CN106296044A - power system risk scheduling method and system - Google Patents

Abstract

The present invention relates to a power system risk scheduling method and system, which acquires the architecture data and new task load section data of the power system; according to the architecture data and the new task load section data, the preset initial knowledge matrix is modified by bacteria foraging reinforcement learning algorithm Perform iterative update to obtain the corresponding risk scheduling objective function value and the updated knowledge matrix. According to the updated knowledge matrix corresponding to the minimum value of the risk scheduling objective function, the new task is optimized online, and the risk scheduling optimization result is obtained and output. The optimal knowledge matrix in the source task is used as the initial matrix of the new task to realize knowledge transfer, and the new task is optimized online by using the reinforcement learning of bacterial foraging based on knowledge transfer. Through transfer learning, the speed of online learning is greatly improved, and the online dynamic optimization of risk scheduling problems is realized. When the scale of the problem is further expanded, it can still guarantee a faster solution speed, and can adapt to the rapid optimization of large-scale and complex risk scheduling.

Description

Power system risk scheduling method and system

technical field

The invention relates to the technical field of power grids, in particular to a method and system for power system risk scheduling.

Background technique

In recent years, with the development of regional power grid interconnection and high-voltage long-distance large-capacity transmission, the safe and stable operation of the power system is facing more severe challenges. In order to better balance system security and economic benefits, and enhance the level of dispatching operations against operational risks, the risk theory of power system is introduced in power generation optimization, and a lot of research has been done on risk dispatching.

The traditional power system risk scheduling method is to apply intelligent algorithms such as genetic algorithm (GA), quantum genetic algorithm (QGA), artificial bee colony (ABC), particle swarm optimization (PSO), etc. various optimization problems in power systems. However, the optimization of similar tasks by such intelligent algorithms is carried out in isolation, cannot effectively preserve the experience and knowledge of past tasks, lacks self-learning ability, and needs to be re-initialized every time a new task is executed, resulting in low optimization efficiency and difficulty in adapting Rapid optimization of large-scale complex risk scheduling.

Contents of the invention

Based on this, it is necessary to provide a power system risk scheduling method and system that can adapt to the rapid optimization of large-scale and complex risk scheduling to address the above problems.

A power system risk scheduling method, comprising the following steps:

Obtain the architecture data of the power system and the load section data of the new task;

According to the framework data and the new task load profile data, the preset initial knowledge matrix is iteratively updated through the bacterial foraging reinforcement learning algorithm to obtain the corresponding risk scheduling objective function value and the updated knowledge matrix; the initial The knowledge matrix is the optimal knowledge matrix in the source task;

According to the updated knowledge matrix corresponding to the minimum value of the risk scheduling objective function, the new task is optimized online, and the risk scheduling optimization result is obtained and output.

A power system risk scheduling system, comprising:

The task data acquisition module is used to acquire the structure data of the power system and the load section data of the new task;

The knowledge matrix update module is used to iteratively update the preset initial knowledge matrix through the bacterial foraging reinforcement learning algorithm according to the architecture data and the new task load section data, to obtain the corresponding risk scheduling objective function value and the updated The knowledge matrix; The initial knowledge matrix is the optimal knowledge matrix in the source task;

The risk scheduling optimization module is used to perform online optimization of new tasks according to the updated knowledge matrix corresponding to the minimum value of the risk scheduling objective function, and obtain and output the risk scheduling optimization results.

The above power system risk scheduling method and system obtains the architecture data of the power system and the new task load section data; according to the architecture data and the new task load section data, iteratively updates the preset initial knowledge matrix through the bacterial foraging reinforcement learning algorithm, The corresponding risk scheduling objective function value and the updated knowledge matrix are obtained. According to the updated knowledge matrix corresponding to the minimum value of the risk scheduling objective function, the new task is optimized online, and the risk scheduling optimization result is obtained and output. The optimal knowledge matrix in the source task is used as the initial matrix of the new task to realize knowledge transfer, and the new task is optimized online by using the reinforcement learning of bacterial foraging based on knowledge transfer. Through transfer learning, the speed of online learning is greatly improved, and the online dynamic optimization of risk scheduling problems is realized. When the scale of the problem is further expanded, it can still guarantee a faster solution speed, and can adapt to the rapid optimization of large-scale and complex risk scheduling.

Description of drawings

Fig. 1 is a flowchart of a power system risk scheduling method in an embodiment;

Fig. 2 is the knowledge acquisition schematic diagram of the bacteria foraging reinforcement learning algorithm based on knowledge transfer in an embodiment;

FIG. 3 is a schematic diagram of dimensionality reduction based on knowledge extension in an embodiment;

Fig. 4 is a schematic diagram of knowledge transfer in an embodiment;

Fig. 5 is a test system topology diagram in an embodiment;

Fig. 6 is a structural diagram of a power system risk dispatching system in an embodiment.

detailed description

In one embodiment, a power system risk scheduling method, as shown in Figure 1, includes the following steps:

Step S120: Obtain the structure data of the power system and the new task load profile data.

The architecture data of the power system may specifically include data such as bus nodes, transmission lines, transformers, and generators. The new task load section data includes one or more load sections, and each load section is regarded as a new task. Obtain the architecture data of the power system and the new task load section data for subsequent risk scheduling optimization.

Step S130: According to the framework data and the new task load section data, iteratively update the preset initial knowledge matrix through the bacterial foraging reinforcement learning algorithm to obtain the corresponding risk scheduling objective function value and the updated knowledge matrix.

The initial knowledge matrix is the optimal knowledge matrix in the source task. The optimal knowledge matrix in the source task is used as the initial matrix of the new task to realize knowledge transfer, and the action selection is performed by combining the bacterial group with the random search mode of the bacterial foraging optimization algorithm and the probabilistic space action selection strategy to realize the bacterial foraging based on knowledge transfer. The new task is optimized online using the Transfer Bacteria Foraging Optimization (TBFO) algorithm.

The specific type of the initial knowledge matrix is not unique. In this embodiment, the initial knowledge matrix is a Q matrix. In the Q learning algorithm, the element Q(s, a) in the Q matrix represents the expectation of the cumulative reward value obtained by choosing action a in state s. The matrix records the agent's knowledge of the process of mapping states to actions. The Q matrix is used as the knowledge matrix for recording group optimization information. By analyzing the similarity between different optimization tasks, the knowledge matrix of the source task is used to form the initial knowledge matrix of the new task, and the online dynamics of different time-section tasks are realized by knowledge transfer. optimized to ensure optimal reliability.

In the TBFO algorithm, the bacterial group obtains the action strategy for a specific environmental state from the initial knowledge matrix, and uses the feedback information obtained from repeated experiments to update the original knowledge, forming an inherent response to the specific state, so that the bacterial group foraging The energy value accumulated during the process reaches a maximum.

In one embodiment, step S130 includes step 131 to step 136 .

Step 131: According to the framework data and new task load section data, the control bacteria perform tropism, migration and replication operations under the guidance of the initial knowledge matrix.

Under the guidance of the initial knowledge matrix, bacteria acquire knowledge through tropism operation, migration operation and replication operation. In the TBFO algorithm, all bacteria will search the foraging area according to the initial knowledge matrix, and feed back the obtained rewards to the knowledge matrix. As shown in Figure 2, TBFO divides bacteria into two states, tropism and migration, according to the operation being performed. In a single iterative cycle of the algorithm, the two states are assigned to a certain proportion of individual bacteria. After the two groups of bacteria perform each operation, the energy values of all bacteria are calculated and sorted, and the replication operation is entered to make the bacteria group feed. The energy value accumulated in reaches the maximum. In the new iteration cycle, the bacterial status is redistributed according to the energy value in the previous iteration. Bacteria with higher energy values keep their area unchanged and perform tropism operations, while bacteria with lower energy values perform migration operations.

Specifically, based on the ranking of energy values, the dominant individuals in the flora are placed in the tendency state, and still undertake the task of local search. Its tendency behavior can be expressed by the following formula:

${\mathrm{<span\; class="notranslate">\; \&theta;</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; l</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; \&theta;</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; l</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span>\frac{\mathrm{<span\; class="notranslate">\; \&Delta;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span>}{\sqrt{{\mathrm{<span\; class="notranslate">\; \&Delta;</span>}}^{\mathrm{<span\; class="notranslate">\; T</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; \&Delta;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span>}}$

In the formula, θi ( ^j ,k,l) is the position of bacterial individual i after the migration operation of the first generation, the replication operation of the kth generation, and the trending operation of the jth generation; Δ represents the unit in the random direction determined after swimming vector.

C _k (i) can be a fixed step size or a variable step size. In the present embodiment, C _k (i) is the inertial step size of non-linear decrease, and C _k (i) update mode is shown in the following formula:

${\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; e</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&lsqb;</span>\frac{\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; k</span>}}{\mathrm{<span\; class="notranslate">\; c</span>}\mathrm{<span\; class="notranslate">\; l</span>}\mathrm{<span\; class="notranslate">\; the\; y</span>}}<span\; class="notranslate">\; -</span>{<span\; class="notranslate">\; (</span>\frac{\mathrm{<span\; class="notranslate">\; k</span>}}{\mathrm{<span\; class="notranslate">\; c</span>}\mathrm{<span\; class="notranslate">\; l</span>}\mathrm{<span\; class="notranslate">\; the\; y</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; \&rsqb;</span>$

In the formula: C _k (i) is the inertial step size at the kth iteration, C ₀ is the initial walking step size, C _e is the final walking step size, and cly is the maximum number of iteration steps.

For bacteria in the migrating state, when the migration probability P _ed is met, the bacteria perform roulette selection according to the action probability matrix; otherwise, the bacteria migrate according to the action corresponding to the largest knowledge element (greedy strategy):

In the formula: the superscript i represents the i-th controllable variable, corresponding to the i-th sub-knowledge matrix, i∈M; M is the set of controllable variables; the superscript j represents the j-th bacterium, j∈N, N is the bacteria group set; P _ed is the migration probability; r is a random number between 0 and 1; a _s is the action selected by the probability matrix P ⁱ in the global scope. When the migration conditions are met, the bacteria perform pseudo-random roulette selection according to the action probability matrix P ⁱ ; the update method of P ⁱ is as follows:

$\left\{\begin{array}{c}{\mathrm{<span\; class="notranslate">\; e</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; a</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{{\mathrm{<span\; class="notranslate">\; Q</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; a</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; \&beta;</span>}\underset{{\mathrm{<span\; class="notranslate">\; a</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; \&Element;</span>{\mathrm{<span\; class="notranslate">\; \&Lambda;</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}}{\mathrm{<span\; class="notranslate">\; max</span>}}{\mathrm{<span\; class="notranslate">\; Q</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; a</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; )</span>}\\ {\mathrm{<span\; class="notranslate">\; P</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; a</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; e</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; a</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>}{\underset{{\mathrm{<span\; class="notranslate">\; a</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; \&Element;</span>{\mathrm{<span\; class="notranslate">\; \&Lambda;</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}}{<span\; class="notranslate">\; \&Sigma;</span>}{\mathrm{<span\; class="notranslate">\; e</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; a</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; )</span>}\end{array}\right.$

In the formula: β is the difference coefficient, which is used to amplify the difference of Q ⁱ matrix elements; e ⁱ belongs to the intermediate calculation matrix.

In one embodiment, a crossover process is introduced in the duplication operation, and the crossover method is as follows:

θ ^i+S/2 (j,k,l)=rθ ⁱ (j,k,l)+(1-r)θ ^i+S/2 (j,k,l)

In the formula: S is the number of individual bacteria, i ∈ [1, S/2], r is a random number in [0, 1].

Step 132: According to the tropism operation, migration operation and replication operation of the bacteria, calculate the power flow value of the power system under the ground state and preset fault.

After the bacterial tropism operation, migration operation and replication operation are over, the power flow value of the power system in the ground state and preset fault is calculated according to the corresponding results. The ground state means that the system does not have a system failure, and the specific type of preset failure is not unique.

Step 133: Calculate the value of the risk scheduling objective function according to the power flow value of the power system under the base state and the preset fault.

In the TBFO algorithm, the immediate reward value reflects the direction of optimization, and the bacterial group obtains the optimal strategy by iteratively optimizing the knowledge matrix in order to expect to obtain the maximum cumulative reward function value. In the mathematical model of risk scheduling, the objective function is the reciprocal of the reward function of the algorithm, and it is expected to minimize the objective function through optimization. In this embodiment, the reward function is designed as follows:

${\mathrm{<span\; class="notranslate">\; R</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; c</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; R</span>}}<span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; c</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; v</span>}}}$

Among them, F _C is the fuel cost described by the nonlinear function, and I _R is the system safety risk index described by the nonlinear utility function. C _V is the degree of violation of the overall constraints of the system in the base state, c ₁ and c ₂ are used to match the magnitude relationship between fuel costs and risk indicators, and ω ₁ and ω ₂ are used to reflect the degree of emphasis on the corresponding goals.

Step 134: Redistribute the bacterial status according to the value of the risk scheduling objective function. After calculating the value of the risk scheduling objective function, the bacterial status is redistributed according to the value of the risk scheduling objective function.

Step 135: Iteratively update the initial knowledge matrix according to the redistributed bacterial status to obtain an updated knowledge matrix. In one embodiment, step 135 includes step 11 and step 12 .

Step 11: Reduce the dimension of the initial knowledge matrix to obtain multiple sub-knowledge matrices.

As shown in Figure 3, in order to effectively solve the "curse of dimensionality" problem, knowledge extension is used to reduce the dimensionality, and the initial knowledge matrix Q is divided into multiple sub-knowledge matrix Q ⁱ , corresponding to each variable one by one. The variables are connected through the knowledge matrix, and the elements in the adjacent matrix are relevant knowledge, that is to say, the action space A _i of x _i is the state space S _i+1 of x _i +1. Only after the action of the variable _xi is determined, the action of _xi+1 can be selected based on the selection result, thus forming a chain extension among related knowledge and realizing the decomposition and dimensionality reduction of the knowledge matrix.

Step 12: Update multiple sub-knowledge matrices according to the redistributed bacterial status to obtain an updated knowledge matrix. The multiple sub-knowledge matrices are updated, and an updated knowledge matrix can be obtained from the updated multiple sub-knowledge matrices.

The bacteria group is used as a multi-agent to update the knowledge matrix collaboratively. All bacteria share a knowledge matrix, and multiple knowledge elements can be updated simultaneously in a single iteration, which greatly speeds up the efficiency of optimization. After each trial and error exploration, the reward value of each subject will be evaluated. After the introduction of bacterial group collaboration, the update method of the sub-knowledge matrix Q ⁱ is as follows:

${\mathrm{<span\; class="notranslate">\; \&rho;</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; R</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&gamma;</span>}\underset{{\mathrm{<span\; class="notranslate">\; a</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; \&Element;</span>{\mathrm{<span\; class="notranslate">\; A</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}}{\mathrm{<span\; class="notranslate">\; max</span>}}{\mathrm{<span\; class="notranslate">\; Q</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; Q</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; )</span>$

${\mathrm{<span\; class="notranslate">\; Q</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; Q</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; \&alpha;\&rho;</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; j</span>}}{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}$

In the formula: R(s ^ij _k , s ^ij _k+1 , a ^ij _k ) represents the reward function value obtained when the k-th iteration selects action a _k in state s _k and transfers to state s _k+1 ; α is the learning factor, and γ is the discount factor.

In another embodiment, step 135 includes step 21 and step 23 .

Step 21: Calculate the active power deviation between each source task and the new task in the initial knowledge matrix according to the redistributed bacterial state.

The active power deviation is defined as the similarity between the source task and the new task, and the active demand is divided into multiple load sections from small to large:

[P _Ds1 ,P _Ds2 ),[P _Ds2 ,P _Ds3 ),...[P Dsi-1 ,P _Dsi )...,[P Dsn _{- 1} ,P _Dsn )

Step 22: Sort the source tasks according to the active power deviation from large to small, and obtain the previously preset number of source tasks. The specific value of the preset number is not unique, and in this embodiment, the preset number is two.

Step 23: Update the initial knowledge matrix according to the acquired source task to obtain an updated knowledge matrix.

Taking two source tasks for matrix update as an example, first calculate the contribution coefficients of the transfer learning of the two source tasks, and then update the initial knowledge matrix according to the transfer coefficients to obtain the knowledge matrix of the new task.

Specifically, assuming that the active demand of the new task x is P _Dx , P _Di and P _Dk are the two section loads closest to task x in the source task, and satisfy P _Di <P _Dx <P _Dk , then the two source tasks The contribution coefficients η ₁ and η ₂ of P _Di and P _Dk to transfer learning can be calculated by the following formula:

$\left\{\begin{array}{c}{\mathrm{<span\; class="notranslate">\; \&eta;</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; D.</span>}\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; D.</span>}\mathrm{<span\; class="notranslate">\; j</span>}}}{{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; D.</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; D.</span>}\mathrm{<span\; class="notranslate">\; j</span>}}}\\ {\mathrm{<span\; class="notranslate">\; \&eta;</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; D.</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; D.</span>}\mathrm{<span\; class="notranslate">\; x</span>}}}{{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; D.</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; D.</span>}\mathrm{<span\; class="notranslate">\; j</span>}}}\end{array}\right.$

Using the linear migration method, the knowledge matrix of the new task x can be obtained:

${\mathrm{<span\; class="notranslate">\; Q</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; \&eta;</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; Q</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; \&eta;</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}{\mathrm{<span\; class="notranslate">\; Q</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}$

Utilize the knowledge with high similarity to the new task, use the source task section information closest to the new task load requirements for migration, avoid negative interference caused by invalid knowledge on the learning quality and rate of the new task, and improve calculation accuracy.

It can be understood that, in one embodiment, it is also possible to reduce the dimensionality of the initial knowledge matrix to obtain multiple sub-knowledge matrices, and then use the knowledge with high similarity to the new task to update the multiple sub-knowledge matrices to obtain an updated knowledge matrix .

Step 136: Determine whether the iterative update satisfies a preset condition.

The specific type of preset condition is not unique. In this embodiment, the preset condition is k>k _max or Among them, k _max represents the preset maximum number of iterations; is the knowledge matrix The 2-norm of , reflects the degree of deviation of the knowledge matrix in the two iterations before and after.

Judging whether the iterative update meets the preset conditions, if not, then use the updated knowledge matrix as the initial knowledge matrix, and return to step 131, and update the knowledge matrix again; if so, the iterative update ends, and the finally obtained knowledge matrix is used as The optimization matrix needed for new task optimization.

Step S140: Perform online optimization of new tasks according to the updated knowledge matrix corresponding to the minimum value of the risk scheduling objective function, and obtain and output the risk scheduling optimization result.

After the iterative update of the initial knowledge matrix, the updated knowledge matrix corresponding to the minimum value of the risk scheduling objective function is used as the optimization matrix to optimize the new task online, and the risk scheduling optimization result is obtained and output. The specific way of outputting the risk scheduling optimization result is not unique, it may be output to a memory for storage, or it may be output to a display for display.

In addition, in one embodiment, before step S130, the power system risk scheduling method further includes step 110.

Step 110: Receive the source task for training, and obtain the optimal knowledge matrix as the initial knowledge matrix.

Step 110 may be before step S120 or after step S120. The TBFO algorithm performs a series of source tasks in the pre-learning stage to obtain the optimal knowledge matrix, and mines the initial knowledge from it to prepare for future related new tasks. As shown in Figure 4, the relevant initial knowledge from the source task will be used in online optimization. According to the similarity between the source task and the new task, the initial knowledge matrix of the source task _QS will be transferred to the initial knowledge matrix of the new task _QN .

In order to facilitate a better understanding of the above power system risk scheduling method, a detailed explanation will be given below in conjunction with specific embodiments.

A certain reliability test system is taken as the simulation object of risk scheduling. The base capacity of the system is selected as 100MVA. There are 24 busbar nodes, 34 transmission lines/transformers and 32 generators in the system. Its topology is shown in Figure 5. Among all 10 generator nodes, the generator node 21 with the largest stand-alone capacity is designated as the balance node of the whole system, and the remaining 9 nodes are PV (voltage control) nodes.

In order to test the adaptability of the algorithm to optimize different load levels, this embodiment conducts risk scheduling optimization simulations for 96 sections. In this embodiment, a typical daily load curve is selected, and a section is divided every fifteen minutes according to the time sequence, and section 1 to section 96 are obtained.

Based on the above platform, follow the steps below to optimize risk scheduling.

(1) Select the active power output _PG of the generator at the PV node as the control variable, the action variable space A (A _PG1 , A _PG2 ,..., A _PGi ) is in one-to-one correspondence with the control variable space, and i is the unit on the PV node total. The action space of the previous variable is the state space of the next variable. The sub-knowledge matrices corresponding to each variable state-action space are Q ^PG1 , Q ^PG2 , . . . , Q ^PGi . Under the guidance of knowledge matrix, bacteria acquire knowledge through tropism operation, migration operation and replication operation.

(2) The calculation of the objective function of risk scheduling depends on nonlinear power flow calculation. If the standard BFO algorithm is used, let N _ed , N _re and N _c denote the operands of migration, replication and trend behavior respectively, the maximum number of walks is N _s , and the expected fault set contains N _p faults, then the number of power flow calculations can be more Up to N _ed N _re N _c N _s (N _p +1) times, making the solution process extremely slow. By improving the optimization mode of the algorithm, the nested loop of the original algorithm is removed, and the efficiency of the algorithm is improved. The bacterial swarm combines the random search mode of the BFO algorithm with the probability space action selection strategy to perform action selection.

(3) Units with the same fuel cost coefficient on the same node are classified as a control variable, and the active output of 31 units is divided into 13 variables as control variables. The output of the former unit is used as the state space of the latter unit. Among them, the state space of the first unit is the active power of the current section, thus reducing the dimension of the knowledge matrix.

(4) In the TBFO algorithm, the immediate reward value reflects the direction of optimization, and the bacterial group obtains the optimal strategy by iteratively optimizing the knowledge matrix in order to expect to obtain the maximum cumulative reward function value.

(5) The active power deviation is defined as the similarity between the source task and the new task, and the active demand is divided into multiple load sections from small to large.

In order to avoid the negative interference of invalid knowledge on the learning quality and rate of the new task, knowledge with high similarity to the new task should be used as much as possible during the learning process. In this embodiment, only the two source task section information closest to the load requirements of the new task are used. to migrate. Assuming that the active demand of the new task x is P _Dx , P _Di and P _Dk are the two section loads closest to task x in the source task, and satisfying P _Di <P _Dx <P _Dk , then the transfer of the two source task pairs Learning the contribution coefficient, and then using the linear transfer method, the knowledge matrix of the new task x can be obtained.

After the iterative update of the initial knowledge matrix is completed, the updated knowledge matrix corresponding to the minimum value of the risk scheduling objective function is used as the optimization matrix to carry out online for new tasks, and the risk scheduling optimization results are obtained and output.

The above power system risk scheduling method uses the optimal knowledge matrix in the source task as the initial matrix of the new task to realize knowledge transfer, and uses knowledge transfer-based bacterial foraging reinforcement learning to optimize the new task online. Through transfer learning, the speed of online learning is greatly improved, and the online dynamic optimization of risk scheduling problems is realized. When the scale of the problem is further expanded, it can still guarantee a faster solution speed, and can adapt to the rapid optimization of large-scale and complex risk scheduling.

In one embodiment, a power system risk scheduling system, as shown in FIG. 6 , includes a task data acquisition module 120 , a knowledge matrix updating module 130 and a risk scheduling optimization module 140 .

The task data acquisition module 120 is used to acquire the architecture data of the power system and the new task load section data. The architecture data of the power system may specifically include data such as bus nodes, transmission lines, transformers, and generators, and the new task load section data includes one or more load sections. Obtain the architecture data of the power system and the new task load section data for subsequent risk scheduling optimization.

The knowledge matrix update module 130 is used to iteratively update the preset initial knowledge matrix through the bacterial foraging reinforcement learning algorithm according to the architecture data and the new task load section data, to obtain the corresponding risk scheduling objective function value and the updated knowledge matrix.

The initial knowledge matrix is the optimal knowledge matrix in the source task. The optimal knowledge matrix in the source task is used as the initial matrix of the new task to realize knowledge transfer, and the action selection is performed by combining the random search mode of the bacterial foraging optimization algorithm with the bacterial foraging optimization algorithm and the action selection strategy of probability space, and realizes the new task by using the TBFO algorithm. Online optimization.

The specific type of the initial knowledge matrix is not unique. In this embodiment, the initial knowledge matrix is a Q matrix. The Q matrix is used as the knowledge matrix for recording group optimization information. By analyzing the similarity between different optimization tasks, the knowledge matrix of the source task is used to form the initial knowledge matrix of the new task, and the online dynamics of different time-section tasks are realized by knowledge transfer. optimized to ensure optimum reliability.

In one embodiment, the knowledge matrix update module 130 includes a first processing unit, a second processing unit, a third processing unit, a fourth processing unit, a fifth processing unit and a sixth processing unit.

The first processing unit is used to control the bacteria to perform tropism, migration and replication operations under the guidance of the initial knowledge matrix according to the architecture data and the new task load section data.

Under the guidance of the initial knowledge matrix, bacteria acquire knowledge through tropism operation, migration operation and replication operation. Specifically, based on the ranking of energy values, the dominant individuals in the flora are placed in the tendency state, and still undertake the task of local search. Its tendency behavior can be expressed by the following formula:

${\mathrm{<span\; class="notranslate">\; \&theta;</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; l</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; \&theta;</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; l</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span>\frac{\mathrm{<span\; class="notranslate">\; \&Delta;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span>}{\sqrt{{\mathrm{<span\; class="notranslate">\; \&Delta;</span>}}^{\mathrm{<span\; class="notranslate">\; T</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; \&Delta;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span>}}$

C _k (i) can be a fixed step size or a variable step size. In the present embodiment, C _k (i) is the inertial step size of non-linear decrease, and C _k (i) update mode is shown in the following formula:

${\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; e</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&lsqb;</span>\frac{\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; k</span>}}{\mathrm{<span\; class="notranslate">\; c</span>}\mathrm{<span\; class="notranslate">\; l</span>}\mathrm{<span\; class="notranslate">\; the\; y</span>}}<span\; class="notranslate">\; -</span>{<span\; class="notranslate">\; (</span>\frac{\mathrm{<span\; class="notranslate">\; k</span>}}{\mathrm{<span\; class="notranslate">\; c</span>}\mathrm{<span\; class="notranslate">\; l</span>}\mathrm{<span\; class="notranslate">\; the\; y</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; \&rsqb;</span>$

For bacteria in the migrating state, when the migration probability P _ed is met, the bacteria perform roulette selection according to the action probability matrix; otherwise, the bacteria migrate according to the action corresponding to the largest knowledge element (greedy strategy):

When the migration conditions are met, the bacteria perform pseudo-random roulette selection a _S according to the action probability matrix P ⁱ ; the update method of P ⁱ is as follows:

$\left\{\begin{array}{c}{\mathrm{<span\; class="notranslate">\; e</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; a</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{{\mathrm{<span\; class="notranslate">\; Q</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; a</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; \&beta;</span>}\underset{{\mathrm{<span\; class="notranslate">\; a</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; \&Element;</span>{\mathrm{<span\; class="notranslate">\; \&Lambda;</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}}{\mathrm{<span\; class="notranslate">\; max</span>}}{\mathrm{<span\; class="notranslate">\; Q</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; a</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; )</span>}\\ {\mathrm{<span\; class="notranslate">\; P</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; a</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; e</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; a</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>}{\underset{{\mathrm{<span\; class="notranslate">\; a</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; \&Element;</span>{\mathrm{<span\; class="notranslate">\; \&Lambda;</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}}{<span\; class="notranslate">\; \&Sigma;</span>}{\mathrm{<span\; class="notranslate">\; e</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; a</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; )</span>}\end{array}\right.$

In one embodiment, a crossover process is introduced in the replication operation, and the crossover method is as follows:

θ ^i+S/2 (j,k,l)=rθ ⁱ (j,k,l)+(1-r)θ ^i+S/2 (j,k,l)

The second processing unit is used to calculate the power flow value of the power system in the base state and preset fault according to the tropism operation, migration operation and replication operation of the bacteria.

After the tropism operation, migration operation and replication operation of bacteria are over, the power flow value of the power system in the ground state and preset fault is calculated according to the corresponding results. The ground state means that the system does not have a system failure, and the specific type of preset failure is not unique.

The third processing unit is used to calculate the value of the risk scheduling objective function according to the power flow value of the power system in the base state and the preset fault.

In the TBFO algorithm, the immediate reward value reflects the direction of optimization, and the bacterial group obtains the optimal strategy by iteratively optimizing the knowledge matrix in order to expect to obtain the maximum cumulative reward function value. In the mathematical model of risk scheduling, the objective function is the reciprocal of the reward function of the algorithm, and it is expected to minimize the objective function through optimization. In this embodiment, the reward function is designed as follows:

${\mathrm{<span\; class="notranslate">\; R</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; c</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; R</span>}}<span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; c</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; v</span>}}}$

Among them, F _C is the fuel cost described by the nonlinear function, and I _R is the system safety risk index described by the nonlinear utility function. C _V is the degree of violation of the overall constraints of the system in the base state, c ₁ and c ₂ are used to match the magnitude relationship between fuel costs and risk indicators, and ω ₁ and ω ₂ are used to reflect the degree of emphasis on the corresponding goals.

The fourth processing unit is used to redistribute the bacteria status according to the value of the risk scheduling objective function. After calculating the value of the risk scheduling objective function, the bacterial status is redistributed according to the value of the risk scheduling objective function.

The fifth processing unit is configured to iteratively update the initial knowledge matrix according to the redistributed bacterial status to obtain an updated knowledge matrix.

In one embodiment, the fifth processing unit includes a dimensionality reduction unit and a matrix update unit.

The dimension reduction unit is used to reduce the dimension of the initial knowledge matrix to obtain multiple sub-knowledge matrices. Dimensionality reduction is carried out by knowledge extension, and the initial knowledge matrix Q is divided into multiple sub-knowledge matrices Q ⁱ , which correspond to each variable one by one.

The matrix update unit is used to update multiple sub-knowledge matrices according to the redistributed bacterial states to obtain updated knowledge matrices. The multiple sub-knowledge matrices are updated, and an updated knowledge matrix can be obtained from the updated multiple sub-knowledge matrices.

The bacteria group is used as a multi-agent to update the knowledge matrix collaboratively. All bacteria share a knowledge matrix, and multiple knowledge elements can be updated simultaneously in a single iteration, which greatly speeds up the efficiency of optimization. After each trial and error exploration, the reward value of each subject will be evaluated. After the introduction of bacterial group collaboration, the update method of the sub-knowledge matrix Q ⁱ is as follows:

${\mathrm{<span\; class="notranslate">\; \&rho;</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; R</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&gamma;</span>}\underset{{\mathrm{<span\; class="notranslate">\; a</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; \&Element;</span>{\mathrm{<span\; class="notranslate">\; A</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}}{\mathrm{<span\; class="notranslate">\; max</span>}}{\mathrm{<span\; class="notranslate">\; Q</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; Q</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; )</span>$

${\mathrm{<span\; class="notranslate">\; Q</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; Q</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; \&alpha;\&rho;</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; j</span>}}{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}$

In another embodiment, the fifth processing unit includes a calculation unit, an extraction unit and an update unit.

The calculation unit is used to calculate the active power deviation between each source task and the new task in the initial knowledge matrix according to the redistributed bacterial state. The active power deviation is defined as the similarity between the source task and the new task, and the active demand is divided into multiple load sections from small to large:

[P _Ds1 ,P _Ds2 ),[P _Ds2 ,P _Ds3 ),...[P Dsi-1 ,P _Dsi )...,[P Dsn _{- 1} ,P _Dsn )

The extraction unit is used to sort the source tasks according to the active power deviation from large to small, and obtain a preset number of source tasks. The specific value of the preset number is not unique, and in this embodiment, the preset number is two.

The update unit is used to update the initial knowledge matrix according to the acquired source task to obtain the updated knowledge matrix.

Taking two source tasks for matrix update as an example, first calculate the contribution coefficients of transfer learning of the two source tasks, and then update the initial knowledge matrix according to the transfer coefficients to obtain the knowledge matrix of the new task. The contribution coefficients η ₁ and η ₂ of the two source tasks P _Di and P _Dk to transfer learning can be calculated by the following formula:

$\left\{\begin{array}{c}{\mathrm{<span\; class="notranslate">\; \&eta;</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; D.</span>}\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; D.</span>}\mathrm{<span\; class="notranslate">\; j</span>}}}{{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; D.</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; D.</span>}\mathrm{<span\; class="notranslate">\; j</span>}}}\\ {\mathrm{<span\; class="notranslate">\; \&eta;</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; D.</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; D.</span>}\mathrm{<span\; class="notranslate">\; x</span>}}}{{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; D.</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; D.</span>}\mathrm{<span\; class="notranslate">\; j</span>}}}\end{array}\right.$

Using the linear migration method, the knowledge matrix of the new task x can be obtained:

${\mathrm{<span\; class="notranslate">\; Q</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; \&eta;</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; Q</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; \&eta;</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}{\mathrm{<span\; class="notranslate">\; Q</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}$

Utilize the knowledge with high similarity to the new task, use the source task section information closest to the new task load requirements for migration, avoid the negative interference of invalid knowledge on the learning quality and rate of the new task, and improve the calculation accuracy.

It can be understood that, in an embodiment, the fifth processing unit may also include a dimension reduction unit and a matrix update unit, and the matrix update unit includes a calculation unit, an extraction unit, and an update unit. First, reduce the dimensions of the initial knowledge matrix to obtain multiple sub-knowledge matrices, and then use the knowledge with high similarity to the new task to update the multiple sub-knowledge matrices to obtain the updated knowledge matrix.

The sixth processing unit is used to judge whether the iterative update satisfies the preset condition, and when the iterative update does not meet the preset condition, use the updated knowledge matrix as the initial knowledge matrix, and control the first processing unit again according to the architecture data and the new task Load section data, control bacteria to perform tropism operation, migration operation and replication operation under the guidance of the initial knowledge matrix.

The specific type of preset condition is not unique. In this embodiment, the preset condition is k>k _max or Determine whether the iterative update meets the preset conditions. If not, use the updated knowledge matrix as the initial knowledge matrix for iterative update again. If so, the iterative update ends, and use the final knowledge matrix as the optimization matrix required for new task optimization. .

The risk scheduling optimization module 140 is used to perform online optimization of new tasks according to the updated knowledge matrix corresponding to the minimum value of the risk scheduling objective function, and obtain and output the risk scheduling optimization result.

After the iterative update of the initial knowledge matrix, the updated knowledge matrix corresponding to the minimum value of the risk scheduling objective function is used as the optimization matrix to optimize the new task online, and the risk scheduling optimization result is obtained and output. The specific way of outputting the risk scheduling optimization result is not unique, it may be output to a memory for storage, or it may be output to a display for display.

In addition, in one embodiment, the power system risk scheduling system further includes a matrix training module.

The matrix training module is used to iteratively update the preset initial knowledge matrix through the bacterial foraging reinforcement learning algorithm in the knowledge matrix update module 130 according to the architecture data and the new task load section data, to obtain the corresponding risk scheduling objective function value and the updated Before the knowledge matrix, the source task is received for training, and the optimal knowledge matrix is obtained as the initial knowledge matrix. The TBFO algorithm performs a series of source tasks in the pre-learning stage to obtain the optimal knowledge matrix, and mines the initial knowledge from it to prepare for future related new tasks.

The above power system risk scheduling system uses the optimal knowledge matrix in the source task as the initial matrix of the new task to realize knowledge transfer, and uses bacterial foraging reinforcement learning based on knowledge transfer to optimize the new task online. Through transfer learning, the speed of online learning is greatly improved, and the online dynamic optimization of risk scheduling problems is realized. When the scale of the problem is further expanded, it can still guarantee a faster solution speed, and can adapt to the rapid optimization of large-scale and complex risk scheduling.

The technical features of the above-mentioned embodiments can be combined arbitrarily. To make the description concise, all possible combinations of the technical features in the above-mentioned embodiments are not described. However, as long as there is no contradiction in the combination of these technical features, should be considered as within the scope of this specification.

The above-mentioned embodiments only express several implementation modes of the present invention, and the descriptions thereof are relatively specific and detailed, but should not be construed as limiting the patent scope of the invention. It should be pointed out that those skilled in the art can make several modifications and improvements without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent for the present invention should be based on the appended claims.

Claims (10)

1. A power system risk scheduling method, characterized in that, comprising the following steps: Obtain the architecture data of the power system and the load section data of the new task; According to the framework data and the new task load profile data, the preset initial knowledge matrix is iteratively updated through the bacterial foraging reinforcement learning algorithm to obtain the corresponding risk scheduling objective function value and the updated knowledge matrix; the initial The knowledge matrix is the optimal knowledge matrix in the source task; According to the updated knowledge matrix corresponding to the minimum value of the risk scheduling objective function, the new task is optimized online, and the risk scheduling optimization result is obtained and output. 2. The power system risk scheduling method according to claim 1, wherein the initial knowledge matrix is a Q matrix. 3. The power system risk scheduling method according to claim 1, characterized in that, according to the architecture data and the new task setting data, the preset initial knowledge matrix is iterated through the bacterial foraging reinforcement learning algorithm Updating, the step of obtaining the corresponding risk scheduling objective function value and the updated knowledge matrix includes the following steps: According to the framework data and the new task load profile data, control bacteria to perform tropism, migration and replication operations under the guidance of the initial knowledge matrix; According to the tropism operation, migratory operation and replicative operation of the bacteria, calculate the power flow value of the power system under the ground state and preset fault; Calculate the value of the risk scheduling objective function according to the power flow value of the power system in the base state and the preset fault; Redistribute the bacteria state according to the risk scheduling objective function value; Iteratively updating the initial knowledge matrix according to the redistributed bacterial state to obtain an updated knowledge matrix; Judging whether the iterative update meets the preset conditions; If not, then use the updated knowledge matrix as the initial knowledge matrix, and return the section data based on the architecture data and the new task load, and control the bacteria to perform tropism operations under the guidance of the initial knowledge matrix , steps of migratory operation and replicative operation. 4. The power system risk scheduling method according to claim 3, wherein the step of iteratively updating the initial knowledge matrix according to the redistributed bacterial state to obtain an updated knowledge matrix comprises the following steps : performing dimensionality reduction on the initial knowledge matrix to obtain multiple sub-knowledge matrices; The plurality of sub-knowledge matrices are updated according to the redistributed bacterial states to obtain an updated knowledge matrix. 5. The power system risk scheduling method according to claim 1, characterized in that, according to the framework data and the new task load section data, the preset initial knowledge matrix is carried out by bacteria foraging reinforcement learning algorithm Before the step of iteratively updating to obtain the corresponding risk scheduling objective function value and the updated knowledge matrix, the following steps are also included: The source task is received for training, and the optimal knowledge matrix is obtained as the initial knowledge matrix. 6. A power system risk dispatching system, characterized in that it comprises: The task data acquisition module is used to acquire the structure data of the power system and the load section data of the new task; The knowledge matrix update module is used to iteratively update the preset initial knowledge matrix through the bacterial foraging reinforcement learning algorithm according to the architecture data and the new task load section data, to obtain the corresponding risk scheduling objective function value and the updated The knowledge matrix; The initial knowledge matrix is the optimal knowledge matrix in the source task; The risk scheduling optimization module is used to perform online optimization of new tasks according to the updated knowledge matrix corresponding to the minimum value of the risk scheduling objective function, and obtain and output the risk scheduling optimization results. 7. The power system risk scheduling system according to claim 6, wherein the initial knowledge matrix is a Q matrix. 8. The power system risk scheduling system according to claim 6, wherein the knowledge matrix updating module comprises: The first processing unit is used to control the bacteria to perform tropism, migration and replication operations under the guidance of the initial knowledge matrix according to the architecture data and the new task load profile data; The second processing unit is used to calculate the power flow value of the power system in the base state and preset fault according to the tropism operation, migration operation and replication operation of the bacteria; The third processing unit is used to calculate the risk scheduling objective function value according to the power flow value of the power system in the base state and the preset fault; The fourth processing unit is used to redistribute the bacterial status according to the value of the risk scheduling objective function; The fifth processing unit is configured to iteratively update the initial knowledge matrix according to the redistributed bacterial state to obtain an updated knowledge matrix; The sixth processing unit is used to judge whether the iterative update meets the preset condition, and when the iterative update does not meet the preset condition, use the updated knowledge matrix as the initial knowledge matrix, and control the first processing unit again according to The architecture data and the new task load section data control the bacteria to perform tropism, migration and replication operations under the guidance of the initial knowledge matrix. 9. The power system risk scheduling system according to claim 8, wherein the fifth processing unit comprises: A dimension reduction unit, configured to reduce the dimension of the initial knowledge matrix to obtain multiple sub-knowledge matrices; A matrix updating unit, configured to update the plurality of sub-knowledge matrices according to the redistributed bacterial states to obtain updated knowledge matrices. 10. The power system risk scheduling system according to claim 6, further comprising a matrix training module, the matrix training module is used to update the knowledge matrix according to the architecture data and the new task load section data , iteratively updates the preset initial knowledge matrix through the bacterial foraging reinforcement learning algorithm, and obtains the corresponding risk scheduling objective function value and the updated knowledge matrix. Before receiving the source task for training, the optimal knowledge matrix is obtained as the initial knowledge matrix.