Power system-oriented network security user calculation evaluation model
By adopting the network security user calculation evaluation model in the power system, using artificial bee colony algorithm to optimize the Gaussian process model, combining neural network and singular value decomposition technology, the subjectivity and complexity of network security evaluation of the power system are solved, and high Accuracy and low complexity network security assessment.
Patent Information
- Application Number
- CN202411339931.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-25
- Publication Date
- 2024-11-12
AI Technical Summary
The existing power system network security assessment methods have large subjective factors, unobjective results and high complexity, making it difficult to achieve accurate and efficient assessment of the risk of network attacks in power system.
Adopting a network security user computing evaluation model for power systems, including information construction units, security perception units, optimization units and evaluation units, the Gaussian process model is optimized through artificial bee colony algorithm, combined with neural networks and singular value decomposition technology to realize the power system Accurate assessment of network security.
It realizes high-precision evaluation of network security of power system, reduces time complexity, and provides a network security assessment tool suitable for power system, which can effectively identify and match network security levels.
Smart Images

Figure CN118944978A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of power system network security, and in particular to a network security user computing evaluation model for power systems. Background Art
[0002] In recent years, the high integration of information technology and communication technology has promoted the intelligence of power grids, but also led to the serious threat of cyber attacks on power grids. In recent years, cyber attacks on power systems around the world have occurred from time to time. Cyber attackers can effectively tamper with the operating data in the power system, which directly affects the decision-making of the power dispatching center, resulting in overload, load shedding and even cascading failures of transmission lines. Therefore, from the perspective of power system defense, it is necessary to quantify and analyze the potential risks of cyber attacks in order to formulate reasonable power system dispatching and defense plans. How to accurately assess the operating status of its network has become an important issue that power system designers need to solve. In response to the research on this issue, domestic and foreign scholars have given some solutions with reference significance, that is, using risk assessment models to perform necessary analysis and prediction. For example, the method of using tradeoff diagrams or vulnerability trees is used to assess the possibility of system attacks from a probability perspective. However, there are large subjective factors in the evaluation process of this method, and it is difficult to obtain objective and neutral evaluation results; by introducing professional databases, risk templates and risk analysis technologies, an automatic tool for qualitative and quantitative analysis of power systems is provided. However, the evaluation process of this tool is highly complex, and its application scope has certain limitations. Summary of the invention
[0003] The purpose of the present invention is to provide a network security user computing evaluation model for power systems to solve the problems existing in the above-mentioned background technology.
[0004] To achieve the above objectives, the present invention provides a network security user computing evaluation model for power systems, comprising:
[0005] An information construction unit, used to obtain operation data in the power system and perform preprocessing;
[0006] A security perception unit is used to construct a security perception model and perform prediction training on it according to network security data until the migration success rate of the security perception model reaches a preset requirement; the security perception unit is also used to input the processed power system operation data into the security perception model and output the power security identification result;
[0007] The optimization unit uses the artificial bee colony algorithm to optimize the Gaussian process-based model;
[0008] An assessment unit matches the network security assessment level according to the power safety identification result.
[0009] Preferably, prediction training is performed on the network security data until the migration success rate of the security perception model reaches a preset requirement, specifically including:
[0010] Extracting the original sample set from the original training set of the security perception model, performing singular value decomposition on the original sample set to obtain a predicted sample; constructing a test sample based on the predicted sample and the prediction result of the predicted sample by the target model;
[0011] Insert the test sample into the original training set to generate a test training set, and perform neural network training based on the test training set to obtain the offset model;
[0012] The prediction samples are predicted based on the offset model, and whether the offset model is successfully offset is determined based on the prediction results; the training is stopped when it is determined that the offset success rate reaches the preset requirement.
[0013] Preferably, in order to optimize the objective function of the Gaussian process as much as possible, the initial value of the hyperparameter is randomly selected, and then the hyperparameter of the Gaussian process is adjusted and optimized using the artificial bee colony algorithm according to the corresponding prediction results. In this process, for the training data R = (x, y), the formula of its likelihood probability distribution p (y|x) is:
[0014] p(y|x)=∫p(y|g,x)p(g|x)dg
[0015] Among them, the prior probability distribution of function g and output y both obey Gaussian distribution; g|x~N(0,σ 2 ),y|g~N(g,σ 2 gX); the marginal likelihood function of the hyperparameter logp(y|x) is as follows:
[0016]
[0017] Preferably, in the operation of the artificial bee colony algorithm, the initial honey source is is a D-dimensional vector, and D = R + 2; introduce S hired bees, and the maximum number of times they can mine nectar is t L ,The maximum number of algorithm cycles is MAX, and its mathematical expression is:
[0018]
[0019] Among them, f is the fitness function of the nectar source. If the absolute value of the fitness function is smaller, the quality of the nectar source is higher.
[0020] Preferably, the execution steps of the artificial bee colony algorithm include:
[0021] S1, employed bees search and record nectar source information, and transmit this information to waiting observer bees;
[0022] S2. Among the received information, the observing bees select the current solution with a certain probability and search for local solutions in its vicinity; when the local solution is better than the current solution, the local solution replaces the current solution, and let p i represent the probability result of the i-th nectar source. Then, the probability calculation formula for the observing bees is as follows:
[0023]
[0024] Using the original nectar source X, the neighboring solution X' of the original nectar source, and a random number between [-1, 1] obtain a new nectar source V as follows
[0025]
[0026] S3. If the mining times of the artificial bee colony algorithm exceed MAX, then abandon the current solution and convert the employed bees into scout bees; in the manner of S2, the scout bees continue to randomly search for new nectar sources near the original nectar source X.
[0027] Preferably, during the process of the observing bees searching for new nectar sources, if a new nectar source is selected as the current nectar source information, then let the mining times item = 0; otherwise, let the mining times item = item + 1.
[0028] Preferably, after several mining operations, the neighboring nectar sources of the current nectar source are nearly exhausted, that is, it is difficult to improve the quality of the nectar source. When the nectar source mining times item < t L If the number of loops l ≥ MAX, then end the entire artificial bee colony algorithm and output the final hyperparameter values; if l < MAX, then continue to mine the nectar source information; when the nectar source mining times item ≥ t L then randomly generate a new nectar source, reset the mining times item, and at the same time abandon the current nectar source, and jump to the step of calculating the fitness function of the nectar source.
[0029] Preferably, the network security assessment levels include the security level, low-risk level, medium-risk level, high-risk level, and severe-risk level.
[0030] Therefore, the present invention adopts the above-mentioned network security user calculation and evaluation model for the power system. By integrating the relevant theories of the artificial bee colony algorithm and the Gaussian process, the optimized network security evaluation model is not only applicable to the power system, but also has a low time complexity and a high evaluation accuracy.
[0031] Next, through the drawings and embodiments, the technical solutions of the present invention will be further described in detail. Description of the Drawings
[0032] Figure 1 is a structural schematic diagram of a network security user calculation and evaluation model for the power system according to the present invention;
[0033] Figure 2 A schematic diagram of model optimization of an embodiment of a network security user computing evaluation model for a power system according to the present invention;
[0034] Figure 3 The time complexity statistics of the two evaluation models in the embodiments of the present invention are as follows;
[0035] Figure 4 The following are the statistical results of the evaluation accuracy of the two evaluation models in the embodiments of the present invention. DETAILED DESCRIPTION
[0036] Example
[0037] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.
[0038] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0039] like Figure 1-2 As shown, a network security user computing evaluation model for power systems includes:
[0040] An information construction unit, used to obtain operation data in the power system and perform preprocessing;
[0041] The security perception unit is used to build a security perception model and perform prediction training on it based on network security data until the migration success rate of the security perception model reaches the preset requirements; specifically, it includes:
[0042] Extracting the original sample set from the original training set of the security perception model, performing singular value decomposition on the original sample set to obtain a predicted sample; constructing a test sample based on the predicted sample and the prediction result of the predicted sample by the target model;
[0043] Insert the test sample into the original training set to generate a test training set, and perform neural network training based on the test training set to obtain the offset model;
[0044] The prediction samples are predicted based on the offset model, and whether the offset model is successfully offset is determined based on the prediction results; the training is stopped when it is determined that the offset success rate reaches the preset requirement.
[0045] The safety perception unit is also used to input the processed power system operation data into the safety perception model and output the power safety identification result;
[0046] The optimization unit uses the artificial bee colony algorithm to optimize the Gaussian process-based model;
[0047] In order to optimize the objective function of the Gaussian process as much as possible, the initial values of the hyperparameters are randomly selected, and then the artificial bee colony algorithm is used to adjust and optimize the hyperparameters of the Gaussian process according to the corresponding prediction results. In this process, for the training data R = (x, y), the formula for its likelihood probability distribution p (y|x) is:
[0048] p(y|x)=∫p(y|g,x)p(g|x)dg (1)
[0049] Among them, the prior probability distribution of function g and output y both obey Gaussian distribution; g|x~N(0,σ 2 ),y|g~N(g,σ 2 gX); the marginal likelihood function of the hyperparameter logp(y|x) is as follows:
[0050]
[0051] In the operation of the artificial bee colony algorithm, the initial nectar source is is a D-dimensional vector, and D = R + 2; introduce S hired bees, and the maximum number of times they can mine nectar is t L ,The maximum number of algorithm cycles is MAX, and its mathematical expression is:
[0052]
[0053] Among them, f is the fitness function of the nectar source. If the absolute value of the fitness function is smaller, the quality of the nectar source is higher.
[0054] The execution steps of the artificial bee colony algorithm include:
[0055] S1, employed bees search and record nectar source information, and transmit this information to waiting observer bees;
[0056] S2. In the received information, the observation bee selects the current solution with a certain probability and searches for the local solution near it; when the local solution is better than the current solution, the local solution replaces the current solution. Let p i represents the probability result of the i-th nectar source, then the probability calculation formula of the observed bee is:
[0057]
[0058] Using the original nectar source X, the neighboring solution X' of the original nectar source, and a random number between [-1,1] Get new nectar source V as follows
[0059]
[0060] S3. If the number of mining times of the artificial bee colony algorithm exceeds MAX, the current solution is abandoned and the employed bees are converted into scout bees; according to the method of S2, the scout bees continue to randomly search for new nectar sources near the original nectar source X.
[0061] In the power system, Gaussian process is usually used in the risk assessment model of network security, and the optimization degree of its hyperparameters greatly affects the evaluation accuracy of the model. Generally speaking, the conjugate gradient method is often used to optimize the hyperparameters, but this method is prone to fall into local repeated cycles, making it difficult to obtain the global optimal solution. In order to avoid this shortcoming, this embodiment introduces an artificial bee colony algorithm with global search capabilities to optimize the hyperparameters, thereby greatly improving the accuracy of the evaluation model. The implementation steps are:
[0062] 1. Initialization
[0063] In the initialization stage of the model, the initial honey source X is randomly generated i =(S i ,σ gi ,σ δi ,)(i=0,...,S-1), the maximum number of nectar sources to be mined is t L = S × D, set the mining times item = 0, set the maximum number of outer loops to MAX, and record the number of loops l = 1;
[0064] 2. Calculate the fitness function
[0065] Using the multiple parameters after initialization and the hyperparameters of the honey source, an evaluation model based on the Gaussian process is established to output the optimized target value. Then, the fitness value of the honey source is calculated using formula (3) to evaluate the quality of the hyperparameters.
[0066] 3. Hired bees
[0067] The current nectar source and formula (5) are used to randomly generate a new nectar source, and the fitness function value of the nectar source is calculated based on the parameters of the Gaussian process model. If the fitness of the new nectar source is better than that of the current nectar source, the new nectar source is used to replace the current nectar source to continue executing the algorithm; otherwise, the current nectar source is retained. In addition, during the dancing stage, the employed bees convey the nectar source information to the observing bees;
[0068] 4. Observe bees
[0069] At this stage, the observing bees utilize the obtained nectar source information and fully exploit the neighboring nectar sources of the current nectar source through the probability calculation formula (4). Then, a new nectar source is randomly generated using formula (3), and the fitness function values of the current nectar source and the new nectar source are compared to save the nectar source information with better quality;
[0070] 5. Update the exploitation times
[0071] If the new nectar source is selected as the current nectar source information in step 4, then set the exploitation times item = 0; otherwise, set the exploitation times item = item + 1;
[0072] 6. Scout bees and condition judgment
[0073] After several exploitations, the neighboring nectar sources of the current nectar source are nearly exhausted, that is, it is difficult to improve the quality of the nectar source. When the nectar source exploitation times item < t L If the loop times l ≥ MAX, then end the entire artificial bee colony algorithm and output the final hyperparameter values; if l < MAX, then continue to exploit the nectar source information; when the nectar source exploitation times item ≥ t L A new nectar source is randomly generated, the exploitation times item is reset, and the current nectar source is discarded, then jump to the calculation step of the fitness function of the nectar source;
[0074] 7. Model establishment
[0075] Use the optimized hyperparameters to train the data, and form a risk assessment model for network security based on the Gaussian process.
[0076] The evaluation unit matches the network security evaluation level according to the power security identification result. The network security evaluation levels include the security level, low - risk level, medium - risk level, high - risk level, and severe - risk level.
[0077] To verify the effectiveness of the method in this embodiment, under the same simulation conditions such as the same initial values and inputs, use Matlab software to conduct multiple simulations and experiments respectively on the risk assessment models based on the genetic algorithm (Genetic Algorithm) and the artificial bee colony algorithm, and count the time complexity and evaluation accuracy of these two models. To objectively evaluate the time complexity and evaluation accuracy of the models, the same basic parameters are set in the algorithm programs of these two models. The specific situation is shown in Table 1.
[0078] Table 1 Parameter design of the model simulation program
[0079] parameter Parameter Value meaning S 50 Population size D 88 Data Dimensions MAX 10000 Maximum number of evaluations Time 20 Simulation repetition times <![CDATA[U b ]]> 10 Individual value upper limit <![CDATA[L b ]]> <![CDATA[1×10 -7 ]]> Individual lower limit
[0080] In the simulation process, ABC represents the evaluation model based on the artificial bee colony algorithm, and GA represents the evaluation model based on the genetic algorithm. In addition, this embodiment selects 150 training data and 70 test data respectively, and repeatedly experiments and compares the two models from the perspective of time complexity and evaluation accuracy. It should be noted that this embodiment uses the fitness function to measure the evaluation accuracy of the model, and the time complexity is measured by the time consumed. The specific statistical results are as follows: Figure 3 , Figure 4 shown.
[0081] exist Figure 3 In the figure, the horizontal axis represents the number of times the evaluation model is run, and the vertical axis represents the time consumed by the model, in seconds. The dotted line and the solid line respectively record the time consumed by the evaluation model based on the genetic algorithm and the artificial bee colony algorithm at different evaluation times. From the trend of the curve, it can be seen that when the number of evaluations is <5000, the time consumed by the artificial bee colony algorithm is slightly less than that of the genetic algorithm. When the number of evaluations continues to increase, the time consumed by the artificial bee colony algorithm is significantly less than that of the genetic algorithm. This shows that the time complexity of the evaluation model based on the artificial bee colony algorithm is lower.
[0082] and Figure 3 Similar, in Figure 4 The horizontal axis represents the number of times the evaluation model is run, the vertical axis represents the fitness function value of the final solution of the model, the dotted line represents the evaluation model based on the genetic algorithm, and the solid line represents the model proposed in this paper. Figure 4 It can be seen that when the initial solutions are the same and the two models are run the same number of times, the fitness function value of the optimal hyperparameters obtained by the model proposed in this article is strictly smaller than that of the evaluation model based on the genetic algorithm, which indicates that the evaluation model of this embodiment has higher evaluation accuracy.
[0083] Therefore, the present invention adopts the above-mentioned network security user calculation evaluation model for the power system. Compared with the classic genetic algorithm model, the network security risk assessment model based on artificial bee colony algorithm optimization has lower time complexity and higher evaluation accuracy.
[0084] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.
Claims
1. A network security user computing evaluation model for power systems, characterized in that: include: An information construction unit, used to obtain operation data in the power system and perform preprocessing; A security perception unit is used to construct a security perception model and perform prediction training on it according to network security data until the migration success rate of the security perception model reaches a preset requirement; the security perception unit is also used to input the processed power system operation data into the security perception model and output the power security identification result; The optimization unit uses the artificial bee colony algorithm to optimize the Gaussian process-based model; An assessment unit matches the network security assessment level according to the power safety identification result.
2. The network security user computing evaluation model for power systems according to claim 1 is characterized in that: The prediction training is carried out based on the network security data until the migration success rate of the security perception model reaches the preset requirements, including: Extracting the original sample set from the original training set of the security perception model, performing singular value decomposition on the original sample set to obtain a predicted sample; constructing a test sample based on the predicted sample and the prediction result of the predicted sample by the target model; Insert the test sample into the original training set to generate a test training set, and perform neural network training based on the test training set to obtain the offset model; The prediction samples are predicted based on the offset model, and whether the offset model is successfully offset is determined based on the prediction results; the training is stopped until the offset success rate reaches the preset requirement.
3. The network security user computing evaluation model for power system according to claim 1 is characterized in that: The artificial bee colony algorithm is used to optimize the hyperparameters of the Gaussian process. In this process, for the training data R = (x, y), the formula for its likelihood probability distribution p (y | x) is: p(y|x)=∫p(ylg,x)p(g|x)dg Among them, the prior probability distribution of function g and output y both obey Gaussian distribution; g|x~N(0,σ 2 ),y|g~N(g,σ 2 gX); the marginal likelihood function of the hyperparameter logp(y|x) is as follows:
4. The network security user computing evaluation model for power system according to claim 3 is characterized in that: In the operation of the artificial bee colony algorithm, the initial nectar source is is a D-dimensional vector, and D = R + 2; introduce S hired bees, and the maximum number of times they can mine nectar is t L ,The maximum number of algorithm cycles is MAX, and its mathematical expression is: Among them, f is the fitness function of the nectar source.
5. The network security user computing evaluation model for power system according to claim 4 is characterized in that: The execution steps of the artificial bee colony algorithm include: S1, employed bees search and record nectar source information, and transmit this information to waiting observer bees; S2. In the received information, the observation bee selects the current solution and searches for local solutions near it; when the local solution is better than the current solution, the local solution replaces the current solution. Let p i represents the probability result of the i-th nectar source, then the probability calculation formula of the observed bee is: Using the original nectar source X, the neighboring solution X' of the original nectar source, and a random number between [-1,1] Get new nectar source V as follows S3. If the number of mining times of the artificial bee colony algorithm exceeds MAX, the current solution is abandoned and the employed bees are converted into scout bees; according to the method of S2, the scout bees continue to randomly search for new nectar sources near the original nectar source X.
6. The network security user computing evaluation model for power systems according to claim 5 is characterized by: When the observer bee is searching for a new nectar source, if it selects the new nectar source as the current nectar source information, the number of mining times item=0; otherwise, the number of mining times item=item+1.
7. The network security user computing evaluation model for power systems according to claim 6 is characterized by: After several rounds of exploitation, the neighboring nectar sources of the current nectar source are nearly exhausted. When the number of nectar source exploitation times item < t L If the number of loop times l ≥ MAX, the entire artificial bee colony algorithm is terminated and the final hyperparameter values are output; if l < MAX, the nectar source information continues to be exploited. When the number of nectar source exploitation times item ≥ t L A new nectar source is randomly generated, the exploitation times item is reset, and the current nectar source is discarded at the same time, and it jumps to the calculation step of the fitness function of the nectar source.
8. The network security user computing evaluation model for power systems according to claim 1 is characterized by: Network security assessment levels include security level, low risk level, medium risk level, high risk level and severe risk level.
