calculate and update local estimated variables and communication compression auxiliary variables based on the decrypted data; The system retrieves encrypted information pushed by neighboring nodes and decrypts and restores the encrypted information using a synchronization key. Based on the decrypted data, it calculates and updates gradient tracking variables and communication compression auxiliary variables, and updates the dynamically scaling variables that decay with iteration and the chaotic key.
[0013] Furthermore, the distributed economic scheduling under the directed network is as follows: ; in, Indicates the overall power generation situation. It is a generator The individual output value, It is a locally closed and convex feasible set; while It is a generator only Known local cost function; and Each generator represents a separate generator. The minimum and maximum generating capacity; for any objective function The following assumptions need to be met: Each local cost function All Strongly convex functions, i.e. and satisfy: ; in, , Represents the local gradient; in the relatively internal region There exists at least one point that satisfies the supply and demand equilibrium constraint. ,in .
[0014] Furthermore, the method for constructing the unbalanced directed network and the communication weight matrix is as follows: Constructing an unbalanced directed network ; in Represents the generator set of an unbalanced directed network. Represents an edge set; If the generator Able to directly supply to the generator Send a message, then the edge ; and They represent generators The inner and outer neighbor sets; if If so, then the network is unbalanced; make Represents network A row-weighted randomized matrix, Representation matrix The Line 1 The elements of the column, if ,but ,otherwise ; Represents network A column-weighted matrix, if ,but ,otherwise ; The network satisfies the following assumptions: Unbalanced directed networks It is strongly connected, and each generator Both contain self-loops; communication matrix The condition is that the row is random. It is a column random matrix.
[0015] Row random matrices can be constructed according to the following rules. Sum of random matrices : ; In the formula, and They represent generators in-degree and out-degree; The unbiased communication compressor used in this algorithm satisfies the following assumptions: The compressor is unbiased, and the compression operator... : ; and its variants ; The constants satisfy: , , .
[0016] To solve the globally constrained economic scheduling problem, we first construct its dual problem and then transform it into a consensus-based distributed optimization problem. (Lagrange function) The definition is as follows: ; in, , Let the Lagrange multiplier vectors be represented; then the dual function is: ; in Yes The convex conjugate function of this pair of functions: ; Therefore, the dual form of the distributed economic scheduling problem can be expressed as: ; This can be reformulated as a consensus-based distributed optimization problem: ; in, Indicates generator The dual objective function; based on the Fenchel duality between strong convexity and Lipschitz continuous gradient. of Strong convexity means have Smoothness, and satisfies the supremum condition of convex conjugate; according to Danskin's theorem, The gradient is: ; From this, we can derive the expression for the dual gradient as follows: ; Strong convexity and the Slater condition assumptions ensure strong duality and a unique primal optimal solution. Through consensus-based distributed optimization problems, we can obtain... Furthermore, the original variables are recovered from the dual variables using the Fenchel-Young equation. Therefore, solving the consensus-based distributed optimization problem is equivalent to solving the distributed economic scheduling problem, and we focus on the distributed algorithm for this dual problem. The dual objective function of this algorithm The following assumptions are satisfied: Global dual objective function satisfy Conditions, among which It is a constant; that is, for all .
[0017] Furthermore, in each iteration, each generator node adopts an adaptive momentum strategy, using the first-order momentum term to track historical gradient information and adaptively adjusting the iteration step size based on the second-order momentum term, subsequently updating the local estimated variables, including: All generators Number of iterations Set to 1, and set a maximum number of iterations. As a criterion for stopping the algorithm's updates; Based on network topology Generated row and column random matrices and The weight parameters are used as weights during network communication and distributed to each generator to establish the push-pull communication framework of the algorithm. Based on the Lipschitz continuity coefficient L of the objective function and the shearing limit of the adaptive step size and Number of generators Calculate all the auxiliary constants below to determine the iteration step size. and compression control parameters Selection range: ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; in, For row random communication matrix The left eigenvector with an eigenvalue of 1; For column random communication matrix The right eigenvector with an eigenvalue of 1; It is composed of invertible matrices , The induced vector norms, for any vector satisfy , ; Is it that the induced norm , satisfy , The constant; This enables the communication matrix. and Satisfying the inequalities respectively and The constant, where , It is a projection matrix. , , It is the identity matrix. It is any possible constant; , These are constants related to the compressor; Based on the auxiliary constants and Lipschitz continuity coefficients calculated above Adaptive step size shearing boundary and Number of generators Row random communication matrix Left eigenvector with eigenvalue of 1 Random communication matrix Right eigenvector with eigenvalue of 1 ,constant , , , Compression control parameters The selection range is calculated as follows: ; ; ; Based on the auxiliary constants and Lipschitz continuity coefficients calculated above Adaptive step size shearing boundary and Number of generators Row random communication matrix Left eigenvector with eigenvalue of 1 Random communication matrix Right eigenvector with eigenvalue of 1 ,constant , , , iteration step size The selection range is calculated as follows: ; in, ; ; ; ; ; ; Each generator Using historical momentum variables and historical gradient tracking variables Update the first-order momentum variable according to the following formula to establish a convergence acceleration mechanism for the algorithm and speed up the convergence rate: ; Where, constant This is the parameter for the exponential moving average, used to control the rate of momentum decline; Each generator Use historical gradients to track variables Update the second-order momentum variable according to the following formula. Establish the robustness of the algorithm for gradient sparse scenarios: ; ; ; Among them, the function This is a shearing function used to cut off... The scope is limited to , To avoid the impact of extreme values; function Construct a diagonal matrix based on the input values. ; Used to control the rate of exponential decay; Each generator Using the updated first-order momentum variables and second-order momentum variables according to the iteration step size Update intermediate variables : .
[0018] Furthermore, the step of retrieving ciphertext information pushed by neighboring nodes and decrypting and restoring the ciphertext using a synchronization key, and then calculating and updating local estimated variables and communication compression auxiliary variables based on the decrypted data, includes: Each generator To each neighbor Retrieve decrypted and compressed local estimation information ,in Dynamic scaling variables By using iterative decay, the accumulation of absolute compression error can be effectively controlled; dynamo Decrypt and reconstruct the received information, and then update its communication-compressed auxiliary variables related to the local estimated variables: ; ; ; ; in, It is a positive constant that controls the update rate of auxiliary variables, which helps to mitigate the accumulation of errors caused by compression; matrix It is row-random, ensuring consistency in unbalanced directed networks; dynamo Update the local estimated variable according to the following formula. and original variables : ; ; in, It is a positive constant used to control the consensus adjustment of local estimates. Due to generator capacity constraints The defined local gradient.
[0019] Furthermore, the step of retrieving the encrypted information pushed by neighboring nodes and decrypting and restoring the encrypted information using the synchronization key, and then calculating and updating the gradient tracking variable and communication compression auxiliary variable based on the decrypted data, includes updating the dynamically scaling variable that decays with iteration and the chaotic key. Calculate intermediate gradient variables , ; wherein, the intermediate gradient variable This is used to enable the algorithm to effectively track global gradient information, thereby driving all nodes in the system to reach a global consensus; Each generator To each neighbor Retrieve decrypted and compressed local estimation information ,in ; Each generator Decrypt and reconstruct the received information, and then update its communication compression auxiliary variables related to the gradient tracking variables: ; ; ; ; in, It is a positive constant that controls the update rate of auxiliary variables, which helps to mitigate the accumulation of errors caused by compression; matrix It is randomized to ensure consistency in unbalanced directed networks; Each generator Update gradient tracking variables: ; in, It is a normal number used to control the consensus adjustment of local estimates; Each generator updates the dynamic scaling variable and the chaos key: ;
[0020] in, , They are two positive numbers.
[0021] Based on the above algorithm, the present invention also provides a robust compressed distributed economic scheduling system based on chaotic encryption, the system comprising: The network construction module is used for distributed economic scheduling in a directed network. It constructs an unbalanced directed network based on the network topology and then constructs row random matrices and column random matrices based on the directed network for communication between generators. The chaotic encryption module is used to introduce a four-dimensional hyperchaotic Lorentz system and combine it with an XOR encryption mechanism to construct a lightweight chaotic encryption strategy, as well as to encrypt and decrypt information in communication networks. The initialization preprocessing module is used to initialize all generator variables, including local estimation variables, gradient tracking variables, communication compression auxiliary variables, dynamic scaling variables, first-order momentum variables, second-order momentum variables, and the initialization chaos key. The adaptive momentum module is used to enable each generator node to adopt an adaptive momentum strategy in each iteration. It uses the first-order momentum term to track historical gradient information and adaptively adjusts the iteration step size according to the second-order momentum term, and then completes the update of the local estimated variables. The local information compression module is used to pull the ciphertext information pushed by neighboring nodes, decrypt the ciphertext using the synchronization key, and calculate and update the local estimated variables and communication compression auxiliary variables based on the decrypted data. The gradient information compression module is used to pull the ciphertext information pushed by neighboring nodes, decrypt the ciphertext using the synchronization key, and calculate and update the gradient tracking variable and communication compression auxiliary variable based on the decrypted data, as well as update the dynamically scaling variable and chaotic key that decay with iteration. The output module is used to determine when the preset maximum number of iterations has been reached, and then outputs the globally optimal estimated variable.
[0022] The beneficial effects of this invention are: Under the push-pull directed network communication framework, gradient tracking technology is innovatively integrated. Each generator node can effectively track global information without building a strict bidirectional symmetrical communication link, thus successfully breaking through the problem of efficient optimization under directed communication networks, relaxing the strict restrictions of the algorithm on the underlying communication topology, and greatly improving its applicability in complex real power grid environments. By integrating the adaptive momentum acceleration mechanism with robust communication compression technology, the algorithm improves the convergence rate by adaptively adjusting the step size using the momentum term, while significantly compressing the amount of data exchanged between nodes. This not only significantly reduces the bandwidth burden caused by high-frequency communication in the system, but also ensures faster convergence accuracy. Its reliability and efficiency in complex physical engineering scenarios have been fully verified by the FPGA underlying hardware platform.
[0023] A bottom-level protection strategy based on chaotic encryption, namely the RCDADT-CE algorithm, is introduced. The local data and gradient tracking information of the generator nodes are encrypted before communication. Therefore, it can effectively resist external communication interception without sacrificing the convergence accuracy of the algorithm or injecting additional disturbance noise, which greatly reduces the risk of system privacy leakage and achieves efficient unity of distributed economic scheduling and bottom-level data security protection.
[0024] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description
[0025] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein: Figure 1 This is a schematic diagram of the overall process of the method in this invention; Figure 2 This is a flowchart illustrating the algorithm in this invention; Figure 3 This is a schematic diagram showing the economic dispatch convergence performance and power balance experimental results of the algorithm in this invention in a microgrid system; Figure 4 A schematic diagram showing the experimental results of the impact of core control parameters on the convergence performance of the algorithm of this invention; Figure 5 This is a schematic diagram illustrating the experimental results comparing the convergence performance of the algorithm in this invention with other mainstream distributed algorithms. Figure 6 This is a schematic diagram showing the economic dispatch convergence performance and power balance experimental results of the algorithm in this invention under a large-scale power grid; Figure 7 This is a schematic diagram of the economic scheduling experimental results of the algorithm in this invention under a dynamic time-varying load scenario; Figure 8 This is a schematic diagram illustrating the privacy protection performance of the algorithm in this invention against communication interception and state inference attacks. Figure 9 This is a schematic diagram of the underlying system architecture deployment of the algorithm in this invention on an FPGA hardware acceleration platform; Figure 10 This is a schematic diagram of the experimental results of the microgrid economic dispatch operation verification based on the FPGA platform in this invention; Figure 11 This is a schematic diagram showing the experimental results comparing the numerical errors of FPGA fixed-point computation and CPU floating-point computation in this invention. Detailed Implementation
[0026] The present invention will be further described below with reference to specific embodiments. The accompanying drawings are for illustrative purposes only, representing schematic diagrams rather than actual physical objects, and should not be construed as limiting the scope of this patent. To better illustrate the embodiments of the present invention, some components in the drawings may be omitted, enlarged, or reduced, and do not represent the actual dimensions of the product. It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.
[0027] This invention addresses the challenge of existing distributed optimization algorithms that combine adaptive step-size techniques. These algorithms typically assume an undirected communication network, thus heavily relying on symmetric double-random matrices for information exchange. This makes them difficult to apply directly to unbalanced directed networks with limited links and clear directions in real-world power systems. The invention innovatively integrates gradient tracking technology within a push-pull directed network communication framework. This allows each generator node to effectively track global information without needing to construct strictly bidirectional symmetric communication links, thus successfully overcoming the challenge of efficient optimization in directed communication networks. It also relaxes the strict restrictions on the underlying communication topology, significantly expanding its applicability in complex real-world power grid environments.
[0028] This invention addresses the problem that existing traditional distributed economic scheduling algorithms are easily constrained by high communication bandwidth overhead during frequent iterative interactions, and their convergence speed is often limited in large-scale networks, making them difficult to deploy in practical large-scale smart grids or microgrid clusters. It integrates an adaptive momentum acceleration mechanism with robust communication compression technology. By adaptively adjusting the step size using the momentum term to improve the algorithm's convergence rate, it significantly compresses the amount of data exchanged between nodes. This design not only significantly reduces the bandwidth burden caused by high-frequency communication in the system but also ensures faster convergence accuracy. Its reliability and efficiency in complex physical engineering scenarios have been fully verified using an FPGA underlying hardware platform.
[0029] This invention addresses the problem that existing distributed algorithms lack underlying security mechanisms during high-frequency information exchange between nodes, making core operating parameters easily exposed in the communication link and posing a high risk of data privacy leakage. It introduces a low-level protection strategy based on chaotic encryption (i.e., the RCDADT-CE algorithm) to encrypt the local data and gradient tracking information exchanged between generator nodes before communication. Therefore, this invention can effectively resist external communication interception without sacrificing algorithm convergence accuracy or injecting additional perturbation noise, significantly reducing the risk of system privacy leakage and achieving a highly efficient unification of distributed economic scheduling and underlying data security protection.
[0030] The technical solution disclosed in this invention is used to solve the optimal scheduling problem arising from distributed scheduling using an unbalanced directed network within an economic dispatch system comprising several generators, where each generator in the system interacts with information through the directed network. Details are as follows.
[0031] refer to Figure 1 and Figure 2 As shown, a robust compressed distributed economic scheduling algorithm based on chaotic encryption is used for distributed economic scheduling in directed networks, including: S100. Construct a directed network based on the network topology. And construct a row random matrix based on this unbalanced directed network. Sum of random matrices Used for communication between various generators.
[0032] Specifically, the distributed economic scheduling under the directed network is as follows: ; in, Indicates the overall power generation situation. It is a generator The individual output value, It is a locally closed and convex feasible set; while It is a generator only Known local cost function; and Each generator represents a separate generator. The minimum and maximum generating capacity; for any objective function The following assumptions need to be met: Each local cost function All Strongly convex functions, i.e. and satisfy: ; in, , Represents the local gradient; in the relatively internal region There exists at least one point that satisfies the supply and demand equilibrium constraint. ;in, .
[0033] Specifically, the method for constructing the unbalanced directed network and the communication weight matrix is as follows: Constructing an unbalanced directed network ; in Represents the generator set of an unbalanced directed network. Represents an edge set; If the generator Able to directly supply to the generator Send a message, then the edge ; and They represent generators The inner and outer neighbor sets; if If so, then the network is unbalanced; make Represents network A row-weighted randomized matrix, Representation matrix The Line 1 The elements of the column, if ,but ,otherwise ;akin Represents network A column-weighted matrix, if ,but ,otherwise ; The network satisfies the following assumptions: Unbalanced directed networks It is strongly connected, and each generator Both contain self-loops; the communication matrix satisfies It is a random row. It is a column random matrix.
[0034] Row random matrices can be constructed according to the following rules. Sum of random matrices : , In the formula, and They represent generators In-degree and out-degree.
[0035] The unbiased communication compressor used in this algorithm satisfies the following assumptions: The compressor is unbiased, and the compression operator... satisfy: ; and its variants ; The constants satisfy: , , .
[0036] To solve the globally constrained economic scheduling problem, we first construct its dual problem and then transform it into a consensus-based distributed optimization problem. (Lagrange function) The definition is as follows: ; in, , Let the Lagrange multiplier vectors be represented; then the dual function is: ; in Yes The convex conjugate function of this pair of functions: ; Therefore, the dual form of the distributed economic scheduling problem can be expressed as: ; This can be reformulated as a consensus-based distributed optimization problem: ; in, Indicates generator The dual objective function; based on the Fenchel duality between strong convexity and Lipschitz continuous gradient. of Strong convexity means have - Smoothness, and satisfies the supremum condition of convex conjugate; according to Danskin's theorem, The gradient is: ; From this, we can derive the expression for the dual gradient as follows: ; Strong convexity and the Slater condition assumptions ensure strong duality and a unique primal optimal solution. Through consensus-based distributed optimization problems, we can obtain... Furthermore, the original variables are recovered from the dual variables using the Fenchel-Young equation. Therefore, solving the consensus-based distributed optimization problem is equivalent to solving the distributed economic scheduling problem, and we focus on the distributed algorithm for this dual problem. The dual objective function of this algorithm The following assumptions are satisfied: Global dual objective function satisfy Conditions, among which It is a constant; that is, for all .
[0037] S200 sets up a four-dimensional hyperchaotic Lorentz system and combines it with an XOR encryption mechanism to construct a lightweight chaotic encryption strategy. This strategy encrypts compressed data before generator node communication, and the receiving end decrypts and restores the ciphertext based on the synchronization key after acquiring the data, thereby completing the subsequent system status update.
[0038] Specifically, the hyperchaotic Lorentz system in step S200 is an important extension of the classic three-dimensional Lorentz model; by introducing a fourth state variable, the system's dimension is increased to four, resulting in more complex dynamic behavior. Compared to standard chaotic systems, hyperchaotic systems have two or more positive Lyapunov exponents. Therefore, their trajectories exhibit stronger divergence characteristics in phase space, making long-term behavior more difficult to predict.
[0039] when At that time, its mathematical model can be expressed as: ; ; ; ; in, These are the four state variables of the hyperchaotic system.
[0040] Under certain parameters, the system will generate two positive Lyapunov exponents, thus ensuring a hyperchaotic state with high topological entropy.
[0041] Initially, the two communicating parties securely negotiated control parameters. and the initial state vector As the master key.
[0042] This algorithm does not employ traditional numerical scaling, but instead directly uses the corresponding numerical values. Extract the underlying 256 binary bits from memory.
[0043] By performing a bitwise XOR operation on these four 64-bit data streams, the system can losslessly capture minute changes in the mantissa and exponent bits, thereby generating a mask with high randomness and full bandwidth (64 bits) as follows. : , in This indicates the extraction operator.
[0044] After generating the binary mask, the encryption and decryption processes will be performed within the communication protocol to ensure data privacy; before transmission, the sender will configure the binary mask... and compressed data Perform a bitwise XOR operation between them to obtain the encrypted ciphertext. The details are as follows: , in .
[0045] Upon receiving the ciphertext Subsequently, the target generator uses its locally synchronized chaotic system to regenerate the exact same binary mask. .
[0046] By analyzing the ciphertext and the same mask By performing an XOR operation, the receiver can decrypt and perfectly reconstruct the original data. The details are as follows: .
[0047] S300. Initialize the variables for each generator, including locally estimated variables. ,in express dimensional vector, gradient tracking variable Communication compression auxiliary variables Dynamic scaling variables , It is an arbitrary constant, a first-order momentum variable. and second-order momentum variables Simultaneously, securely initialize the chaotic key. .
[0048] Specifically, step S300 initializes the variables for all generators, including: For all generators All variables are initialized as follows: Local estimated variables With gradient tracking variables Dynamic scaling variables , It is an arbitrary constant, a first-order momentum variable. With second-order momentum variables The communication compression auxiliary variable is initialized to The chaotic key is initialized to .
[0049] S400. In each iteration, each generator node adopts an adaptive momentum strategy, utilizing the first-order momentum term. Track historical gradient information and based on the second momentum term The iteration step size is adaptively adjusted, and then the local estimated variables are updated.
[0050] Specifically, step S400 includes: All generators Number of iterations Set to 1, and set a maximum number of iterations. As a criterion for stopping the algorithm's updates; Based on network topology Generated row and column random matrices and The weight parameters are used as weights during network communication and distributed to each generator to establish the push-pull communication framework of the algorithm. Based on the Lipschitz continuity coefficient L of the objective function and the shearing limit of the adaptive step size and Number of generators Calculate all the auxiliary constants below to determine the iteration step size. and compression control parameters Selection range: ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; in, For row random communication matrix The left eigenvector with an eigenvalue of 1; For column random communication matrix The right eigenvector with an eigenvalue of 1; It is composed of invertible matrices , The induced vector norms, for any vector satisfy , ; Is it that the induced norm , satisfy , The constant; This enables the communication matrix. and Satisfying the inequalities respectively and The constant, where , It is a projection matrix. , , It is the identity matrix. It is any possible constant; , These are constants related to the compressor; Based on the auxiliary constants and Lipschitz continuity coefficients calculated above Adaptive step size shearing boundary and Number of generators Row random communication matrix Left eigenvector with eigenvalue of 1 Random communication matrix Right eigenvector with eigenvalue of 1 ,constant , , , Compression control parameters The selection range is calculated as follows: , , ; Based on the auxiliary constants and Lipschitz continuity coefficients calculated above Adaptive step size shearing boundary and Number of generators Row random communication matrix Left eigenvector with eigenvalue of 1 Random communication matrix Right eigenvector with eigenvalue of 1 ,constant , , , iteration step size The selection range is calculated as follows: ; in, ; ; ; ; ; ; Each generator Using historical momentum variables and historical gradient tracking variables Update the first-order momentum variable according to the following formula to establish a convergence acceleration mechanism for the algorithm and speed up the convergence rate: ; Where, constant This is the parameter for the exponential moving average, used to control the rate of momentum decline; Each generator Use historical gradients to track variables Update the second-order momentum variable according to the following formula. Establish the robustness of the algorithm for gradient sparse scenarios: ; ; ; Among them, the function This is a shearing function used to cut off... The scope is limited to , To avoid the influence of extreme values; function Construct a diagonal matrix based on the input values. ; Used to control the rate of exponential decay; Each generator Using the updated first-order momentum variables and second-order momentum variables according to the iteration step size Update intermediate variables : .
[0051] S500: Retrieve the encrypted information pushed by neighboring nodes and decrypt the encrypted information using the synchronization key to restore it. Local estimation of variables is completed based on the decrypted data. Communication compression auxiliary variables Calculation and update.
[0052] Specifically, S500 includes: Each generator To each neighbor Retrieve decrypted and compressed local estimation information ,in Dynamic scaling variables By decaying with iteration, the accumulation of absolute compression error can be effectively controlled; dynamo Decrypt and reconstruct the received information, and then update its communication-compressed auxiliary variables related to the local estimated variables: ; ; ; ; in, It is a positive constant that controls the update rate of auxiliary variables, which helps to mitigate the accumulation of errors caused by compression; matrix It is row-random, ensuring consistency in unbalanced directed networks; dynamo Update the local estimated variable according to the following formula. and original variables : , , in, It is a positive constant used to control the consensus adjustment of local estimates. Due to generator capacity constraints The defined local gradient.
[0053] S600: Retrieve the encrypted information pushed by neighboring nodes and decrypt the encrypted information using the synchronization key to restore it. Subsequently, gradient tracking variables were completed based on the decrypted data. Communication compression auxiliary variables The calculation and updating of the dynamic scaling variable, which decays with each iteration, are also updated. and the Chaos Key .
[0054] Step S600 includes: Calculate intermediate gradient variables , ; wherein, the intermediate gradient variable This is used to enable the algorithm to effectively track global gradient information, thereby driving all nodes in the system to reach a global consensus; Each generator To each neighbor Retrieve decrypted and compressed local estimation information ,in ; Each generator Decrypt and reconstruct the received information, and then update its communication compression auxiliary variables related to the gradient tracking variables: ; ; ; ; in, It is a positive constant that controls the update rate of auxiliary variables, which helps to mitigate the accumulation of errors caused by compression; matrix It is randomized to ensure consistency in unbalanced directed networks; Each generator Update gradient tracking variables: ; in, It is a normal number used to control the consensus adjustment of local estimates; Each generator updates the dynamic scaling variable and the chaos key: ; ; in, , They are two positive numbers.
[0055] In practice, steps S400 to S600 are repeated until the preset maximum number of iterations is reached. Output the globally optimal estimated variable.
[0056] Specifically, increment the iteration count by 1; if the iteration count is less than the maximum iteration count... Return to the steps above and execute. Calculate until The algorithm stops running and returns. As the optimal solution of the objective function.
[0057] In the following implementation examples, we tested the effectiveness of the proposed algorithm and system in a distributed low-rank matrix recovery scenario.
[0058] In some embodiments, distributed adaptive economic scheduling simulation and hardware verification experiments based on a standard 118 bus test system (54 generators) and a 5 bus test system (5 generators) were designed to systematically verify the effectiveness, security and practical engineering advantages of the proposed algorithm and system in this specific complex scenario.
[0059] In this embodiment, each generator The cost function is defined in the following quadratic form: ; in, , and These are local cost coefficients; these coefficients only apply to the generator. It is self-visible. The core objective of this optimization is to minimize the global generation cost in an unbalanced directed network structure while ensuring that the system load demand is met.
[0060] A robust compressed distributed economic scheduling system based on chaotic encryption, the system comprising: The network construction module is used for distributed economic scheduling in a directed network. It constructs an unbalanced directed network based on the network topology and then constructs row random matrices and column random matrices based on the directed network for communication between generators. The chaotic encryption module is used to introduce a four-dimensional hyperchaotic Lorentz system and combine it with an XOR encryption mechanism to construct a lightweight chaotic encryption strategy, as well as to encrypt and decrypt information in communication networks. The initialization preprocessing module is used to initialize all generator variables, including local estimation variables, gradient tracking variables, communication compression auxiliary variables, dynamic scaling variables, first-order momentum variables, second-order momentum variables, and the initialization chaos key. The adaptive momentum module is used to enable each generator node to adopt an adaptive momentum strategy in each iteration. It uses the first-order momentum term to track historical gradient information and adaptively adjusts the iteration step size according to the second-order momentum term, and then completes the update of the local estimated variables. The local information compression module is used to pull the ciphertext information pushed by neighboring nodes, decrypt the ciphertext using the synchronization key, and calculate and update the local estimated variables and communication compression auxiliary variables based on the decrypted data. The gradient information compression module is used to pull the ciphertext information pushed by neighboring nodes, decrypt the ciphertext using the synchronization key, and calculate and update the gradient tracking variable and communication compression auxiliary variable based on the decrypted data, as well as update the dynamically scaling variable and chaotic key that decay with iteration. The output module is used to determine when the preset maximum number of iterations has been reached, and then outputs the globally optimal estimated variable.
[0061] In this embodiment, an economic dispatch experiment is conducted using a standard smart grid testing system. To simulate the heterogeneous characteristics of generator units in a real distributed dispatch scenario, this embodiment distributes different generation cost coefficients and physical output constraints non-uniformly. The presence of multiple generator nodes leads to significant differences in cost objectives and power generation capacity among different nodes. In this simulation of a microgrid scenario, the total global power demand of the entire system is set to 120 MW, and the local secondary cost coefficients of each node are respectively in the range [range missing]. , , The values are discrete within the boundary, and the lower and upper limits of the physical constraints are respectively at... (MW) and (MW) range; when At that time, it was used to simulate a large-scale complex power grid scenario, and the total global power demand of the entire system was set to 4242 (MW), with the cost coefficients of each node falling within the range of... , , The values are independent within the range, while the lower and upper limits of the physical constraints are respectively at... (MW) and (MW) range is defined. Two typical directed communication networks are constructed as communication topologies for the distributed system in the experiment: one includes... A directed strongly connected network of generators; the second is a network containing An unbalanced directed network of generators. Specifically, this embodiment first constructs a directed communication strongly connected network. Then according to The communication relationships between intelligent agents are constructed using a row random matrix according to the following rules. Sum of random matrices These parameters are used as weights during network communication and distributed to each agent to establish the algorithm's push-pull communication framework. , In the formula, and They represent generators In-degree and out-degree.
[0062] The parameter settings for each algorithm in the distributed economic scheduling algorithm RCDADT-CE proposed in this invention are as follows: The algorithm of this invention is used for calculation. For all generator nodes... iteration step size Set to 0.03; local estimator variables are initialized to Gradient tracking variables are initialized to ,in The first-order momentum variable is initialized to... Second-order momentum variable The dynamic scaling variable is set to The communication compression auxiliary variable is initialized to... The remaining hyperparameters are set to... , , , , .
[0063] Subsequently, the distributed economic scheduling implementation example is completed according to the following steps: The number of iterations for each generator Set it to 1, and then set a maximum number of iterations. As a criterion for stopping the algorithm's iterative updates; Each generator Using historical momentum variables and historical gradient tracking variables Update the first-order momentum variable according to the following formula to establish a convergence acceleration mechanism for the algorithm and increase the convergence rate of the algorithm: , Where, constant This is the parameter for the exponential moving average, used to control the rate of momentum decline; Each generator Use historical gradients to track variables Update the second-order momentum variable according to the following formula. Establish the robustness of the algorithm for gradient sparse scenarios: , , , Among them, the function This is a shearing function used to cut off... The scope is limited to , To avoid the influence of extreme values; function Construct a diagonal matrix based on the input values. ; Used to control the rate of exponential decay; Each generator Using the updated momentum variables and second-order momentum variables according to the iteration step size Update intermediate variables : ; Each generator To each neighbor Retrieve decrypted and compressed local estimation information ,in Dynamic scaling variables By decaying with iteration, the accumulation of absolute compression error can be effectively controlled; dynamo Decrypt and reconstruct the received information, and then update its communication-compressed auxiliary variables related to the local estimated variables: , , , , in, It is a positive constant that controls the update rate of auxiliary variables, which helps to mitigate the accumulation of errors caused by compression; matrix It is row-random, ensuring consistency in unbalanced directed networks; dynamo Update the local estimated variable according to the following formula. and original variables : , , in, It is a positive constant used to control the consensus adjustment of local estimates. Due to generator capacity constraints The defined local gradient.
[0064] Calculate intermediate gradient variables : , Wherein, the intermediate gradient variable This is used to enable the algorithm to effectively track global gradient information, thereby driving all nodes in the system to reach a global consensus; Each generator To each neighbor Retrieve decrypted and compressed local estimation information ,in ; Each generator Decrypt and reconstruct the received information, and then update its communication compression auxiliary variables related to the gradient tracking variables: , , , , in, It is a positive constant that controls the update rate of auxiliary variables, which helps to mitigate the accumulation of errors caused by compression; matrix It is randomized to ensure consistency in unbalanced directed networks; Each generator Update gradient tracking variables: , in, It is a normal number used to control the consensus adjustment of local estimates; Each generator updates the dynamic scaling variable and the chaos key: , , in, , They are two positive numbers; Increment the iteration count by 1. If the iteration count is less than the maximum iteration count, return to step (3) and continue executing the algorithm until... The algorithm stops running and returns. As the optimal solution of the objective function.
[0065] Each generator will As the optimal economic scheduling scheme.
[0066] In the network environment constructed in this simulation experiment, the relevant parameters in the embodiments are obtained based on the current data partitioning method, network topology, and experimental task settings. It should be noted that in actual applications, due to differences in different network topologies, the number of generators, the degree of data heterogeneity, and task objectives, the appropriate parameter settings may also change.
[0067] In actual implementation, each generator node in the system relies on the underlying directed communication network. Encrypted communication is performed based on the topological connections, and the local estimated variables are continuously updated according to the RCDADT-CE algorithm steps. Gradient tracking variables First-order momentum variable Second-order momentum variable And each auxiliary variable, when the maximum number of iterations is reached Output the optimal solution at the time This is the optimal economic scheduling scheme.
[0068] This embodiment uses the original variables, dual variables, global power balance, and total system generation cost as core evaluation indicators to comprehensively assess the effectiveness and convergence performance of the scheduling scheme obtained based on the distributed algorithm. The convergence trajectories of the original and dual variables focus on quantifying the convergence accuracy and state consistency level of each generator node during the collaborative interaction process; the global power balance focuses on verifying the algorithm's ability to meet the stringent physical supply and demand constraints of the power grid; and the evolution of the total system generation cost directly reflects the degree to which the global scheduling scheme approaches the theoretical minimum generation cost. These indicators accurately characterize the algorithm's optimal economic scheduling solution process from multiple dimensions, thus comprehensively and objectively reflecting the actual performance of the proposed algorithm in complex microgrid distributed economic scheduling scenarios.
[0069] Figure 3 The steady-state convergence performance of the proposed algorithm in the economic scheduling scenario of a 5-bus test system is fully demonstrated. Specifically, Figure 3 (a) and (b) show that the original and dual variables of the generator set can converge quickly and smoothly to the optimal allocation value, and the incremental costs of each node ultimately achieve a high degree of coordination and consistency. Figure 3(c) and (d) further confirm that as the system variables approach the optimal trajectory, the entire network achieves a precise balance between supply and demand power, and the total power generation cost steadily decreases to the lowest level globally.
[0070] Figure 4 A detailed analysis was conducted on the core control parameters (momentum acceleration coefficient). With iteration step size The specific impact on the algorithm's convergence performance. Specifically, Figure 4 (a) shows that the convergence speed of the algorithm is positively correlated with the speedup factor. It exhibits the fastest convergence trend at that time. Meanwhile, Figure 4 (b) reveals that while increasing the step size can significantly shorten the convergence time, a larger step size can also cause significant system oscillations. In summary, while aggressive parameter settings can improve optimization efficiency, they also increase the risk of instability. Therefore, in practical applications, a careful balance needs to be struck between convergence speed and the smoothness of system operation.
[0071] Figure 5 The convergence performance of the algorithm of this invention was compared with that of other mainstream distributed algorithms. The results show that although the convergence rate of the algorithm of this invention is slightly slower than some uncompressed algorithms (such as the distributed primal dual gradient algorithm and the consensus-based economic scheduling algorithm), it not only successfully maintains strict linear convergence, effectively avoiding the severe oscillations (such as the distributed Lagrange momentum algorithm) and slow convergence (such as the fully distributed economic scheduling algorithm and the gradient-based push-sum algorithm) problems present in other algorithms, but more importantly, it significantly reduces communication transmission overhead by introducing a compression mechanism. In summary, the algorithm of this invention perfectly achieves a comprehensive balance between convergence rate and communication efficiency in an unbalanced directed network environment.
[0072] Figure 6 The convergence behavior and economic scheduling performance of the proposed algorithm on a 118 bus test system are comprehensively demonstrated. Specifically, Figure 6 (a) and (b) show that even under complex directed topologies, the system node variables can converge rapidly and synchronously within a very small number of iterations; Figure 6 (c) and (d) further confirm that the system accurately achieves global supply and demand balance and minimizes total operating costs. In summary, these results strongly demonstrate that the proposed algorithm possesses excellent robustness and extremely high optimization efficiency when dealing with large-scale restricted networks.
[0073] Figure 7 The algorithm's adaptive tracking capability in the face of dynamic, time-varying loads was verified. Specifically, Figure 7 (b) demonstrates that, in the face of successive demand surges of 4,000, 6,000 and 5,000 MW, the total power generation of the system can quickly and accurately track load changes; Figure 7(a) shows that the dual variables of each node can quickly re-achieve synchronous convergence after load switching. In summary, this algorithm demonstrates excellent dynamic adjustment and optimization performance in fluctuating real-world power grid environments.
[0074] Figure 8 The privacy protection capabilities of the proposed distributed algorithm in the economic scheduling process were fully verified. Specifically, Figure 8 (a) and (b) show that even if communication between nodes is intercepted, attackers cannot extract any useful information from the chaotic ciphertext; Figure 8 (c) and (d) further confirm that even under the worst-case attack assumption (an attacker attempts to infer the state using partial known information), the inferred variable trajectory deviates completely from the actual optimal target. In summary, this algorithm effectively resists attacks from external eavesdroppers and honest, curious nodes, ensuring the absolute security of the core data of the power generation equipment.
[0075] Figures 9 to 11 The deployment architecture and performance verification results of the algorithm of this invention on the FPGA underlying hardware platform are fully presented. Specifically, Figure 9 The overall hardware system architecture, including a multi-generator data path and a chaotic encryption engine, was demonstrated. Figure 10 The actual operation effect of the aforementioned 5-bus test system in the economic scheduling scenario on the FPGA was verified. All system variables converged robustly to the optimal solution and accurately achieved supply and demand balance. Figure 11 The numerical errors of FPGA fixed-point computation and CPU floating-point computation were further compared, and the results showed that the hardware computation errors of each variable were effectively controlled near the least significant bit (1 LSB). In summary, this series of verifications fully demonstrates that the present invention has successfully achieved efficient and highly reliable underlying hardware acceleration deployment while strictly ensuring the computational accuracy of the algorithm.
[0076] This invention innovatively integrates an adaptive momentum mechanism with push-pull communication compression technology within a distributed economic scheduling framework in a non-convex environment with a directed communication topology. It also incorporates a chaotic encryption strategy to provide low-level encryption protection for compressed state variables and gradient tracking information in inter-node interactions. Therefore, without sacrificing algorithm convergence accuracy or injecting additional perturbation noise, this invention not only effectively protects data privacy in multi-node interactions but also significantly reduces the bandwidth overhead of high-frequency communication in large-scale power grids or microgrid clusters. Furthermore, testing on an FPGA-based hardware platform has successfully verified the efficiency and reliability of this method in complex physical engineering applications.
[0077] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.
Claims
1. A robust compressed distributed economic scheduling algorithm based on chaotic encryption, characterized in that, Distributed economic scheduling for directed networks includes: A directed network is constructed based on the network topology, and row random matrices and column random matrices are constructed based on this unbalanced directed network for communication between generators; A four-dimensional hyperchaotic Lorentz system is set up and combined with an XOR encryption mechanism to construct a lightweight chaotic encryption strategy. This strategy encrypts the compressed data before the generator node communicates, and the receiving end decrypts and restores the ciphertext based on the synchronization key after acquiring the data, thereby completing the subsequent system state update. Initialize the variables for each generator, including local estimation variables, gradient tracking variables, communication compression auxiliary variables, dynamic scaling variables, first-order momentum variables, and second-order momentum variables, and initialize the hybrid key. In each iteration, each generator node adopts an adaptive momentum strategy, using the first-order momentum term to track historical gradient information and adaptively adjusting the iteration step size based on the second-order momentum term, and then completing the update of the local estimated variables. Retrieve encrypted information pushed by neighboring nodes, decrypt and restore the encrypted information using the synchronization key, and calculate and update local estimated variables and communication compression auxiliary variables based on the decrypted data; The system retrieves encrypted information pushed by neighboring nodes and decrypts and restores the encrypted information using a synchronization key. Based on the decrypted data, it calculates and updates gradient tracking variables and communication compression auxiliary variables, and updates the dynamically scaling variables that decay with iteration and the chaotic key.
2. The robust compressed distributed economic scheduling algorithm based on chaotic encryption according to claim 1, characterized in that, The distributed economic scheduling under the directed network is as follows: ; in, Indicates the overall power generation situation. It is a generator The individual output value, It is a locally closed and convex feasible set; while It is a generator only Known local cost function; and Each generator represents a separate generator. The minimum and maximum generating capacity; for any objective function The following assumptions need to be met: Each local cost function All Strongly convex functions, i.e. and satisfy: ; in, , Represents the local gradient; in the relatively internal region There exists at least one point that satisfies the supply and demand equilibrium constraint. ;in, .
3. The robust compressed distributed economic scheduling algorithm based on chaotic encryption according to claim 2, characterized in that, The method for constructing an unbalanced directed network and the communication weight matrix is as follows: Constructing an unbalanced directed network ; in Represents the generator set of an unbalanced directed network. Represents an edge set; If the generator Able to directly supply to the generator Send a message, then the edge ; and They represent generators The inner and outer neighbor sets; if If so, then the network is unbalanced; make Represents network A row-weighted randomized matrix, Representation matrix The Line 1 The elements of the column, if ,but ,otherwise ; Represents network A column-weighted matrix, if ,but ,otherwise ; The network satisfies the following assumptions: Unbalanced directed networks It is strongly connected, and each generator Both contain self-loops; the communication matrix satisfies It is a random row. A columnar random matrix; Row random matrices can be constructed according to the following rules. Sum of random matrices : ; In the formula, and They represent generators in-degree and out-degree; The unbiased communication compressor used in this algorithm satisfies the following assumptions: The compressor is unbiased, and the compression operator... satisfy: ; and its variants ; The constants satisfy: , , , To solve the globally constrained economic scheduling problem, we first construct its dual problem and then transform it into a consensus-based distributed optimization problem. (Lagrange function) The definition is as follows: ; in, , Let the Lagrange multiplier vectors be represented; then the dual function is: ; in Yes The convex conjugate function of this pair of functions: ; Therefore, the dual form of the distributed economic scheduling problem can be expressed as: ; Rephrased as a consensus-based distributed optimization problem: ; in, Indicates generator The dual objective function is based on the Fenchel duality between strong convexity and the Lipschitz continuous gradient. of Strong convexity means have - Smoothness, and satisfies the supremum condition of convex conjugate; According to Danskin's theorem The gradient is: ; From this, we can derive the expression for the dual gradient as follows: ; Strong convexity and the Slater condition assumptions ensure strong duality and a unique primal optimal solution. Through consensus-based distributed optimization problems, we can obtain Furthermore, the original variables are recovered from the dual variables using the Fenchel-Young equation. Therefore, solving the consensus-based distributed optimization problem is equivalent to solving the distributed economic scheduling problem, and we focus on the distributed algorithm for this dual problem. The dual objective function of this algorithm The following assumptions are satisfied: Global dual objective function satisfy Conditions, among which It is a constant; that is, for all .
4. The robust compressed distributed economic scheduling algorithm based on chaotic encryption according to claim 1, characterized in that, In each iteration, each generator node employs an adaptive momentum strategy, using a first-order momentum term to track historical gradient information and adaptively adjusting the iteration step size based on the second-order momentum term. The subsequent update of the locally estimated variables includes: All generators Number of iterations Set to 1, and set a maximum number of iterations. As a criterion for stopping the algorithm's updates; Based on network topology Generated row and column random matrices and The weight parameters are used as weights during communication and distributed to each generator to establish the push-pull communication framework of the algorithm. Based on the Lipschitz continuity coefficient L of the objective function and the shearing limit of the adaptive step size and Number of generators Calculate all the auxiliary constants below to determine the iteration step size. and compression control parameters Selection range: ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; in, For row random communication matrix The left eigenvector with an eigenvalue of 1; For column random communication matrix The right eigenvector with an eigenvalue of 1; It is composed of invertible matrices , The induced vector norms, for any vector satisfy , ; Is it that the induced norm , satisfy , The constant; This enables the communication matrix. and Satisfying the inequalities respectively and The constant, where , It is a projection matrix. , , It is the identity matrix. It is any possible constant; , These are constants related to the compressor; Based on the auxiliary constants and Lipschitz continuity coefficients calculated above Adaptive step size shearing boundary and Number of generators Row random communication matrix Left eigenvector with eigenvalue of 1 Random communication matrix Right eigenvector with eigenvalue of 1 ,constant , , , Compression control parameters The selection range is calculated as follows: ; ; ; Based on the auxiliary constants and Lipschitz continuity coefficients calculated above Adaptive step size shearing boundary and Number of generators Row random communication matrix Left eigenvector with eigenvalue of 1 Random communication matrix Right eigenvector with eigenvalue of 1 ,constant , , , iteration step size The selection range is calculated as follows: ; in, ; ; ; ; , ; Each generator Using historical momentum variables and historical gradient tracking variables Update the first-order momentum variable according to the following formula to establish a convergence acceleration mechanism for the algorithm and speed up the convergence rate: , Where, constant This is the parameter for the exponential moving average, used to control the rate of momentum decline; Each generator Use historical gradients to track variables Update the second-order momentum variable according to the following formula. Establish the robustness of the algorithm for gradient sparse scenarios: ; ; ; Among them, the function This is a shearing function used to cut off... The scope is limited to , To avoid the influence of extreme values; function Construct a diagonal matrix based on the input values. ; Used to control the rate of exponential decay; Each generator Using the updated first-order momentum variables and second-order momentum variables according to the iteration step size Update intermediate variables : 。 5. The robust compressed distributed economic scheduling algorithm based on chaotic encryption according to claim 1, characterized in that, The process of retrieving ciphertext information pushed by neighboring nodes, decrypting and restoring the ciphertext using a synchronization key, and calculating and updating local estimated variables and communication compression auxiliary variables based on the decrypted data includes: Each generator To each neighbor Retrieve decrypted and compressed local estimation information ; Dynamic scaling variables By decaying with iteration, the accumulation of absolute compression error can be effectively controlled; dynamo The received information is decrypted and reconstructed, and then its communication compression auxiliary variables related to the local estimated variables are updated as follows: ; ; ; ; in, It is a positive constant that controls the update rate of auxiliary variables, which helps to mitigate the accumulation of errors caused by compression; matrix It is row-random, ensuring consistency in unbalanced directed networks; dynamo Update the local estimated variable according to the following formula. and original variables : , , in, It is a positive constant used to control the consensus adjustment of local estimates. Due to generator capacity constraints The defined local gradient.
6. The robust compressed distributed economic scheduling algorithm based on chaotic encryption according to claim 5, characterized in that, The process involves retrieving ciphertext pushed by neighboring nodes, decrypting and restoring the ciphertext using a synchronization key, and calculating and updating gradient tracking variables and communication compression auxiliary variables based on the decrypted data. Updating the dynamically scaling variable that decays with iteration and the chaotic key includes: Calculate intermediate gradient variables , ; wherein, the intermediate gradient variable This is used to enable the algorithm to effectively track global gradient information, thereby driving all nodes in the system to reach a global consensus; Each generator To each neighbor Retrieve decrypted and compressed local estimation information ,in ; Each generator Decrypt and reconstruct the received information, and then update its communication compression auxiliary variables related to the gradient tracking variables: ; ; ; ; in, It is a positive constant that controls the update rate of auxiliary variables, which helps to mitigate the accumulation of errors caused by compression; matrix It is randomized to ensure consistency in unbalanced directed networks; Each generator Update gradient tracking variables. ; in, It is a normal number used to control the consensus adjustment of local estimates; Each generator updates the dynamic scaling variable and the chaos key: ; ; in, , They are two positive numbers.
7. A robust compressed distributed economic scheduling system based on chaotic encryption, characterized in that, The system includes: The network construction module is used for distributed economic scheduling in a directed network. It constructs an unbalanced directed network based on the network topology and then constructs row random matrices and column random matrices based on the directed network for communication between generators. The chaotic encryption module is used to introduce a four-dimensional hyperchaotic Lorentz system and combine it with an XOR encryption mechanism to construct a lightweight chaotic encryption strategy, as well as to encrypt and decrypt information in communication networks. The initialization preprocessing module is used to initialize all generator variables, including local estimation variables, gradient tracking variables, communication compression auxiliary variables, dynamic scaling variables, first-order momentum variables, second-order momentum variables, and the initialization chaos key. The adaptive momentum module is used to enable each generator node to adopt an adaptive momentum strategy in each iteration. It uses the first-order momentum term to track historical gradient information and adaptively adjusts the iteration step size according to the second-order momentum term, and then completes the update of the local estimated variables. The local information compression module is used to pull the ciphertext information pushed by neighboring nodes, decrypt the ciphertext using the synchronization key, and calculate and update the local estimated variables and communication compression auxiliary variables based on the decrypted data. The gradient information compression module is used to pull the ciphertext information pushed by neighboring nodes, decrypt the ciphertext using the synchronization key, and calculate and update the gradient tracking variable and communication compression auxiliary variable based on the decrypted data, as well as update the dynamically scaling variable and chaotic key that decay with iteration. The output module is used to determine when the preset maximum number of iterations has been reached, and then outputs the globally optimal estimated variable.
